Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Designing health care provider payment systems to reduce potentially preventable medical needs and patient harm: a simulation study
(USC Thesis Other)
Designing health care provider payment systems to reduce potentially preventable medical needs and patient harm: a simulation study
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Designing Health Care Provider Payment Systems to Reduce Potentially
Preventable Medical Needs and Patient Harm: A Simulation Study
By John Franklin
A dissertation submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy (Industrial and Systems Engineering)
Faculty of the USC Graduate School
University of Southern California
August 8, 2017
1
Acknowledgments
“ T h is w as r eally eas y an d I di dn ’t n eed a n y h elp ” . . . s aid n o d oct oral s t u dent ev er . I am
forever indebted, first and foremost, to my two dissertation co-chairs, Najmedin Meshkati and
Shinyi Wu. Najm, thank you for your mentorship and friendship for nearly 10 years now. I might
n ev er h av e s t ar t ed t h is P h D w it h out y our en c our agem e n t and u n f ailin g beli ef in me (d on ’t w orry , t h at ’s a t h an kf u l s t at emen t ). A n d Sh i n yi , t h an k y o u als o f or y our me n t orsh ip and f rie n ds h ip f or
what feels like nearly as long. I might never have finished this PhD without your patience,
experience, and steady hand helping me reach the finish line.
I am also indebted to the third member of my dissertation committee, Alexander Capron.
Alex, the fruits of your thoughtful criticism can be found throughout this dissertation. Thank you
for lending me so much of the health care wisdom you have accumulated, I promise to pay it
forward.
I want to thank the three USC professors, Robert Myrtle, Michael Nichol, and Gregory
Stevens, who suffered through several one on one meetings with me and a long afternoon
presentation and discussion to lend me their expertise and to help refine and strengthen my
simulation model.
I am also thankful for the many nurses, doctors, and health care administrators who over
the years have taught me so much about the health care system and its nuances, most of whom I
met thanks to the USC Keck School of Medicine and its forward-thinking encouragement of
collaboration and innovation with the rest of the USC community. I want to especially thank Drs.
Philip Lumb and Carol Peden, from whom I have learned so much, starting with our work on the
Health Systems Improvement Collaborative and continuing with its current incarnation, the USC
Center for Health System Innovation.
I also want to thank the USC Viterbi School of Engineering and especially the Daniel J.
Epstein Department of Industrial and Systems Engineering for their generous ongoing financial
and administrative support without which I could never have completed this study.
And finally, I want to thank the many friends and family who have seen me through all the
lows and highs of this process and sustained me the whole way with their love and support.
Especially one in particular, my best friend and fiancée. Rosy, I love you, thank you for everything.
An d d o n ’t w orry : it ’s ov er n ow , o n t o a n ew c h ap t er .
2
Abstract
It is difficult to pay for prevention in health care. Most payment system alternatives to fee-
for-service (FFS) try to encourage prevention by creating financial incentives that reward
providers for reducing the medical needs and patient harm that lead to demand for care and
health expenditures. But due to uncertainty in medical treatment and health outcomes, providers
(especially smaller ones) might invest in preventive interventions only to fail to see expected
returns. The chance for financial incentives not to cover the cost to providers of prevention
creates financial risk for those providers who invest in it, and this financial risk could be a barrier
that makes prevention hard to sustain. I created a simulation model to test the theoretical
contribution of variation in patient needs to the financial risk of providers under various payment
system alternatives who are attempting to practice prevention (w h at I c all “ ris k m it igat ion t re at m en t ” or RM T f or t h e p u rp oses of t h e mo del). T h e r es u lt s of t h e mo del s u gg es t t h at p ay in g
for RMT directly with its own up-front payment while paying for all other care needs with FFS is
associated with less provider financial risk than alternative payment systems that pay for RMT
indirectly out of payer or provider savings. FFS plus a direct payment for RMT was unambiguously
superior to paying for RMT through ACO shared savings (with or without shared losses). Paying
for RMT through full capitation was associated with more provider financial risk than FFS plus
direct payment for RMT but also less payer costs. Choosing between these two payment systems
to fund RMT interventions will require consideration of provider size, provider cost structure, and
t h e a m oun t of v ar iat ion i n medic al n eed s a n d p at ien t h ar m in t h e p rov ider ’ s p at ien t p op u lat ion as well as concerns about the unintended financial incentives of FFS vs capitation.
3
Table of Contents
Acknowledgments 1
Abstract 2
Table of Contents 3
List of Figures 7
List of Tables 9
Introduction 13
Chapter 1 - The Problem of Payment 17
1.1 - Past and Present Criticisms of Fee-for-Service 17
1.2 - Major Payment Reform Responses to Fee-for-Service 22
1.2.1 - The Health Maintenance Organization (HMO) Movement 22
1.2.2 - The Managed Care Organization (MCO) Movement 24
1.2.3 - The Accountable Care Organization (ACO) Movement 25
1.2.4 - Common Threads of Payment Reform 26
1.3 - Potential Problems with Payment Reform Design Thinking 28
1.3.1 - Reducing High Health Care Costs 29
1.3.2 - F in di n g t h e “ Swe et Sp ot” of P rov ider F i n an c i al Liability for Patient Demand 32
1.3.3 - Paying for Performance 34
Chapter 2 - Medical Need 37
2.1 - What is Medical Need? 37
2.2 - A Brief History of Medical Need in the U.S. 39
2.3 - Medical Need Today 43
2.4 - Morbidity and Morbidity Risk 45
2.5 - Treating Morbidity Risk 49
2.5.1 - Chronic Care Management / Coordination 50
2.5.2 - Perioperative Risk Management 52
2.5.3 - End of Life Care Management / Coordination 53
2.5.4 - Iatrogenic Risk Management 55
Chapter 3 - Simulation Model Design 58
3.1 - Why Simulation? 61
4
3.2 - Simulation Model Design 63
3.2.1 - Module 1: Patient Needs Generation 64
3.2.2 - Module 2: Provider Financial Accounting 73
3.2.3 - Module 3: Provider Risk Averse Behavior 77
3.2.4 - Module 4: Risk Mitigation Treatment 82
3.3 - Simulation Inputs, Outputs, and Experiments 85
3.3.1 - Simulation Objective 1 88
3.3.2 - Simulation Objective 2 90
3.3.3 - Simulation Objective 3 92
3.4 - Expert Panel 92
Chapter 4 - Simulation Model Results 95
4.1 - Simulation Results: Simulation Objective 1 95
4.1.1 - Baseline Fee-for-service vs Capitation: Without Risk Averse Provider Behavior 95
4.1.2 - Baseline Fee-for-service vs Capitation: Adding Risk Averse Provider Behavior 96
4.1.3 - Baseline Fee-for-service with Shared Savings 100
4.1.4 - Baseline Fee-for-service with Shared Savings and Losses 101
4.1.5 - Baseline Capitation with Risk Adjustment 102
4.2 - Simulation Results: Simulation Objective 2 104
4.2.1 - FFS with Shared Savings: Adding Risk Mitigation Treatment 105
4.2.2 - FFS with Shared Savings and Losses: Adding Risk Mitigation Treatment 106
4.2.3 - Capitation: Adding Risk Mitigation Treatment 106
4.2.4 - Capitation with Risk Adjustment: Adding Risk Mitigation Treatment 107
4.3 - Simulation Results: Simulation Objective 3 108
4.3.1 - FFS plus paid RMT (2.5% cost) 109
4.3.2 - FFS plus paid RMT (3% cost) 110
4.3.3 - FFS plus paid RMT (5% cost) 110
4.4 - Summary and Contextualization of Results 111
4.4.1 - Paying for RMT (2.5% Cost) with Shared Savings or with Direct Payment 112
4.4.2 - Paying for RMT (3% Cost) with Shared Savings and Losses or with Direct Payment 115
4.4.3 - Paying for RMT (5% Cost) with Full Capitation or with Direct Payment 118
4.4.4 - Paying for RMT (5% Cost) with Risk Adjusted Capitation or with Direct Payment 122
5
4.5 - Sensitivity Analysis 124
4.6 - Special Scenarios 127
4.7.1 - FFS with Shared Savings and Losses Plus RMT, No Savings or Losses Thresholds 128
4.7.2 - Random Effect RMT 129
4.7.3 - L ower F ixed a n d “ Ov er t im e” Cost s 130
Chapter 5 - Discussion 135
5.1 - Simulation Model Recap 136
5.1.1 - Shared Savings and Financial Risk 137
5.1.2 - Full Capitation vs FFS Plus RMT-Only Capitation 139
5.1.3 - Risk Adjusted Capitation 143
5.2 - Implications 144
5.2.1 - Primary Care 145
5.2.2 - Patient Safety 147
5.3 - Limitations 150
5.3.1 - Simulation Model Limitations 150
5.3.2 - General Limitations / Caveats 152
5.4 - Future Research 153
5.5 - Conclusion 154
References 155
Appendix 167
A.1 - Batch Repetitions Full Tables 167
A.2 - Comparative Financial Performance of FFS and Capitation without Risk Aversion 174
A.3 - Comparative Financial Performance of FFS and Capitation when Adding Risk Aversion 179
A.4 - Simulation Objective 1: PID Values for Baseline Payment Systems without RMT 185
A.4.1 - Baseline Fee-for-service (FFS) PID Values 185
A.4.2 - Baseline Capitation PID Values 185
A.4.3 - Baseline FFS with Shared Savings PID Values 186
A.4.4 - Baseline FFS with Shared Savings and Losses PID Values 186
A.4.5 - Determining the Scaling Factor for PID under Capitation with Risk Adjustment 187
A.4.6 - Baseline Capitation with Risk Adjustment PID Values (Scaled) 188
A.5 - Simulation Objective 2: PID Values for Payment Systems Paying Indirectly for RMT 189
6
A.5.1 - FFS with Shared Savings Plus RMT (2.5% Cost) PID Values 189
A.5.2 - FFS with Shared Savings and Losses Plus RMT (3% Cost) PID Values 189
A.5.3 - Capitation Plus RMT (5% Cost) PID Values 190
A.5.4 - Capitation with Risk Adjustment Plus RMT (5% Cost) PID Values 190
A.6 - Simulation Objective 3: PID Values for Payment Systems Paying Directly for RMT 191
A.6.1 - FFS Plus Paid RMT (2.5% Cost) PID Values 191
A.6.2 - FFS Plus Paid RMT (3% Cost) PID Values 192
A.6.3 - FFS Plus Paid RMT (5% Cost) PID Values 192
A.7 - Sensitivity Analysis 193
A.7.1 - Prior Year Weight for Existing Patient Needs 193
A.7.2 - Patient Turnover Rate 194
A.7.3 - Fraction Fixed Costs 194
A.7.4 - “ Ov er t im e” Varia ble Cost F ra c t io n 195
A.7.5 - “ Ov er t im e” I n it iali za t ion T h re sh ol d 195
A.7.6 - Max Demand Modification 196
A.8 - Special Scenarios: PID Values 196
A.8.1 - FFS with Shared Savings and Losses Plus RMT, No Savings/Losses Thresholds, PID
Values 196
A.8.2 - FFS with Shared Savings Plus Random Effect RMT (2.5% Cost) PID Values 197
A.8.3 - FFS with Shared Savings and Losses Plus Random Effect RMT (3% Cost) PID Values197
A.8.4 - Capitation Plus Random Effect RMT (5% Cost) PID Values 198
A.8.5 - Risk Adjusted Capitation Plus Random Effect RMT (5% Cost) (Scaled) PID Values 198
A.8.6 - FFS Plus Random Effect Paid RMT (5% Cost) PID Values 199
A.8.7 - F F S w it h Sh ar ed S av in gs P lus RM T (2 . 5 % C ost ) Low er F ixed a n d “ Ov er t im e” Cost s P I D
Values 200
A.8.8 - F F S w it h Sh ar ed S av in gs and L oss es P lus RM T (3% C ost ) Low er F ixe d a n d “ Ov er t im e” Costs PID Values 200
A.8.9 - Cap it at ion P lus R MT (5% C ost ) Low er F ixe d a n d “ Ov er t im e” Cost s P I D Value s 201
A.8.10 - Ris k Adj u st ed Ca p it at ion P lus RM T (5 % C ost ) (Sc aled) Lowe r F ixed and “ Ov er t im e” Costs PID Values 201
A.8.11 - F F S P lus P a id RM T (5% C ost ) Low er F ixed an d “ Ov er t im e” Cost s P I D Value s 202
7
List of Figures
Figure 1.1: The relationship between patient populations, their medical needs and patient harm, resulting
medical care, and net profit (created using Insight Maker) 20
Figure 1.2: International comparison of health expenditures (via Squires & Anderson, 2015) 31
Figure 2.1: The instrumental definition of medical need 38
Figure 2.2: Medical need equivalents 38
Figure 2.3: Top 10 causes of death, 1900 and 2010 (via Jones et al., 2012) 40
Figure 2.4: The relationship between population, morbidity risk, and morbidity (created using Insight
Maker) 45
Figure 2.5: Histogram of total U.S. health care expenditures, log transformed. Data from the 2013 Medical
Expenditure Panel Survey (made using R Survey) 47
Figure 3.1: A conceptual framework of morbidity risk, morbidity, and payment (created using Insight
Maker) 59
Figure 3.2: Histogram of total U.S. health care expenditures, log transformed. Data from the 2013 Medical
Expenditure Panel Survey (made using R Survey) 66
Figure 3.3: Simplified diagram of patient needs generation (created using Insight Maker) 70
Figure 3.4: Simplified diagram of simulated financial accounting (created using Insight Maker) 74
Figure 3.5: Simplified diagram of provider risk-averse behavior (created using Insight Maker) 77
Figure 3.6: Simplified diagram of needs generation with RMT (created using Insight Maker) 84
Figure 3.7: Simplified diagram of full simulation model (created using Insight Maker) 86
Figure 4.1: Baseline FFS (left) vs. capitation (right), no risk aversion, average days cash on hand, full 10 years
96
Figure 4.2: Baseline FFS (left) vs capitation (right), average patient risk selection per year, first 3 years 97
Figure 4.3: Baseline FFS (left) vs. capitation (right), average PID per year per capita, first 3 years 98
Figure 4.4: Change in absolute PID from FFS to capitation, first 3 years (left) and full 10 years (right) 99
Figure 4.5: Change in PID when adding shared savings to baseline FFS, first 3 years (left) and full 10 years
(right) 101
Figure 4.6: Change in PID when adding shared savings and losses to baseline FFS, first 3 years (left) and full
10 years (right) 102
Figure 4.7: Change in scaled PID when adding risk adjustment to baseline capitation, first 3 years (left) and
full 10 years (right) 103
Figure 4.8: Change in PID when adding RMT (2.5% cost) to FFS with shared savings, first 3 years (left) and
full 10 years (right) 105
Figure 4.9: Change in PID when adding RMT (3% cost) to FFS with shared savings and losses, first 3 years
(left) and full 10 years (right) 106
Figure 4.10: Change in absolute PID when adding RMT (5% cost) to capitation, first 3 years (left) and full 10
years (right) 107
Figure 4.11: Change in PID (scaled) when adding RMT (5% cost) to risk adjusted capitation, first 3 years
(left) and full 10 years (right) 108
Figure 4.12: Change in PID when adding paid RMT (2.5% cost) to FFS 109
Figure 4.13: Change in PID when adding paid RMT (3% cost) to FFS 110
Figure 4.14: Change in PID when adding paid RMT (5% cost) to FFS 111
F i gu re 4.15 : Δ P I D f ro m bas el i n e F F S w hen i mp l em e nting RMT (2.5% cost) under FFS w/ SS (left) or paid
8
RMT (right), first 3 years 112
F i gu re 4.16: Δ P I D f ro m F F S + p ai d R M T ( 2. 5 % cost) to FFS w/ SS + RMT, first 3 years 113
F i gu re 4.17: Δ p a ye r c os t s f ro m i dea l F F S w hen i m p l em e n t i n g R MT ( 2 .5% c os t ) un de r F F S w/ SS (left) or
paid RMT (right), first 3 years 115
F i gu re 4.18 : Δ P I D f ro m bas e l i n e F F S w hen i m p l em e n t i n g R MT ( 3 % cost) u n der F F S w/ SS/SL (left) or paid
RMT (right), first 3 years 116
F i gu re 4.19 : Δ P I D f ro m F F S + p ai d R M T ( 3% cos t ) to F F S w / S S / S L + R MT , fi r st 3 years 116
F i gu re 4.2 0 : Δ p aye r c os t s f ro m i dea l F F S w hen i m p l em e n t i n g R MT ( 3 % cost) u n der F F S w / S S / S L ( l ef t ) o r
paid RMT (right), first 3 years 118
F i gu re 4.2 1: Δ P ID f ro m bas el i n e F F S w hen i mp l em en t i n g RM T ( 5 % cost) u n der c ap i t ati on ( l ef t ) o r p ai d R M T
(right), first 3 years 119
F i gu re 4.2 2 : Δ P I D f ro m F F S + p ai d R M T ( 5 % c os t ) to ca p i t ati on + R MT , fi rs t 3 ye ar s 119
F i gu re 4.2 3: Δ p a ye r c os t s f ro m bas eli n e F F S w he n i mp l em ent i n g R MT ( 5 % c os t ) un der cap i t a t i on ( l ef t ) o r
paid RMT (right), first 3 years 121
F i gu re 4.2 4 : Δ P I D f ro m bas eli n e F F S w hen i m p l em e n t i n g R MT ( 5 % c os t ) un der r i s k a dju s t ed c ap i t a t i on
(scaled) (left) or paid RMT (right), first 3 years 122
F i gu re 4.2 5 : Δ P I D f ro m F F S + p ai d R M T ( 5 % c os t ) to r i s k a dju s t e d ca p i t ati on + R M T ( s c aled) , fi rs t 3 ye ar s 122
F i gu re 4.2 6: Δ p aye r c os t s f ro m bas eli n e F F S w he n i mp l em ent i n g R MT ( 5 % c os t ) un der r i s k a dju s t ed
capitation (left) or paid RMT (right), first 3 years 124
Figure 4.27: Sensitivity analysis, % change in average PID per year per capita for every 20% change in input
parameters, median FFS provider, full 10 years 126
Figure 4.28: Sensitivity analysis, % change in average PID per year per capita for every 20% change in input
parameters, median capitation provider, full 10 years 127
F i gu re 4.2 9 : Δ P I D , re movi n g thresholds for FFS with shared savings and losses plus RMT (3% cost) 128
F i gu re 4.30: Δ P I D f or ca p i t a t i on ( l ef t ) a n d p ai d R MT ( ri ght) from regular to random RMT (5% cost) 129
F i gu re 4.31: Δ P I D , re g ul ar t o f i xe d/ ove rti me cos t r ed uc t i on, F F S w / SS + R MT ( 2 .5% c ost) (left) and FFS w/
SS/SL + RMT (3% cost) (right), first 3 years 130
F i gu re 4.32 : Δ P ID , re gu l ar t o f i xe d/ ove rti me cos t r ed uc t i o n, capitation + RMT (5% cost) (left) and risk
adjusted capitation + RMT (5% cost) (scaled) (right), first 3 years 131
Figure 4.33: Δ P I D , re g ul ar t o f i xe d/ ove rti me cos t r ed uc t i on, F F S + p ai d R MT ( 5 % co s t ) , fi rs t 3 ye ar s 131
F i gu re 4.34: Δ P ID w he n i mp l em ent i n g RMT (5% cost) from baseline FFS to reduced fixed/overtime costs
capitation (left) and risk adjusted capitation (right), first 3 years 132
F i gu re 4.35 : Δ P I D w hen i m p l em ent i n g R MT ( 5 % cost) f r om bas eli n e F F S to r ed uc ed f i xe d/ ove rti me cos t s paid RMT, first 3 years 132
Figure 5.1: A conceptual framework of morbidity risk, morbidity, and payment (created using Insight
Maker) 148
Figure A.1: Baseline FFS (left) vs. capitation (right), no risk aversion, average loss per capita 176
Figure A.2: Baseline FFS (left) vs. capitation (right), no risk aversion, max loss per capita 177
Figure A.3: Baseline FFS (left) vs. capitation (right), no risk aversion, average days cash on hand 178
Figure A.4: Baseline FFS (left) vs. capitation (right), no risk aversion, lowest days cash on hand 179
9
List of Tables
Table 2.1: U.S. population and health data, select years (via CMS, 2015b and CDC, 2016b) 42
Table 2.2: Lognormal model vs. actual total health expenditures (non-zero) for various normal scores 47
Table 3.1: Lognormal models vs. actual data from MEPS 2013 for various normal scores 68
Table 3.2: CVs for population mean expenditures in normal vs scaled lognormal models 69
Table 3.3: Distribution of actual average population medical needs (as equivalent expenditures) in
simulation 73
Table 3.4: Baseline simulation input parameters 85
Table 3.5: 15 batch repetitions, average PID per year per capita, baseline FFS, 5000 patients, first 3 years 87
Table 3.6: 15 batch repetitions, average PID per year per capita, baseline capitation, 5000 patients, first 3
years 87
Table 4.1: Baseline FFS (with risk aversion), average days cash on hand, full 10 years 99
Table 4.2: Baseline capitation (with risk aversion), average days cash on hand, full 10 years 100
Ta bl e 4. 3 : Mag n i t u de Δ P I D f ro m bas eli n e F F S w he n i mp l em ent i n g R MT ( 2. 5 % c os t ) un der s hare d s avi n gs or paid RMT for median 5,000 patient provider, first 3 years 113
Table 4.4: Ratios of reduction in PID from various provider size increases, [FFS w/ SS] / [paid RMT], for
median provider when implementing RMT (2.5% cost), first 3 years 114
Ta bl e 4. 5 : Magn i t ud e Δ P ID f ro m bas eli n e F F S w h en implementing RMT (3% cost) under shared savings and
losses or paid RMT for median 5,000 patient provider, first 3 years 117
Table 4.6: Ratios of reduction in PID from various provider size increases, [FFS w/ SS/SL] / [paid RMT], for
median provider when implementing RMT (3% cost), first 3 years 117
Ta bl e 4. 7: Ma gn i t ud e Δ P I D f ro m bas eli n e F F S w hen i m p l em e n t i n g R M T ( 5 % c os t ) un der ca p i t a t i on or p ai d
RMT for median 5,000 patient provider, first 3 years 120
Table 4.8: Ratios of reduction in PID from various provider size increases, [capitation] / [paid RMT], for
median provider when implementing RMT (5% cost), first 3 years 120
Ta bl e 4. 9 : Magn i t ud e Δ P I D f ro m bas eli n e F F S w hen i m p l em e n t i n g R M T ( 5 % c os t ) un der r i s k a dj usted
capitation (scaled) or paid RMT for median 5,000 patient provider, first 3 years 123
Table 4.10: Ratios of reduction in PID from various provider size increases, [risk adjusted capitation
(scaled)] / [paid RMT], for median provider when implementing RMT (5% cost), first 3 years 123
Table 4.11: Baseline simulation input parameters 125
Ta bl e 4. 1 2: Ma gn i t ud e Δ P I D f ro m bas eli n e F F S w hen implementing RMT (5% cost) under capitation, risk
adjusted capitation (scaled), or paid RMT for median 5,000 patient provider with reduced fixed/overtime
costs, first 3 years 133
Table 4.13: Ratios of reduction in PID from various provider size increases, [risk adjusted capitation
(scaled)] or [capitation] / [paid RMT], for median provider with reduced fixed/overtime costs when
implementing RMT (5% cost), first 3 years 134
Table A.1: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 5000 Patients, First 3 Years 167
Table A.2: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 5000 Patients, Full 10 Years 168
Table A.3: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 50000 Patients, First 3 Years 169
Table A.4: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 50000 Patients, Full 10 Years 170
Table A.5: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 5000 Patients, First 3 Years 171
Table A.6: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 5000 Patients, Full 10 Years 172
Table A.7: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 50000 Patients, First 3 Years
10
172
Table A.8: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 50000 Patients, Full 10 Years
173
Table A.9: Baseline FFS (no risk aversion), number of years with a loss, first 3 years 174
Table A.10: Baseline FFS (no risk aversion), number of years with a loss, full 10 years 174
Table A.11: Baseline capitation (no risk aversion), number of years with a loss, first 3 years 174
Table A.12: Baseline capitation (no risk aversion), number of years with a loss, full 10 years 175
Table A.13: Baseline FFS (with risk aversion), number of years with loss, first 3 years 179
Table A.14: Baseline FFS (with risk aversion), number of years with loss, full 10 years 180
Table A.15: Baseline capitation (with risk aversion), number of years with loss, first 3 years 180
Table A.16: Baseline capitation (with risk aversion), number of years with loss, full 10 years 180
Table A.17: Baseline FFS (with risk aversion), average loss per capita, first 3 years 180
Table A.18: Baseline FFS (with risk aversion), average loss per capita, full 10 years 181
Table A.19: Baseline capitation (with risk aversion), average loss per capita, first 3 years 181
Table A.20: Baseline capitation (with risk aversion), average loss per capita, full 10 years 181
Table A.21: Baseline FFS (with risk aversion), max loss per capita, first 3 years 182
Table A.22: Baseline FFS (with risk aversion), max loss per capita, full 10 years 182
Table A.23: Baseline capitation (with risk aversion), max loss per capita, first 3 years 182
Table A.24: Baseline capitation (with risk aversion), max loss per capita, full 10 years 182
Table A.25: Baseline FFS (with risk aversion), average days cash on hand, first 3 years 183
Table A.26: Baseline FFS (with risk aversion), average days cash on hand, full 10 years 183
Table A.27: Baseline capitation (with risk aversion), average days cash on hand, first 3 years 183
Table A.28: Baseline capitation (with risk aversion), average days cash on hand, full 10 years 183
Table A.29: Baseline FFS (with risk aversion), lowest days cash on hand, first 3 years 184
Table A.30: Baseline FFS (with risk aversion), lowest days cash on hand, full 10 years 184
Table A.31: Baseline capitation (with risk aversion), lowest days cash on hand, first 3 years 184
Table A.32: Baseline capitation (with risk aversion), lowest days cash on hand, full 10 years 185
Table A.33: Baseline FFS, average PID per year per capita, first 3 years 185
Table A.34: Baseline FFS, average PID per year per capita, full 10 years 185
Table A.35: Baseline capitation (with risk aversion), average PID per year per capita, first 3 years 186
Table A.36: Baseline capitation (with risk aversion), average PID per year per capita, full 10 years 186
Table A.37: Baseline FFS with shared savings, average PID per year per capita, first 3 years 186
Table A.38: Baseline FFS with shared savings, average PID per year per capita, full 10 years 186
Table A.39: Baseline FFS with shared savings and losses, average PID per year per capita, first 3 years 187
Table A.40: Baseline FFS with shared savings and losses, average PID per year per capita, full 10 years 187
Table A.41: Regular capitation, no patient risk selection, average PID per year per capita, first 3 years 187
Table A.42: Regular capitation, no patient risk selection, average PID per year per capita, full 10 years 187
Table A.43: Magnitude increase in PID when turning off risk selection for capitated providers, first 3 years
188
Table A.44: Magnitude increase in PID when turning off risk selection for capitated providers, full 10 years
188
Table A.45: Capitation with risk adjustment, average PID per year per capita (scaled), first 3 years 189
Table A.46: Capitation with risk adjustment, average PID per year per capita (scaled), full 10 years 189
Table A.47: FFS with shared savings plus RMT (2.5% cost), average PID per year per capita, first 3 years 189
Table A.48: FFS with shared savings plus RMT (2.5% cost), average PID per year per capita, full 10 years 189
11
Table A.49: FFS with SS/SL plus RMT (3% cost), average PID per year per capita, first 3 years 190
Table A.50: FFS with SS/SL plus RMT (3% cost), average PID per year per capita, full 10 years 190
Table A.51: Capitation plus RMT (5% cost), average PID per year per capita, first 3 years 190
Table A.52: Capitation plus RMT (5% cost), average PID per year per capita, full 10 years 190
Table A.53: Risk adjusted capitation plus RMT (5% cost), average PID per year per capita (scaled), first 3
years 191
Table A.54: Risk adjusted capitation plus RMT (5% cost), average PID per year per capita (scaled), full 10
years 191
Table A.55: FFS plus paid RMT (2.5% cost), average PID per year per capita, first 3 years 191
Table A.56: FFS plus paid RMT (2.5% cost), average PID per year per capita, full 10 years 191
Table A.57: FFS plus paid RMT (3% cost), average PID per year per capita, first 3 years 192
Table A.58: FFS plus paid RMT (3% cost), average PID per year per capita, full 10 years 192
Table A.59: FFS plus paid RMT (5% cost), average PID per year per capita, first 3 years 192
Table A.60: FFS plus paid RMT (5% cost), average PID per year per capita, full 10 years 192
Table A.61: Baseline simulation input parameters 193
Table A.62: % change in PID per 20% change in prior year weight, median provider, first 3 years 194
Table A.63: % change in PID per 20% change in prior year weight, median provider, full 10 years 194
Table A.64: % change in PID per 20% change in patient turnover rate, median provider, first 3 years 194
Table A.65: % change in PID per 20% change in patient turnover rate, median provider, full 10 years 194
Table A.66: % change in PID per 20% change in fraction fixed costs, median provider, first 3 years 195
Table A.67: % change in PID per 20% change in fraction fixed costs, median provider, full 10 years 195
Ta bl e A.68 : % ch a n ge i n P I D p er 20 % ch a n ge i n “o ver t i me ” cos t f ra c t i o n , me di an p ro vi der , fi rs t 3 ye ar s 195
Ta bl e A.6 9 : % ch a n ge i n P I D p er 20 % ch a n ge i n “o ver t i me ” cos t f ra c t i o n , me di an p ro vi der , ful l 10 ye ar s 195
Ta bl e A. 70: % c han ge i n PI D p er 20 % ch a n ge i n “o ver t i me ” i n i t i ali za t i on t hre s ho l d, m edi an p ro vi der , fi rs t 3
years 196
Ta bl e A. 71: % c ha n ge i n P ID p er 20 % c han ge i n “ove rti m e” i n i t i ali za t i on t hre s hol d, me di an p ro vi der , f ul l 10 years 196
Table A.72: % change in PID per 20% change in max demand modification, median provider, first 3 years 196
Table A.73: % change in PID per 20% change in max demand modification, median provider, full 10 years196
Table A.74: FFS with shared savings and losses, no thresholds, average PID per year per capita, first 3 years
197
Table A.75: FFS with shared savings and losses, no thresholds, average PID per year per capita, full 10 years
197
Table A.76: FFS with shared savings plus random RMT (2.5% cost), average PID per year per capita, first 3
years 197
Table A.77: FFS with shared savings plus random RMT (2.5% cost), average PID per year per capita, full 10
years 197
Table A.78: FFS with SS/SL plus random RMT (3% cost), average PID per year per capita, first 3 years 198
Table A.79: FFS with SS/SL plus random RMT (3% cost), average PID per year per capita, full 10 years 198
Table A.80: Capitation plus random RMT (5% cost), average PID per year per capita, first 3 years 198
Table A.81: Capitation plus random RMT (5% cost), average PID per year per capita, full 10 years 198
Table A.82: Risk adjusted capitation plus random RMT (5% cost) (scaled), average PID per year per capita,
first 3 years 199
Table A.83: Risk adjusted capitation plus random RMT (5% cost) (scaled), average PID per year per capita,
full 10 years 199
12
Table A.84: FFS plus random paid RMT (5% cost), average PID per year per capita, first 3 years 199
Table A.85: FFS plus random paid RMT (5% cost), average PID per year per capita, full 10 years 199
Ta bl e A.86 : FF S w i t h s hare d s avi n gs p l us R MT ( 2 .5% c os t ) , re du c e d f i xe d a n d “over t i me ” cos t s , aver age P I D per year per capita, first 3 years 200
Ta bl e A.87 : FF S w i t h s hare d s avi n gs p l us R MT ( 2 .5% c os t ) , re du c e d f i xe d a n d “over t i me ” cos t s , aver age P I D per year per capita, full 10 years 200
Ta bl e A.88 : FF S w i t h s hare d s avi n gs and l os s es p l us R M T ( 3% cos t ) , re du c ed f i xe d a n d “over t i me ” cos t s ,
average PID per year per capita, first 3 years 200
Ta bl e A.8 9 : FF S w i t h s hare d s avi n gs and l os s es p l us R M T ( 3% cos t ) , re du c ed f i xe d a n d “over t i me ” cos t s ,
average PID per year per capita, full 10 years 201
Ta bl e A. 9 0 : Ca p i t ati on p l us R MT ( 5 % cost) , re du c e d f i xe d a n d “over t i me ” cos t s , a ver age P I D p er ye ar p e r
capita, first 3 years 201
Ta bl e A. 9 1: Ca p i t a t i on p l us R MT ( 5 % cost) , re du c e d f i xe d a n d “over t i me ” cos t s , a ver age P I D p er ye ar p e r
capita, full 10 years 201
Ta bl e A. 9 2 : R i s k a dj usted ca p i t ati on p l us R MT ( 5 % c os t ) ( s c aled) , re du c e d f i xe d a n d “over t i me ” cos t s ,
average PID per year per capita, first 3 years 201
Table A.93: Risk adjusted capitation plus RMT (5% cost) (scaled), reduced f i xe d a n d “over t i me ” cos t s ,
average PID per year per capita, full 10 years 202
Table A.94: FFS plus paid RMT (5% cost), reduced f i xe d a n d “over t i me ” c os t s , aver a ge PI D p er ye ar p er capita, first 3 years 202
Table A.95: FFS plus paid RMT (5% cost), reduced fix ed a n d “over t i me ” c os t s , aver a ge PI D p er ye ar p er capita, full 10 years 202
13
Introduction
In most of the current US health care system, treating illness is more profitable than
promoting health. (Frieden & Mostashari, 2008, p. 950)
Despite serious and widespread efforts to improve the quality of health care, many
patients still suffer preventable harm every day...No hospitals or health systems have
achieved consistent excellence throughout their institutions. (Chassin & Loeb, 2013, p.
459)
The hardest thing about paying for prevention in health care is the uncertainty. Some
m edic al n eed s a n d p at ien t h ar m ar e c om p let ely p re v en t ab le. T h at does n ’t n ec es sa ril y make
preventing them inexpensive or easy, but it at least creates some certainty in the process. Most,
however, are only potentially preventable. Uncertainty in both the effectiveness of medical
treatment as well as the onset of patient morbidity can make it difficult to know how well or even
whether a preventive intervention is working, especially at the individual health care provider
level.
This creates a payment problem. We typically pay health care providers for specific
treatment services, targeted at specific patient needs for care, and for specific expected outcomes.
U su ally t h ose t re at m en t s h elp , and s o m et im es t h ey don ’t ; t h is is a n orm al , ac c ep t ed p ar t of t h e
dynamics of health and health care. But flip to paying for prevention and things get weird.
Preventive services try to reduce patient risk of future care needs, which are much less tangible
than present needs. And then the same dynamic still applies; often the prevention works, and
som et im es it does n ’t . I t h in k t h is dy n amic is h ar de r t o ac c e pt in prevention than it is in regular
t re at m en t . Wh e n w e p ay f or p re v en t ion , w e w a n t i t t o work . An d y et s o m et i m es it does n ’t . An d
be c au se of t h e u n c er t ain n at u re of p at ien t mo rb idi t y it ’s h ar d t o sa y t o wh at degree it ’s t h e
p rov ider ’s f au lt . E xp lo ring w a ys to design provider payment systems that get around this problem
is the primary goal of this study.
Despite this uncertainty there is one thing most health care experts agree on: current fee-
for-service provider payment systems do not pay well for prevention. The quote by Frieden and
Mostashari above could be read as suggesting that health care providers purposefully avoid
prevention so that they might profit from cure, but the reality is not so sinister. The real problem
is that, when treatment is more profitable than prevention, preventive interventions can cause
financial trouble for the providers who implement them, whether they work or not. If prevention
doesn’t w ork, p rov ider s h av e w as t ed re sou rc es t h at t h ey c ould h av e ot h er w is e u se d t o t re at patients. If it does work, the same providers are faced with lower patient demand for treatment
and thus lower profits. While this would certainly improve patient health initially, over the long
run it would put such providers out of business. This payment catch-22 seriously hinders system-
wide preventive efforts in health care.
14
There have been many attempts over the past several decades to reform health care
provider payment systems, and most of those efforts have explicitly cited this problem with
prevention as one driving motivation. But though they have made some important strides, these
payment reforms have not yet succeeded in creating a truly prevention-oriented health care
delivery system. Chapter one details the arc of these payment system reforms with a particular
lo ok at w h y t h ey h av en ’t h ad muc h s u c c es s t h u s f a r in en c our a gin g t h e p re v en t ion of medic al
needs and patient harm. The primary takeaway is that the zealous focus in most payment reforms
on changing provider financial incentives in order to reduce health care costs has often clashed
with the goal of needs and harm prevention. The struggles of these reforms provide some valuable
lessons about designing provider payment systems that encourage prevention.
Chapter two delves into theory on better prevention-oriented payment system designs. It
focuses on patient medical needs and how they have changed over the past century, in some ways
complicating the task of needs and harm prevention but also opening up new potential design
opportunities. One theme this chapter continually returns to is the relationship between
uncertainty in patient medical needs and payment. By focusing provider payments on the
treatment of patient medical risk, which tends to be less variable and uncertain over time than
the actual needs and harm that result from it, we might be able to use payment to encourage
more prevention without causing the unintended consequences of other prevention-oriented
payment reforms that make providers financially liable for the ill health of their patients.
Chapter three introduces a simulation model to test the body of theory presented in
chapters one and two. At the heart of the model is the idea that, if payment systems are currently
a barrier to better prevention in health care, then in redesigning them care must be taken not to
exacerbate other barriers to prevention, including in this case the financial risk providers
undertake in order to provide their services. Chapter four examines the results of the simulation
model. These include comparisons between different payment systems in how they create
f in an c ial r is k f or p rov ider s w h o im p lem e n t “ ris k m i t igat ion t re at m en t ” in ad dit ion t o t ra dit ion al
treatment for morbid needs and harm. And chapter five discusses these simulation results in the
context of the entire study and offers some thoughts on moving forward in redesigning health
care provider payment systems to be the forces for prevention and maximizing patient health that
they ought to be.
Before jumping into the first chapter, I want to briefly address my decision to look at the
prevention of both medical needs and patient harm. In one sense I think these terms are
somewhat redundant. A medical need typically implies some amount of patient morbidity that
requires care. Patient morbidities tend to be harmful. So a preventable medical need is essentially
also a preventable patient harm. And vice versa in the patient safety sense of harms. Safety-related
h ar m s t o p at ien t s, if t h ey don ’t r es u lt in p at ien t de at h , t yp ic ally c re at e a dd i t ion al p at i ent needs
for care. So in my mind preventable medical needs and patient harm are often synonymous.
There are good reasons to distinguish between them though. There are many preventable
m edic al n eed s t h at ar en ’t s t ric t ly r elat ed t o p at ien t s af et y. P rimar y care physicians tend to worry
ab out t h eir p at ien t s’ p re v en t ab le ill n es se s, c o m p lic at ion s, a n d a c c ident s, i n c lud in g p re v en t ab le
15
es c alat ion s in di ab et es or as t h m a (Freu n d e t al . , 2 0 1 3) and t h eir elder ly p at ien t s’ mit igable r is ks of falls (Michael et al., 2010). And these preventable medical needs are quite a burden. Researchers
have estimated that providing all of the recommended clinical preventive services would require
nearly eight hours every working day for the average primary care physician, before counting
other clinical responsibilities (Yarnall, Pollak, Østbye, Krause, & Michener, 2003). Yet if primary
care physicians in the U.S. were able to provide these services to most of their patients who could
benefit from them, this alone could prevent 50,000-100,000 deaths among those less than 80 years
old (Farley, Dalal, Mostashari, & Frieden, 2010) as well as an even greater amount of non-lethal
disease and disability (Ogden, Richards, & Shenson, 2012) every year.
Patient harm, on the other hand, is typically associated with patient safety rather than
preventable medical needs. And whether related to wrong site surgeries, hospital-acquired
infections, errors during care transitions, or improperly administered medications, there are many
safety-related harms in health care that ought to be reduced (Chassin & Loeb, 2013). Recent
estimates suggest that safety-related patient harms may result in over 250,000 deaths (Makary &
Daniel, 2016) and even more non-lethal morbidities (James, 2013) every year in the U.S.
Nevertheless, from a whole-systems perspective medical needs and patient harm are
closely linked, and this is important when talking about paying for prevention. There are related
factors that can often lead to both needs and harm, involving interactions between actors,
organizations, and systems across the full spectrum of health care. For example, a patient who
acquires nosocomial pneumonia in the hospital while on a ventilator has certainly experienced a
preventable safety-related harm (Ducel, Fabry, and Nicolle, 2002). But the exacerbation of the
p at ien t ’s c h ron ic obs t ru c t iv e p u lm on ar y dis eas e t h at p u t h im in t h e h osp it al in t h e f irs t p lac e is a
medical need that was itself likely preventable through better clinical disease management. And
even the presence of the disease itself likely could have been prevented had he been successfully
dissuaded from habitual smoking (Rabe et al., 2007). In designing provider payment systems that
are more prevention- orie n t ed, it ’s imp ort an t t o t h i n k ab out h ow t h es e n eed s and harm can be
prevented as early in the cycle as possible.
This brings up an important difference between needs and harm in health care and harms
in other industries. In the airline industry, for example, the demand for aviation services comes
from growing populations who increasingly have the means and desire to travel. Harms to airline
passengers and crew come from errors during the delivery of aviation services, errors that the
industry has made incredible strides to systematically reduce and eliminate (U.S. Department of
Transportation, 2014) and that in a good safety culture should never happen. There are analogous
errors in health care that should also never happen, like wrong site / wrong patient surgeries and
stage 3 or 4 pressure ulcers (AHRQ, 2016a). But most needs and harm in health care are not
completely preventable, only potentially. And even fully preventable medical errors often occur
during treatment for potentially preventable medical needs. Unlike in aviation, the demand for
health care services comes largely from needs and harm. With the exception of completely
elective cosmetic services, patient needs and harm, whether in the form of influenza, chronic
kidney disease, depression, hospital-acquired infections, automobile accidents, or gun violence,
are the reason that health care exists at all.
16
This means that truly holistic efforts to reduce preventable medical needs and patient
harm are tantamount to reducing the preventable need and demand for health care itself (Fries et
al., 1993). Health care is unique among all industries in this regard. It would be nonsensical to try
to reduce the demand for air travel as a way of preventing harm and achieving high-reliability in
aviation. Yet in health care such a tactic is not only sensible but ethically incumbent upon an
industry charged with maximizing the health of patient populations. Ironically, the health care
system should attempt, as early and often as it reasonably can, to reduce the preventable medical
needs and patient harm that lead to the need and demand for its services in the first place.
But predominant fee-for-service (FFS) payment systems punish providers who reduce
patient demand, which means they punish providers who reduce medical needs and patient
h ar m . T o c re at e “ h ealt h c ar e a s if h ealt h mat t er ed” (F rie den & M ost as h ar i, 2 0 0 8 ) t h es e p ay m e n t systems must be changed. The next chapter looks at why this has been so difficult.
17
Chapter 1 - The Problem of Payment
In a health care environment characterized by high and rising costs and lower-than-
desired quality, reforming the way we pay health care providers has taken on an urgent tone. Fee-
for-service, the most common past and present method for paying health care providers, has
known flaws. It carries with it a financial incentive to overdiagnose and overtreat as well as a
disincentive to deliver many preventive services. The resulting delivery system often produces
avoidable patient needs and harm such as preventable illnesses and complications, unnecessary
treatments, and iatrogenic (health care-associated) harms, along with all of their attendant
financial costs.
Over the past several decades, most payment system reforms have responded to these
problems with fee-for-service by substituting other financial incentives to motivate more
desirable provider behavior. Reformers believe that through some combination of these
alternative incentives, total health care costs can be significantly reduced and overall quality
significantly improved. Yet success has proven elusive; these new financial incentives have the
potential to cause unintended consequences, and so far they have done little to stem rising costs.
This chapter will detail the historical criticisms of fee-for-service and the most popular
payment reform responses to it. When it comes to encouraging prevention, I believe that these
payment reforms have a few common problems. Most notably, they have focused on using
financial incentives to spur provider behavior toward goals like cost containment in ways that
may distract from and even undermine prevention.
1.1 - Past and Present Criticisms of Fee-for-Service
Criticisms of the use of fee-for-service (FFS), which uses piece rate payments for health
care provider services, have been around for a long time. Over one hundred years ago George
Bernard Shaw (1911) wrote:
That any sane nation, having observed that you could provide for the supply of bread by
giving bakers a pecuniary interest in baking for you, should go on to give a surgeon a
pecuniary interest in cutting off your leg, is enough to make one despair of political
humanity. (p. 5)
Modern criticisms are no less harsh. Ellwood et al. (1971) argued in the early 1970s for the creation
of health maintenance organizations (HMOs) as an alternative to FFS-based ones, noting:
The way that health care is financed today works against the consumer's interest. Since
payment is based upon the number of physician contacts and hospital days used, the
18
greater the number of contacts and days, the greater the reward to the provider. (p. 292)
And more recently, Fisher et al. (2009) have argued for the creation of accountable care
organizations (ACOs) as an alternative to the still mostly FFS-based ones today, writing:
[FFS is] a payment system that rewards volume, growth, and intensity, regardless of value
(and that penalizes providers who adopt cost-saving innovations) [...] Under current fee-
for-service (FFS) payment, the local capacity of the delivery system is an important
determinant of cost and quality, and there are few incentives to provide low-cost, high-
quality care or to make local capacity decisions that support efficient care. (p. w220 &
w221)
Despite the longstanding and consistent rhetoric against it, FFS has nevertheless continued to be
the dominant form of provider reimbursement in the U.S. and elsewhere.
The past and present criticisms of FFS generally fall into two overlapping but distinct
categories:
1. Overtreatment: By paying health care providers for each service they provide, FFS
incentivizes both more and more-intense care than is necessary and encourages the
avoidance of preventive measures that may decrease future care needs.
2. Skewed treatment: By paying more for some services and less or none for others, FFS
offers an incentive to both over-deliver more lucrative services and under-deliver less
lucrative ones, often regardless of efficacy.
The first criticism, that FFS payments encourage overtreatment or unnecessary care, is a
serious concern. Even if the technical quality of care delivered is good, unnecessary care has the
potential to cause avoidable harms to patients. Examples of these harms include false-positive
diagnoses, drug-resistant infections due to societal abuse of antibiotics, and unwanted intensive
end-of-life care. Unnecessary care is also expensive and wasteful; Berwick & Hackbarth (2012)
estimate that it accounted for between $158 to 226 billion, or 6 to 8% of total U.S. health care
expenditures in 2011.
The health care required to treat medical needs and patient harm that result from unsafe,
poor quality, or insufficiently preventive prior care could also be considered a form of unnecessary
care (necessary in the moment, but potentially preventable with sufficient prior action). Examples
of this include care for hospital-acquired infections, hospital readmissions caused by poor care
transitions, and preventable escalations of chronic disease. Berwick and Hackbarth (2012) refer to
these as failures of care delivery and coordination, and they estimate that these accounted for
between $127 to 199 billion, or 5 to 7% of total health care expenditures in 2011. The link between
FFS payments and this type of overtreatment is less direct. The concern is not that providers will
purposefully deliver poor quality care to generate more needs and harm and thus higher payment.
Rather, the nature of FFS means that delivering high quality preventive care can result in lower
19
future reimbursements, which makes investing in that kind of care financially penalizing for
providers (Frieden & Mostashari, 2008). I n t h is w ay w e c an s t ill say t h at F F S “ en c our ages” t h is kind of overtreatment.
The second criticism, that FFS skews treatment decisions often without regard to efficacy
or merit, is also serious. Private insurers try to negotiate FFS rates (and Medicare tries to set
them) so that health care providers earn fair margins for each service they provide. In addition,
Medicare and private insurers try to reimburse well for services that are highly effective and less
well or not at all for services of poor or limited effectiveness. The ideal result is that these
payments incentivize providers to deliver the combination of evidence-based care that is most
effective and valuable to each of their patients at any given time, though the incentive to over-
treat would still exist (Rich, Lake, Valenzano, & Maxfield, 2013).
Setting the levels of FFS payments to achieve this ideal result, however, is difficult. Even
the most well-intentioned fee schedules can have imbalances in the margins between services, as
Me dic ar e’s s t ruggle to optimize its resource-based relative-value scale (RBRVS), which attempts
to set fair rates for services, makes clear (Ginsburg & Berenson, 2007). High margin services with
generally limited effectiveness may crowd out more effective services with lower margins. And
generally effective services with high margins may be applied ineffectively to patients less likely to
benefit from them. The result is that these imbalances in the margins of FFS payments can result
in both over- and under-delivery of services, both of which can be undesirable.
The classic example of how skewed treatment in FFS can cause problems is the gap in
incomes between primary care and specialty care (Bodenheimer, Berenson, & Rudolf, 2007).
Me dic ar e’s RBRVS ha s h i st oric ally r eim bursed for medical procedures (e.g., diagnostic imaging
and colonoscopies) at a higher margin than for cognitive / management services that take the
same amount of time. For some procedures this is a logical attempt to account for the fact that
they often require larger investments in technology than cognitive work. For others it reflects the
v iew s of s p ec ialis t p h ys ic i an s w h o believ e t h at p roc edu ra l w ork is of t en m ore “ in t en se ” (i. e . , requires more skill, effort, judgment, or stress). But because specialists perform far more
procedures than primary care providers, and because the productivity of procedural work has
been increasing over time, the gap between specialist and primary care provider incomes has
widened considerably, causing a decrease in the number of new doctors choosing careers in
primary care. The resulting shortage of primary care physicians and its concomitant impact on
patient access to primary care threatens to cause an increase in hospitalizations and emergency
department visits, especially among low income communities (Bodenheimer & Pham, 2010; Shi,
2012).
Imbalance in the margins for FFS payments also causes problems within primary care
itself (Berenson & Rich, 2010). In the days when primary care largely revolved around treating
acute illness, a fixed FFS payment for each patient visit was sensible. But today the main burdens
in primary care are chronic diseases and their complications. Managing these ailments requires
services that are complex, highly individualized, and long-term in scope, including such activities
as care planning and follow-up, population health monitoring, and managing transitions from
hospital to home. Too often these activities strain the capabilities of a typical face-to-face office
20
visit. But office visits are typically the bulk of what FFS pays for in primary care; not emails, phone
calls, or other non visit- ba se d a c t iv it ies . An d t h er e’s go od re as o n f or t h is . T h e c ost of c at alog u in g , accounting for time spent, and submitting claims for all of these services would likely not be
worth the potential reimbursement. Not to mention, as Berenson and Rich (2010) note, the moral
hazard of a system which pays physicians for every email they exchange with their patients. But
even though it makes sense under FFS to pay for office visits and little else, the results are a
suboptimal primary care system which performs many office visits and struggles to deliver
effective care coordination, preventive services, and chronic disease management, with patients
losing out.
Figure 1.1: The relationship between patient populations, their medical needs and patient harm, resulting
medical care, and net profit (created using Insight Maker)
I t ’s p oss ib le t o ill u st ra t e t h es e p roblems w it h F F S w it h a si m p le m odel i n t h e f orm of a
causal loop diagram (Figure 1.1). In this model, patient population could refer to the size of the
21
n at ion a l p at ien t p op u lat ion , a p ar t ic u lar re gio n ’s p op u lat ion , or ev en j u st t h e p op u lat ion s er v ed
by an in div idu al p h ys ic ian. T h e s u m of ev er y in div idu al p at ien t ’s medical needs and harms equals
the total medical needs and patient harm of the population, i.e., the reason why medical care
exists. Without them there would be no need for medical care, and their associated costs to
society would be zero (Fries et al., 1993). Of course, there will always be needs for medical care.
But the way we pay for care has an important effect on the level of these needs in a population. By
rewarding overtreatment and skewed treatment, FFS tends to result in a higher level of medical
needs and patient harm in a population.
As the model shows, although medical needs and patient harm tend to lead to medical
care, in theory medical care then reduces the level of these needs, resulting in a balancing
relationship between the two (symbolized by the red minus sign with arrows around it). But the
degree to which these two stocks influence each other can vary. In particular, the degree to which
needs and harm lead to medical care is at least partly a function of demand, and the degree to
which medical care reduces needs and harm is at least partly a function of its effectiveness. Health
care providers have a significant amount of control (though not complete control) over both the
demand for and effectiveness of their services.
Providers can increase demand for their services in many ways, including increasing
diagnostic testing, improving patient access to their services, increasing marketing of their
services, and just prescribing more treatment. As long as the margin for a service is positive, by
in c re as in g de m a n d f or it a p rov ider c an in c re as e p rof it s. T h is is n ’t n ec es sa rily a ba d t h in g; if a
service is highly effective and a particular patient could truly benefit from it, this is exactly the
incentive society wants. But this is also how the incentive in FFS to over-treat works; as long as
the margin on a service is positive, providers will profit even if the service is unnecessary. As for a
service with a low or negative margin, the incentive can work in the opposite direction. Providers
can reduce demand for that service using the same mechanisms: less diagnostic testing, reducing
patient access, and just not prescribing or offering treatment. There need not be any malicious
intent; if a provider is reimbursed poorly for an effective service, it may simply not be financially
feasible to provide it.
T h e mo del a ls o sh ows w h y F F S on it s ow n does n ’t c re at e a s t ron g in c en t iv e t o p rov ide
effective care. For the most part, providers make money after services are provided regardless of
how effective those services were (though there are attempts to change that by tying FFS
payments more strongly to performance). Furthermore, care that is less effective might result in
more future needs and harm along with more care and profit. For a cynical analogy, compare
medical needs and patient harm to a renewable resource such as a fishery. The more needs that
p rov ider s “ h ar v es t ” (i. e. , r emo v e f rom t h e s ys t em vi a e f f ec t iv e care) or alternatively the more that
providers and the public health system degrade the processes that generate new needs (e.g.,
through better prevention of chronic disease, accidents, and other needs and harm), the less
needs there will be for the health care system to use to generate net profits. Luckily there are
other incentives for providing effective care that typically fill in this gap, including professional
norms and competition over quality (though because of information asymmetries and imperfect
ac c es s it is n ’t alw ay s eas y f or p at ien t s t o sh op ar oun d f or h ighly ef f ec t iv e p r ovid er s ). Nevertheless,
22
the fact that FFS at baseline has a weak incentive to provide effective care is a common criticism.
I t ’s n ot all ba d t h ough. U n der F F S , p rov ider s at l east have an incentive to seek out and
treat patients with higher individual needs for care, as long as the care they need is profitable. For
very sick patients with treatable illnesses (and the means to pay) this a good influence. It ensures
that providers will compete to provide them with good access to needed services. In addition, the
system has every incentive under FFS (and all payment systems, for that matter) to try to
len gth e n p at ien t s’ liv es b y re du c in g t h eir r is k of mo rt alit y, bec au se if t h ey d ie their needs and
demand for care go w it h t h em. Wh il e t h is is n ’t alw ay s a goo d t h in g – intensive end-of-life care
t h at a p a t ien t does n ’t t ru ly w an t is a p ote n t ial p roblem – in general I think that an inclination to
prolong life is a good incentive for a health care system.
Of course, this is just a model and as such it is a simplification of reality. In particular, not
all health care providers respond to the profit motive in the same way. Some may be highly
motivated to earn more profit, while others may only care about covering their fixed costs and
avoiding a loss. And it would be difficult to believe that any providers purposefully reduce the
effectiveness of their care so as to generate future income; the problem is more that providing
highly effective and preventive care tends to be financially punishing. The value of models though
is that they allow us to see more clearly the basic structure of the system, including the different
elements involved and how they generally interact. They also help to more easily identify the
potential obstacles involved in changing the system for the better.
1.2 - Major Payment Reform Responses to Fee-for-Service
1.2.1 - The Health Maintenance Organization (HMO) Movement
Although the problems with FFS have been known for a long time, the first serious
attempt at a policy response in the U.S. was the HMO movement that began in the early 1970s. At
the beginning of the decade U.S. national annual health expenditures were $74.6 billion (in 2014
dollars), approximately 6.9% of GDP, and rising fast at a rate of about 10-14% per year (Centers for
Medicare & Medicaid Services [CMS], 2015b). This was a very high amount of expenditures at the
t im e, grea t er t h an i n any oth er c oun t ry , and c er t ainl y h igh e n ough f or F F S’s in c en t iv e for
overtreatment to be a major concern.
But equally concerning was the tendency for FFS to skew treatment. Ellwood et al. (1971)
w er e p ar t ic u larly w orri ed ab out t h e U . S. f ede ra l g o v er n m e n t ’s r ol e i n ad m i n is t er in g an F F S sy st em
through Medicare and Medicaid and what they saw as the excessive amount of federal regulation
required to make it work, including regulation about the price and merit of an enormous quantity
of services. They felt that this regulation and the regular updating it would require was and would
continue to be extremely difficult to manage, and furthermore would mean that these federal
regulatory decisions, whether right or wrong, would come to dominate the investment and
organizational planning decisions made by private health care providers potentially to the
23
detriment of long- t er m i n n ova t ion i n t h e indus t ry . “ I n t h e lo n g ru n , ” t h ey b eliev ed, “t h e on ly
feasible solution is one which will make the industry self-regulating, so that constant intervention
an d t in ke ri n g will n ot be n ec es sa ry ” (p. 293).
Their solution, health maintenance organizations (HMOs), was based on a rare health
care delivery model at the time called the prepaid group practice (Mechanic, 2004). Such
organizations, including Kaiser Permanente (of whom they wrote approvingly) were not paid FFS,
but rather accepted a lump sum annual payment per patient (known as capitation) which they
used to provide all necessary care. In effect these prepaid group practices were both providers and
insurers. Ellwood et al. (1971) considered this model to be a market-based self-regulatory
approach that, if adopted widely, would not only help to slow rising health care costs but would
unleash market competition among health care providers that would lead to better and more
innovative delivery systems and a greater focus on prevention:
The health maintenance strategy is based on the promotion of a highly diversified and
competitive health maintenance industry. Internal self-regulation would be encouraged by
providing economic and professional incentives directed toward maintaining health rather
than merely providing services when illness occurs. [...] [The HMO] agrees to provide
comprehensive health maintenance services to its enrollees in exchange for a fixed annual
fee...The economic incentives of both the provider and the consumer are aligned by means
of their contractual agreement, which assures that the provider will share the financial risk
of ill health with the consumer [emphasis added]. (Ellwood et al., 1971, p. 295).
Of course they re c og n ized t h at s om e r egulat ion w ould s t ill be r equ ire d. N o t in g t h at “ m an y
ar e w ar y of a m ar ke t ap p r oac h t o he alt h c ar e d eliv er y” , t h ey p rop osed t h at :
Health maintenance contracts would be awarded only to responsible organizations whose
structure, resources, and performance demonstrate the capacity to provide quality health
services. A performance reporting system [emphasis added] of proven reliability would be
developed and installed to provide...individual consumers...with accurate information on
the comparative performance of alternate sources of health care. (Ellwood et al., 1971, p.
297).
This kind of performance reporting is important in systems that ask health care providers to share
f in an c ial liab ilit y f or t h eir p at ien t s’ ill h ealt h . I n t h e s ame w ay that a sick patient facing the
prospect of high out-of-pocket costs for a treatment may choose not to seek it, a provider facing
similar costs may choose not to provide it, in both cases to the detriment of the patient.
Performance measurement systems try to hold providers accountable to their contracted
agreement to provide all necessary care to their patients by discouraging them from skimping on
care, and thus they are a crucial element of payment systems that create shared financial liability.
24
1.2.2 - The Managed Care Organization (MCO) Movement
By the beginning of the 1990s, U.S. annual health care expenditures had increased to
$721.4 billion (in 2014 dollars), about 12.1% of GDP (compared to 6.9% in 1970) (CMS, 2015b). By
this point concerns ab out F F S’s t e n dency t o sk ew t re at m en t de c is ion s were almost entirely
overshadowed by concerns about its incentive for overtreatment, especially because of the
associated higher financial costs. Despite the passage of the Health Maintenance Organization
Act of 1973, which provided federal assistance and cleared state legislative roadblocks for the
formation of new prepaid group practices like Kaiser (Mueller, 1974), by the 1990s this pure idea
of an HMO had not yet managed to spread much beyond the original organizations it was
modeled after. However, other organizations had begun to incorporate elements of it as part of
the managed care movement, including the use of capitation payments and other financial
incentives for providers to rein in health care costs (Mechanic, 2004).
While traditional prepaid group practices collect capitation payments from patients and
their employers in the form of premiums, and thus technically serve as an insurer, they are
unusual in that they also either directly employ or form exclusive contracts with groups of
physicians, who are typically paid a salary, to provide services only to their members. In this way
these practices are essentially a hybrid of an insurer and a health care provider. In contrast, most
n ew “ HMOs” du rin g the managed care movement were only insurers, and eventually the term
managed care organization (MCO) became more commonly used to try and reflect this
difference. Instead of employing physicians, these MCOs contracted with independent physician
networks – who were allowed to see outside patients and contract with other MCOs as well – and
negotiated discounted service fees in return for only allowing their patients to see in-network
physicians. Eventually, consumer backlash against the limited choice of providers in MCOs saw
the rise of somewhat more-flexible preferred provider organizations (PPOs), which were
essentially the same thing as MCOs but with the exception that they allowed their patients to see
providers outside the network for a higher out-of-pocket charge (Mechanic, 2004).
These MCOs and PPOs typically collect capitation payments in the form of premiums, just
like traditional HMOs and other insurers , b u t t h ey don ’t alw ay s p ay t h eir p r ovid er n et w orks in
the same way. Early on they most often paid their providers FFS, and they used a variety of
additional incentives and tactics to put pressure on them to controls costs. These included
utilization review (denying a claim for a service they deemed unnecessary after it had already
been provided), precertification (requiring insurer approval for a service before it was provided),
and income withholding (withholding a part of FFS payments to be distributed later as a bonus if
providers kept costs down) (Berwick, 1996). The unpopularity of these methods led to a provider
backlash, after which MCOs increasingly started to just pay their provider networks capitation
instead. Eventually though, even these capitation payments became unpopular because of the
downward pressure they put on provider profits in the face of rapidly increasing practice expenses
(Frakt & Mayes, 2012). Providers increasingly turned to mergers and acquisitions as a way of
mitigating the risk of poor financial performance imposed by capitation, and as a result of their
increased market power they eventually ended up negotiating away from capitated contracts (and
25
back toward FFS ones) altogether.
1.2.3 - The Accountable Care Organization (ACO) Movement
As of 2014 U.S. annual health expenditures have risen to $3.031 trillion, or about 17.5% of
GDP (CMS, 2015b). Even though the annual growth in these expenditures has slowed in recent
years, this is still about 50% more than the next highest spending country (France, 11.6% of GDP)
and about double what the U.K. spends (8.8% of GDP) (Squires & Anderson, 2015). These high
expenditures have dominated the current debate about provider payment reform, and nearly all
re f orm p rop osals ar e n ow ev alu at ed on t h eir ab ilit y t o “b en d t h e c u rv e” of h e alt h c ar e c ost growt h . In addition, policymakers are placing increased attention on improving the value of U.S. health
care; even though Americans pay more for their health care than all other countries, our medical
outcomes are generally no better for it (Squires & Anderson, 2015). Once again, the use of
financial incentives to motivate provider behavior toward these ends is the preferred method of
choice.
Of the many provisions of the Affordable Care Act of 2010, one of the most well-publicized
with regards to payment reform relates to the creation of a new type of provider organization for
Medicare patients, the accountable care organization (ACO). Other private insurers have also
followed suit in trying to their own ACOs (Berenson & Burton, 2012). An ACO is something of a
hybrid between the pure version of an HMO envisioned by Ellwood et al. (1971) and the
alternative versions that dominated during the managed care movement. Like pure HMOs, ACOs
are generally provider networks that collaborate to manage a defined population of patients.
Unlike pure HMOs, ACOs are not also insurers; they generally still submit claims to and receive
FFS payments from insurers who typically also manage claims for other ACOs and non-ACOs. In
addition, patients in an ACO can see any physicians covered by their insurance and are not
limited to just their ACO physicians, similar to PPO managed care.
Unlike most managed care plans, however, insurers contracted with ACO physicians do
not generally use utilization review, precertification, or income withholding to try and control
their prescribing behavior. Rather, these insurers offer ACO providers a cut of any savings they
generate for the insurer (in comparison to previous years), savings that providers can create by
reducing the total health care utilization of their assigned patient population. In some cases
in su re rs w ill also f in a n c ia ll y p en alize p rov ider s w h ose p at ien t s’ ex p en dit u re s ar e h igher t h an expected. This shared savings (and sometimes shared losses) mechanism is essentially a form of
soft capitation. In addition, these shared savings payments are usually predicated on the ACO
maintaining certain quality metrics; if they fail to satisfy these metrics, they may receive no bonus
ev en if t h ey r edu c e t h eir p at ien t s’ ex p en se s.
26
1.2.4 - Common Threads of Payment Reform
The common thread running through all of these payment reforms over past several
decades has been the use of financial incentives (especially variations of provider financial liability
f or p at ien t out c om es ) t o m oti v at e p rov ider s t o red u c e t h eir p a t ien t s’ h ealt h care utilization and
expenditures and more recently also to improve quality of care. Despite the differences in these
reforms over the years, this basic design principle has been the same. What has differed has been
the type and magnitude of the incentives used. Sometimes these incentives have been purely
monetary: bonuses for shared savings and high quality, penalties for increased expenditures and
low quality, utilization review denying reimbursement for services deemed unnecessary, and a
requirement that physicians in capitated contracts pay the cost of lab testing, imaging, or even
specialty care for their patients (Casalino, 1992). Other times these incentives have been non-
monetary but can still be considered financial. For instance, a physician paid capitation who has a
sudden influx of patient demand for his services will not necessarily see his personal income
change as a result, but he will have to work more hours for the same amount of pay. Thus, he still
has a financial incentive to limit the amount of services he provides (Lee, Grumbach, & Jameson,
1990).
The effects of these financial incentives on provider behavior can also be illustrated using
the previous model (Figure 1.1). Though the financial incentives of capitation, managed care, and
ACOs generally influence patient utilization in the opposite direction of FFS, the basic structure
of the system is still the same. The same factors govern the relationship between patient needs
and medical care provided, and health care providers still have basically the same degree of
control over these factors.
As capitation is typically seen as an ideal end state of payment reform efforts to reduce
p at ien t h ealt h c ar e ex p en dit u re s I ’ll g o t h rou gh it s exa m p le in t h e m odel , t h ough m a n y of t h e
HMO/managed care/ACO financial incentives detailed thus far have similar intended and
unintended effects. The provision of medical care tends to generate marginal losses for providers
under capitation, so they have an incentive to reduce patient demand for their services to mitigate
t h ose loss es . T h ey c an do t h is u si n g t h e s ame m ec h an is m s by w h ic h t h ey ’r e ab le t o in c re as e
demand under FFS: reducing diagnostic testing, reducing patient access, as well as just not
prescribing or delivering treatment. In the case of services that are of low value or are
unnecessary, this can be a good thing. And it also certainly helps reduce or at least minimize the
growth of health care costs. The fear is that, in the same way FFS payment systems can encourage
overtreatment, capitated systems may encourage undertreatment, both of which can result in
avoidable patient harms. Moreover, whereas FFS encourages physicians to recruit sicker patients
with high medical needs, capitation can encourage physicians to avoid these patients as they are
likely to demand more services and cause marginal losses. In both cases providers still have an
incentive to prolong patient life; in the case of FFS because patients generate medical needs and
demand for care, while in the case of capitation because they continue to pay capitation
payments.
Unfortunately the connection between capitated payments and care effectiveness is also
27
weak, though perhaps not as weak as with FFS. Both suffer from the same limitation in the ability
of patients to discern what is truly high quality care (did the provider do the best he could?). But
more importantly, the easiest way for providers under capitation to reduce losses is to reduce
demand directly. Increasing care effectiveness might reduce medical need, which in turn might
also reduce demand, but due to the uncertain nature of medicine it might not. And chances are
that these preventive efforts will cost money along the way. Thus, there is some financial risk
associated with providers investing in prevention under capitation (as well under all of the
previous payment reforms mentioned). This is the reason why most payment reforms (including
E ll w oo d e t al . ’s HMO p ro p osal in 197 1) i n c lud e s ep ar at e mec h an is m s f or qu alit y an d p er f orm an c e
measurement independent of the payments themselves, to create a stronger incentive for quality
and prevention. Most modern reforms have folded these measures into the financial incentives as
well in the hope of incentivizing higher service quality and discouraging undertreatment directly
through payment. Although it is theoretically possible to use financial incentives in this way there
ar e s ign if ic an t obs t ac les t o t h is ap p roac h , as I ’ll n o t e in t h e n ext s ec t ion .
T h er e’s ano t h er w ay of lo okin g at t h e r is k of margi n al l oss es f rom h igh p at ie n t demand
as soc iat ed w it h t h es e p ay m en t r ef orm s . F i n a n c ial r is k t ec h n ic ally d oesn ’t c o m e f rom a p ay m e n t method per se; rather, it exists because the actual costs to the provider of patient treatment may
differ from the expected amount when payments were negotiated or set ahead of time (Berwick,
1996). For instance, a physician under a capitation arrangement negotiates (or in the case of
Medicare, has set for her) a payment amount that is meant to cover her expected future expenses
in caring for her patients. She only has some control over the medical needs and patient harm
that drive these expenses though, and these have a certain amount of variation to them over time.
I f p at ien t n eed s h ap p en t o be lowe r t h an ex p ec t ed, t h e p h ys ic ian won ’t h av e t o p rov ide as man y
services and she can pocket the excess payment as extra profit. If they happen to be higher than
expected, however, the physician may incur a loss. The same is actually true in FFS systems, just
in the reverse direction. FFS rates are set or negotiated so that their marginal profit will eventually
result in a net profit for physicians on average, after they have covered their fixed costs. If patient
needs happen to be higher than expected a physician will more than likely end up providing
enough services to cover her fixed costs and then some. If needs are lower than expected,
however, she might not be able to cover her fixed costs which would also result in a loss.
T h e imp ort an t p oin t is t h at a p rov ider ’s f in an c ial r is k u n der eit h er F F S or c ap it at ion -based
payment systems (or some variant in between) is influenced by the difference between its
p at ien t s’ ex p ec t ed an d a c t u al medic al n eed s a n d h ar m , t h ough it s ac t u al n e t p rof it s or losse s ar e
ultimately determined by the amount of services it provides in response. This means that, per the
model in Figure 1.1, the amount of variation in medical needs and patient harm plays an important
role in the effect of payment systems on provider behavior. It may serve to amplify or dampen the
amount that providers influence demand for their services as well as the degree to which they
seek out sicker or healthier patients. Furthermore, the known magnitude of possible variation in
t h eir p at ien t s’ f u t u re n eed s m ay mo t iv at e p rov ider s t o m in i m ize t h e a ss ocia t ed f in an c ial r is k
through consistent utilization patterns – always provide more care, or always provide less care –
even if actual needs end up being within expectation. In other words, financial risk for health care
28
p rov ider s is c re at ed by v a ria t ion in t h eir p at ien t s’ med ical needs and harm, while payment
systems govern how providers may ultimately respond to that financial risk.
1.3 - Potential Problems with Payment Reform Design Thinking
As Dr . Ala n Hil lm a n n ote d, “ T h e c ru c ial qu es t ion is n ot w h et h er f in an c ial i n c en t iv es affect
p h ys ic ians’ d ec is ion s, bu t w h et h er s om e f in an c ial i n c en t iv es distort p h ys ic ians’ j u dgm e n t . ” (Hillman, 1990, p. 891). Numerous studies have shown that financial incentives have at least some
influence on provider treatment behavior and in the expected direction (Hillman, Pauly, &
Kerstein, 1989; Hellinger, 1996; Robinson, 2001; Saver et al., 2004; Shafrin, 2010). This of course
should be intuitive; payment reform advocates have good reason to believe that financial
incentives can work to reduce excessive volumes of care. But financial incentives come with the
risk of working too well and causing unintended consequences. In the same way that we believe
FFS might cause patient harms through overtreatment, financial incentives to reduce care
volumes might cause similar harms through undertreatment. It has never been proven that either
FFS or capitation are uniformly worse for quality of care, though actually proving so in a research
study would be exceedingly difficult anyway (Hellinger, 1996). But there are good reasons to be
wary of the potential negative effects of financial incentives to reduce care volumes regardless
(Hillman, 1990; Berwick, 1996; Armour et al., 2001).
Payment reformers are certainly aware of and have attempted to account for the potential
problems with systems that place some financial liability on providers for the demand of their
patients. Ellwood et al. (1971) made this clear in their recognition that HMOs would require
certain performance measures and consumer protections to make sure that they did not skimp on
care. And Fisher et al. (2009) do the same in calling for improved performance measures for
ACOs, noting that even if financial incentives don ’t c au se s ign if ic an t p roblems w it h qu alit y
patients would still desire such measures to be confident that ACOs are truly giving them better
value. But the unpopularity and ultimate retreat of the managed care movement showed that all it
takes are a few isolated cases of providers denying care for seemingly profit-motivated reasons,
plus a load of bad press, to send even the best-intentioned financial incentives back to the
drawing board (Mechanic, 2004).
Given the plethora of policy efforts involving these financial incentives over the past
se v er al de c ad es , it ’s w ort h ex amin i n g t h e t h in ki n g b eh in d t h e p ay m en t r ef o rms t h at c on t in u e t o
use them. There are generally three main and often parallel lines of thought:
1. Shifting from payment systems based on FFS, which come with financial incentives for
providers to increase care volumes, to alternative systems with financial incentives for
providers to reduce care volumes, will significantly reduce the high cost of health care in
the U.S., if not by reducing it in the absolute then at least by reducing its annual growth.
2. Through experimentation we can find an ideal amount of financial liability to place on
providers for the ill health of their patients that will motivate them to neither over-treat
29
nor under-treat and ideally to invest in prevention too, although this amount may differ
from one provider to another.
3. E v en if w e c an ’t f in d t h e id eal a m ou n t of f in an c ial l iab ilit y t o us e, w e c an u se qu alit y an d
performance measurement tied to financial incentives to make up the difference and at
least ensure that providers do not skimp on care.
There are some potential problems with each of these lines of thought. None of these problems
rule out the use of financial incentives as tools for positive change in U.S. health care but should
rather serve as sobering reminders of their limits. Moreover, they should prompt some questions.
Are provider financial incentives a fundamental solution or are they better used on top of other
more fundamental reforms? And when it comes to reducing potentially preventable medical
needs and patient harm, are there other ways of thinking about provider payment system reform
that might be a better place to start?
1.3.1 - Reducing High Health Care Costs
The potential problems with using payment reforms based on provider financial
incentives to address the high cost of health care in the U.S. are perhaps the most straightforward
to address. Much of the impetus for both HMOs and ACOs has come from repeated observations
that the cost and quality of health care around the U.S. is not uniform (Fisher et al., 2009). For
many years John Wennberg and colleagues at the Dartmouth Atlas Project have shown that there
is tremendous unwarranted geographic variation in Medicare utilization in the U.S.; that is,
variation that is not well explained by patterns of illness (Dartmouth Atlas of Health Care, 2016).
Furthermore, regions with higher volumes of care actually tend to have slightly worse health
outcomes and quality of care compared to regions with lower volumes, and this variation is
significantly associated with differences in supply-side factors such as the number of hospital beds
per capita, the number of specialists per capita, and greater use of diagnostic testing and
procedures (Wennberg, Fisher, Goodman, & Skinner, 2008). In other words, the value of care
(quality/quantity) varies significantly in the U.S. Wennberg (2010) notes, for instance, that if the
value of care for chronically ill Medicare patients were the same nationwide as it is at some of the
most valuable providers – such as the Mayo Clinic, Geisinger Clinic, and Cleveland Clinic – the
U.S. could save 30-40% on the cost of caring for these patients.
There is some reason to be skeptical of these numbers. A similar Institute of Medicine
(IOM) study (now the National Academy of Medicine) of unwarranted geographic variation in
health care volumes in the U.S. found significantly less of it (about 40% less) when adjusting
better for differences in provider wages between regions as well as using more sophisticated risk
adjustment for patient health status (Rosenthal, 2012). There are also additional questions on
whether these studies properly adjust for socioeconomic status. But at least some unwarranted
variation in volumes of care probably does exis t in t h e U . S. , a n d i t ’s c er t ainl y r eas on ab le t o as k
whether variation in volumes above some minimum is wasteful. What is somewhat doubtful is
30
that FFS is a major factor causing this variation, as well as that FFS and this variation have a major
impact on U.S. annual health expenditures.
First, since Medicare FFS is used to reimburse health care providers in all regions in the
U.S., including those with high and low costs, it cannot explain unwarranted variation in volumes
of care on its own. In addition, unwarranted variation in care volumes is not a unique feature of
FFS systems. A similar study of the English NHS system, which does not use FFS at all, also found
significant unwarranted variation in care between regions (Public Health England, 2015), although
t h e magn it u de of t h e v ar iat ion w as ge n er ally les s t h an t h at in U . S. M edic ar e (M ay s, 2 0 11 ). I t ’s likely t h at t h er e a re mo re f u n da m en t al f ac t ors c au si n g t h is u n w ar ra n t ed v ar i at ion , a lt h ough it ’s possible that FFS payments in the presence of thes e f ac t ors c ould e xa c er ba t e it . I n h is ar t ic le, “ T h e
Co st Co n u n dr u m ” , At u l G aw an de (2 0 0 9 ) lan d ed o n t h e “ ec on o m ic c u lt u re ” of h ealt h c ar e
professionals in an area as one of these likely factors. Some communities, he noted, seem to have
spontaneously developed cultures of accountability for the total care of their patients, reflecting a
structural desire to avoid overtreatment and its risk of patient harms. Others, unfortunately, have
n ot. But p er h ap s, ev en t h ough F F S c an ’t b e a f u n da m en t al c au se of u n w ar ra n t ed variation in care
volumes, the fact that it also does nothing to discourage this variation means that it is still due for
a replacement.
However, even if a payment system based on provider financial incentives can be found
that reduces unwarranted variation in care volumes, there are still reasons to believe that such a
system, on its own, will struggle to reduce high U.S. health care costs. Given worries about these
h igh a n d r is in g cost s, m a n y ad v oca t es of p ay m en t re f orm c it e FF S’s i n c en t iv e f or ov er t re atment as
a significant cause (The National Commission on Physician Payment Reform, 2013). But most
health economists agree that high health care prices are a better explanation (Anderson,
Reinhardt, Hussey, & Petrosyan, 2003). International comparisons show that the U.S. has a
generally similar supply of hospital beds and physicians as well as similar utilization rates for
major services compared to other OECD countries, yet it has much higher health care expenses.
The logical conclusion is that higher prices, perhaps driven by higher administrative costs and
higher provider incomes, are to blame, not higher volumes of services, although higher utilization
rates for certain diagnostic technologies and higher obesity rates may also play a smaller role
(Ginsburg, 2008; Squires & Anderson, 2015).
T h is c on c lus ion t h at F F S is n ot f u n da m e n t ally t o b lam e f or t h e U . S . ’s h igh h ealt h c ar e
costs should be intuitive; many other countries also have reimbursement systems that use FFS
and yet have much lower costs. For instance, in 2013 Canada and the U.S. spent $4,569 and $9,086
per capita, or 10.7% and 17.1% of GDP, respectively, on health care. And like the U.S., Canada
mostly reimburses its providers using FFS (Mossialos, Wenzl, Osborn, & Sarnak, 2015). Pozen and
Cutler (2010) studied this difference in costs between the U.S and Canada, and found that
differences in care volumes and intensity could only explain 14% of higher U.S. expenditures; the
rest was explained by higher administrative costs and provider incomes. I t ’s p oss ib le t h at F F S c an encourage higher administrative costs and provider incomes too, in addition to higher care
v ol u m es , b u t t h at s t ill doesn ’t ex p lain w h y t h es e c ost s ar e s o m u c h h igher in t h e U . S. t h a n in Canada. In fact, the majority of the national health systems profiled by Mossialos et al. (2015) use
31
FFS to pay for at least some services, and all of them spend much less on health care (see Figure
1.2).
Figure 1.2: International comparison of health expenditures (via Squires & Anderson, 2015)
What these less costly countries have in common is not that they have moved away from
FFS, but rather that they have enacted nationwide policies to try to cap total health care
expenditures across the board instead of relying on financial incentives to influence individual
p rov ider b eh av ior. As M ar m or and Obe rland er (2 0 12 ) n ote , “ I n m ost in du st ria lized de m ocr ac ies , h ealt h c ar e s p en ding is c on t rol led ‘u p st re am’ t h rou gh b u dget in g , f ee s c h edu les, an d s ys t emwid e
lim it s o n medic al c ap ac it y.” (p . 12 16 ) . N on e of these have anything to do with the way that
payment systems influence individual provider care volumes; rather, they represent a collective
agreeme n t o n t h e p ar t of t h es e c oun t rie s’ elec t ed p ol ic ymak er s ab out h ow m u c h t h ey w an t t o
spend on health care nationally. Some kind of payment reform is still vitally important to reduce
the growth and absolute level of health care costs in the U.S., but it likely needs to be a system-
wide payment reform that puts a cap on rising health care expenditures and does not simply try
to reduce health care costs through isolated programs to change individual provider behavior
(Ginsburg, 2013).
32
I n s ys t ems t h in kin g t h er e is a n ame f or t h is k in d of p roblemat ic de si gn t h i n king . I t ’s c alled
seeking the wrong goal (Meadows, 2008). Payment reforms in the U.S. have typically tried to use
financial incentives to encourage providers to deliver less care, all in the name of lowering
national health care costs. But without simultaneous reform that puts a system-wide cap on the
total costs of care (volume * prices), these financial incentives will be fruitless. On their own,
provider financial incentives to reduce volumes of care cannot control costs. In fact, in isolation
these financial incentives can encourage providers to merge and acquire other providers in order
to spread financial risk, resulting in increased market power and the ability to negotiate higher
prices, which is exactly what happened during the managed care movement (Dranove, Simon, &
White, 2002).
Reducing total costs is the right goal for nationwide policy reforms aimed at reducing or
capping the total use of national economic resources for health care. It is likely the wrong goal for
isolated payment reforms aimed at changing the behavior of individual health care providers. A
better goal for such payment reforms is to try to increase the quality of care delivery and to
reduce potentially preventable medical needs and patient harm. Many of the provider payment
reforms tried in the last four decades have some merit in this regard and are worth exploring and
refining further. But by tying these reforms to a goal – reducing health care costs – that by design
they will struggle to achieve, we risk throwing the baby out with the bathwater when they fail to
meet it.
1.3.2 - Finding the “Sweet Spot” of Provider Financial Liability for Patient Demand
The second common vein of design thinking for payment reform is about finding financial
incentives that, rather than encouraging providers to reduce care volumes in the name of cost
control, instead just encourage them to provide the Goldilocks amount of care: not too much, not
t oo lit t le, b u t j u st r ight . F ra kt and M ay es (2 0 12 ) c al l t h is t h e s ear c h f or t h e “ s w eet s p ot” in t h e
amount of financial liability placed on providers for the ill health of their patients. Moreover, they
note that this sweet spot should ideally encourage just the right size of provider too in addition to
the right amount of care. Not so small that they struggle with the financial liability, but not so
large that they gain too much negotiating leverage over insurers to raise prices. The Affordable
Care Act contains a myriad of provisions for payment experiments that are in effect an attempt to
find this sweet spot. The thinking goes: FFS offers incentives for overtreatment, while capitation
offers incentives for undertreatment. Is there something in between FFS and capitation that
encourages just the right amount of treatment without encouraging providers to merge and grow
too large?
The model in Figure 1.1 illustrates why we believe that FFS incentivizes overtreatment.
Providers have some control over the demand for their services, and when they know that a
service will result in a marginal profit, they have an incentive to increase demand for that service.
Similarly, the same figure shows why we believe financial incentives to reduce care volumes (like
capitation) encourage undertreatment. When providers know that a service will result in a
33
marginal loss, they have an incentive to reduce demand for that service (using the same
mechanisms by which they can also increase demand). Assuming that we are capable of
discerning whether a provider has either increased or reduced demand too much for a service
(resulting in overtreatment or undertreatment, respectively), when we use financial incentives to
m it igat e b oth of t h es e p oss ib ilit ies w h at w e’r e es se n t ially t ry in g t o d o is mi n im ize t h e ef f ec t t h a t an ex p ec t ed p rof it or loss h as on a p rov ider ’s in c li n at ion t o m a n ip u lat e d em an d. P ar ad oxic ally , the search for a sweet spot of provider financial liability for their care decisions is essentially a
se ar c h f or p ay m en t s ys t e m s t h at don ’t ac t u ally incent iv ize p rov ider b eh av ior a t all, at leas t w it h respect to the volume of care they provide. That way, the theory goes, any care they provide will
be based solely on the merit of that care for a patient, and will not be the result of undue
influence from methods of payment (Lee et al., 1990).
The trouble with this idea is that all services usually result in either a marginal profit or
loss, no matter the payment system. The provision of an additional service will almost always
h av e a t leas t s om e imp ac t on a p rov ider ’s f in a n c es ( eit h er di re c t ly by c h an gin g a p rov ider ’s in c om e or i n d ire c t ly b y c h an gi n g t h e v alu e of a p rov ider ’s t im e). T h u s, t h e se ar c h f or n eu t ra l
provider financial incentives is more realistically a search for payment systems that reduce the
influence of profit or loss on provider decision making (Berwick, 1996). But financial incentives
work to influence provider behavior precisely because providers are supposed to be motivated by
exp ec t ed p rof it s or losse s. Wh en it c o m es t o f in a n c ial in c en t iv es , it ’s h ar d t o h av e y our c ak e a n d
eat it too.
More ove r, it ’s n ot c lear w h et h er t h er e a c t u ally e xis t s an amo u n t of p rov ider f in an c ial
liab ilit y f or t h eir p at ien t s’ c ar e v ol u m es t h at is s im u lt an eous ly: 1) l ar ge e n ough t o m oti v at e
providers to reduce unnecessary care; 2) not so large that it creates an incentive to reduce
necessary care; and 3) not so large that it encourages providers to merge and gain market power
in an attempt to mitigate the associated financial risk, which would result in higher prices and the
dimin is h ed p ower of #1 . I t ’s p ossible that such an amount of provider financial liability exists, but
it would likely be different for providers of different sizes, in different locations, and treating
different case mixes of patients. Setting this amount correctly for every unique provider would be
a rather difficult task, and the fact that it might be different for different providers could be seen
as unfair.
Due to the difficulty in setting financial incentives at an appropriate level, there has
increasingly been a push to use non-financial incentives to promote desirable provider behavior
instead (Berenson & Rice, 2015). Luckily medicine is full of useful non-financial incentives,
in c lud in g, “ sc re en i n g a n d s elec t ion , ex p lic it p re sc rip t ion of de si re d p er f orm a n c e, m o n it orin g o f complia n c e , and i n c u lca t ion of n or m s an d c u lt u ra l exp ec t at ion s” (Robin so n , 2 0 0 1 , p . 1 65 ) as w el l
as clinical rules like treatment protocols and practice guidelines (Hillman, 1991). Such non-
financial incentives may do just as well to encourage providers to deliver appropriate volumes of
high quality care without the potential unintended consequences of financial incentives.
As Robin s on ( 2 0 0 1) n ote s, “ E v en t h e mo st s op h is t ic at ed m ec h an is m f or p ay in g p h ys ic ians
merely attenuates and does not eliminate the incentives for overtreatment, undertreatment, and
oth er s ocia ll y u n des ira ble b eh av iors ” (p . 16 4) . T h e f ac t r emain s t h at any w ay of p ay in g hea lt h c ar e
34
providers will result in financial incentives that influence the volume of care they provide, with
the risk that the y c re at e t oo muc h i n f luence . Wit h p rop er p re c au t ion t h is is n ’t alw ay s a b ad t h in g; if these incentives can be modified to encourage more evidence-based care and less care of
questionable value, the influence of money on provider behavior can be used for good (Lake,
Rich, Valenzano, & Maxfield, 2013). Moreover, in cases where the clinical decision is clear- c u t , it ’s unlikely that financial incentives will sway physicians away from the clinically prudent course of
action, regardless of the effect on their fin an c es . As Hill m an ( 1990) n ote s, “ t h e ef f ec t of t h es e
f in an c ial i n c en t iv es may be mo st imp ort an t i n c as es w h er e t h e c orre c t de c i si on is n ot obv ious ” (p . 893).
If it is too difficult or impossible to eliminate financial incentives for overtreatment and
undertreatment in a payment system, an interesting question is whether payment systems ought
to be biased toward one or the other. Both can cause patient harms, and it would be hard to say
which one might cause more. However, if indeed the potential for over- or undertreatment is
greatest in cases where the clinical imperative is least certain, and assuming due diligence has
been made to ascertain the best plan of care for the patient, it is likely that both providers and
patients would prefer a system that is biased toward action over inaction. Their fears of harm
resulting from too little care tend to be greater than their fears of harm from too much (Stoddard,
Reed, & Hadley, 2003; Gallagher, St. Peter, Chesney, & Lo, 2001).
1.3.3 - Paying for Performance
The third common line of design thinking for payment reform concerns the use of quality
and performance metrics combined with financial incentives to ensure that providers provide the
right amount of high quality care, regardless of whether or not policymakers have found the sweet
spot amount of financial liability for patient demand. These kinds of financial incentives are most
commonly called pay-for-performance, or P4P, and as noted earlier they are considered to be a
necessary component of payment systems designed to reduce volumes of care in order to ensure
that the providers involved are not undertreating. Often the line between these financial
incentives and other financial incentives designed to reduce care volumes can be blurred. The
previously mentioned shared savings component of ACOs, which rewards providers if they reduce
t h eir p at ien t s’ c ar e ex p en se s, is a d ire c t in c en t iv e t o red u c e c ar e v ol u m es in on e s e n se , b u t because it is predicated on having met certain performance criteria it is also a P4P incentive.
A deep review of the merits and demerits of P4P would be beyond the scope of this study.
Suffice to say that the current empirical evidence about the efficacy of P4P is mixed and suggests
t h at it w on ’t alw ay s m it igat e t h e p ote n t ial f or u n der treatment under alternative payment systems
(Van Herck et al., 2010). There is evidence that for specific, well-defined, and easily measurable
performance measures, P4P can influence those measures in the desired direction. But what is
specific, well-defin ed, a n d e as ily meas u ra ble may n ot alw ay s be w h at is mo st imp ort an t , and it ’s possible that incentivizing such measures can unintentionally result in reduced provider effort
toward less measurable elements of quality care.
35
Donabedian (1966) famously noted that outcomes are the best possible indicators of good
performance but that they are also the most difficult to measure and account for properly. And
sometimes true outcomes of interest may take years to manifest, especially in the case of
preventive care. For this reason, measures of process in health care are much more frequently
used for P4P incentives. These measures of process may be useful for incentivizing certain care
practices, such as diabetic screening. But they may be less useful in cases where what we actually
care about is a related but not always highly correlated outcome of interest, such as properly
managed blood glucose (Frieden & Mostashari, 2008).
Furthermore, as Donabedian (1966) also noted, there are many factors that drive
outcomes that physicians have little to no control over. Trying to account for the existence of
t h es e f ac t ors is p roc es s c a ll ed ris k ad j u st m en t , a n d i t ’s an im p ort an t s t ep t o t ry t o m ak e s u re t h at providers are being rewarded fairly (or not being punished unfairly) by P4P incentives.
Notwithstanding the predictive power of risk adjustment methodologies, which can vary
substantially in their ability to assign causes of variation for patient outcomes, doctors and
hospitals supply diagnoses, one of the main components of these algorithms. The P4P incentive
then becomes a choice of whether to invest in attaining the outcome measure or to game the
system through upcoding and overdiagnosis. Unfortunately, there is some evidence that providers
often choose the latter (Himmelstein & Woolhandler, 2014).
Finally, there is evidence from behavioral economics that suggests that P4P incentives
have the potential to unintentionally reduce the quality of care in some cases. P4P payments,
which by definition are devices of extrinsic motivation, are commonly thought to be additive to
existing intrinsic motivators. But behavioral economics research has shown that these extrinsic
motivators may crowd out and undermine intrinsic motivators, such as purpose or altruism, and
that this effect is most pronounced for work – like health care – which is cognitively complex and
highly intrinsically motivating (Himmelstein, Ariely, & Woolhandler, 2014). Furthermore, P4P
incentives that are penalties, such a fine for patient expenses in an ACO above a target level, may
produce more pronounced behavior than rewards, even if the magnitude of both is the same, due
to loss-aversion generally being a stronger motivator than profit (Kao, 2015).
All three of these lines of design thinking for provider payment reform involve using
financial incentives to motivate provider behavior by primarily influencing the amount of care
they deliver. And all three have the potential for unintended consequences as a result. The model
in Figure 1.1 is simple but it shows a health care system governed by strong relationships and
powerful actors. Financial incentives try to push on one part of the system: the relationship
between care delivered and provider finances. And in response the system pushes back through
entrenched care processes over which providers have significant control. In systems thinking
terminology this phenomenon is called policy resistance (Meadows, 2008), and it makes changing
the behavior of systems rather difficult. The usual way around it involves aligning the goals of the
actors in the system in a direction they can all rally behind. Financial incentives, which by nature
tend to create goals for individual providers rather than for the system as a whole, can be
incredibly polarizing as a result.
36
More ove r, it ’s p oss ib le t h at t h e f ocu s on in f luencin g p rov ider c ar e v ol u m es h as t ak en attention away from other potentially more worthy goals, such as using payment methods to help
increase the amount of innovation and structural quality in care delivery systems (Berwick, 1996).
At the end of the day, the strongest evidence that a change in course is needed is that after several
dec ad es of t ry in g, dow n st re am ef f ort s t o m oti v at e p rov ider b eh av ior s im p ly h av en ’t w orke d t o
produce desired system-wide outcomes (Marmor & Oberlander, 2012). Are there alternative ways
of paying health care providers that might not generate so much resistance and have the potential
to significantly improve provider prevention of medical needs and patient harm? The next
chapter begins to tackle this question.
37
Chapter 2 - Medical Need
The previous chapter critically examined efforts in the U.S. over the past several decades
to reform health care provider payment systems. In most of these reforms, the primary goal has
been to try to reduce high health care costs by moving away from fee-for-service and its
incentives for overtreatment. The results of these reforms, however, have been mixed. Despite its
problems, fee-for-service has been a staple of health care payment systems globally since long
before the debate began about its merits, and it still continues to be the dominant style of
payment today. The complex array of factors behind rising U.S. health care costs are beyond the
scope of this study, but on its own FFS does not seem to be one of them. Policies that shift away
from FFS may still be useful as part of a comprehensive plan to contain costs, but in isolation they
are likely limited tools for this purpose.
Health care in a developed country is ultimately a public responsibility even if it is mostly
privately funded. The communal nature of health insurance, as well as the generally accepted
humanitarian mandate that anyone in serious need of care – regardless of ability to pay – must be
treated makes this clear. And like any public responsibility there will be a continuous debate
about the amount of resources society ought to allocate to it, a debate that is ultimately broader,
more complex, and more contentious than simply comparing the merits of alternative payment
sy st ems. T h er e is n o get t in g arou n d t h is ; “ how sh o u ld w e p ay ?” is n ot a goo d s u bs t it u t e f or t h e
que st ion , “ how much sh o u ld w e p ay ?” .
The need for a broader debate about cost containment in health care is nothing new.
However, one thing in health care has changed substantially over the past century, and that is the
overall burden of medical need among the population. The most common medical afflictions and
complications in the U.S. at the turn of the 20th century are largely unrecognizable today. And
while changing needs and harm do not offer a simple explanation for rising costs they are still the
fundamental force driving all demand for health care services (Fries et al., 1993). By looking
deeper at these needs and how they have changed over time, it might be possible to think about
payment system reform in a different light. Does the nature of modern medical need offer any
guidance for the design of payment systems toward goals other than just cost containment; in
particular better prevention of medical needs and patient harm? The rest of this chapter consists
of a focused review of the literature to explore this question.
2.1 - What is Medical Need?
Before looking at how medical needs in the U.S. have changed over time it would be
p ru dent t o def in e w h at is mean t b y “ m edic al n eed ” , and Harr is on , Y oun g, B u t ow, and S ol o m o n (2013) have written a thorough review doing just that. They emphasize first and foremost that
there is not one correct way of defining medical need. For instance, citing from Liss (1993) they
38
note that one definition is simply that of ill health, a physiological tension or disequilibrium that
m ight b e r ec t if ied t h rou gh a h ealt h in t er v en t ion . B u t t h is de f in it ion does n ’t t ak e into ac c oun t limitations in health care resources, which often dictate that not all needs can be fully met.
F u rt h er m ore, it als o d oes n ’t ac c oun t f or n eed s f or p re v en t ion , w h ic h may n ot be measu ra ble a s a
disturbance of health but are no less important. They go on to detail a number of different
definitions of medical need with these various perspectives in mind.
Figure 2.1: The instrumental definition of medical need
For the purposes of this study, the most useful definition of medical need that they
mention is that of a gap between an actual and a desired state of health (see Figure 2.1). Liss (1993)
calls this the instrumental definition, because medical need is defined as the need for a medical
intervention (to reduce the gap). Crucially, under this definition all of the actors in the health care
system must collectively determine what the goal – the desired health state – actually is. For
instance, physicians may prefer to set this goal as the highest achievable state of health for each of
their patients using existing medical technology. Yet policymakers and whole societies must
contend with limited resources. They may wish to limit this goal at the population level, rationing
health care in a way that most efficiently and equitably maximizes the health of the population
ev en if t h at mean s n ot p ay in g f or s o m e v alu ab le t re at m en t s. O n t h e ot h er h an d, p at ien t s’ de si re d
h ealt h s t at es ar e imp ort a n t t oo , ev e n if t h ey ’r e im p oss ib le t o rea c h w it h ex i st in g m edic al
technology or would require overly-expensive interventions. Without considering these
sometimes unrealistic patient desires, the system would never focus on developing new
interventions to treat previously untreatable needs, or on making existing interventions more
efficient. Thus, a compromise goal must be reached from all of these perspectives.
Figure 2.2: Medical need equivalents
There are many actors that may try to reduce medical need, including public health
professionals, employers, community health organizations, health insurers, and patients
themselves. But given this study ’s f ocu s on p rov ider p ay m en t s ys t ems, t h is c h ap t er w ill f ocu s on
Actual
Health
State
Desired
Health
State
Medical Need
Medical
Need
Medical
Services
Medical
Resources
39
medical need from the perspective of health care providers and the health care delivery system
specifically. Donabedian (1974) provides a useful framework for doing so with his concept of
medical need equivalents (see Figure 2.2). He notes that while needs in health care can be
measured as a deviation from a desired health state, whether caused by the burden of disease,
accidents, or health risks, they can also be defined according to the services that would be needed
to treat those needs, as well as by the resources that would be necessary to provide those services.
For instance, the amount of need in the form of cardiac disease in a population could be
measured directly through a patient survey of cardiac disease burden, but it could also be
approximated by the amount of cardiac care services provided, as well as by the dollars spent on
those cardiac services. Even though these three dimensions are different and may vary in
magnitude from one another, they are nonetheless related and each one may serve as a useful
proxy for medical need.
The factors that tend to drive medical need, whether measured directly or as the amount
of services or resources devoted to them, are important in the discussion of provider payment
systems. After all, these factors are the reason why health care exists at all. They have an outsized
influence on health care workflows and organizational designs; the reason hospitals exist and
operate the way they do is because, ideally, they are the most effective way of resolving hospital-
related medical needs. And payment plays a role in this as well. It essentially sets limits on what
kinds of workflows are financially viable. Health care providers must design their services within
these limits to provide sustainable care to their patients (Berwick, 1996). Thus, the question for
payment reform ought to be: Would modern medical needs, especially those related to
preventable medical needs and patient harm, be better served by alternative health care
workflows? And if so, how could payment systems be redesigned to best facilitate those
workflows? Exploring this question would first benefit from an understanding of how medical
needs have changed since the inception of modern payment systems.
2.2 - A Brief History of Medical Need in the U.S.
...disease is never static. Just as organisms evolve to keep up with changing environmental
c on dit io n s (t h e “ Red Qu een E f f ec t ” ), medic in e s t ru gg les t o keep u p w it h t h e c h an gin g
burden of disease. (Jones, Podolsky, & Greene, 2012, p. 2338)
As part of its bicentennial issue, the New England Journal of Medicine noted how vastly
the landscape of medical needs had shifted during its time in publication (Jones et al., 2012). At
the time of its inaugural issue in 1812, lists of causes of death included oddities like apoplexy and
drinking cold water, among more recognizable causes like infectious diseases, accidents, and
various kinds of newborn and maternal mortality. These medical needs were overwhelmingly
acute; they tended to arise suddenly and were often followed quickly by either recovery or death.
By 1900 this was still largely the case; infectious diseases – pneumonia, tuberculosis, and
gastrointestinal disease – were the top 3 causes of death (see Figure 2.3), and even the 4th cause,
40
heart disease, was also primarily due to an infectious variant (today heart disease is primarily
caused by atherosclerosis, a chronic, non-infectious risk factor). These acute illnesses and
accidents preyed upon both the young and old equally, preventing the majority of individuals
from reaching their elder years.
Figure 2.3: Top 10 causes of death, 1900 and 2010 (via Jones et al., 2012)
There were signs of change in 1900, however; cancer and senility, chronic conditions
associated primarily with older individuals, were also among the top 10 causes of death. This was a
time of transition; the industrial revolution was in full swing, and a combination of rising
incomes, better nutrition, and better hygiene was beginning to combat many traditional acute
medical needs. Eventually, better public health systems and medical science reduced the
mortality of these acute needs even further. Between 1900 and 1960, average life expectancy rose
from 47.3 to 69.7 years (CDC, 2016b) and articles in the New England Journal of Medicine began
regularly mentioning centenarians (Jones et al., 2012). Epidemiologists today refer to this time as
t h e s ec on d s t age of t h e ep idemiol og ic t ra n si t io n , “ t h e a ge o f r ec eding p an de m ic s” , du e t o t h e
massive, rapid, and unprecedented demographic change that occurred following these reductions
in acute mortality (Omran, 1971; Olshansky & Ault, 1986).
It was also during this time that large, structured systems of provider payment, in the
form of third party health insurance, were invented and popularized in the U.S. (Thomasson,
2003; Showstack, Blumberg, Schwartz, & Schroeder, 1979; Morrisey, 2013). Historically physicians
41
and hospitals had typically negotiated only with their patients for payment. Most often they
charged a piece rate (fee-for-service) for each visit or procedure they performed, and they often
adjusted these rat es b as e d on t h eir p at ien t s’ ab ilit y t o p ay (p oo r p at ien t s w er e f re que n t ly c h ar ged
less than wealthier ones). Before about 1920, however, the efficacy of medical services was mixed.
Me dic al s c ien c e w as s t ill relat iv ely p rimit iv e b y t o da y’ s st an da rd s, so av e rage household demand
for health care (and thus health insurance as well) was low. Households instead often purchased
sickness or disability insurance to protect against losses in job income during illness.
After 1920, however, rising efficacy of acute medical treatment led to increased demand
for medical services, which in turn also increased the cost of these services (Thomasson, 2003).
Households began to face scenarios where their illnesses were both more reliably treatable and
more expensive to treat. Health insurance provided a convenient solution, allowing patients to
pay a premium to transfer the risk of future medical need to an insurer. There are many
interesting details about how health insurance evolved in the U.S., from its beginnings in prepaid
group practices (what Kaiser Permanente is today), to the emergence of the non-profit Blue Cross
and Blue Shield health plans, to the role of employers and the tax code in its wider proliferation,
and then finally to the rise of for-profit commercial health insurers that dominate among the non-
elderly today. But for the purposes of this study, the most important takeaway is that health
insurance formed at a time of high and rising demand for the treatment of acute medical needs.
In theory, either fee-for-service (FFS) or capitation-style payment systems can pay
adequately for acute medical care, and in fact capitation was more common early on. The
Depression made demand for health care unpredictable and often low, so prepaid group practices,
which received a payment for health services up front and on a regular, capitated basis, allowed
physicians and hospitals to more easily maintain operations (Thomasson, 2003). But after the
Depression, FFS-style insurance plans grew far more quickly in popularity and eventually FFS
became the dominant payment system (Showstack et al., 1979). There are many explanations for
why this occurred, but the simplest is that FFS fit well for the needs and values of the time.
Because the vast majority of providers already accepted FFS payments from their patients it was
straightforward to transition to accepting them from insurers as well. Furthermore, FFS was (and
still is) generally seen as a fair reimbursement system for acute medical care. Acute medical needs
are often a one-off affair, so a single payment for services rendered (as opposed to a continuous
payment over time) makes intuitive sense. Under FFS it was also easier for patients to have free
choice of any provider that would accept their insurance, and providers were financially rewarded
for seeking out and treating sick patients. FFS was therefore fairly popular with the public as well.
An d f in al ly, m ost p h ys ic ians j u st di dn ’t like t h e p r ep aid group p ra c t ic e mo del a n d t h e c ap it at ed
payments associated with it. Many felt it was a threat to their incomes and their independence
from hospitals (Morrisey, 2013).
By 1960, 68.3% of Americans had some form of private health insurance, up from 9.1% in
1940 and basically none in 1920 (Morrisey, 2013). And after the passage in 1964 of the legislation
establishing Medicare and Medicaid, public health insurance for the elderly and the poor,
respectively, the percentage of Americans with any health insurance rose even higher. Notably,
Medicare and Medicaid followed the vast majority of private insurers and chose to use an FFS-
42
based payment system (Showstack et al., 1979).
By this time in the early 1960s, medical need too had changed substantially. While
patients still suffered from acute illnesses and accidents, the proliferation of health insurance,
combined with the advancement of acute medical care (on top of previous successes in public
health), had reduced the morbidity of these needs considerably. Overall death rates stabilized at
historically low levels. The U.S. had entered the so-called third stage of the epidemiologic
t ra n si t ion , t h e “ age of de g en er at iv e a n d ma n - m ad e dis eas es ” (Om ra n , 197 1) , m ar ke d b y an increasingly older population age structure. The needs patients most often died from now were
chronic needs associated with old age, such as heart disease, cancer, and dementia.
These new mortality trends persisted for a short while. But with the diminished threat
from acute medical needs, health care and public health systems eventually set their sights on
better treatment for increasingly predominant chronic needs. And b y al l ac c oun t s t h ey ’v e h ad enormous success. Between 1960 and 2014, mortality from chronic illnesses decreased so
significantly that average life expectancy rose from 69.7 to 78.8 years, with much of the gain due
to increased longevity after age 65 (see Table 2.1). Epidemiologists have taken to calling this
p er iod t h e f our t h s t age of t h e ep idemiol o gic t ra n si t ion , t h e “ age of de laye d d egen er at iv e d is eas es ” (Olshansky & Ault, 1986). Such successes have fueled much speculation about the theoretical
upper limits on average human life expectancy, with some experts estimating it could eventually
reach 90 years, though they admit it would take some time to get there (Fries, Bruce, &
Chakravarty, 2011).
1960 1990 2014
U.S. Population (millions) 186 254 318
Average Life Expectancy at Birth 69.7 75.4 78.8
Average Life Expectancy, Age 65 14.3 17.2 19.3
Average Life Expectancy, Age 75 N/A 10.9 12.2
GDP (amount in billions, 2014
dollars)
543 5,980 17,348
National Health Expenditures
(amount in billions, 2014 dollars)
27.2 721.4 3,031.3
As % of GDP 5.0 12.1 17.5
Per capita (2014 dollars) 146 2,843 9,523
Table 2.1: U.S. population and health data, select years (via CMS, 2015b and CDC, 2016b)
These successes are not without cost, however. In theory, if medical advances significantly
delay t h e mo rt alit y f rom c h ron ic illn es se s, bu t don ’t als o c om p ar at iv ely d elay t h e a ss ocia t ed
43
morbidity, then people will both live longer and suffer from disease longer. This has been called
t h e “ f ailu re s of s u c c es s” h yp oth es is (G ru en be rg, 19 7 7 ), w h er eb y delay ed m or t alit y ac t u ally e n ds u p causing both more total morbid harm to patients as well as higher lifetime health care costs.
Given that health care spending in the U.S. has increased substantially since 1960, from $146
annually per capita to $9,523 (both in 2014 dollars, see Table 2.1), this hypothesis might offer a
partial explanation. Variations of it still pop up today, often as an argument against investing in
prevention. For instance, using a simulation model, van Baal et al. (2008) estimated that if obesity
were better prevented it would actually result in greater lifetime health care costs due to formerly
obese patients living longer and suffering from more morbidity as they age.
On the other hand, if it were possible to delay the morbidity from chronic medical needs
by a greater amount than the mortality, total lifetime morbidity (and thus also lifetime health
c ar e c ost s) c ould b e r edu c ed. Th is is t h e “ c om p re s si on of m orbi dit y” h yp oth es is of f er ed by Jam es Fries (1980), which says that, if indeed there is a theoretical upper limit to average life expectancy,
then efforts to successfully prevent or delay morbidity from occurring will promote the greatest
amount of health and prevent the most medical needs and patient harm over a lifetime. This
hypothesis is echoed today by researchers calling for greater investment in interventions to delay
aging (e.g., to delay the onset of all age-related disease and disability at once) as opposed to
predominant interventions which primarily focus on treating or delaying the effects of individual
chronic diseases. For instance, Goldman et al. (2013), using their Future Elderly Model
microsimulation, show that even modest efforts to delay aging now could reap enormous
population health benefits in the next 50 years, and that the fiscal implications would be
manageable through minor modifications to existing entitlement programs.
T im e w ill t ell wh ic h h yp oth es is , “ f ailu re s of s u c c es s ” or “ c om p re ss ion of m or bi dit y” , w ill
prevail. Yet the evidence, as well as the moral and fiscal imperative, seems to be shifting toward
the latter. In a recent review, Fries et al. (2011) examined longitudinal studies of lifestyle
interventions as well as general aging trends and concluded that significant compression of
morbidity is possible, though not inevitable; it will only occur through a comprehensive effort for
better morbidity prevention. Researchers from the delayed aging camp agree (Nikolich-Zugich et
al., 2015). Moreover, they note that the alternative, the projected high increases in chronic illness
and disability over the next several decades due to rising elderly populations, is fiscally
unsustainable and morally untenable. The question of how best to organize health care delivery
and public health systems to address modern medical need promises to be one of the defining
problems of the 21st century.
2.3 - Medical Need Today
In summary, medical need today is far more chronic than acute compared to the
beginning of the 20th century. About 86% of all health care spending in the U.S. is for patients
with one or more chronic illnesses, and these illnesses make up 7 of the top 10 causes of death
(CDC, 2016a). Yet this way of describing modern medical need is problematic. First, the line
44
between chronic and acute needs can be blurry. Chronic diseases often cause acute complications,
such as when chronic heart disease results in a heart attack, or when advanced AIDS results in a
greater likelihood of acute infections. And second, this categorization of medical needs furthers
the same disease-centric paradigm of health care delivery that delayed aging researchers argue is
no longer sufficient to treat the dominant patient needs of the next several decades (Nikolich-
Zugich et al., 2015; Goldman, Gaudette, & Cheng, 2016).
I think that a better way to describe modern medical need, offered by Acheson (1978), is
by the phenomena underlying need, which he groups into two categories: morbidity and the risk
of morbidity. Morbidity includes the pain, disability, and risk of death associated with active
illnesses, accidents, complications, and medical errors. Risk of morbidity, on the other hand, is
related to but distinct from morbidity. It represents the risk factors that may cause morbidity
eventually but that are not guaranteed to at any given time. These risk factors can be unique to
specific morbidities, but the most prominent morbidities today – those caused by age-related
chronic illnesses – share many common risk factors, including obesity, hypertension,
hyperlipidemia, diabetes, smoking, poor nutrition, lack of exercise, and the aging process itself
(CDC, 2016a; Fries et al., 2011, Nikolich-Zugich et al., 2015). Rising obesity in particular threatens
to undo much of the epidemiological progress made over the past several decades because of its
role in generating so many modern morbidities (Gaziano, 2010).
Using these two categories, what has changed most about medical need over the past
century is the increasing recognition of the imperative to treat morbidity risk and not just
m orbi dit y alo n e. I n a w ay , t h e t h re at of t h e “ f ailu re s of s u c c es s” h yp oth es is h as man da t ed it ; greater success in treating individual morbidities, absent similar success in treating general
morbidity risk, will make the dire population health projections of the next several decades –
longer lifespans accompanied by greater lifetime amounts of patient needs and harm – come true.
Thus, a 21st century health care system serious about prevention must be designed to do as much
as possible to reduce the risk of morbidity, in addition to its traditional role of treating
morbidities themselves as they arise.
Risk of morbidity need not only be driven by patient health characteristics. As noted in
the introduction to this study, medical services themselves often come with the risk of causing
new morbidities. Recent successes in patient safety research indicate that much of this health
care-related morbidity risk is mitigable. For example, recent efforts have shown considerable
success in reducing central line-associated bloodstream infections (Pronovost, Marsteller, &
Goeschel, 2011). Furthermore, some safety-related morbidities, so-called “ n e v er ev en t s” s u c h as wrong site surgeries, medication errors, and patient suicides while under the supervision of a
health care facility, are inexcusable and can and should be reduced to zero (AHRQ, 2016a). These
kinds of safety-related morbidity risks are important medical needs under this framework as well.
However, though there are isolated examples of innovation, the U.S. health care and
medical science research systems are still mostly centered on the reactive treatment of single
disease morbidities. They rarely focus on the treatment of patient morbidity risk generally
(Goldman et al., 2016) and they have been slow to reduce the risk of morbidities caused by health
care services themselves (Chassin, 2013). This is paralleled by the general lack of fundamental
45
changes in provider payment systems, systems which still mostly use FFS as the primary mode of
reimbursement. As noted earlier in this chapter, these two inertial outcomes are related; payment
systems set limits on health care delivery, so if pa y m en t s ys t ems do n ’t f u n d amen t ally c h a n ge , h ealt h c ar e d eliv er y c an ’t f u n da m en t al ly c h an ge eit h er .
Does the increasing prominence of morbidity risk as a medical need offer any opening for
the redesign of provider payment systems to foster better treatment of it (i.e., better prevention of
medical needs and patient harm)? I will argue that it does because of one key difference between
morbidity risk and morbidity itself: how much each tends to vary over time.
2.4 - Morbidity and Morbidity Risk
Figure 2.4: The relationship between population, morbidity risk, and morbidity (created using Insight Maker)
Recall the figure in the previous chapter showing the relationship between the patient
population, medical needs and patient harm, medical care, and net profit. If we separate medical
needs and patient harm into their underlying phenomena, morbidity risk and morbidity itself,
and split medical care into the treatment of risk and morbidity separately, it looks like Figure 2.4
46
(ignoring for now the effect of payment on demand for these medical services; this will be
dis c u ss ed m ore in t h e n e xt c h ap t er ). I t ’s imp ort an t t o n ote t h e p rogres si on of n eed : i n div idu als in a population first acquire some risk of morbidity, and then this risk gives rise to individual
morbidities over time. Reducing the risk of morbidity has typically been called prevention, while
reducing morbidity itself is usually called treatment, but these labels can be confusing. For
instance, a coronary artery bypass graft (CABG) may be performed both as treatment for angina as
w ell as t o t ry and p re v en t a f u t u re h ear t at t a c k. F or t h e p u rp oses of t h i s m odel it ’s s im p ler t o ref er to any health care intervention targeting a type of medical need as a treatment for that need (even
if it treats both kinds of need at once).
Besides the fact that morbidity risk always precedes the morbidities that arise from it, the
other key difference between these two types of medical need relates to how much they tend to
vary over time. Medical needs and harms have always been highly variable and uncertain
phenomena; after all, the reason health insurance exists is to protect patients from that
uncertainty. Recall that we can measure medical need by using health care expenditures as a
proxy (Donabedian, 1974). According to data from the Medical Expenditure Panel Survey (MEPS),
a nationally-representative survey of the U.S. civilian noninstitutionalized population (AHRQ,
2015), in 2013 the top 5% of the population ranked by total health expenditures accounted for
48.7% of all expenditures, while the top 20% consumed 81.5%. On the other hand, the least costly
50% of the population accounted for only 2.9% of all expenditures, including 15.6% of the
population that had no health expenditures at all. Mean annual expenditures were $4,436 but the
median was only $909, indicating significant positive skew. And furthermore, there was
significant mobility between strata of expenditures. Among the top 5% of individuals ranked by
total expenditures in 2012 only about 34% of them repeated that ranking in 2013, while among the
top 20% of individuals this repeat rate was 54% (Cohen, 2015).
Bec au se of t h e s ign if ic an t p osit iv e s ke w of h ealt h e xp en dit u re s it ’s of t en ea si er t o v is u alize their distribution after a log transformation (with $1 added to each data value to allow
visualization of individuals with no expenditures). The pattern of health expenditures in Figure
2.5 is extremely common (Diehr, Yanez, Ash, Hornbrook, & Lin, 1999). In any given year there is
typically a significant amount of individuals with little or no health expenditures, while the highly
p osit iv e s ke w mean s a s m all g rou p ac c oun t s f or t h e lion ’s s h ar e . F u rt h er m or e, t h e s h ap e of t h e
distribution of expenditures for those individuals with at least some need (expenditures > $0) can
be described as somewhat lognormal, meaning that the expenditures seem to vary
multiplicatively around a geometric mean (rather than additively around an arithmetic mean as
in a typical normal distribution). Put mathematically, non-zero expenditures tend to follow the
f orm x = μ
g
σ
g
z
, w h er e μ
g
an d σ
g
are the geometric mean (median) and geometric standard
deviation, respectively, and z is the geometric equivalent of the normal score (Kirkwood, 1979). In
the 2013 MEPS data, aft er r emo v in g i n div idu als w it h n o exp en dit u re s, μ
g
= $ 1 , 37 8 and σ
g
= 5.50. A
v alu e of σ
g
that high is indicative of quite a lot of variation. Table 2.2 compares the expected
expenditures at various normal scores according to this lognormal model vs. actual expenditures
at those percentiles. The model seems to deviate from reality for extremely high total
expenditures, which may be due to a lack of data points (the 95% confidence interval for the
47
99.9% quantile of expenditures is from $122,864 to $237,357) or to the fact that the expenditures
are not perfectly lognormally distributed.
Figure 2.5: Histogram of total U.S. health care expenditures, log transformed. Data from the 2013 Medical
Expenditure Panel Survey (made using R Survey)
σ
g
-3
(0.1%)
σ
g
-2
(2.3%)
σ
g
-1
(15.9%)
σ
g
0
(50.0%)
σ
g
1
(84.1%)
σ
g
2
(97.7%)
σ
g
3
(99.9%)
Model $8 $46 $251 $1,378 $7,574 $41,643 $228,961
Actual $5 $50 $244 $1,389 $8,075 $36,576 $154,440
Table 2.2: Lognormal model vs. actual total health expenditures (non-zero) for various normal scores
Though these data show the distribution of expenditures due to all medical needs, there
are good reasons to believe that the vast majority of the measured variation is due to treatment of
morbidity and not morbidity risk. The first reason is intuitive. Treatment of morbidity risk implies
that a decision to intervene was made in advance of some expected future morbidity. Thus, risk
mitigation treatment is necessarily a proactive service that health care providers can provide on a
regular, minimally varying schedule of their choosing. The second reason is essentially
tautological; to call something a risk actually implies that it is less variable than what may result
from the risk. Risk, after all, is basically an estimate of an expected outcome, a combination of the
probability of it occurring and its likely severity. These expectations are essentially averages;
48
t h ey ’r e b u ilt f rom analys es of man y obs er v at io n s ( m orbi dit ies ) ove r t im e , s o t h ey ’r e r elat iv ely
insensitive to the fluctuations of individual observations in the short run and thus will vary less
due to purely mathematical reasons. Finally, the third reason is that we can actually measure the
amount of variation in medical needs that is due to identifiable medical risk. This process is most
frequently used for risk adjustment, whereby quality measures and payment rates are adjusted to
reflect differences in the amount of expected medical use and outcomes among disparate patient
populations. The R
2
value of risk adjustment models, which corresponds to the proportion of
variation predictable by the model, is typically less than 20% (often much less) for models of all
individual health care expenditures based on patient health characteristics (Diehr et al., 1999;
Cuc c iar e & O’Don ohu e , 2 0 0 6; New h ous e, M a n n i n g , Keeler, & Sloss, 1989). This means that at least
80% (and likely more) of the variation in total patient medical needs is most likely due to the
random nature of morbidity.
This has potentially significant implications for the design of health care provider
payment systems. Remember that variation in medical needs is the primary reason why health
insurance exists. Insurers protect their patient members from this potentially costly variation by
taking on its associated financial risk themselves, in exchange for a regular premium. But this
insurance offers no such protection for health care providers, whose net incomes are subject to
the whims of varying patient demand for services. Providers gain significant protection against
financial risk at the population level due to the law of large numbers; the average demand of a
population of patients will vary much less than the demand of individual patients, with
decreasing variation in average demand as the size of the population increases. Yet all payment
systems are still associated with some financial risk for providers. The truth is that any provider
who treats morbidity must contend with financial risk simply because of how variable morbidity
can be. The way providers are paid only determines the scenarios under which variation in
morbidity results in profits or loses. Under FFS, they gain when morbidity swings high and lose
w h en it s w in gs l ow; u n de r c ap it at ion it ’s v ic e v e rs a .
Treatment of morbidity risk, on the other hand, may not involve as much provider
financial risk because of how little morbidity risk varies over time in comparison to morbidity
itself. This is a potentially useful design opportunity. Capitation style systems were noted in the
previous chapter as a way of funding greater innovation in health care delivery systems, since the
payment is not dependent on delivering billable services such as in-person visits. But the financial
incentives of capitation, combined with the normal financial risk associated with morbidity
treatment, can cause undesirable provider behavior. If capitation could be directed specifically at
treatment for morbidity risk, it might be possible to fund innovative new health care delivery
workflows that result in substantial reductions in preventable medical needs and patient harm,
without causing the potential unintended consequences of capitation when used more broadly to
pay for regular morbidity treatment as well. Such a system would not be perfect – the risk of
either over- or undertreating morbidity itself would still exist, depending on how morbidity
treatment was paid for – but it might offer a better, more prevention-oriented baseline than
existing systems. A delivery system with a strong focus on reducing morbidity risk and the right
funding to make it sustainable over the long term could theoretically accomplish an enormous
49
amount of needs and harm prevention, especially for those patients whose morbidity risk is most
readily treatable.
2.5 - Treating Morbidity Risk
In order for paying health care providers to reduce the risk of morbidity among their
patient population to be valuable, that risk must actually be clinically treatable. As noted earlier,
the bulk of medical needs today, those related to chronic illnesses, are ultimately caused by just a
few modifiable behavioral risk factors (smoking, poor eating habits, lack of exercise, and alcohol
abuse). Providers may or may not be able to make a dent in these patient behaviors, but
regardless the outcomes of such efforts would likely take a long time to become apparent,
especially among currently healthy patients, and would hinge greatly on good patient compliance.
Such long term and uncertain prospects for intervention may explain why general health behavior
modification programs have traditionally fallen under the purview of public health agencies
instead of health care providers.
Luckily, there are many activities health care providers can do to reduce morbidity risk
that are faster-acting and more certain, and thus are better candidates to be included as part of a
st an da rd t re at m en t p rogram. T h es e a c t iv it ies w ill be r ef er re d t o c ol lec t iv ely a s “ ris k m it igat ion treatm en t ” (RM T ) f or t h e r es t of t h e s t u dy , to help identify them as health care provider-specific
clinical interventions and to differentiate from te rms lik e “ p re v en t ion ” a n d “ c lin ic al p re v en t ion ” that often mean different things depending on the audience. RMT includes any clinical
interventions designed to reduce patient risk of future morbidity within a normal clinical window
(typically less than 3 years), including changes in practice workflows to reduce the likelihood of
health care-associated harm. Similar to clinical interventions that treat morbidity directly, RMT is
likely limited in capability and subject to diminishing returns. Many patient risks of morbidity are
low hanging fruit that are easily treated, while others are more difficult to treat or may be
treatable with less certainty. Still others may be treatable eventually but are currently untreatable
given the existing state of medical science.
RMT may target patient medical risk generally or only under specific circumstances. A
well-known example of general RMT is care coordination / care management for patients with
chronic illnesses, which attempts to reduce patient risk of harmful disease complications. Some
care management programs specifically focus on high-cost / high-risk patients, who often have
multiple chronic illnesses or are part of a high-risk demographic. RMT is also increasingly
happening during surgical episodes. The most prominent recent examples, such as surgical
checklists, have typically focused on reducing the risk of preventable harms during the surgery
itself, but there are ongoing efforts to reduce patient risk as part of the full perioperative process
as well. End-of-life care, which is so often fraught with painful, avoidable harm and great expense,
is a third area being targeted by RMT efforts. And finally, there have been considerable efforts to
reduce iatrogenic risk in general, i.e., the risk of preventable patient harms that can occur in the
course of any medical treatment (such as harms due to health care-associated infections).
50
These RMT areas are not mutually exclusive, nor are they exhaustive, but they broadly
capture the most prominent current clinical interventions that are trying to reduce preventable
medical needs and patient harm. The following sections get into each in further detail.
2.5.1 - Chronic Care Management / Coordination
Care management / coordination for chronically ill patients is one of the better-known
examples of RMT. Interventions specific to better chronic disease management have gained
prominence since the introduction of the chronic care model (Bodenheimer, Wagner, and
Grumbach, 2002) which advocated for the redesign of primary care along six key dimensions:
community resources and policies, health care organization (including provider payment), self-
management support, delivery system design, decision support, and clinical information systems.
In addit ion t o bet t er ma n agem e n t of c h ron ic ill n es s, it ’s imp ort an t t o n ote t h at t h e mo del s t re ss es better management of care itself, including helping patients navigate complex care networks,
transition from one care setting to another, and seek out community / social support.
Interventions simply targeting high-risk / high-cost patients are usually related though may
som et im es of f er ad dit ion al in t er v en t io n s, i n c lud in g “h ot - sp ott in g” or t ry in g t o f ocu s on t h e s ocio -
demographic factors behind persistently high-cost patients (Gawande, 2011).
The potential benefits of RMT among the chronically ill are thought to lie mostly in
reductions in future acute care – emergency room visits and hospitalizations – and their related
spending. Researchers have estimated, for example, that the top 10% most-costly Medicare
patients account for 73% of all acute care spending in the Medicare population. Furthermore,
t h ey ’v e n ote d t h at w it h in t h is h igh -cost group, 10% of their acute care spending (approximately
$9.1 billion in 2010) is potentially preventable through better care management (Joynt, Gawande,
Orav, & Jha, 2013). It is noteworthy that the kind of care management the authors refer to only
includes interventions that would have an almost immediate effect on risk, such as better
management of asthma and type 2 diabetes. Interventions with a somewhat longer time frame,
such as those to manage cardio- and cerebrovascular disease, were not included in these
estimates, so including them could potentially reduce acute care and spending even further.
A recent report from the U.S. Congressional Budget Office (CBO) was skeptical of the
value of such care management / coordination programs, noting in a review of 34 of them (all of
which were conducted by the Centers for Medicare and Medicaid Services (CMS) among the
Medicare population) that, on average, they had no effect on hospital admissions or regular
Medicare expenditures (expenditures before program fees were accounted for) (Nelson, 2012). The
report offers some nuance to this skepticism though; the programs differed tremendously in both
design and the patients they targeted, and there was considerable variation in the resulting
outcomes, including much uncertainty about program effects due to in many cases the small
numbers of patient enrollees.
A f ew p rograms s t oo d ou t ev en amids t t h is u n c er t ainty . M as sa c h u se t t s G en er al Ho sp it al’s (MGH) program, which focused primarily on high-risk Medicare patients, achieved a statistically
51
significant overall return on investment (ROI) for Medicare of about 2.6, mostly through
reductions in patient hospitalizations (McCall, Cromwell, & Urato, 2010; Urato et al., 2013). Some
others achieved at least cost neutrality, reducing regular Medicare expenditures by as much as the
cost of the program, and also typically through reductions in patient hospitalizations (Brown,
Peikes, Peterson, Schore, & Razafindrakoto, 2012). Despite what seems like disappointment
among many researchers regarding these cost-neutral programs, even cost-neutral RMT
interventions are a great success since it means that many patient morbidities leading to
hospitalizations were prevented without adding any additional costs to Medicare. Furthermore,
these programs were early experiments mostly done on small patient populations; if they could be
scaled and improved they would likely eventually be able to gain economies of scale that do result
in net Medicare savings.
Nelson (2012) noted that the care coordination programs that focused on high-risk
patients tended to be the most successful (produced the most savings and/or reductions in acute
c ar e n eed s). Whet h er a p rogram wa s “ at r is k” f or it s c ar e man age m e n t f ees – that is, whether or
n ot a p rogra m w ould h av e t o p ay b ac k t o Med ic ar e it s p rogram f ees if it di d n ’t at leas t r edu c e
expenditures to achieve cost-neutrality – did not seem to make a difference. Brown et al. (2012)
further identified six care coordinator practices among the programs that focused on high-risk
patients that were associated with the best outcomes:
● Supplementing telephone calls to patients with frequent in-person meetings
● Occasionally meeting in person with providers
● Acting as a communications hub for providers
● Delivering evidence-based education to patients
● Providing strong medication management
● Providing timely and comprehensive transitional care after hospitalizations
There is undoubtedly much more work to be done in crafting care management /
coordination programs that do a better job of treating risk. It is worth noting that there are
significant payment impediments in this regard, however. Most of the previously mentioned
Medicare care coordination programs were special one-off demonstrations where providers were
paid a negotiated up-front fee per-patient per-month (essentially capitation but only for care
management services). This fee was not inconsequential, ranging from $80 to $444 per-patient
per-month and $235 on average (Nelson, 2012). It was expected to completely cover the cost of
complex, experimental care management services including telemedicine, coordination among
multiple providers, and even occasional home visits for some patients, services that are difficult or
even impossible to reimburse under traditional fee-for-service (FFS) (Berenson & Rich, 2010).
Though many of these programs have since been folded into accountable care
organization (ACO) demonstrations, there are still few regular, national efforts to pay these sums
for care management, either by Medicare or private payers. The closest analogues are payments
for patient-centered medical homes (PCMH) which typically involve paltry per-patient per-month
payments on top of an FFS base and are not nearly sufficient for advanced care management
52
interventions (Landon, 2015). One promising program by Medicare may yet change this. Its
Comprehensive Primary Care (CPC) and recently-announced CPC+ initiatives strive to not only
provide higher up-front payments to primary care physicians to pay for care management, but to
also provide support for related practice redesign. Furthermore, these initiatives have gotten
private payers involved as well, so that primary care practices might receive similar payments for
c ar e c oo rd in at io n f rom m ost of t h eir p at ien t s’ in su r er s, sign if ic an t ly b ol st er in g t h eir ab ilit y t o
sustain operational changes (Sessums, McHugh, & Rajkumar, 2016). This program will be
discussed further in the discussion chapter at the end of this study.
2.5.2 - Perioperative Risk Management
Though care management for patients with chronic illnesses is probably the most famous
example of risk mitigation treatment in medicine today, there are other areas undergoing similar
innovation. One of these is perioperative risk management, or treatment of the risk of
preventable needs and harm related to surgery. Much of this has focused on preventing medical
er rors du ring s u rger y, esp ec ially t h ose a ss ocia t ed w it h “ n ev er ev en t s” s u c h as w ron g -site surgeries
an d medic at io n er rors (A HRQ, 2 0 16 a ); I ’ll add re ss t h ose s ep ar at ely i n t h e s ec t ion o n ia t roge n ic risk management. This section will focus on efforts to treat surgical risks that are not typically
associated with medical errors.
Patients are exposed to many risks of harm during the course of a surgical episode,
defined by the period starting with their decision for surgery and ending with their return to
reasonable function post-surgery. Some of these risks are related to patient health behaviors. For
instance, an active smoker who quits at least 8 weeks before a surgery is half as likely to develop
postoperative respiratory complications as o n e w h o doesn ’t (G h af er i, Bir km ey er , & Dimic k, 2 0 0 9) . Other risks are dependent on the type of surgery – some surgeries are inherently riskier than
others – yet they can often still be lessened by certain kinds of perioperative care (medical care
provided before, during, and after the surgery itself). For instance, for a long time patients have
been instructed not to eat or drink anything after midnight before a surgery, but evidence is
increasingly showing that a carbohydrate drink about 2 hours before surgery can reduce
postoperative discomfort and may even reduce length of stays (LOS) for major abdominal
surgeries (Awad, Varadhan, Ljungqvist, & Lobo, 2013). And finally, some perioperative risks can
simply be mitigated by better perioperative clinical workflows. A well-functioning preoperative
clinic can identify patient risks that call for modifications to the chosen surgery, special pre- or
post-surgical preparation, or even recommendations against surgery that lead to significant
reductions in postoperative morbidity and mortality (Bader & Hepner, 2009; Wijeysundera &
Sweitzer, 2015).
In recent years there have been a number of efforts to combine these perioperative risk
management efforts into comprehensive systems of care that span the entire pre-, intra-, and
postsurgical continuum. In Europe, which has been innovating in this area the longest, these
models of perioperative care are most commonly called enhanced recovery after surgery (ERAS)
53
(Ljungqvist, 2014). ERAS is fairly popular in Europe; the U.K. for instance underwent a massive
p rogram t o p ro m ote E RA S t h rou ghout t h e c oun t ry ’s NHS h osp it als an d f ou n d t h at it r es u lt ed in meaningful reductions in LOS for colorectal, orthopedic, and urological elective surgeries
(Simpson et al., 2015). Some parts of the U.S. have also begun experimenting with ERAS, though it
is not yet as widespread. In addition, a similar model called the perioperative surgical home
(PSH), meant to be analogous to the patient-centered medical home (PCMH) model in primary
care, has also started to take hold in some U.S. medical institutions. These PSH efforts, though
early, have shown substantial promise to improve surgical care quality and reduce preventable
patient morbidity and its costs (Kash, Zhang, Cline, Menser, & Miller, 2014).
In a recent example, Dr. Peter Pronovost and colleagues at Johns Hopkins Hospital
implemented an ERAS- st yle p rogram (wh at t h ey c alled a n “ i n t egrat ed re c ove ry p at h w ay ” or “ I RP ” )
for colorectal surgery patients (Wick et al., 2015). Previous efforts to reduce surgical site infections
(SSIs) in colorectal surgery using their Comprehensive Unit-Based Safety Program (CUSP), which
was initially created to reduce the incidence of catheter-related bloodstream infections in the
ICU, had successfully reduced the SSI rate from 27% to 18% over 3 years, yet this was still higher
than desired. Their IRP intervention expanded on CUSP by adding standardized and fully-
coordinated pre-, intra-, and postoperative care protocols, including preoperative patient
counseling, intraoperative fluids management, and early postoperative mobilization and
resumption of oral intake. As a result, SSIs dropped even further, from 18.8% to 7.3%, and mean
LOS dropped from 7 to 5 days, all while patient experience measures increased and variable direct
costs of surgery decreased.
There is a lot happening in perioperative risk management right now, and as evidence-
based care protocols continue to be developed over the coming years there are likely to be many
changes in perioperative practice. On the payment front, perioperative risk management is better
positioned than most other types of risk treatment thanks to many years of bundled, up-front
payments to hospitals from Medicare and many private insurers, which have given hospitals some
flexibility to experiment with and implement perioperative risk management on their own.
Further seismic shifts in bundled payments for surgeries from Medicare, including current
bundles for hip and knee replacement surgery (CMS, 2016a) and imminent bundles for acute
myocardial infarction, coronary artery bypass graft (CABG), and surgical hip/femur fracture
episodes (CMS, 2016d), may provide even more stimulus for perioperative risk management due
to financial penalties for readmissions and complications up to 90 days post-discharge. Payment
for perioperative risk management and these Medicare bundles, including some potential
unintended consequences, will be discussed further in the discussion chapter.
2.5.3 - End of Life Care Management / Coordination
In 2014 about 2.6 million people in the U.S. died, and of these 80% were on Medicare at
the time. About 25% of traditional Medicare spending goes to beneficiaries in their last year of
life, making end of life care one of the most expensive parts of the U.S. health care system (KFF,
54
2016). This high expense is understandable given the complex and chronically-ill status of most
Medicare patients in their last year of life, yet it also potentially reflects a significant amount of
patient need and harm that in many cases may be avoidable or mitigable.
Treating the risks of harmful morbidity for end of life patients shares some similarities
with general care management / coordination for chronic illness outlined earlier, but it also has
enough unique challenges that it is worth mentioning separately (IOM, 2015; Gawande, 2014).
One particular challenge is that it is often difficult to recognize when a complex, chronically ill
patient is approaching end of life. Such a patient may undergo expensive and often painful
treatments with small but tantalizing chances of extending life. Sometimes these treatments are
successful but often they are in vain, and they can end up adding misery and pain to the end of
life stage for both the patient and his or her family members. It is for this reason that site of death
has been proposed as a health care quality measure, because despite the fact that the vast
majority of adults, when faced with a terminal illness, would prefer to die peacefully at home,
instead most die at health care institutions during or soon after receiving intensive treatment
(Teno et al., 2013). It is also for this reason that end of life care management / coordination
requires an especially patient-centered emphasis on advance care planning and regular physician-
patient communication to elicit and manage expectations.
The preventable needs and harm that can arise from a lack of or poor quality RMT during
end of life care are numerous. Some of them, just as can happen for non-terminally ill chronic
disease patients, arise from fragmented care delivery. Despite the fact that end of life patients are
typically dying from advanced, complex illnesses, there are still clinical measures that can be
taken to reduce the likelihood of those illnesses escalating into painful complications and hospital
stays and ensure that patient quality of life is as high as possible until the end (IOM, 2015). But
doing so properly typically requires coordination between multiple health care providers and
other social services; without it patients may slip through the cracks.
Other preventable needs and harm can arise from a lack of advanced planning. End of life
patients can often suddenly and unexpectedly lose control of their ability to make decisions about
their care, leaving it to family members and caregivers. Medicare very recently began paying for
advance care planning (ACP) consultations between physicians and Medicare patients so that
more patients might be able to create plans for their end of life care before they lose the ability to
make decisions about it themselves (CMS, 2016c). Despite this, ACP can be emotionally difficult
and is stigmatic for some, and many physicians with end of life patients are not trained to provide
it (IOM, 2015). Furthermore, needs and harm associated with end of life can extend beyond
patients to their family members and caregivers as well, who must often bear the outsized
emotional, psychological, and financial costs of end of life care.
Types of RMT for end of life care can vary significantly, often depending on the setting
where the care takes place, and like many parts of health care it is an area undergoing significant
experimentation and change. A recent review of palliative care (care that focuses on relief of pain
and other harmful symptoms over curative measures) in the ICU found strong evidence in the
literature that, as opposed to standard care, it tends to result in decreased LOS in both the ICU
and the hospital generally, as well as, surprisingly, either no change in or sometimes even lower
55
mortality (Aslakson, 2014). Another review of studies on palliative care performed prior an ICU
admission suggests that such care might reduce the likelihood of an ICU admission by up to 37%
(Khandelwal et al., 2015). Other types of end of life RMT can occur in the home. A study of a
home-based primary care (HBPC) intervention in Washington, DC found that the intervention
reduced Medicare costs and their associated harms by 17%, mostly through a reduction in hospital
and skilled nursing facility (SNF) costs, for high-risk elderly patients over a 2 year period
compared with a matched control group (De Jonge et al., 2014). And a comprehensive program by
Sutter Health in California that includes care coordination and home-based palliative care (what
t h ey c all “ ad v an c ed illn es s m an a gem e n t ” ) m ay b e r edu c in g Med ic ar e c ost s an d t h eir as soc iat ed
harms by an even greater amount, though published data are still forthcoming (Meyer, 2011).
Unfortunately, payment (or sometimes lack thereof) for end of life care causes similar
barriers as in standard chronic care management / coordination. In a recent report on end of life
care, the Institute of Medicine (IOM) (now the National Academy of Medicine) noted payment as
the number one barrier to providing high quality comprehensive care for patients with advanced
illnesses nearing end of life (IOM, 2015). There have been some successful payer pilot programs to
liberalize payments for hospice and palliative care services and enable more comprehensive end
of life care management (Spettell et al., 2009). But overall there are few widespread payer efforts
to better enable this kind of care. As a result, institutions like Sutter Health have had to provide
such programs at a significant financial loss, even though payer savings likely outweigh program
costs (Meyer, 2011). Recent studies showing an increasing trend since the year 2000 in ICU use in
the last month of life, late referrals to hospice care, and repeat hospitalizations in the last 90 days
of life demonstrate the urgency in solving this payment barrier for comprehensive end of life care
(Teno, Freedman, Kasper, Gozalo, & Mor, 2015).
2.5.4 - Iatrogenic Risk Management
Finally, advances in iatrogenic risk management are a fourth major area of ongoing RMT
efforts. Iatrogenic risk refers to the risk of patient needs and harm associated with the provision of
(or failure to provide) medical services. Thus, it is an overlapping element of all the previously-
mentioned areas of RMT, as well as a concern in all areas of standard morbidity treatment. This is
an area most often associated with medical errors, though what is and is not a medical error can
be a contentious issue. Unambiguous and completely preventable medical errors are often called
“ n ev er ev en t s” or “ se n t in e l ev en t s” ; t h es e a re p roblems s u c h as w ron g sit e / w ron g p at ien t surgeries, medication errors, and patient suicide or self-harm while under the care of a health care
facility (AHRQ, 2016a). Other iatrogenic harms, such as health care-associated infections (HAIs)
and problems with care transitions, are often labeled medical errors but may vary in
preventability. A recent review concluded, for instance, that 55-70% of major HAIs may be
preventable using current evidence-based techniques (Umscheid et al., 2011). Just enforcing
evidence-based practice in this case would absolutely be a great thing and would prevent a lot of
harm and costs, though it would likely be difficult to say in any individual HAI case that it was
56
p re v en t ab le w it h c er t aint y. Regard less , ev en if an i at rogen ic h ar m is n ’t alw a ys c om p let ely
preventable, if there are evidence-based interventions known to at least reduce the risk of that
harm occurring then they ought to be provided in every applicable case.
Managing iatrogenic risks is challenging as they are numerous and varied. Risks of never
events may seem like the most straightforward to treat since they are considered 100%
preventable; Medicare and many private insurers even stopped paying for any costs associated
with them years ago, partly to encourage health care institutions to make necessary system
changes to avoid them (AHRQ, 2016a). Yet because these events are generally rare such a financial
penalty may not be enough. Given the low risk of any never event happening in a given year, a
h osp it al mig h t h av e u n sa f e d eliv er y p roc es se s or p oo r sa f et y c u lt u re and n o t r ealiz e it u n t il it ’s t oo lat e. T h is is on e r eas o n w h y t h er e’s s u c h a p u sh f or h osp i tals especially to adopt the principles of
high reliability organizations (HROs), like aviation and nuclear power companies, who are well-
versed in techniques to mitigate low probability but dangerous risks (Chassin & Loeb, 2013).
Iatrogenic harms other than never events are typically much more common and are
therefore generally much more harmful and costly in the aggregate. Even though they may not be
100% preventable there are still significant efforts underway to redesign care processes to try to
reduce their risk. One of the brightest success stories has been in the prevention of central line-
associated bloodstream infections (CLABIs), as noted earlier in this chapter (Pronovost et al.,
2011). Rates of other hospital-acquired conditions (HACs), including other HAIs, pressure ulcers,
adverse drug events, and falls, have been trending downward overall since 2010 though generally
not as impressively as the reductions in CLABIs (and not at all in some cases) (AHRQ, 2016b).
Payment for iatrogenic risk management has mostly consisted of financial penalties in the
form of nonpayment for the added costs of treating iatrogenic harms. In addition to nonpayment
for never events, in 2008 Medicare stopped paying for costs related to certain HAIs (Peasah,
McKay, Harman, Al-Amin, & Cook, 2013). This was associated with a slight decrease in the
probability of acquiring these harms, though given that there were also many concurrent HAI
prevention efforts it is impossible to say whether nonpayment was a motivating factor. Because
harms such as readmissions after surgeries and hospitalizations within 30 days are often at least
partly iatrogenic, Medicare has also been penalizing hospitals for several types of these since 2012
through its Hospital Readmissions Reduction Program (which is a different program from the
previously-mentioned surgical bundles) (CMS, 2016e). These financial penalties are likely quite
motivating, though there are concerns they may have unintended consequences. Moreover, they
often target iatrogenic risks related to hospitalizations and surgeries while payment incentives to
manage other iatrogenic risks, such as those in primary care or involving care transitions from
on e f ac ilit y t o an oth er , ar e lac king . I ’ll delv e i n t o t h es e mo re i n t h e d is c u ss i on c h a pter.
Pronovost and Bo-Linn (2012) have noted that focused efforts to reduce individual types of
preventable patient needs and harm, though they have been successful in limited cases, are likely
to prove disappointing overall since these needs tend to be interdependent. Needs and harm,
whether fully iatrogenic, partially iatrogenic, or unrelated to quality of care, tend to arise from
complex interactions of risk factors related to patient acuity, inherent riskiness of a medical
57
intervention, and poor system quality. As they note, truly tackling patient morbidity risk in a
holistic manner will require innovation in entire systems of care, including identifying all of the
risks of needs and harm for each type of patient-intervention combination, establishing best
evidence-based care to reduce those risks, and standardizing and automating as much of that care
as possible to ensure that it is consistently provided. In effect, successfully treating patient
morbidity risk will require as much or more reengineering of care processes and systems as was
required to develop successful treatments for the morbidities themselves. In this way it is difficult
to draw boundaries between the various RMT areas outlined thus far since they are all ultimately
c on n ec t ed. An d i t ’s likely that these necessary innovations in systems of care will require
significant concurrent innovations in provider payment to support them.
Medical needs in the U.S. have changed significantly over the past century, especially
since large-scale health care provider payment systems were initially conceived and popularized.
Needs have grown more chronic as average lifespans have increased, but more specifically the
need to treat the risk of harmful morbidity, and not just morbidity itself, has grown. Because the
risk of morbidity behaves in a fundamentally different way than the morbidities it causes – it
varies significantly less over time – this presents an opportunity to reexamine these old payment
system designs in a new light. It may be possible to design payment systems that specifically
encourage better morbidity risk reduction – i.e., better prevention of medical needs and patient
harm – without causing the unintended consequences of payment reforms that have focused on
all medical care and its associated costs at once.
The devil, of course, is in the details. The next chapter details the design of a simulation
model that tests ways in which this difference in variation between morbidity and morbidity risk
might be used practically toward the design of more prevention-oriented provider payment
systems.
58
Chapter 3 - Simulation Model Design
Patient morbidities may occur despite the best efforts to prevent them. And those
morbidities that are successfully prevented may require years of healthy patient behavior and
environments to do so. Though these public health outcomes are important goals for society as a
whole, they are largely beyond the near term influence of individual health care providers. That
said, there is still a small but significant amount of patient morbidities that are potentially
preventable by health care providers in the short term. Whether these morbidities are caused by
provider error, natural complications of disease, or even just health care system complexity and
fragmentation, they are patient needs and harms that providers ought to try to prevent.
As noted in the previous chapter, treatable patient morbidities are a type of patient need
for medical care. And if, in the short term, health care providers are able to potentially prevent
some of these morbidities, the risk of these morbidities itself can also be considered a type of
patient need for care (Acheson, 1978). These two types of patient medical needs, morbidity risk
and morbidity itself, are functionally related; the former influences the latter, though in a
probabilistic (rather than certain) way. Patient morbidities vary naturally over time, so any
attempt to mitigate the risk of these morbidities may succeed on average but fall below
expectations in any given year. Any health care provider who attempts to treat both of these types
of needs among its patient population must contend with this complex system.
The innovative health care interventions that were summarized in the previous chapter,
including care management / coordination for the chronically ill, perioperative care management,
end of life care management, and interventions to reduce medical errors and other iatrogenic
risks, all have enormous potential to reduce potentially preventable medical needs and patient
harm. Since these kinds of interventions effectively treat the risk of morbidity itself, I call them
risk mitigation treatment, or RMT, to better differentiate from other prevention-related efforts
practiced by non-health care providers. Together they offer significant promise toward the
development of a more holistic health care system that tries as much to mitigate the risk of future
patient morbidities as it does to treat them when they arise.
But as noted in the first chapter, sustainably paying for these RMT interventions is
difficult under predominant fee-for-service (FFS) provider payment systems in the U.S. First,
many RMT interventions benefit from atypical care pathways, such as telemedicine and
population health management outside of a patient visit, that are often difficult or cost-
prohibitive to bill for under FFS. Second, FFS payment systems can sometimes promote delivery
system fragmentation which makes RMT difficult and may even be a cause of preventable patient
harm itself. And third, if an FFS health care provider delivering RMT is also responsible for
t re at in g it s p at ien t s’ f u t u r e mo rb idit ies , w h e n RM T r edu c es t h e r is k of those morbidities it also
likely r edu c es t h e p rov ide r’ s f u t u re r ev en u es . T h is c an make su st aini n g RMT in t er v en t ion s , w h ic h themselves usually require some ongoing costs, difficult even for the most altruistic providers.
Figure 3.1 is a conceptual framework that illustrates this complex system of
59
aforementioned relationships between morbidity risk, morbidity, treatment of each, and
payment. This framework also introduces the concept of provider leverage over these
relationships. Health care providers do not exist in a vacuum, simply responding to patient
medical needs as they arise. They in fact have some influence over the makeup of morbidity risk
in their patient populations, and they have a lot of influence over how morbidity risk and
morbidity are translated into demand for care and eventual treatment. Most of this influence is
normal and expected, but as noted in the first chapter there is concern that the way providers are
paid can push this influence in an undesirable direction. FFS may encourage providers to create
more demand for their treatment than is necessary, and capitation may encourage providers to
avoid sicker patients who have higher morbidity risk and to be too stingy with their treatment.
And if morbidity treatment has a consistently grea t er in f luence ov er a p rov ider ’s f in an c ial
p er f orm an c e t h an RM T , t h e f orm er may c o m e t o d om i n at e a p rov ider ’s de c is ion m ak in g t o t h e
neglect of the latter.
Figure 3.1: A conceptual framework of morbidity risk, morbidity, and payment (created using Insight Maker)
One way in which health care experts worry about the effect of provider payments on
these leverage points relates to profit seeking behavior. Providers who are very profit oriented
may influence these leverage points (either upward or downward, depending on the payment
system) in undesirable ways to gain higher profits (this is typically why it is thought that FFS
encourages care volume over value). But even for highly altruistic providers there is another
worry. Both morbidity risk and morbidities themselves tend vary significantly in patient
populations over time. Health care providers typically have to invest in a substantial amount of
resources – buildings, equipment, and labor – in order t o add re ss t h eir p a t i en t s’ medic al n eed s
and demand for care, and these resources have ongoing costs that do not change in the short
60
term whether demand is high or low (i.e., they are effectively fixed costs in the short term).
Natural variation in patient needs that causes demand to fall outside of expected norms can
generate financial risk for health care providers. And this financial risk may also create an
incentive for providers to influence their leverage points in undesirable ways, not simply from a
desire to gain more profits, but rather from a worry about staying in business.
Co n t ra ry t o p op u lar u sa g e, w h er e “ f in an c ial r is k” is u su ally r es er v ed f or de sc rib in g
providers under capitation or alternative payment systems who ar e a t r is k if t h eir p at ien t s’ demand is too high, it can just as easily describe providers under FFS who are at risk if their
p at ien t s’ de m an d i s t oo l ow. Bec au se c ap it at ion p ay m en t am oun t s ar e s et in ad v an c e of c ar e a n d
do not change in response to demand, whereas FFS payments do change based on demand,
capitation is typically thought of as more financially risky for the provider. This can also stem
from the fact that the incentives behind capitation (provide less care) are perhaps more
worrisome to patients and health care experts than the incentives behind FFS (provide more
care). But generally high short-term fixed costs in health care (which must also be determined in
advance and do not change in response to demand) ensure that there is ample financial risk for
health care providers regardless of payment system. In fact, because it has a more predictable
payment, capitation has the potential to be less financially risky than FFS (Gapenski & Langland-
Orban, 1996) and as noted in the previous chapter it was the payment system of choice for
hospitals during the Depression for this exact reason.
The more patient needs and demand for care tend to vary, the higher provider financial
risk can be. Providers can mitigate the effects of this kind of financial risk through the law of large
n u m be rs ; a large p at ien t p op u lat ion w ill t en d t o ha v e les s n at u ra l v ar iat ion in it s p at ien t s’ av er age
needs than a smaller one (assuming they are drawn from a similar patient distribution). But it
is n ’t alw ay s p oss ib le or d es ira ble for providers to grow larger in an attempt to mitigate financial
ris k; l ar ger p rov ider s ar en ’t an o p t ion i n s m a ll er c o m m u n it ies , a n d e v en w h e n t h ey ar e a n op t io n many health care experts worry that incentives for providers to grow larger will lead to more
mergers, consolidating provider market power and driving up prices. Yet providers with smaller
patient populations are still at a greater financial risk from outlier years; this is why current
provider payment theory tends to recommend that only larger providers be paid capitation
(Conrad, 2015).
This is unfortunate, because smaller providers make up a significant proportion of the
health care providers in the U.S. Furthermore, the implication that payment systems like
capitation may be problematic for smaller providers does not bode well for other alternatives to
FFS, such as shared savings and losses in accountable care organizations (ACOs). In theory, like
capitation, these payment systems allow providers to be reimbursed for interventions like RMT
t h at , as men t io n ed ear lier , of t en c o n t ain ca re p at h w ay s like t elem edic in e t h at ar en ’t as ea si ly
financed in regular FFS. And since these FFS alternatives are often seen as a bridge to capitation
or even a desirable payment system in their own right, they have the potential to promote the
innovation and proliferation of RMT interventions more widely. But if these payment systems also
cause a significant amount of financial risk, especially for smaller providers who have
implemented RMT interventions, they could make sustaining such interventions difficult.
61
This is where there is one potentially useful difference between morbidity risk and
morbidity when it comes to payment. As mentioned in the previous chapter, both types of needs
vary naturally over time, but from a mathematical standpoint morbidity varies a lot more than
morbidity risk. Because these types of needs have different amounts of natural variation, it ’s possible that they are responsible for causing different amounts of financial risk for providers. If
capitation or other payment alternatives were used to pay only for RMT, this might produce less
financial risk for providers (especially smaller ones) than if they were used to pay for morbidity
treatment as well. This lowered financial risk could reduce a barrier to provider adoption and
sustainment of RMT and thereby effectively help to reduce potentially preventable medical needs
and patient harms.
In order to obtain some idea about the role of variation in patient needs and demand in
creating financial risk for health care providers and how that might change when they implement
RMT interventions under various payment systems, I constructed a simulation model. From its
results I hope to comment on the theoretical sustainability of RMT interventions – from the
perspective of minimizing financial risk – for providers of various sizes, as well as to give an idea
about which types of payment systems might make sustaining RMT easier or harder, especially for
smaller providers. Since finding ways to mitigate the risk of future patient morbidity is crucial to
needs and harm prevention in health care, I believe it is important to try to say which types of
payment systems might be best suited for the job.
The primary research question I am interested in addressing with this simulation is the
following:
Does a direct capitation payment to implement provider interventions that reduce
potentially preventable medical needs and patient harm create less provider financial
risk in comparison to other payment systems (including full capitation and ACO shared
savings), and if so how much less?
There are three specific simulation objectives in line with this research question:
1. Quantify how much financial risk, measured as provider manipulation of demand, is
produced by natural variation in patient medical needs among health care providers under
various payment systems at baseline (without risk mitigation treatment).
2. Quantify how much financial risk is produced when adding a risk mitigation treatment
intervention that is paid for indirectly (from shared savings or capitation).
3. Quantify how much financial risk is produced when the risk mitigation treatment
intervention is paid for directly (with a specific RMT capitation payment).
3.1 - Why Simulation?
Natural variation in patient needs and demand is a large potential source of financial risk
62
for health c ar e p rov ider s, bu t it is n ’t t h e on ly on e . Mos t p rov ider s in r eal lif e d on ’t o n ly h av e t o
account for the varying needs of their patient panels. They also have to account for the size of
those panels, as well as whether their patients are seeking care with them or with their
competitors. This market risk due to competition adds another element to financial risk that may
also drive even altruistic and nonprofit providers to engage in risk-averse behavior to try and stay
in business. And then, of course, any providers who are particularly profit seeking could also
decide to engage in the same behavior simply to make more profit. If it were possible in real life to
detect when an individual provider was trying to risk select for more favorable patients or induce
demand directly, it would be hard to determine how much of the incentive for them to do so was
driven by financial risk (either from variation in patient needs or market risk) or from a profit
motive. A simulation allows us to create a controlled environment where we can eliminate other
factors that might drive providers to manipulate demand, so that, in this case, I can measure a
p rov ider ’s f in an c ial r is k du e s ol ely t o n at u ra l v ar iat ion i n it s p at ien t s’ n eed s.
Of c our se , in r eal lif e it ’s also ef f ec t iv ely impossible to determine when an individual
provider is engaging in risk-averse manipulation of patient demand, especially from patient
utilization data. Providers play a crucial and expected role in creating patient demand for care.
Because of their expertise, they are typically the ones who determine diagnoses and courses of
treatment on behalf of their patients. Whether or not they are creating an undesirable amount of
demand (either too little or too much) can be very difficult to determine at the individual
provider level unless it is an egregious departure from evidence-based standard practice. And if
perceived financial risk is motivating providers to undesirably modify patient demand but only by
a small amount, they may not even believe that what they are doing is inappropriate
(Himmelstein, Ariely, & Woolhandler, 2014) even though it can have significant implications for
t h eir f in an c es as w ell as f or t h eir p at ien t s’ h ealt h a n d s at is f ac t ion . Or c o n v er se ly, if p rov ider s don ’t m odif y t h eir p at ien t s’ demand in response to financial risk it could potentially threaten their
bottom line, but at least in that case the financial risk could be more easily measured as variation
in financial performance.
That said it ’s c er t ainl y p oss ib le t o l oo k at ag gregat e data and determine that, in general,
some amount of provider manipulation of patient demand is probably occurring. It has been
known for a while, for example, that there is significant geographic variation in average per capita
Medicare patient expenditures even after adjusting for differences in regional prices and patient
h ealt h , and t h at t h is v ar ia t ion is lar ge ly d u e t o dif f er en t lev els of “ su p p ly s en si t iv e” (i. e. , ea si ly
provider-modified) care (Wennberg, Fisher, & Skinner, 2002). There is also some aggregate
evidence that Medicare Advantage providers risk select their patients (selectively seek out and
enroll healthier ones) to some degree, even after payments are adjusted to reflect patient risk
(Brown, Duggan, Kuziemko, & Woolston, 2011). Thus it seems reasonable that, at a minimum,
providers may choose to manipulate demand to some degree in response to financial risk, even if
in r eal lif e t h is ma n ip u lat ion c an ’t b e m eas u re d a t t h e indiv idu al p rov ider lev el. A s im u lat io n , however, does allow us to measure it based on our assumptions about how providers might
manipulate demand in response to financial risk, so it provides a useful theoretical tool in this
case. I t ’s imp ort an t t o n ot e t h at , de sp it e t h e f ac t t h at f in an c ial r is k is b ein g m eas u re d in the
63
simulation in relation to provider risk-averse manipulation of patient demand, there are other
risk-averse behaviors that providers could also engage in, either in addition to or instead of
demand manipulation. A later section in this chapter will detail why I ’v e c h osen t h is p ar t ic u lar
method for measuring financial risk.
Finally, simulations are an inexpensive way to conduct experiments that would be overly
costly (either in time or money) in the real world. RMT is still a developing area of medicine, so
simulation allows us to test alternative assumptions about its theoretical possible future system
effects. And with simulation we can also easily vary different input parameters and assumptions
to test system sensitivity under uncertainty in them, as well as to test different real world
scenarios (such as a patient population with less or more variation in needs, or a provider with a
different cost structure).
Because this simulation does not represent all of the factors that could lead to provider
behavior to manipulate patient demand, and furthermore because this behavior is extremely
difficult to measure in practice, it cannot (nor is it intended to) produce validatable measures of
provider demand manipulation that could be compared to real world data. Rather, the purpose of
this simulation is to ask what-if questions about the theoretical role that natural variation in
patient needs plays in contributing to financial risk among providers of various sizes and under
different payment system designs, especially for those providers who are implementing RMT
interventions. Given the likely impossibility of determining this theoretical role from real world
data, simulation is a useful tool to offer some insight, and it provides some starting direction and
theory for those wishing to design payment systems that will encourage RMT and thereby aid
providers in practicing prevention.
3.2 - Simulation Model Design
The simulation model was designed as a Monte Carlo simulation within a system
dynamics framework using a free web-based software called Insight Maker (Fortmann-Roe, 2014).
System dynamics is a simulation technique that explores the behavior of whole populations at
once, rather than of individuals, and as such could be considered to be at the opposite end of the
spectrum from techniques such as agent-based simulation. Thus, in this simulation patients are
aggregated together in a single population, and the outcomes of the simulation are measured at
the population level. This is probably a good way of conceptualizing financial risk since impactful
variation in financial outcomes typically comes from the fluctuating demand of a whole patient
population rather than from any individual patient. Monte Carlo refers to the fact that the
simulation is random within defined parameters, and statistical quantiles of outcomes of interest
are determined based on many repetitions of simulation runs.
A single provi der ’s p at ien t p op u lat ion is c re at ed in ea c h s im u lat ion r u n , and t h e s ys t em’s
behavior – t h e medic al n eed s gen er at ed a n d t h e p rov ider ’s r es u lt in g f in a n c ial p er f orm an c e a n d
risk-averse behavior – is observed over time. As opposed to predictive models based on patient
attributes, such as those that use demographics or level of illness to predict costs, this simulation
64
simply attempts to generate a realistic pattern of total patient population needs based on survey
data describing the likelihood, mean, and variance of patient health expenditures. Also observed
is t h e p rov ider ’s b eh av ior in r es p on se t o t h es e n ee ds u n der v ar ious p ay m en t s ys t em sc en ar ios a n d
for providers of various sizes. It is expected that providers with larger patient populations will
have lower variation in patient needs, so one goal of the simulation is to determine how much
increases in provider size affect resultant risk-averse behaviors, as this may indicate how much of
an incentive providers have under various payment systems to try to increase in size.
There are four primary modules within the simulation: patient needs generation, provider
financial accounting, provider risk-averse behavior, and risk mitigation treatment. The following
sections go into each of these modules in detail.
3.2.1 - Module 1: Patient Needs Generation
Needs in the simulation are measured according to the equivalent amount of expenditures
t h at w ould b e n eed ed t o t re at t h em, f ol lo w i n g Do n ab edia n ’s (197 4) m ode l o f medic al n eed equivalents (as noted in the previous chapter). Data on patient medical expenditures comes from
t h e Agen c y f or Healt h c ar e Re se ar c h and Qu alit y’ s Me dic al E xp en dit u re P an el Su rv ey (M E P S) , 2 0 13
Household Component (AHRQ, 2015). In that year, the sample mean of patient medical
expenditures was $4,436, and the sample standard deviation of expenditures was $12,733. This is a
nationally-representative survey of the U.S. civilian non-institutionalized population, all ages. The
sample size is about 35,000 patients, and patients from lesser-represented demographic and
diagnostic groups are oversampled to improve survey representativeness. Information collected in
the survey includes: demographic information, health conditions and status, types of health
service utilization, total health care expenditures (including payer and patient portions of
expenditures), access to and satisfaction with care, health insurance status, and income and
employ m en t . T o t h e b es t de gree p oss ib le, p at ien t s’ h ealt h c ar e p rov ider s ar e also s u rv ey ed t o
supplement and/or replace information obtained from the patients.
Simulated needs are generated using a two part method that first calculates the fraction of
patients in the population that will have health care expenditures in the current period, and then
generates those expenditures (before any provider modification) for those patients (Diehr, Yanez,
Ash, Hornbrook, & Lin, 1999). As noted in the previous chapter, typical distributions of patient
health care expenditures are complex, featuring significant multi-modality and often including a
large fraction of non-users of health care. Two part models are a useful way of trying to capture
this distribution of expenditures more accurately. Expenditures could also have been generated
adequately using a one part method, but splitting it into two parts allows both factors (the chance
of having expenditures and size of those expenditures that do occur) to be calculated
independently, which can be useful for testing alternative assumptions about each. In the MEPS
2013 survey, 84.4% of the population had at least some expenditures in the year, meaning
conversely that 15.6% of the population had no health care expenditures at all.
The fraction of the patient population that has any health care expenditures in a period is
65
modeled in the simulation as a binomial random variable (with a normal approximation to
improve simulation speed) according to the following formula in Insight Maker:
[Fraction with Needs] = RandNormal([Population] * [Yearly Chance of Needs],
sqrt([Population] * [Yearly Chance of Needs] * (1-[Yearly Chance of Needs])))/[Population]
This is just the standard formula for the normal approximation of a binomial, divided by the
population size to create the fraction of patients in the population who have expenditures in each
simulation period. This assumption of a binomial distribution for the fraction of health care users
is simply the minimum amount of variation one should expect in this measure given a specific
baseline fraction of health care users and a specific patient population size. It is also likely a
conservative assumption of this amount of variation. For instance, a binomial distribution would
in c lud e t h e a ss u m p t ion t h at ea c h p a t ien t ’s c h an c e of ex p er ien c in g a h ea lt h c ar e ex p en dit u re in the year is independent of all other patients, which is certainly not true for many. Families often
get sick together, there are demographic factors that families and communities share that make
them more likely to experience needs for care, and so on. Any interdependence among patients in
their chance of incurring health care expenditures would likely increase the variation in a
p op u lat ion ’s [ F ra c t ion w i t h Need s]. So t h e b i n o m i al as su m p t ion s h ould b e c on si der ed a b as eli n e, conservative estimate of the year to year distribution of [Fraction with Needs] for a population.
And as a result, even for the smallest patient population size modeled (5,000 patients), when
using a baseline fraction of health care users of 84.4% the standard error of the fraction of actual
health care users in any given year is only 0.5%, with this value decreasing significantly for larger
patient population sizes. So this first part of the model only contributes a small amount to the
actual year to year variation in generated patient needs.
The second part of the model accounts for the unmodified expenditures of those patients
who have them. Modeling these expenditures can be tricky. As noted in the previous chapter, in
addition to multi-modality, distributions of health care expenditures also tend to have significant
rightward skew and heteroscedasticity, and the data from MEPS is no exception. Even though
health care expenditures are almost certainly non-normal, the most unbiased estimators of the
mean and variance of expenditures in a whole patient population are still the sample mean and
variance. In the MEPS 2013 survey, the sample mean and standard deviation of expenditures for
patients who had them (i.e., >$0 in expenditures) were $5,256 and $13,704, respectively. As Diehr
et al. (1999) note, due to the non-normality of health care expenditure data, ordinary least squares
(OLS) analyses based on the sample mean and variance will tend to produce regression
coefficients with standard errors that are too small, potentially inflating significance. Yet for large
enough data sets, as they also note, you could often double or triple the standard errors without
changing your conclusions due to the typically strong significant effects. This is why ordinary
least squares (OLS) is still the most common regression method of choice for large sets of health
care expenditure data.
For precise modeling of a distribution of patient health care expenditures that are non-
normal, however, the sample mean and variance can potentially be less than ideal even though
66
they are unbiased. Briggs, Nixon, Dixon, and Thompson (2005) note that, for instance, if
expenditures are actually lognormally distributed, the lognormal mean of expenditures will be a
better estimator of population mean expenditures than the usual sample mean. They also note,
however, that if expenditures are not truly lognormally distributed then the lognormal mean
quickly becomes a very poor estimator. Thus, they recommend caution in using it, as the sample
mean is often a perfectly adequate estimator given a large enough sample size. For this reason,
when modeling patient health care expenditures in the simulation, I use the usual sample mean of
expenditures from MEPS ($5,256 for patients with >$0 in expenditures).
Figure 3.2: Histogram of total U.S. health care expenditures, log transformed. Data from the 2013 Medical
Expenditure Panel Survey (made using R Survey)
The sample variance is a slightly different story. Briggs et al. (2005) do not comment on
the precision of the lognormal variance versus the sample variance as estimators of the population
variance, likely because in most practical applications such a precise estimator is unnecessary.
Typical applications of health care expenditure modeling, such as risk adjustment, ultimately care
about modeling total or average costs of large populations as a whole. Via the central limit
theorem (CLT), average expenditures w ill n ot v ar y si gn if ic an t ly f or lar ge p op u lat ion s, s o t h er e’s no need for a particularly precise estimate for population variance in most cases. However, in the
case of this simulation model, I simulate some populations that are relatively small at 5,000 and
67
10,000 patients. In addition, I simulate risk-averse health care provider behavior that directly
depends on the amount of variation in generated patient needs. Because of this, I believe it is
worth trying to obtain a somewhat more precise estimate of the population variance for health
care expenditures.
Health care expenditure data is said to be lognormal if ln(expenditures) are normally
distributed (Diehr et al., 1999). As noted in the previous chapter, a log transformation of the
MEPS expenditure data does indeed appear visually to be normal for those patients who had
expenditures (see this again in Figure 3.2), so the lognormal distribution seems like a logical place
to start when building a model of patient expenditures. The sample mean and variance of
ln(expenditures) for patients with >$0 in health care expenditures in the 2013 MEPS data are
7 . 2 2 8 1 an d 2 . 9 0 5 , r es p ec t iv ely (I ’ll n ote t h em as x
log
and s
2
log
). Back transforming x
log
yields the
sample geometric mean (median) for the lognormal model of costs, x
g
, via the following equation:
(1): x
g
= exp(7.2281) = $1,378
This is quite close to the actual median in the MEPS data, $1,389, so it seems that the lognormal
model describes the geometric central tendency of the data well.
If the data were truly lognormal, you could transform x
g
into the arithmetic mean, x , by
multiplying it by exp(s
2
log
/2). Doing so yields:
(2): x = 1378 * exp(2.905/2) = $5,888
Unfortunately this is higher than the actual sample arithmetic mean of $5,256. If the data were
truly lognormally distributed it is possible that the arithmetic mean derived from a lognormal
model could be a more precise estimator than the usual sample mean (Briggs et al., 2005), but
given the size of the MEPS sample (~35,000 patients) it is more likely that the data is simply not
completely lognormal.
Since I would prefer to use the actual sample arithmetic mean of expenditures (for
patients with >$0 in expenditures) in my needs generating model, the arithmetic mean of the
model should ideally reflect this value. This can be accomplished in the lognormal model of
patient expenditures by scaling the estimate of s
2
log
accordingly so that equation (2) above is equal
to $5,256, like so:
(3): x = 1378 * exp(s
2
log-scaled
/2) = $5,256
Solving for s
2
log-scaled
yields a value of 2.678, a drop from its previous value, 2.905. The resulting
scaled lognormal model, where x
log
and s
2
log
equal 7.2281 and 2.678, respectively, seems to
adequately describe both the geometric and arithmetic means of expenditures in the MEPS
sample data. The distribution of both the scaled and unscaled lognormal models are compared to
actual values from the MEPS 2013 data for various normal scores in Table 3.1.
68
Normal
Score
-3
(0.1%)
-2
(2.3%)
-1
(15.9%)
0
(50.0%)
1
(84.1%)
2
(97.7%)
3
(99.9%)
Actual $5 $50 $244 $1,389 $8,075 $36,576 $154,440
Lognormal
(scaled)
$10 $52 $268 $1,378 $7,077 $36,354 $186,750
Lognormal
(unscaled)
$8 $46 $251 $1,378 $7,574 $41,643 $228,961
Table 3.1: Lognormal models vs. actual data from MEPS 2013 for various normal scores
The scaled lognormal model seems to fit the expenditure data fairly well, especially in the
extreme upper quantiles, though not perfectly of course; the true distribution of the underlying
data is likely far more complex. But given that the underlying data does seem to be nearly
lognormal and it would be beneficial for the purpose of this simulation model to have a more
precise estimate of the population variance of expenditures, a scaled lognormal model that
describes both the geometric and arithmetic means of the MEPS sample data well ought to be
suitable. For any lognormal distribution, the variance can be calculated by the equation:
(4): σ
2
= (exp (σ
2
log
) - 1) * ex p (2 *μ
log
+ σ
2
log
)
Substituting the appropriate estimators, we come up with an estimate for σ, t h e p op u lat ion standard deviation of health care expenditures (for those patients with >$0 of expenditures), of
$19,352. Compared to the regular sample standard deviation (again only for patients with >$0 of
expenditures) of $13,704, this estimate is somewhat higher.
We can find the overall population standard deviation (which includes patients with no
expenditures as well) by taking into account the fraction of patients who had expenditures
(84.4%). This overall standard deviation, based on the scaled lognormal model, is
sqrt(0.844*19352^2) = $17,779, which is again higher than the regular sample standard deviation
(normal model) of $12,733 from the MEPS data. Using the sample mean of patient expenditures
($4,436) this translates to coefficients of variation (CVs) of 4.01 and 2.87 for the two models,
respectively. Both of these CV values are within the range of values typically reported in other
patient population samples, though they are higher than those typically reported for populations
with only Medicare patients since the MEPS data includes patients of all ages from the general
population. For example, reported CVs for individual health care expenditures in Medicare
populations typically range from 1.32 to 2.60 (DeLia, Hoover, & Cantor, 2012), while populations
including non-Medicare patients tend to have more extreme variation, with reported CVs ranging
from 2.76 to 6.46 (Friedberg, Buendia, Lauderdale, & Hussey, 2013; DeLia, 2016). Since I believe it
describes the variation in individual expenditures in the 2013 MEPS population a bit more
p re c is ely t h an t h e n orm al mo del a n d i s w it h in t h e ra n ge o f oth er r ep ort ed v alu es , I ’v e op t ed t o
use the values from the scaled lognormal model (mean = $4,436, standard deviation = $17,779, CV
69
= 4.01 for all patients) to simulate variation in patient needs.
As n ote d e ar lier , du e t o t h e c en t ra l lim it t h eorem t h is h igher c oef f ic ien t of v ar iat ion w on ’t have much of an absolute effect on total and average population expenditures for large patient
p op u lat ion s, bu t it ’ s worth comparing the two models for the various patient population sizes
modeled in the simulation to see the difference. Via the central limit theorem, average patient
expenditures in each population have an expected value equal to the mean of individual
expenditures, and they will be approximately normally distributed with a standard deviation
equ al t o σ/s qrt (n ) , w h er e n is t h e patient population size. Table 3.2 compares the coefficients of
variation for the average of patient expenditures in a population , equ al t o ( σ/ sqr t (n ))/ μ , f or t h e
normal and scaled lognormal models at the various patient population sizes tested in the
simulation (shown here as % since these average expenditure CVs are much smaller than the CVs
for individual expenditures noted in the previous paragraph). As you can see, the coefficients of
variation for average expenditures in both models approach each other in absolute terms at larger
population sizes. But for smaller population sizes, the absolute difference in CVs between the
models is more significant.
Population Size CV: Normal Model
CV: Scaled Lognormal
Model
5,000 4.06% 5.67%
10,000 2.87% 4.01%
25,000 1.82% 2.53%
50,000 1.28% 1.79%
Table 3.2: CVs for population mean expenditures in normal vs scaled lognormal models
We can also compare these values representing variation in average expenditures with
analogous values from the literature. Using the standard deviation of individual patient
expenditures from the scaled lognormal model ($17,779), we can calculate that the standard
deviation of average expenditures for a provider population consisting of 5,000 patients would be
17779/sqrt(5000) = $251. Or in other words, random samples of 5,000 patients from the MEPS
sample population would produce average expenditures that likely vary by a standard deviation of
about $251; larger patient population sizes would vary less. Yet a recent study of the differences in
average patient expenditures between ACO and local non-ACO Medicare populations in similar
geographic areas found that, even after standard adjustments for population risk (factoring out
differences between populations due to level of illness), the standard deviation of these
differences in average expenditures was still $371 (Rose, Zaslavsky, & McWilliams, 2016). It is
impossible to say for sure how much of this variation was due to differences in care efficiency
between ACOs versus random variation in patient needs. Nevertheless, given that the overall
MEPS population would be expected to have higher natural variation in expenditures than the
70
Medicare populations from this study, this at least suggests that a standard deviation of $17,779
for patient expenditures is not an unrealistic value to use in generating varying population needs.
Figure 3.3: Simplified diagram of patient needs generation (created using Insight Maker)
Combining both parts, total patient population health care needs (calculated as equivalent
patient expenditures) are modeled as a random normal variable (via the central limit theorem)
with the following general formula in Insight Maker:
[Total Needs] = RandNormal([Population] * [Fraction with Needs] * [Average Needs],
sqrt([Population] * [Fraction with Needs] * [Needs Variance]))
[Average Needs] and [Needs Variance] refer to the average and variance of patient expenditures
from the scaled lognormal model for patients with >$0 in expenditures. [Total Needs] are
generated separately in three situations in the simulation: for the initial general patient
population as a whole at the start of each simulation run; for the existing patient population
(those patients that remain in the general population from the previous period) each period; and
for the new patient population (those patients who are new to the general population due to
turnover) each period (see Figure 3.3). Initial patient needs at the start of a run are calculated
exactly according to the [Total Needs] formula above. This means that the initial patient
population in each simulation run is effectively a random sample from the scaled lognormal
model of the patient population in the MEPS sample. Needs for the existing patient population
ar e c alc u lat ed w it h a slig h t mo dif ic at ion b as ed o n t h e “ st ic kiness ” of n eed s f rom o n e p er iod t o t h e
71
next (detailed further be lo w ). N ew p at ien t n eed s a re c alc u lat ed si m ilar t o t h e initi al p op u lat ion ’s needs but can be modified slightly if the provider is practicing risk selection for new patients
(detailed in a later section).
The sizes of the existing and new patient populations in each period are influenced by
turnover. Patient turnover is calculated as a binomial random variable (via a normal
approximation) based on a base turnover rate according to the following function:
[Fraction New Patients] = RandNormal([Population] * [Patient Turnover], sqrt([Population]
* [Patient Turnover] * (1 - [Patient Turnover])))/[Population]
Here again the assumption of independence for the binomial distribution also likely
underestimates the true variation in patient turnover year to year. Not only are families likely to
m ove t o o t h er p rov ider s or p lan s t og et h er , b u t if a doc t or leav es a p rov ider or p lan it ’s p oss ib le
that many of her patients will follow. Thus, the real variability of year to year patient turnover at
any given provider is probably higher than the standard binomial would assume, but without
more data about variation in year to year turnover rates, the binomial offers the minimum level of
variation one should expect.
Different providers have been shown to have quite a variety turnover rates. A 2015 study of
Me dic ar e P io n eer ACOs , w h ic h r ep re se n t M edic ar e’s lar gest and m ost de v el op ed ac c oun t ab le c ar e
organizations, found that between 2012 and 2013 turnover rates at these ACOs ranged from 20% to
54%, with an average of about 38% (L&M Policy Research LLC, 2015; Sullivan, 2015). There is very
little data on turnover rates for patients with private insurance, but what little there is suggests
they are sometimes even higher (Sullivan, 2015) (interestingly, physician turnover rates in the
Pioneer ACOs were similar to patient turnover rates, about 27% on average, pointing to a
potential link between physician and patient turnover).
It is unlikely that the natural patient turnover rate varies this much at any one particular
provider year to year; providers likely have different organizational structures and referral
patterns that tend to encourage more or less patient turnover. For instance, the Massachusetts
General Hospital (MGH) care coordination program outlined in the previous chapter has been
folded into a Pioneer ACO that has managed to maintain somewhat lower patient turnover in
recent years (18-27%) likely due to having a much wider network of affiliated providers, though
these turnover rates have still made accountable care challenging (Hsu et al., 2016).
The base turnover rate can be easily modified in the simulation to test various patient
turnover scenarios, but based on this data a rate of 35% is used at baseline. Patient population size
is held constant throughout the simulation to limit variation in needs to just what is expected
naturally, therefore all patients who leave the provider due to turnover are replaced with an
equivalent amount of new patients in each period.
After [Fraction New Patients] is calculated, patient needs (in the form of equivalent
expenditures) are calculated separately for the existing and new patient populations in each
simulation period. Needs for the existing patient population are calculated first according to the
[Total Needs] formula above as [Existing Patients Needs Baseline]. Afterward, they are weighted
72
ba se d on t h e p re v ious p er iod’s t ota l p op u lat ion n ee ds (s ee Fig u re 3 . 3 again) according to the
following formula:
[Existing Patient Needs Adjusted] = (1 - [Prior Year Weight]) * [Existing Patient Needs
Baseline] + [Prior Year Weight] * [Prior Year Total Needs] * (1 - [Fraction New Patients])
T h e [P rior Y ear T ota l Nee ds ] in t h is f orm u la is t h e p at ien t p op u lat ion ’s u n m odif ied t ota l n eed s
(i.e., before any demand inducement on the part of the provider) in the previous year. Also, an
assumption in using this formula is that turnover is completely random, i.e., the existing patient
p op u lat ion is n ot m ore or less s ic k on av er age t h an t h e p re v ious y ear ’s t ota l p op u lat ion . [P ri or
Year Weight] is the weighting factor.
Health care needs are not independent year to year within the same patient population,
bu t r at h er ar e a n c h ored a t leas t s om ew h at t o t h e p re v ious y ear ’s n eed s. T h is is es p ec ially t h e c as e
for patients with chronic illnesses, who typically need similar ongoing care year to year.
Commonly reported values for this weighting factor in large patient samples range from roughly
40-50% (Duncan, Loginov, & Ludkovski, 2016; Ellis, Fiebig, Johar, Jones, & Savage, 2013). This is
also consistent with the previously-referenced study on ACO costs (Rose et al., 2016), which found
that about 45% of the variation in mean per capita ACO patient expenditures could be explained
by measu re d p op u lat ion r is k (an d t h u s, if a p op u lat ion ’s r is k did n ot c h an ge si gn i ficantly from
on e y ear t o t h e n ext , ab o u t 45% of t h at p op u lat ion ’s n eed s w ould b e t h e s ame). T h is f ac t or c an b e
easily modified in the simulation to test various assumptions of fixed effects on patient needs, but
based on this literature a [Prior Year Weight] of 45% is used at baseline.
New patient population needs are also calculated according to the [Total Needs] formula
ab ove . T h es e n eed s ar e g en er at ed in dep e n dently o f t h e p re v ious y ear ’ s p op u lat ion n eed s, u n like
the case with existing patients. In addition, providers have an opportunity to risk select new
patients, which has an effect on their cost generation; more on that in a later section. Total
population needs in the current simulation period are just the sum of existing and new patient
population needs.
It is assumed that the data from MEPS used to simulate patient medical needs represents
a “ n at u ra l” dis t rib u t ion of n eed s an d d e m a n d a c c ording t o a u su al s t an da rd of c ar e, i. e. , b ef ore
taking into account any provider behaviors that would purposefully modify demand upward or
downward beyond this standard such as patient risk selection or provider induced demand. This
is obviously not likely the case; the expenditure data in MEPS technically represents realized
patient demand for care, and thus it is possible that this demand has already been modified by
providers via these two mechanisms in some way. Yet there is no way in practice to determine
af t er t h e f ac t ex a c t ly h ow m u c h of a p at ien t ’s de m an d f or c ar e w as du e t o a n at u ra l amo u n t of medical need and how much might be due to any undesirable inducement by a provider since it is
providers who are prescribing the care in the first place. Thus, it being necessary for the purposes
of the simulation model to have a baseline distribution of unmodified patient needs (so that
provider inducement of demand can be measured on top of this baseline), the MEPS data is used
to represent this baseline. Interestingly, because risk selection and demand inducement are
73
modeled in the simulation as responses to financial risk, the resulting patient demand has less
variation than the underlying baseline MEPS distribution. So if indeed financial risk generates this
kind of provider behavior in practice, then the MEPS data would represent a distribution of
patient demand fo r m edi c al c ar e t h at v ar ies less t h an t h e “ n at u ra l” m edic al n eed s t h at u n der lie it . Nev er t h eless , s in c e it is imp oss ib le t o ac t u ally me as u re t h at “ n at u ra l” dis t ri bu t ion of medic al
needs, the MEPS data is used as-is for this baseline.
After using the base distribution of patient population needs from the scaled lognormal
m odel m e n t ion ed p re v ious ly a n d t h en ac c oun t in g f or t h e “ st i c kiness ” of n eed s f or t h e ex is t in g
patient population from one period to the next (45% weighting at baseline), as well as for patient
turnover (35% at baseline), the actual resulting distribution of unmodified (i.e., before allowing
providers to modify patient demand in response to financial risk) year-to-year average population
needs in the simulation is approximately normal with the following characteristics for each
patient population size:
Population Size Mean Standard Deviation CV
5,000 $4,436 $196 4.41%
10,000 $4,436 $140 3.15%
25,000 $4,436 $88 1.98%
50,000 $4,436 $62 1.40%
Table 3.3: Distribution of actual average population medical needs (as equivalent expenditures) in simulation
Because the simulation is random these values change slightly from run to run, but Table 3.3
reflects the general distribution of unmodified average needs for each patient population size
tested. These distributions are unaffected by the type of payment system as they only reflect
underlying patient needs. In practice, providers in the simulation can modify actual patient
demand for care slightly, through a combination of risk selection for new patients as well as direct
inducement (up or down) of demand. When providers do this, the resulting distribution of needs
will change slightly depending on the type of payment system (due to risk selection), and the
distribution of actual payer costs will also change due to induced demand (except under regular
capitation where the payment stays the same).
3.2.2 - Module 2: Provider Financial Accounting
Provider revenues are straightforward to generate in the simulation and depend on the
type of payment system. Under regular FFS they are equal to the total amount of patient
population needs (expenditures) generated in each simulation period, plus any extra induced
demand. Under capitation, revenues are just equal to the baseline expected patient population
74
expenditures, $4,436 per capita in each period. Other types of payment, such as ACO shared
savings and losses, add additional revenue components and will be detailed separately in the
section on experiments.
Figure 3.4: Simplified diagram of simulated financial accounting (created using Insight Maker)
Provider costs are calculated by assuming a baseline fraction of fixed vs. variable costs
which can be modified to simulate different organizational cost structures. Fixed costs at a health
care institution, at least in an operational sense (as opposed to an accounting sense) in the short
term, typically include salaried labor and benefits, buildings and utilities, medical equipment
purchases, and depreciation. Variable costs include medical supplies and medications. Often
physicians are not salaried but are rather paid FFS or capitation directly (usually because of self-
employment), meaning their pay (technically a cost to the health care institution that employs
them) could theoretically be considered a variable cost. Yet there is good reason to believe that
physicians themselves have expectations about the level of income they ought to receive – a
reference income – and thus will tend to treat their pay at least somewhat like a fixed cost. They
may do so by trying to price their services in a way that recovers this expected pay, similar to how
a large health care institution such as a hospital will try to price its services to recover overhead,
or they may adjust their work / practice patterns to achieve their reference income without
modifying their prices. Some of this behavior could be considered societally undesirable;
physicians might merge practices with other physicians in an attempt to raise their prices through
higher market power, or they might modify the quantity of care they provide (induced demand)
in a way that increases their take home pay. There is some evidence in the literature that
physicians do indeed have reference incomes and may be more zealous about taking steps to
achieve these incomes when they are below them than when they are above them, indicating that
loss aversion may play a role in their behavior (Rizzo & Zeckhauser, 2003). Since physicians
typically control the bulk of health care spending and since this simulation attempts to model
provider behavior in response to financial risk, it seems prudent to assume that physician income
is a fixed cost in this simulation even if it is not salaried.
75
Most estimates of fixed versus variable costs in health care institutions do indeed count
physician pay as a fixed cost, and altogether total fixed costs are usually quite high. The literature
is somewhat scarce on this issue, but one of the most widely cited articles on fixed versus variable
costs for hospitals found that, at a particular public hospital that had a mix of emergency visits,
inpatient admissions, and outpatient visits, fixed costs were 84% of total costs (Roberts et al.,
1999). Another author found a similar fraction (85%) among a set of multispecialty group
practices (Weil, 2002). So it seems safe to assume that a typical ACO serving a normal variety of
patient needs would have a similar fraction of fixed costs. There are scenarios under which it
would make sense to test a somewhat lower fraction. For instance, a hospital may decide to take
on a greater proportion of part-time or contract workers, especially nursing staff and technicians,
or it may decide to outsource some care entirely. Any of these actions would decrease the fraction
of costs that could be considered fixed. This would be useful, for example, for a hospital
undertaking a program to lower patient expenditures as part of an ACO (perhaps by doing risk
mitigation treatment). In the event that the hospital was able to reduce patient utilization, they
would save more money by having a higher fraction of variable costs. Conversely, if anticipated
u t ilizat ion r edu c t ion s did n ’t mat er ialize t h ey w oul d lose m ore m on ey , s o t h er e’s s om e r i sk to this
strategy. At baseline in the simulation I use a fixed cost fraction of 85%, in line with the above-
mentioned literature, but I test lower fixed cost fractions in sensitivity testing to simulate the
possibility of lower fixed cost and higher variable cost environments.
Fixed costs in the simulation are based on the expected value of patient needs, $4,436 per
capita (equivalent), assuming a net profit of $0 at this expected level. This means that, at baseline,
fixed costs are equal to 85% of $4,436 ($3,771) per capita in each period, independent of variation
in patient demand. Variable costs would then be equal to 15% of actual patient demand in a given
period. Effectively, setting net profit to $0 at the expected value of patient needs ($4,436 per
capita) means that, under FFS, the provider will have a net profit greater than $0 if needs are
higher than this expected value, and less than $0 if they are lower (vice versa for capitation). The
decision to set net profit to $0 at the expected value of needs is somewhat arbitrary, but roughly
in line with the goals of nonprofit health care institutions to (eventually) spend all revenues they
receive on patient care rather than retain some of them as profits to distribute to shareholders.
This becomes som ew h at m ore c o m p lic at ed w it h t h e intr odu c t ion of “ ove rt im e” c ost s, as de t ailed in the next paragraph, but what is most important is that providers of all sizes are simulated with
this same expectation of $0 in net profit at the expected value of patient needs (i.e., all providers
have the same efficiency). The real value may differ for providers in the real world, which could
cause the financial metrics and risk generated in the simulation to be higher or lower in practice.
Any simulation period that involves much higher-than-expected demand would likely
require staff overtime or other additional labor costs related to being over capacity, and these
would have to be treated essentially as an increased amount of variable costs (or diminished
marginal profit). The simulation includes a mechanism to create these additional overcapacity or
“ ove rt im e” c ost s t h at c an b e a dj u st ed f or di f f er en t organ iza t i on al c h ar ac t er is t ic s. T yp ic ally, it is non-physician staff who are either paid overtime or are hired on a part-time or seasonal basis
during high demand, so it is their portion of costs that are used at baseline in the simulation to
76
re p re se n t an ext ra “ ove rt im e” v ar iab le c ost . I n t h e Rober t s et al. (199 9) s t u dy , di re c t s er v ic e lab or
(i.e., that related only to health care delivery, not including benefits or other support staff)
accounted for 52.4% of costs, with 20.9% from physician direct labor, so roughly 30% of costs
were due to non-physician direct labor. Erring conservatively on the scenario that most of this
extra labor consists of part-time or seasonal staff (and therefore would be paid at regular market
ra t es r at h er t h an 1. 5x or 2 x ov er t im e r at es ), t h e “ ove rt im e c ost f ra c t ion ” in t h e s im u lat ion is j u st set to 30% at baseline. Furthermore, the simulation includes a modifiable threshold at which
t h es e “ ove rt im e c ost s” k ic k in . At b as eli n e t h is t h re sh ol d i s ar bi t ra rily s et t o 2 % ; i. e. , if p at ien t demand is 2% or more above the expected value of $4,436 per capita, extra overtime costs are
applied to 30% of the demand above that 2% threshold (modified thresholds are tested in
sensitivity testing). This 2% threshold is high enough to ensure that most of patient demand is
not subjected to overtime costs, no matter the provider size, but low enough to ensure that there
is a difference between providers of varying sizes (but of the same efficiency) in their likelihood of
incurring overtime costs (since smaller providers are more likely to have outlier patient demand,
they are more likely to incur overtime costs in the simulation). Once again, these values are likely
different and variable in real life settings. What is most important for the purposes of this
simulation is that they are within reasonable bounds and are applied equally to all providers, so
that the conclusions about the magnitude of differences in financial outcomes between providers
of different sizes and under different payment systems are fair.
For simplicity, the model assumes that all health care provided is compensated. It also
assumes that provider investment in new capital or to counteract depreciation is baked into the
existing fixed costs in the model; any excess profits or remaining cash on hand at the end of a
given simulation period are saved for the next period rather than reinvested or given out as
dividends or bonuses. Presumably providers would do this so as to ward against the possibility of
future losses. This is a fairly conservative assumption and is likely somewhat unrealistic, but any
assumption of spending excess profits would only increase volatility in cash on hand and,
presumably, risk-averse provider behavior in response as well. So this is useful as a way of
establishing a baseline level of risk-aversion induced just by variation in patient needs and their
effect on profitability and unrelated to higher-order profit-oriented or investment-related
provider behavior. In each simulation period profits or losses are tallied and added to or
subtracted from a pool of cash on hand. The level of this cash on hand, as well as the magnitude
of profits and losses, are tracked and used to determine provider financial risk aversion (detailed
in the next section). Furthermore, it is assumed that providers are aware of their patient
p op u lat ion ’s t yp ic al t u rno v er r at e, t h e t yp ic al “ s t ic kiness ” of n eed s for their patients who remain
from one period to the next, as well as their own fixed cost structure. With this information
p rov ider s c an c alc u lat e t h e ex p ec t ed v alu e of t h eir p at ien t ’s u n m odif ied n ee ds and de m a n d i n t h e
next simulation period, as well as the resulting financial implications. Providers use this
information about expected financials, as well as their current level of cash, in their determination
of risk aversion.
77
3.2.3 - Module 3: Provider Risk Averse Behavior
Figure 3.5: Simplified diagram of provider risk-averse behavior (created using Insight Maker)
There are two main risk-averse provider behaviors modeled in the simulation. The first is
provider induced demand (PID), whereby providers take direct actions to either increase or
de c re as e p at ien t de m an d f or h ealt h c ar e s er v ic es c om p ar ed t o a “ n at u ra l” le v el of gen er at ed n eed s. In real life they can accomplish this in a number of ways, such as by ordering more or less
diagnostic tests during a medical episode, scheduling recurring patient visits more or less
frequently, and having an increased or decreased propensity to recommend treatment. For
instance, common criticisms that FFS encourages volume over value essentially revolve around
the idea that it can incentivize wasteful PID upward. PID that is either upward (such as that
in c en t iv ized b y F F S) or d own w ar d (s u c h as t h at in c en t iv ized b y c ap it a t ion ) f rom a “ n at u ra l”
amount of appropriate, evidence-based care for a given set of patient needs would generally be
considered a bad thing. Of course, if the standard practice for treatment of a given need was
determined to be excessive and wasteful (perhaps because a more parsimonious treatment had
been developed), then at that point PID downward to a volume of care below standard practice
would be considered a good thing (and vice versa for the case that the standard of care was
det er m in ed t o be t o o st in gy). T h e s im u lat io n m ode l c an ’t make a n y in h er e n t v alu e j u dgm en t s
about whether the PID it generates is good or bad; it can only simulate the magnitude and
dir ec t ion of t h e P I D b as e d on t h e p rov ider ’s p er c eiv ed f in an c ial r is k. Si n c e it s im u lat es a “ n at u ra l”
distribution of unmodified patient needs, whether you believe the resulting PID is good or bad
likely depends on how you feel about the appr op ri at en es s of t h e v ol u m e of t h at “ n at u ra l” st an da rd of care in the first place.
The first chapter detailed the arguments that either more or less care than is necessary can
78
be potentially harmful to patients, although it is not quite clear if one is worse than the other
f rom a m edic al s t an dp o in t . F ro m t h e p at ien t ’s p er s p ec t iv e, h owev er , it is p oss ib le t h at t h ey w ould
be more upset to find out they received less than a typical amount of care (PID downward) than
that they received more care than was necessary (PID upward). Patients are likely loss-averse too,
and may potentially fear receiving too little care more than they would fear receiving too much
care. So it is quite likely that, irrespective of the actual medical consequences (or lack thereof),
patients may perceive PID downward to be worse than an equivalent amount of PID upward.
Though this says nothing about the actual effects of either on patient health, it does speak to the
potentially greater difficulties providers may face in creating patient satisfaction if their financial
incentives encourage them to provide less care.
The second risk-averse provider behavior modeled is risk selection. Risk selection occurs
when providers manage to specifically attract patients that are either more or less sick on average
than they would sans such intervention. In real life they can do this through such means as
selective advertising, focusing on services that typically treat high cost, high margin patients
(such as orthopedic patients), avoiding services that treat high cost, low margin patients (such as
those with certain cancers), or even just refusing to accept sicker patients into their care. Though
risk selection can occur either upward or downward, it is typically only seen as a bad thing if it
occurs downward. Physicians purposefully seeking out sicker patients is considered to be one of
the better incentives of FFS, whereas their incentive to avoid sicker patients in capitation can
potentially limit patient access to care. It is for this reason that capitation payments are often risk
adjusted, whereby the payment is changed to try to reflect average population risk and negate
provider attempts to select only healthy patients (because doing so would, in theory, lower their
capitation payments).
It is assumed at baseline that providers undertake both of these risk-averse behaviors in
response to perceived financial risk. Providers perform PID in the simulation as a consistent
modification of demand for an entire simulation year; i.e., if a provider determines they must
directly modify patient demand by 3% on average in order to alleviate financial risk, then they
consistently modify all patient demand by 3% for the duration of that year (this value will change
as perceived financial risk is recalculated every year). Risk selection in the simulation is performed
only on new patients who arrive due to turnover, and it manifests as a bias in the mean value used
to generate needs for new patients. For instance, if a provider determines that they need to
perform 3% risk selection in order to alleviate financial risk in a given year, they will try to attract
new patients who are 3% sicker or healthier on average (depending on the payment system) just
for that year. Because of the randomness of patient needs, however, providers are not guaranteed
to attract the new patients they want; they essentially just increase the chance of acquiring new
patients who are financially favorable to them. In this way, PID is more direct and predictable
than risk selection, and because PID applies to all patients while risk selection only applies to new
patients, in effect PID accounts for most of modified demand in the simulation.
The level of provider financial risk aversion in the simulation is determined based on two
financial performance measures: days cash on hand at the beginning of the current simulation
period and expected negative cash flow over the duration of the current period. The literature on
79
health care provider financial performance seems to rate cash flow and cash on hand similarly in
im p ort an c e. F or i n st an c e, M oo dy ’s r at in g m et h odo lo gy f or t h e c re dit w ort h in es s of n on p rof it h osp it als as si gn s ab out equ al w eight t o t h e h osp it al’s c as h on h an d a n d c as h f lo w p er f orm an c e
(among other factors not dealt with in this simulation such as governance, market position, and
deb t man age m e n t ) (Mo o dy ’s I n v es t or Ser v ic e, 2 0 12 ). I t s eems r eas o n ab le then to expect that, if a
provider is financially risk-averse, these two factors would be large contributors to that risk
aversion. In the simulation, providers have a preferred level of days cash on hand (arbitrarily set
to 90 days) which is also the same as the amount of cash they start with. They also prefer not to
have negative net profit (cash flow losses) in any simulation period. The amount by which the
p rov ider is b elo w it s p re f er re d lev el of c as h (if at all) p lus t h e p rov ider ’s ex p ec t ed p rof it s over the
current period if t h ey do n ’t in t er v en e are added together to calculate if the provider has an
expected shortfall. Any time this expected shortfall exists, providers will undertake the financial
risk-averse behaviors mentioned previously in proportion to its magnitude (more on that in
practice below). If an expected shortfall does not exist, however, providers will not perform any
financial risk-averse behaviors in the simulation; i.e., these risk-averse behaviors are modeled as
one-sided and only in response an expected shortfall.
An expected shortfall can be represented in percentage terms as an expected operating
shortfall by dividing by the amount of expected revenue over the course of the current simulation
period if the provider were to not engage in risk-averse behavior. This represents the percentage
amount by which the operating margin would have to increase from expectation in order to
eliminate the shortfall (for instance, if the expected operating shortfall was 2% and the expected
operating margin over the current period was -1%, the actual operating margin would have to be
+1% in order to eliminate the shortfall). The amount that providers would actually need to modify
deman d i n orde r t o i n f lue n c e t h e op er at in g m ar gin in t h is w ay is de p en dent on t h e p rov ider ’s marginal profit from changes in demand. Under FFS, for instance, every $1 equivalent increase in
patient demand is associated with a $1 increase in revenue along with an increase in variable costs
equal to that $1 times the variable cost fraction. Which means that the marginal profit from
demand under FFS is just equal to (1 - variable cost fraction) times the change in demand; i.e., the
marginal profit rate is (1 - variable cost fraction). In real life this can vary significantly from one
treatment to another depending on its type and intensity. But in the simulation on average, not
c oun t in g a n y “ ove rt im e” c ost s, t h is margin al p rof it f rom de m a n d r at e u n der F F S is j u st equ al t o
the fixed cost fraction, which at baseline is 85%. So eliminating an expected operating shortfall of
2% would require a 2% / (0.85) = 2.35% increase in patient demand under FFS in the simulation.
Capitation is a bit different, since under regular capitation changes in patient demand do
not affect revenues; by design the per capita capitation payment to providers is the same every
period. Thus, every $1 equivalent change in patient demand under capitation is only accompanied
by the change in variable costs associated with it, so in the simulation on average (and again not
counting any overtime costs) the marginal profit rate from changes in demand under capitation is
equal to -(variable cost fraction), or -15% at baseline (negative because an increase in demand
would result in a decrease in marginal profit, and vice versa). So eliminating an expected
operating shortfall of 2% would require a 2% / (-0.15) = -13.33% change in demand under
80
capitation. This seems like a drastic difference from FFS, but in the simulation operating shortfalls
under capitation tend to be much lower than under FFS, so the calculated magnitudes of needed
change in demand tend to be similar between the two payment systems.
This assumes a linear relationship between expected shortfall and response; i.e., a 4%
expected operating shortfall will produce twice as much risk-averse behavior as a 2% expected
operating shortfall. It is possible that providers in reality would be even more risk averse at higher
expected shortfalls than would be predicted by a linear response, but not knowing of any
literature on the shape of utility curves for provider risk aversion in response to poor financial
indicators, nor how this would differ under FFS versus capitation, I decided to model a simple
linear response. Furthermore, while the simulation calculates expected overtime costs as part of
the provider calculation of expected revenues and profits or losses, these costs are rarely expected
in practice in the simulation (they tend to occur because of random chance outside the expected
v alu e of p at ien t de m an d) . T h er ef ore, s in c e ex p ec t ed ove rt im e c ost s ar e r ar e a n d i t ’s u n c lear h ow
providers would factor them into their risk averse behavior anyway, I do not account for them in
calculating the marginal profit rates used for determining risk aversion; I just use the simple rates
mentioned above for FFS and capitation. For ACO shared savings payment models there is a slight
modification to this that is detailed in the section of this chapter on experiments.
This conceptualization of provider financial risk as the expected amount that demand
would need to change in order to eliminate an expected shortfall shares some similarities with
prior literature but is a bit different. Prior studies simulating provider financial risk have
conceptualized it as variation in profits around an expected value (Gapenski & Langland-Orban,
1998). This is certainly the main dri v er of f in a n c ial ris k bu t it does n ’t c ap t u re t h e d if f er en c es between payment systems in the impact of equivalent shortfalls. As noted above, given same-
sized shortfalls in FFS and capitation, the former requires far less absolute change in demand to
recoup these shortfalls than the latter, meaning shortfalls in capitation are potentially more
difficult to overcome than in FFS on a dollar for dollar basis. Thus, I believe measuring financial
risk as the amount that demand would actually need to change to recoup a shortfall gives a better
apples to apples comparison between payment systems than simply measuring variation in the
shortfalls themselves.
Even though I am measuring provider financial risk in the simulation as the amount of
risk-averse provider manipulation of demand that would be needed to mitigate it, that is not to
say that this is the only or even the most likely course of action that providers would take in
response to financial risk. For instance, providers could also potentially decide to lower the
quality or effectiveness of the care they provide – as opposed to modifying the quantity of care –
as a way to save on costs and mitigate financial risk, which would likely have complex and
uncertain implications for patient health and future needs. They could also choose to simply
ignore the financial risk and deal with whatever financial consequences may come of it. The
purpose of the simulation is not to try to say which courses of action providers are more likely to
take in response to financial risk, but rather to simply quantify financial risk due to varying
patient needs and under various payment circumstances. Measuring this financial risk via one of
its potential behavioral incentives – risk-averse demand modification – simply provides a useful
81
way to quantify its magnitude that is fairly comparable across different payment scenarios. How
providers actually respond to this financial risk in reality could vary significantly from one
provider to another. Needless to say, however, this financial risk is undesirable to providers
regardless of how they actually decide to deal with it.
For simplicity it is assumed in the simulation that, in response to financial risk, providers
will undertake both PID and risk selection in equal percentage amounts in order to modify
demand sufficiently. Because these two actions interact to add up to the total desired percent
adjustment to demand, individually they should each be less than the total (depending on the
turnover rate, since it is assumed that risk selection is only performed on new patients). If the
desired absolute percent demand adjustment is D, the required percent PID and risk selection is
each U, and the turnover rate is T, then the following equation should be true (all percentages as
fractions):
(1): (T(1+U) + (1-T))(1+U) = 1 + D
In other words, increasing the expected demand of new patients by U through risk selection, and
then increasing the demand of all patients by U through PID, should result in a total increase in
expected demand of D. Solving for U via the quadratic formula yields:
(2): U = (-(T+1) + sqrt((T+1)^2 + 4TD)) / 2*T
Thus, for a given required percent demand adjustment of D and expected turnover rate T,
providers in the simulation will perform both PID and risk selection by U in the appropriate
direction depending on the payment system.
It is likely that there are limits to both the ability and willingness of providers to perform
PID and risk selection. In a recent experiment examining provider behavior under FFS and
capitation, for instance, the gap in the quantity of prescribed services between the two payment
types was about 20% for the physician participants (Brosig-Koch, Hennig-Schmidt, Kairies-
Schwarz, & Wiesen, 2015), indicating a potential physician willingness to adjust demand upward
or downward by roughly 10% (they could have adjusted more if they wanted to; the 10% average
was a self-imposed limit). There is also a large body of behavioral economics research indicating
t h at p eop le gen er ally h av e a s m all bu t c on si st en t w illin gn es s t o “c h eat ” , es p ec ially w h en mo n ey is at stake (Himmelstein et al., 2014), which would seem to support a real but limited willingness to
undesirably modify demand on the part of health care providers. Providers are almost certainly
capable of modifying demand more than 10%, but this is probably a good conservative limit. Thus,
D is capped in the simulation at 10%, though in practice it rarely gets that high even for the small
patient populations. Following the equations above, at the cap of D = 10%, with an average
turnover rate T of 35% and assuming both PID and risk selection are enabled, U would be equal to
about 7.27%.
Even though PID and risk selection are assumed to be the same in percentage terms,
because risk selection only affects new patients and PID affects all patients, PID accounts for the
82
bulk of actual modified demand. Using an average turnover rate of 35%, the proportion of
modified demand due to PID would be about 72.7%, with the other 27.3% coming from risk
selection. This makes some logical sense. Since PID is more predictable than risk selection, it
would more consistently help to alleviate financial risk, thus it makes sense for PID to account for
the bulk of the response to risk. If you wanted to model a provider who performed PID but did
not risk select, then U would be always equal to D and you would expect PID to rise, in theory, by
a factor of about 1 / 0.727 = 1.375.
3.2.4 - Module 4: Risk Mitigation Treatment
Risk assessment in health care is typically used to determine risk-adjusted outcomes or
payments that take into account healthier or sicker than average patient populations. It is also
necessary in order to determine the patients who may benefit the most from risk mitigation
treatment (RMT). Both individual patients and patient populations can be assigned a risk score
which corresponds to their relative health (risk of morbidity) compared to the average patient or
population. In this simulation, the risk score of the initial patient population is determined once
at the beginning of each run and then separately just for the population of new patients in each
simulation period. Because all patient information in the simulation is aggregated at the
population level in the form of an amount of needs, it is these needs that are used as the basis for
risk assessment. For example, after the initial patient popul at ion ’s n eed s ar e gen er at ed, t h e
expected value of needs in the next simulation period for that same population can be calculated
as:
[Expected Future Needs] = ([Initial Needs] - [Baseline Expected Needs]) * [Prior Year
Weight] + [Baseline Expected Needs]
This is essentially an expression of the central tendency of needs for this patient population in the
short term. If we then divide by the baseline expected needs, this gives the factor by which future
needs are expected to be either above or below the baseline (average) needs for that patient
population. In other words, this is an expression of the assessed risk of the population, its risk of
morbidity compared to an average population. This is often called a risk score, and is calculated in
the simulation as:
[Risk Score] = 1 + ([Initial Needs] - [Baseline Expected Needs]) * [Prior Year Weight] /
[Baseline Expected Needs]
A risk score of 1.03, for example, would indicate a patient population that was 3% sicker than
average, whereas a score of 0.98 would indicate a population 2% healthier than average. The risk
score for the new patient population in each simulation period is assessed similarly and then
in t egrat ed alo n g wit h t h e ex is t in g p at ien t p op u lat ion ’s r is k sc ore int o a w eig h t ed av er age t ota l
83
risk score.
Once assessed, this risk score does not change for the same population of patients over
the duration of a simulation run, for a few reasons. First, it is a convention of prospective risk
adjustment systems, whether for determining payment or measurement of performance, to risk
adjust based on patient demographics and health characteristics / diagnoses only, and not based
on actual realized patient demand for care. This is to try to avoid the incentive for a provider to
game the risk adjustment system by purposefully increasing the amount of care they provide,
thereby increasing risk-adjusted payments or ensuring more favorable performance metrics for
t h eir p at ien t s. Sin c e t h is s im u lat ion u se s ag gregat ed p op u lat ion s t h at don ’t in c lud e s p ec if ic patient diagnoses, population risk is assessed based only on realized needs, once initially for each
new patient population and then not again for that specific population for the rest of the run, so
that the risk scores can be considered prospective after the initial risk assessment. There is some
evidence that providers engage in other gaming behavior even under prospective risk adjustment
such as upcoding (Himmelstein & Woolhandler, 2014), but this is not considered in this
simulation.
Second, though a spec if ic p at ien t p op u lat ion ’s r is k w ould t en d t o in c re as e n at u ra ll y ove r
time due to aging and escalation of illness, any reasonable assumption of patient turnover would
render this naturally increasing risk irrelevant for the purposes of this simulation. For instance, in
an ACO with an average patient turnover rate of 35%, each patient would be expected to stay with
the ACO for an average of only 3 years. This is enough time for a wide range of actual needs to
occur, and even for individuals with certain chronic illnesses to see significant escalations of those
illn es se s , b u t it ’s n ot en o u gh t im e f or b as ic u n der l yi n g ris k f ac t ors (age, di a gn oses , h ealt h mark er s
like cholesterol and blood pressure) to change naturally for the entire population to the degree
where they significantly and consistently affect the population average of realized needs. Thus,
t h e p rov ider ’s p at ien t p o p u lat ion r is k sc ore in t h e si m u lat ion w i ll o n ly c h an ge f rom p er iod t o
period because, due to patient turnover, the actual patients in that population are changing, and
not because of naturally increasing or decreasing patient risk.
3 years is still enough time to target many patient risk factors via RMT, however. As
mentioned in the previous chapter, there are currently a number of innovative interventions in
health care that are meant to reduce patient risk of future needs in the short term, including
interventions related to chronic care management, patient safety, end of life care, and
perioperative care management. Because these interventions have implications for provider
payment, the ultimate purpose of this simulation is to assess how implementing them affects
provider finances and perceived financial risk. Thus, the last module of the simulation allows the
implementation of an RMT intervention.
When RMT is turned on, it reduces the risk of all patients generating new needs by a
specified percentage. In population terms, this involves biasing the mean value used to generate
both new and existing patient needs downward by the amount of the RMT. Because this only
af f ec t s ris k, it does n ’t gua ra n t ee t h at av er age p at ie n t n eed s w ill ac t u ally g o dow n , b u t it makes it much more likely that they will, especially for larger patient populations. At baseline, the RMT is
set to a value of 5%; i.e., patient population risk is reduced by 5% when RMT is turned on. This
84
value was chosen because, based on the studies of RMT-like interventions mentioned in the
previous chapter, it seems like a realistic possible value, and it has also been used in other
simulations looking at the effects of interventions to produce savings in ACOs (DeLia, 2016).
Figure 3.6: Simplified diagram of needs generation with RMT (created using Insight Maker)
It is expected that RMT would involve some ongoing cost to the provider, and this cost is
expressed on average in the simulation as a percentage of baseline expected patient population
exp en se s. T h is RM T c ost f u rt h er f ol lo w s t h e p rov ider ’s c ost s t ru c t u re , i.e., much of it is fixed based
on t h e p rov ider ’s f ixed c ost f ra c t ion . T h e v ar iab le p ort ion of t h e RM T c ost in t h e s im u lat ion , m eanwh ile , is de p en d en t on t h e p at ien t p op u lat ion ’s c u rr en t r is k sc ore; i. e. , i t is as su m ed t h at a
patient population with a risk score of 1.03 would have RMT variable costs that are 3% higher than
those of the average patient population. RMT costs are also subject to an extra overtime variable
cost if the patient population risk score is above the overtime threshold, though this happens
infrequently at the baseline threshold of 2%. So for instance, using baseline values: if the RMT
were exp ec t ed t o c ost on av er age 5% of a p at ien t p op u lat ion ’s av er age t ota l exp en se s, 5 % * 0 . 8 5 = 4.25% would consist of a fixed cost every period, while the remaining 0.75% would be variable and
dependent o n t h e p op u lat ion ’s r is k sc ore (w h ic h w ould c h an ge ev er y p er iod b ec au se of t u rno v er ). I f t h e p op u lat ion ’s r is k sc ore w er e ev er ab ove 1. 0 2 ( m ore t h an 2 % ab ove av er age), t h er e w ould b e
added overtime costs equal to 30% of the baseline cost of treating the excess risk.
The revenues to pay for these RMT costs might be expected to come out of savings due to
lower patient need; this is the philosophy behind using an ACO shared savings payment system or
capitation to fund such interventions. The simulation also allows for RMT to be paid for directly
with a specific RMT capitation payment. So, for instance, if RMT was set to cost 5% of baseline
expected patient expenses on average, and if RMT payment was turned on, the provider would
receive an RMT capitation payment equal to this average expected cost every period. This would
85
be useful if, say, an RMT intervention were being implemented by a provider paid regular FFS for
patient morbidity needs, but it could also theoretically be used with other payment systems.
Finally, it is assumed that providers implementing RMT will reduce their fixed costs per
capita devoted to the treatment of patient morbidity needs by the same amount that the RMT
intervention is expected to reduce morbidity risk. For example, with an RMT intervention
designed to reduce risk by 5%, a provider would reduce its regular per capita fixed costs by 5% to
reflect the lower level of expected morbidity needs (they would presumably shift some or all of
these fixed cost resources to the RMT intervention, depending on how much it itself costs). This
would also reduce the total demand threshold needed to trigger overtime costs for morbidity
treatment by 5% as well, as this threshold is fundamentally based on fixed cost capacity. Reducing
per capita fixed costs in this way is necessary for sustainable RMT when the RMT is being
implemented by the same provider who also treats it s p at ien t s’ r egular mo rb idit y n eed s. Bec au se t h es e n eed s ar e ex p ec t ed t o dec re as e, if t h e p rov id er does n ’t r edu c e f ixed c ost s dev ote d t o t h em
accordingly it will have too much capacity and unsustainably high fixed costs as a result.
3.3 - Simulation Inputs, Outputs, and Experiments
To recap, the baseline input parameters used in the simulation, regardless of payment
system, are the following:
Input Baseline Value(s) Notes
Patient population size 5000, 10000, 25000, and
50000 (all batched separately)
Held constant over the course
of each simulation run
Prior year weight for existing
patient needs (page 72)
0.45 Constant weight, actual
patient needs (expenditures)
vary
Patient turnover rate (page 71) 0.35 Varies binomially
Fraction fixed costs (page 75) 0.85 Constant
“ Ov er t im e” variable cost
fraction (page 76)
0.30 Constant for any patient
demand above the threshold
“ Ov er t im e” i nitialization
threshold (page 76)
2% above baseline average Constant
Max demand modification
(page 81)
0.10 Constant, per year max
Table 3.4: Baseline simulation input parameters
86
Other details are as follows. Each simulation run is 10 years long with recalculation of simulation
outputs every year. This time frame was chosen because it is long enough to produce steady state
values and yet short enough that it is a believable amount of time over which a provider could be
in operation. Furthermore, each simulation batch consists of 5,000 runs, and the median, quartile,
and decile values of outputs of interest from these runs are gathered. 5,000 runs was chosen as a
trade-off between simulation processing time (less runs are better) and maintaining consistency
in quantile values from batch to batch for the same input parameters (more runs are better).
Figure 3.7: Simplified diagram of full simulation model (created using Insight Maker)
The outputs collected over the course of each simulation run are the following:
● Average per capita patient morbidity needs
● Average per capita payer costs
● Average risk selection per year
● Average per capita provider induced demand (PID) per year
● Other financial metrics, including: number of years with a loss, average losses (per capita),
worst loss (per capita), number of years below desired cash level, average days cash on
hand, and lowest days cash on hand
87
For each output, the median, quartile (25th and 75th percentile), and decile (10th and 90th
percentile) values are obtained from a 5,000 run batch to determine its spread. In addition, for
each output both the short term (first 3 years) and the long term (whole 10 year run) results are
analyzed. Figure 3.7 summarizes the full simulation model, including the previously mentioned
inputs, outputs, and main process elements.
Because each batch consists of 5,000 runs, output quantiles are relatively consistent from
batch to batch. For instance, Table 3.5 shows the mean and standard deviation of the quantile
results for average PID per year per capita, the primary output of interest in the model, from 15
repetitions of the baseline FFS scenario in the model and for only the smallest providers (5,000
patients) over the short term (first 3 years).
Percentile 10th 25th Median 75th 90th
Mean $0.00 $3.97 $40.14 $108.16 $168.88
Std Dev $0.00 $0.51 $2.18 $2.04 $2.47
CV NaN 12.73% 5.43% 1.88% 1.46%
Table 3.5: 15 batch repetitions, average PID per year per capita, baseline FFS, 5000 patients, first 3 years
And Table 3.6 shows the same results for the same size providers and time span but under
capitation payment:
Percentile 10th 25th Median 75th 90th
Mean $0.00 -$3.71 -$60.44 -$151.90 -$240.32
Std Dev $0.00 $0.32 $2.77 $4.67 $5.10
CV NaN 8.58% 4.59% 3.08% 2.12%
Table 3.6: 15 batch repetitions, average PID per year per capita, baseline capitation, 5000 patients, first 3 years
For even the most volatile scenario for FFS and capitation providers (smallest amount of patients,
shortest time span) the PID quantile values do not vary significantly from batch to batch with the
same input parameters. For larger providers and over the long term (10 year average) these
quantiles vary even less (see the appendix section A.1 for full tables). According to the results in
the tables above, when changing input parameters or payment systems in the model we can be
assured that they result in a significant change in the median PID if it is a change of at least 20%
(as y ou’ll se e in t h e n ext c h ap t er , c h an ges in p ay m e n t s ys t ems in the model typically result in
much larger changes in PID than that).
The different simulation experiments are laid out according to the simulation objectives
listed at the beginning of this chapter, and will be addressed in order.
88
3.3.1 - Simulation Objective 1
Quantify how much financial risk, measured as provider manipulation of demand, is
produced by natural variation in patient medical needs among health care providers under
various payment systems at baseline (without risk mitigation treatment).
For this simulation objective, different payment systems are tested at baseline without any
risk mitigation treatment (RMT). This includes the following payment systems:
● Regular fee-for-service (FFS)
● Regular capitation
● FFS with 1-sided ACO shared savings
● FFS with 2-sided ACO shared savings and losses
● Capitation with risk adjustment
FFS and capitation are straightforward as described earlier; FFS providers receive a payment equal
t o t h eir p at ien t s’ r ealiz ed t ota l deman d f or c ar e (i . e . , af t er any r is k se l ection and PID), and
c ap it at ion p rov ider s re c ei v e a c on st an t p ay m en t p e r c ap it a e qua l t o t h eir p a t ien t s’ b as elin e
expected costs based on the MEPS data (under regular capitation this payment is the same for
every patient regardless of risk).
For the FFS with 1 or 2-sided ACO shared savings payment systems, most revenues are still
based on realized patient demand, similar to regular FFS. In addition, however, if patient demand
falls below a certain threshold, providers will still lose revenue from the lowered demand but will
gain a p ort io n of it b ac k in t h e f orm of a sh ar ed sa v in gs p ay m e n t (t h e “ sa v i n gs” ar e s av ed by payers, who then share part of those savings with the provider). Base parameters for both the 1
and 2-sided shared savings payment systems are based on the rules for ACOs in the Medicare
Shared Savings Program (MSSP) (CMS, 2015c). For the 1-sided model, in which there are only
savings, the savings rate is 50%, i.e., as long as providers have cleared a savings threshold, for
every $1 they reduce demand below a benchmark they will receive $0.50 as a savings bonus. Once
they have cleared the threshold for savings providers will receive a shared savings payment for all
sa v in gs b elo w t h e b en c h m ar k, n ot j u st f or s av in gs b elo w t h e t h re sh ol d (“ f irs t doll ar ” s av in gs). T h e
thresholds for savings in the 1-sided model depend on provider population size, and are the
following for the population sizes in the simulation:
● 5,000 patients: 3.9% savings
● 10,000 patients: 3.0% savings
● 25,000 patients: 2.4% savings
● 50,000 patients: 2.2% savings
The reason these thresholds are different depending on population size in the 1-sided model is to
try to minimize the chance of the payer paying out savings inappropriately. Since smaller
89
providers experience more natural varia t ion in t h ei r p at ien t s’ n eed s, t h ey h av e t o ac h iev e h igher thresholds of savings in order to prove that those savings occurred because of genuine provider
efforts and not simply due to random chance.
Of course, while higher thresholds decrease the chance that payers will pay out savings
when none have genuinely occurred (a type I error in statistics), they also increase the chance
that providers will fail to receive savings when they have indeed occurred (a type II error). There
is a tradeoff between these two errors, and some researchers have shown that this tradeoff can
quite often result in smaller providers not receiving deserved savings (DeLia et al, 2012).
Effectively, a type I error represents a financial risk to the payer, while a type II error represents a
financial risk to the provider. Because these threshold values were created for the MSSP based on
Medicare patient data, which tends to have less variation in patient needs than the MEPS data
used in this simulation, the risk of a type I error will be somewhat higher and the risk of a type II
error will be somewhat lower than if higher thresholds that better match the MEPS data were
used (i.e., provider financial risk will be lower). This is fine for the purposes of this simulation, as I
am not addressing payer financial risk and I would like to avoid overstating provider financial risk
due to the chance of not receiving deserved savings.
For the 2-sided ACO model the provider shares in both payer savings below the
benchmark as well as payer losses above the benchmark. Here I also use the parameters from the
MSSP, where the savings rate is 60% and the loss rate is 40%. In addition, under this model every
provider has the same thresholds for savings and losses regardless of size. Any savings of more
than 2% below the benchmark will result in 60% sharing of all savings below the benchmark, and
any patient demand more than 2% above the benchmark will result in 40% sharing in all payer
losses above the benchmark. There are no differences between provider sizes in the benchmarks
here most probably because providers who do not have any actual savings have theoretically the
same chance of receiving underserved savings as they do undeserved losses, so payers are less
worried about the risk of a type I error over time.
In both the 1 and 2-sided ACO models, the benchmark for savings or losses is simply based
on the baseline average per capita patient expenditures from the MEPS data. This is different than
how benchmarks have been determined for most of the MSSP, in which they were different for
each provider based on their historical average patient expenditures and risk adjusted for
providers with healthier or sicker than average patient populations. However, it is more similar to
the direction the program is currently trending, in which benchmarks are starting to be based on
regional averages among many providers (CMS, 2016b). The goal of this change is avoid
penalizing providers who are already efficient (because their historical patient expenditures are
already low) as well as to avoid rewarding providers who ramp up patient expenses just prior to
joining the MSSP in an effort to game the benchmark (Rose et al., 2016). At baseline in the
simulation this benchmark is not risk adjusted (i.e., it is the same per capita regardless of
population risk).
Also in both ACO shared savings models, it is assumed that if expected savings in the next
period would eliminate an expected shortfall, then the provider will not engage in any risk-averse
behavior. However, if expected savings are not enough to eliminate the expected shortfall, then
90
the provider will perform risk selection and PID under the normal FFS protocol without regard for
possible savings. This is due to the structure of shared savings, especially the savings threshold. A
provider faced with a shortfall under FFS with shared savings (at the savings rates mentioned
above) still only has one option if it wants to modify patient demand to try to eliminate the
shortfall, and that is to increase demand. Despite the fact that the provider shares in payer savings
when demand is lower, the provider will still lose money in the short term from decreased
demand u n less t h e s av in gs r at e ex c eed s t h e p rov i der ’s f ra c t ion of f ixed c ost s (Cher n ew , M c G u ire , & McWilliams, 2014, go through the math in greater detail). Furthermore, even if the sharing rate
exceeds the fixed cost fraction, it would still be easier for the provider to eliminate a shortfall by
pushing demand up rather than by reducing it, unless this was substantially outweighed by the
potential for offsetting shared losses (such as under the 2-sided model). Using the baseline shared
savings values listed above, including a fixed cost fraction of 85%, providers will only ever be able
to eliminate expected shortfalls by increasing patient demand, even under a 2-sided model with
the possibility of shared losses. Furthermore, because providers in the simulation are only trying
to eliminate shortfalls and not seek excess profits, in practice they rarely expect to encounter
shared losses even after PID and risk selection (though some inevitably do due to random
chance). However, because there is a threshold for savings, any provider engaging in PID and risk
selection should expect that those behaviors may likely push demand above the threshold (and
thus disqualify them from savings). This is why, if a provider does not expect shared savings to
eliminate an expected shortfall and chooses to engage in risk-averse behaviors, they do so with
t h e ex p ec t a t ion t h at t h ey w on ’t r ec eiv e s av in gs at all (i. e . , t h ey do s o ac c ording t o t h e r egular F F S
rules noted earlier).
Finally, for providers receiving capitation payments with risk adjustment, their capitation
p ay m en t s w ill be a dj u st ed ba se d on t h eir p at ien t p op u lat ion ’s c u rr en t r is k sc ore a s des c rib ed
earlier. It is also assumed when risk adjustment is turned on that providers no longer choose to
perform risk selection. This is probably not entirely accurate; there is some evidence from the
Medicare Advantage program that, in response to risk adjustment, providers still tend to risk
select but only on parameters not measured in the risk adjustment mechanism (Brown et al.,
2011). But in general, risk adjustment in capitation is intended to eliminate the incentive to risk
select healthier patients, since the capitation payments for healthier patients will be lower to
reflect their lower risk. Therefore, it is assumed in the simulation that risk adjustment works as
intended for this purpose, and risk selection is turned off. Providers can still engage in PID,
however, so depending on how much the underlying financial risk changes with the shift to risk
adjusted payments PID will likely still go up (theoretically by a factor of 1.375 compared to regular
capitation if financial risk remains unchanged, as noted earlier).
3.3.2 - Simulation Objective 2
Quantify how much financial risk is produced when adding a risk mitigation treatment
intervention that is paid for indirectly (from shared savings or capitation).
91
For this simulation objective, all of the above-mentioned payment systems (with the
exception of FFS) will be tested with the addition of a risk mitigation treatment (RMT)
intervention that, at baseline, reduces patient risk of future needs by a constant 5% every
simulation period. Though the risk is reduced by 5%, the actual needs themselves are still random
as described earlier. As part of sensitivity testing I will also test scenarios where the RMT
intervention produces an average reduction in risk of 5% but with some variation year to year in
this value.
The goal is to test each payment system alternative to FFS using the most expensive type
of RMT it could theoretically pay for (this is why regular FFS is not tested as, on its own, it offers
no way to pay for RMT). This produces steady state financial values that are, I believe, interesting
and reflective of a more developed provider and payer market where payers are unlikely to pay
significantly more for an intervention than it actually costs. For instance, under regular and risk
adjusted capitation, an RMT intervention that produces a 5% average reduction in risk could be
paid for even if it costs on average 5% of baseline expected patient expenses, as long as 5% of the
fixed cost resources that would have been devoted to regular patient morbidity needs are devoted
to the RMT intervention instead. This is the case because, under capitation, providers keep the
entire capitat ion p ay m en t w h en t h ey r edu c e t h eir p at ien t s’ r is k, so as lo n g a s an RM T in t er v en t ion d oes n ’t c ost mo re on average than the equivalent amount of patient needs that it
mitigates would have cost to treat, it will be financially sustainable. Since under ACO shared
savings providers only keep a portion of the savings they generate, then such payment systems
can only sustainably pay for RMT when the intervention costs are in line with the savings rate. For
instance, in the 1-sided ACO shared savings system tested which has a shared savings rate of 50%,
this payment system could sustainably pay for RMT that costs up to $0.50 for every $1 in patient
demand it mitigates. Similarly, the 2-sided ACO shared savings system which has a 60% savings
rate could pay for RMT that costs $0.60 for every $1 in patient demand it mitigates. Each of these
payment systems is tested with its respective most costly but still sustainable RMT intervention to
look at the effect of a mature, standardized RMT intervention on provider financial risk under
various payment scenarios.
The simulation looks at the early implementation of such an RMT intervention after
calculating a prior year of normal patient needs that occurred before the intervention (see Figure
3.6). It is assumed that, other t h an t h e n at u ra l “ st i c kiness ” of ex is t in g p at ien t n eed s ye ar t o yea r,
there are no other delays in the effect of the RMT on risk. This assumption of fast acting RMT is
meant to conservatively model a best case scenario; to determine the minimum likely amount of
financial risk that the introduction of an RMT intervention might produce. Any assumption of a
great er de lay in t h e ef f ec t of RM T w ould lik ely i n c re as e t h e p rov ider ’s p er c ei v ed f in an c ial r is k
since the provider would have to contend for a longer period of time with higher patient demand
than it has capacity to adequately handle (with the possible exception of FFS with 1-sided shared
savings, as there is no penalty for high demand other than overtime).
92
3.3.3 - Simulation Objective 3
Quantify how much financial risk is produced when the risk mitigation treatment
intervention is paid for directly (with a specific RMT capitation payment).
This final simulation objective deals with the question: what if, instead of paying for an
RMT intervention indirectly out of provider or payer savings from the reduced risk, the
intervention is simply paid for up front with its own capitation payment? This question is most
pertinent for comparisons between regular FFS (with direct paid RMT) and FFS with ACO shared
savings or capitation (RMT paid for indirectly out of savings). It is certainly possible to pay
directly for RMT with its own capitation payment at a fully capitated provider (and reduce the
capitation payment for regular patient morbidity needs by the same amount), but in terms of
accounting the numbers will be identical to those from the previous simulation objective, though
with some small differences for the case of a risk adjusted capitation provider.
For instance, it would be interesting to compare the 1-sided ACO shared savings model
from the previous simulation objective, which with a 50% savings rate can pay indirectly for an
RMT intervention that costs up to $0.50 for every $1 in demand it mitigates on average (i.e., costs
2.5% of baseline expected patient expenses for a 5% reduction in risk), with a regular FFS model in
which the RMT is simply paid up front as a capitation payment (a regular payment worth 2.5% of
baseline expected patient expenses for an RMT intervention expected to reduce risk by 5% on
average). A similar comparison can be made between the 2-sided ACO shared savings model, with
a 60% savings rate, and a regular FFS model with a paid RMT intervention that costs 3% of
baseline expected patient expenses.
Ultimately, this simulation is trying to determine the differences in the amount of
financial risk (as measured by resulting risk-averse provider behavior), produced only as a result
of natural variation in patient needs, between providers implementing an RMT intervention
under various payment system scenarios. Given that the financial risk that accompanies new
payment system innovations is a significant concern among most health care payment experts,
especially for smaller health care providers, and given the rising importance of trying to find a
way to pay sustainably for RMT interventions, the question of how much financial risk is
generated when implementing these interventions is important to try to answer if we wish to
design payment systems that will fund them sustainably.
3.4 - Expert Panel
Because of the nature of the simulation model, in that it cannot be fully validated against
real world data, an expert panel was gathered to assess the scientific integrity and merit of its
design. In addition to the members of the dissertation committee, the following USC professors
volunteered to participate in the panel:
93
● Robert Myrtle, Professor of Public Administration, USC Price School of Public Policy
● Michael Nichol, Professor of Health Policy, USC Price School of Public Policy
● Gregory Stevens, Associate Professor of Clinical Family Medicine, USC Keck School of
Medicine
After meeting individually with the panel members, I gave a presentation to the full panel on
March 14, 2017. I presented the design of the model as outlined in this chapter, of which they were
also provided a draft copy to read prior to the meeting. I had two primary questions for which I
wanted feedback from the panel, though it was an open discussion so other topics were also
discussed. My two main questions were:
1. Are the inputs used in the simulation set to reasonable baseline values?
2. Is the assumption of zero net profit for providers at the average expected level of patient
demand reasonable? (The explanation for why this was set this way is on page 75).
For the first question, discussion centered almost entirely on my choice of the baseline for
the fraction of fixed costs (85%), as my choices for the baseline values of all other inputs seemed
reasonable to the panel. We talked at length about what should be counted as fixed versus
variable costs in my simulation model, as depending on the context many costs that can be
considered fixed in one sense can also be considered variable in another. For instance, over the
long term accountants typically view most costs as variable, as labor, medical equipment, and
even buildings can be increased or decreased to reflect long term changes in demand. Given that
my simulation model focuses more on short term provider reactions to changes in finances, I
assumed that over the horizon in which providers are deciding whether or not (and by how
much) to manipulate patient demand, which is just a single year, that they would rather change
patient demand in order to improve their short term finances than hire or fire staff, buy or sell
medical equipment, or build or demolish hospital wings. On the other hand, there are potential
advantages to be gained by providers who create a profit structure that is built more around
variable costs, as high fixed costs can be a liability when demand is highly variable. Many
providers have in fact pursued these kinds of structures, relying more on seasonal and flextime
staff, for example. The panel agreed that in a short term operational sense, a baseline fixed cost
fraction of 85% was reasonable, but they encouraged me to also explore alternative scenarios in
which providers chose to invest in a lower fixed cost fraction.
For the second question, the panel remarked on how the average provider does not
typically have zero net profit at the average expected level of patient demand. However, because I
make this assumption before providers are allowed to manipulate patient demand, and since
providers in real life also potentially manipulate patient demand to improve their finances, they
agreed that real life provider financial performance is not a great basis for setting this benchmark
for this particular simulation. Since my ultimate goal with this simulation is to compare the
relative financial risk and performance of providers under various payment systems due to natural
94
variation in patient needs, and not to make claims about the exact financial risk generated by any
single payment system in the absolute, they thought it was reasonable for me to set zero net profit
at the average expected level of demand.
We also discussed future applications and extensions of my simulation model. For
instance, right now the model is an open system, looking at a single provider, in which new
patients flow in from the ether due to patient turnover. It would be interesting as an extension of
the model to look at a larger closed loop system, such as the entire California insurance market,
where new patients generally come not from the ether but from other providers and/or payers. In
such a system, for instance, providers could try to risk select their patients, and some might be
more successful than others, but net risk selection for the system as a whole would be zero or
negative since all patients must end up somewhere (or simply have less access). Such a system
would allow the exploration of risk mitigation treatment (RMT) as a tragedy of the commons
problem with potentially longer timeframes for the RMT to take effect. Given turnover in that
system, how important would it be to pay for RMT directly as opposed to paying for it through
capitation or shared savings, given that one provider might invest in reducing the risk of new
needs for its patients while another provider could reap the shared savings that result?
95
Chapter 4 - Simulation Model Results
Recall that the primary research question I am attempting to answer using this simulation
model is the following:
Does a direct capitation payment to implement provider interventions that reduce
potentially preventable medical needs and patient harm create less provider financial
risk in comparison to other payment systems (including full capitation and ACO shared
savings), and if so how much less?
The rest of this chapter lays out the results from the simulation model that help to answer this
question. I will address the results from each of the three simulation objectives in line with this
research question, discussed in the previous chapter, in order. I will also look at other results,
such as the cost to payers of various payment systems, that are pertinent but not directly related
t o t h is r es ear c h qu es t ion . F in a ll y, I w ill p er f or m a s en si t iv it y an alys is t o l oo k at h ow t h e mo del’s results for provider financial risk change in response to changes in the baseline model parameters,
an d I ’ll us e t h e r es u lt s of t h at s en si t iv it y an alys is t o in f or m s o m e s p ec ial mo del s c en ar ios t h at may
also be pertinent to the sustainable funding of risk mitigation treatment interventions.
4.1 - Simulation Results: Simulation Objective 1
Quantify how much financial risk, measured as provider manipulation of demand, is
produced by natural variation in patient medical needs among health care providers under
various payment systems at baseline (without risk mitigation treatment).
4.1.1 - Baseline Fee-for-service vs Capitation: Without Risk Averse Provider Behavior
I t ’s u se f u l t o f irs t c om p ar e t h e r aw f in an c ial p er f orm an c e of h ealt h c ar e p rov ider s in t h e
simulation under FFS and capitation without allowing them to manipulate patient demand as a
response to expected financial shortfalls. This provides a baseline understanding of the workings
of provider finances in the simulation that will make it easier to discern later what is going on
financially when providers decide to engage in patient demand manipulation due to perceived
financial risk.
In summary, with provider risk averse behavior turned off, FFS and capitated providers
experience the same number of years with a cash flow loss, but both the average and worst losses
are greater in magnitude for FFS providers in comparison to capitated ones. In addition, FFS
providers experience greater variation in both their average and lowest days cash on hand over
96
the course of a simulation run than capitated providers. See Figure 4.1 below for an example of
how average days cash on hand vary between FFS and capitated providers for the full 10 year
period (providers start with 90 days cash on hand). For all graphs in this chapter, the black square
represents the median provider, while the blue, red, green, and purple points represent the 10th,
25th, 75th, and 90th percentile provider, respectively. See the appendix section A.2 for the full set
of tables comparing the financial performance of FFS and capitated providers with risk averse
behavior turned off.
Figure 4.1: Baseline FFS (left) vs. capitation (right), no risk aversion, average days cash on hand, full 10 years
Based on these performance measures, capitated providers seem initially to have less
financial risk than FFS providers. However, the underlying profit structures of FFS and capitated
providers are quite different, and as noted in the previous chapter this suggests that they ought to
look at financial risk somewhat differently. Specifically, the marginal profit rate from changes in
patient demand is much higher for FFS providers compared to capitated ones, based on the
si m u lat ion ’s b as eli n e v alu es f or t h e f r action of fixed versus variable costs. Thus, while FFS
providers are subjected to higher variation in raw financial performance in the simulation than
capitated providers, their ability to potentially mitigate the effect of that variation is also much
higher. It is for this reason that I have decided to use the amount of provider manipulation of
demand needed to overcome financial shortfalls as a proxy for financial risk in the simulation, as I
feel it offers a better apples to apples comparison of providers under different payment systems
who have different resultant profit structures.
4.1.2 - Baseline Fee-for-service vs Capitation: Adding Risk Averse Provider Behavior
Allowing providers to manipulate demand, both in response to lower than desired cash on
97
hand as well as expected losses, results in a certain amount of both patient risk selection
(selecting healthier or sicker new patients) as well as direct provider induced demand (PID), as
described in the previous chapter. This demand manipulation in r es p o n se t o p rov ider s’ p er c eiv ed
financial risk can be measured directly in the simulation. For instance, allowing risk-averse
behavior for FFS and capitated providers produces the following distribution of average patient
risk selection per year for the first 3 years:
Figure 4.2: Baseline FFS (left) vs capitation (right), average patient risk selection per year, first 3 years
Here I am showing the absolute values of patient risk selection. FFS providers have an incentive to
seek out sicker (higher risk) patients, so their risk selection values are normally positive. Whereas
capitated providers have an incentive to seek out healthier (lower risk) patients, so their risk
selection values are normally negative. However, as a representation of financial ris k I ’m o n ly
interested in the absolute value of risk selection because it is higher absolute values of risk
selection (regardless of direction) that represent higher provider financial risk.
Because providers perform equal percentage amounts of patient risk selection and
provider induced demand (PID) every time they engage in risk averse demand manipulation in
the simulation, the values for the average amount of PID per year per capita performed by FFS
and capitated providers would be the same in percentage terms. In actual dollar terms per capita
(i.e., the approximate average dollar value of demand per year per capita that is directly induced
upward or downward) they are the following (again for the first 3 years):
98
Figure 4.3: Baseline FFS (left) vs. capitation (right), average PID per year per capita, first 3 years
Once again, FFS providers have an incentive to directly increase demand, so their PID values are
positive, whereas capitation providers have an incentive to directly reduce demand, so their raw
PID values are technically negative. But again , I ’ m p rimar ily int er es t ed in t h e a bs ol u t e v alu e of PID, since it is this absolute value that represents the amount of financial risk felt by the provider.
So the graphs above, as well as throughout this chapter, show the absolute value of PID.
F u rt h er m ore, I ’v e c h ose n t o dis p lay P I D in t er m s of t h e d ol lar v alu e of in du c ed deman d t h at it represents, as I feel that this is a salient way of conceptualizing financial risk and its potential
effects. However, for a point of reference, recall that the average expected patient demand in the
simulation is $4,436 per capita. So approximately every $45 in average per capita PID is equivalent
to a 1% change in average demand.
These measures of provider manipulation of demand (patient risk selection and PID) are
the proxy measures I am using for measuring provider financial risk in the simulation, though as
outlined in the previous chapter they are by no means the only or even necessarily the most likely
risk-averse behavior that providers may choose to engage in. Since the outcomes for patient risk
selection and PID are basically identical in percentage terms, I will limit reporting to just PID to
avoid redundancy. In this way, PID is the main proxy measure I will use for comparing provider
financial risk under various payment systems in the chapter. Just remember that, in addition to
the reported values of PID, there is also an equivalent percentage amount of patient risk selection
happening as well unless specifically noted otherwise in cases where I have turned risk selection
off. However also remember that, as noted in the previous chapter, this PID accounts for the
majority of total provider demand manipulation, about 73%. The appendix beginning with section
A.4 contains tables with the raw PID values for various payment systems in both the short term
(first 3 years) and long term (full 10 years).
Since I am primarily interested in relative differences in provider financial risk across
providers under different payment systems, as opposed to simply measuring financial risk in the
99
ab solut e, I ’ll t yp ic ally ma ke u se of grap h s of t h e d if f er en c e in P I D qua n t iles be t w een p rov ider s
under different payment systems. Such graphs for the difference in the distribution of average
PID per year per capita between FFS and capitated providers, for example, look like the following
(showing the change in short term PID on the left and long term PID on the right):
Figure 4.4: Change in absolute PID from FFS to capitation, first 3 years (left) and full 10 years (right)
These graphs allow us to very clearly see the difference between FFS and capitation when it comes
to financial risk. Note that the positive values indicate that capitation is associated with an
increase in absolute PID (i.e., an increase in financial risk) in comparison to FFS across nearly all
provider quantiles, with generally higher increases among smaller providers and in the short
term.
The benefits of risk-averse behavior on provider financial performance show up most
notably in the results for average days cash on hand. For FFS providers, they show the following
average days cash on hand with risk-averse behavior (with the change from no risk aversion
shown in parentheses):
# Patients 10th 25th Median 75th 90th
5000 87 (+52) 93 (+33) 103 (+19) 119 (+11) 134 (+6)
10000 89 (+38) 93 (+26) 101 (+14) 112 (+7) 125 (+5)
25000 90 (+25) 93 (+16) 98 (+9) 106 (+5) 113 (+2)
50000 90 (+17) 92 (+12) 97 (+7) 102 (+4) 108 (+2)
Table 4.1: Baseline FFS (with risk aversion), average days cash on hand, full 10 years
Whereas the results for capitated providers are somewhat muted in comparison:
100
# Patients 10th 25th Median 75th 90th
5000 88 (+23) 90 (+15) 92 (+8) 95 (+4) 99 (+2)
10000 89 (+14) 90 (+9) 92 (+5) 94 (+2) 97 (+1)
25000 90 (+7) 90 (+5) 92 (+2) 93 (+1) 95 (+1)
50000 90 (+4) 90 (+2) 91 (+1) 92 (+1) 93 (+0)
Table 4.2: Baseline capitation (with risk aversion), average days cash on hand, full 10 years
Remember that providers in the simulation prefer to have at least 90 days cash on hand. Above
this level they will not be risk averse from their level of cash on hand, only from anticipated cash
flow losses. Below this level they will seek to return to having at least 90 days cash on hand. Now
note that, despite the fact that capitated providers at baseline perform more risk-averse
manipulation of demand than FFS ones, they end up with lower average days cash on hand. This
goes to show why capitated providers end up being more risk-averse; they have to be in order to
consistently reach their desired level of cash, whereas FFS providers reach this level far more
easily with less effort. This pattern is consistent in all of the other financial performance measures
as well; see the appendix section A.3 for the rest.
As a result, despite the fact that capitated providers perform more risk-averse behavior
than FFS ones, they apparently need to do so in order to achieve anywhere near the same level of
financial parity (i.e., their behavior does not seem excessive). Thus, it seems like the total amount
of provider manipulation of demand due to risk aversion is working as a good proxy measure for
provider financial risk that is comparable across different payment systems.
4.1.3 - Baseline Fee-for-service with Shared Savings
Recall that FFS with shared savings in the simulation is modeled using a 50% shared
savings rate and with the following thresholds below average patient demand at which savings
kick in for the various provider sizes modeled:
● 5,000 patients: 3.9%
● 10,000 patients: 3.0%
● 25,000 patients: 2.4%
● 50,000 patients: 2.2%
Adding this shared savings element to the baseline risk averse FFS provider results in the
following change in the distribution of provider induced demand in the simulation, again with
the change in short term PID on the left and long term PID on the right (the y-axis is set to the
same scale as Figure 4.4 for easier comparison):
101
Figure 4.5: Change in PID when adding shared savings to baseline FFS, first 3 years (left) and full 10 years
(right)
This means that, as expected, adding a shared savings program reduces financial risk for FFS
providers at baseline. This reduction in financial risk is fairly small, however, and less substantial
than the difference in financial risk between FFS and capitated providers.
4.1.4 - Baseline Fee-for-service with Shared Savings and Losses
Recall that fee-for-service providers responsible for both shared savings and losses are
modeled using a 60% shared savings rate and a 40% shared loss rate in the simulation.
Furthermore, the threshold for both savings and losses is 2%, i.e., savings kick in when demand
reaches 2% below average (but are counted from the first dollar of savings), and losses kick in
when demand reaches 2% above average (but again are counted on a first dollar basis). Adding
shared savings and losses to the baseline FFS providers in the simulation results in the following
change in the distribution of provider induced demand (again the y-axis is set to the same scale as
the previous two figures for easier comparison):
102
Figure 4.6: Change in PID when adding shared savings and losses to baseline FFS, first 3 years (left) and full 10
years (right)
Thus, adding shared savings and losses seems to decrease financial risk for many FFS providers
but it also significantly increases it for others, though again not by as much as the increase from
FFS to capitation and mostly only in the short term.
4.1.5 - Baseline Capitation with Risk Adjustment
Recall that, under capitation with risk adjustment, the capitated per capita payment to
providers no longer remains constant, but rather it changes based on the risk score of the patient
population (sicker patient population = higher risk score). I also assume, since the primary
purpose of risk adjustment is to remove the incentive in capitation for providers to risk select for
healthier patients, that risk adjustment in the simulation does so successfully. Therefore, I turn
off patient risk selection in the simulation when risk adjustment for capitation payments is active.
Because patient risk selection is turned off in risk adjusted capitation, providers will
increase PID to compensate. I therefore ran a scenario of baseline capitation but with risk
selection turned off to see how much PID increased in magnitude compared to normal capitation
where providers are able to risk select. Based on the math from the previous chapter we should
expect this magnitude increase in PID to be about 1.375, but due to emergent properties of the
simulation it was actually an average of 1.457 for the median provider over the full 10 year run.
Thus, I will use 1.457 as an approximate scaling factor, reducing the PID values for risk adjusted
capitation providers by it, so that they can be more appropriately compared to the PID values for
providers under other payment systems who are able to risk select (see the appendix, section
A.4.5, for more information on calculating this scaling factor).
After scaling the PID values for providers under risk adjusted capitation (dividing them by
103
1.457), risk adjustment of capitation payments results in the following increase in absolute PID
compared to regular capitation:
Figure 4.7: Change in scaled PID when adding risk adjustment to baseline capitation, first 3 years (left) and
full 10 years (right)
Note that even after scaling the PID values to reflect the lack of patient risk selection there
is still a fairly significant increase in financial risk when adding risk adjustment to capitation, even
more than the increase in financial risk going from baseline FFS to regular capitation. These
re su lt s do m ak e s om e s en se t h ough. Risk ad j u st m en t does n ’t c h an ge t h e p rov ider ’s f u n da m en t al
p rof it s t r u c t u re . I t ’s s t ill ess en t ially t h e s ame as u nder regular capitation, as any manipulation of
demand on their part still only affects their variable costs, not their revenues, in the short term.
But now, with more variation in revenues it can lead to more variation in financial performance,
similar to F F S , b u t w it h out t h e b en ef it s of F F S’s p rof it s t ru c t u re . T h u s, ris k ad j u st m en t , t h ough it reduces the incentive for capitated providers to risk select for healthier patients, increases the
amount of financial risk actually felt by these providers.
There are some important caveats to this finding, however. In the simulation I assume
that providers do not adjust their fixed costs in response to the anticipated short term changes in
the needs of their patients, since their forecasts may be wrong and patient needs tends to be
volatile year to year. This rigid fixed cost structure is a great potential source of financial risk for
providers under risk adjusted capitation. However, providers could elect to make their cost
structure less fixed and more variable as a w ay of m it igat in g t h is r is k, t h ough it ’s a st ra t egy t h at c ould p oss ib ly b ac kf ire . I ’ll dis c u ss t h is mo re in a s p ec ial s c en ar io at t h e end of t h is c h ap t er as w ell
as in the discussion in the next chapter.
I n t h e n ext s ec t ion , I ’l l g o t h rou gh h ow f in an c ial risk changes when providers implement a
risk mitigation treatment (RMT) intervention that is paid for indirectly out of the payer and/or
p rov ider s av in gs t h at r es u lt f rom t h at in t er v en t ion . A n d i n t h e s ec t ion af t er I ’ll add re ss h ow
104
financial risk changes when RMT is paid for directly up front instead.
4.2 - Simulation Results: Simulation Objective 2
Quantify how much financial risk is produced when adding a risk mitigation treatment
intervention that is paid for indirectly (from shared savings or capitation).
Recall that the risk mitigation treatment (RMT) intervention tested is an intervention that
reduces patient risk of future needs by 5%. In the following payment systems, the RMT
intervention is tested under a variety of assumed costs for the intervention, each corresponding to
the most expensive intervention that the payment system could theoretically support. For this
reason the values for financial risk derived for the payment systems in this section are not strictly
comparable to one another, and instead will be used to compare to a payment system from the
next section where the RMT is paid for directly (as per the third simulation objective).
However, I can say that, in general, if one payment system is theoretically capable of
supporting a more expensive RMT intervention than another for the same reduction in risk, that
is an advantage for RMT in that system. Assuming equal values for provider operational
efficiency, a payment system that can support more expensive RMT interventions means it can
support interventions with a greater variety of returns on investment (ROIs). For example, a
payment system that can support RMT interventions that, on average, produce up to a 1 to 1 ROI
($1 reduction in risk for each $1 spent on the intervention) can also support interventions that
produce higher ROIs (say, a $2 reduction in risk for each $1 spent). A payment system that can
only support RMT interventions that produce up to a 2 to 1 ROI (i.e., lesser expensive
interventions) can still support interventions with ROIs higher than that, but not lower. Such a
payment system that can only support cheaper RMT interventions could still be great for taking
care of low hanging fruit in patient risk of needs and harm, but it would be less helpful for more
expensive to treat types of risk, as well as for the early stages in RMT development when the
interventions are still being innovated on and will likely cost more. The next chapter will discuss
this dynamic further.
The following payment systems tested in the simulation are capable of supporting an RMT
intervention indirectly through some kind of payer and/or provider savings:
● FFS with shared savings
● FFS with shared savings and losses
● Capitation
● Capitation with risk adjustment
I ’ll g o t h rou gh ea c h of t h em t o see h ow adding RMT c h an ges p rov ider s’ p er c ep t ion of f in an c ial
risk. For the exact amount of PID generated by providers engaging in RMT under each payment
system, see the appendix section A.5.
105
4.2.1 - FFS with Shared Savings: Adding Risk Mitigation Treatment
With the tested shared savings rate of 50%, this means that FFS with shared savings could
theoretically support any RMT intervention that produces at least a 2 to 1 ROI. For example, if an
RMT intervention reduced patient demand by $50, under FFS with shared savings and a 50%
savings rate the provider would get $25 in savings from the payer. Thus, as long as the RMT
intervention cost $25 or less it could be sustainably supported by this shared savings payment. In
keeping with this, the RMT intervention in this scenario is modeled to cost on average 2.5% of
average expected patient demand for an average reduction of 5% in patient risk of future needs.
Recall that this also assumes that the provider undertaking this RMT intervention is willing to
reduce its fixed cost resources per capita devoted to regular patient needs by the same amount as
t h e inte rv en t ion ’s ex p ec t ed re du c t ion in p at ien t r i sk (5% ) .
It is assumed that providers can still manipulate patient demand, but that they do so for
t h eir p at ien t p op u lat ion ’s r egular n eed s a n d n ot f or RM T t re at m en t , s o t h e P I D v alu es in t h es e
RMT scenarios represent PID for regular care. After adding an RMT intervention as described
above (2.5% cost), providers under FFS with shared savings show the following change in the
distribution of average provider induced demand per year per capita (change in short term
average PID on the left and long term average PID on the right):
Figure 4.8: Change in PID when adding RMT (2.5% cost) to FFS with shared savings, first 3 years (left) and full
10 years (right)
The increase in financial risk when adding RMT for a provider receiving FFS with shared savings
is pretty significant and is in fact generally greater even than the increase in financial risk going
from baseline FFS to capitation.
106
4.2.2 - FFS with Shared Savings and Losses: Adding Risk Mitigation Treatment
Under the version of shared savings and losses modeled, the savings rate is 60%. This
means that, in theory, it can support any RMT intervention that costs no more than $0.60 for
every $1 in demand it mitigates. Thus, the RMT intervention in this scenario is modeled to cost 3%
of average expected patient demand for a 5% reduction in risk. Because the cost is slightly
different, t h is RM T s c en ar io isn’t dir ec t ly c om p ar ab le t o t h e p re v ious on e, b u t it c an s t ill be
compared to the baseline scenario of FFS with shared savings and losses prior to adding RMT.
After adding the RMT intervention (3% cost), providers under FFS with shared savings
and losses show the following change in the distribution of average PID per year per capita (the y-
axis is set to the same scale as the previous figure for easier comparison):
Figure 4.9: Change in PID when adding RMT (3% cost) to FFS with shared savings and losses, first 3 years
(left) and full 10 years (right)
The increase in financial risk is once again greater than that going from baseline FFS to capitation.
I t ’s als o g re at er t h an t h e incr eas e w h en ad din g t h e RM T (2 . 5% c ost ) i n t er v en t ion t o FFS with only
shared savings in the previous scenario. Needless to say, the increase in financial risk involved in
both scenarios when adding RMT is substantial and potentially problematic.
4.2.3 - Capitation: Adding Risk Mitigation Treatment
Under capitation, because reductions in patient demand do not also reduce revenues, any
RMT intervention can be sustainable as long as it produces at least a 1 to 1 ROI. In this regard
capitation is theoretically the best payment system for paying for RMT interventions indirectly as
it can support the widest variety of intervention costs. Thus, the RMT intervention tested with
107
capitation is one that costs 5% of expected patient demand, on average, for again a 5% reduction
in patient risk. And again, this still requires that providers reduce by 5% their fixed cost resources
per capita devoted to regular patient needs, although presumably these same resources could just
be wholly invested in the RMT intervention instead.
After adding an RMT intervention (5% cost), the change in the distribution of average PID
per year per capita for capitated providers looks like this (y-axis is set to the same scale as the
previous two figures):
Figure 4.10: Change in absolute PID when adding RMT (5% cost) to capitation, first 3 years (left) and full 10
years (right)
Compared to baseline capitation without RMT, the above results show somewhat higher financial
risk, though mostly only in the short term. However, the change in financial risk is not as drastic
as that seen when adding RMT to either the FFS with shared savings or FFS with shared savings
and losses scenarios. And this is despite the fact that the RMT tested in this scenario costs more
than the RMT tested in those two previous scenarios.
4.2.4 - Capitation with Risk Adjustment: Adding Risk Mitigation Treatment
U n der r is k ad j u st ed c ap i t at ion , de sp it e t h e f ac t t h a t t h e p rov ider ’s p a ymen t is n o l on ger
constant, providers still keep any savings they produce by reducing patient risk. Thus, risk
adjusted capitation, like regular capitation, can theoretically support any RMT intervention that
produces at least a 1 to 1 ROI, so it is also tested with an RMT intervention that costs 5% of
expected patient demand for an average reduction in risk of 5%. Furthermore, I assume that the
re du c t ion in r is k du e t o t h e RM T p rogram does n o t als o red u c e t h e p rov ider ’s c ap it at ion p ay m en t (i. e. , t h e c ap it at ion p ay m en t is ad j u st ed ba se d on w h at t h e p a t ien t p op u lat i on ’s r is k w ould h av e
108
been without RMT). This is a necessary assumption in order to allow the RMT intervention to pay
for itself out of provider savings; otherwise the intervention would be self-defeating.
After adding an RMT intervention (5% cost), the change in the distribution of average PID
per year per capita (after scaling downward by a factor of 1.457 to account for the lack of patient
risk selection) for providers under risk adjusted capitation looks like this:
Figure 4.11: Change in PID (scaled) when adding RMT (5% cost) to risk adjusted capitation, first 3 years (left)
and full 10 years (right)
Note that, similar to the change when adding RMT to regular capitation, there is an increase in
PID when adding RMT to risk adjusted capitation but it is again not very large and mostly in the
short term.
4.3 - Simulation Results: Simulation Objective 3
Quantify how much financial risk is produced when the risk mitigation treatment
intervention is paid for directly (with a specific RMT capitation payment).
The purpose of this section is to look at alternative payment scenarios to those in the
previous section in which the same RMT interventions are paid for directly with their own
capitation payment (instead of indirectly out of payer/provider savings), while all other patient
needs are paid for with regular FFS payments. This RMT capitation payment is always set to the
average cost of the RMT intervention; i.e., a provider implementing an RMT intervention that
costs on average 2.5% of expected patient demand will receive a capitation payment every year
exactly equal to this average expected cost. In all of these scenarios this RMT capitation payment
is not risk adjusted; i.e., it stays the same throughout the duration of each simulation run.
109
The following payment systems are examined:
● FFS plus paid RMT (2.5% cost)
● FFS plus paid RMT (3% cost)
● FFS plus paid RMT (5% cost)
The next few sections look at how financial risk changes when providers under baseline FFS
implement these payment systems to pay for an RMT intervention. For the raw PID values
produced by these payment systems, see the appendix section A.6.
4.3.1 - FFS plus paid RMT (2.5% cost)
Under FFS with paid RMT (2.5% cost), providers are paid regular FFS payments for most
of t h eir p at ien t ’s n eed s bu t in ad dit ion t h ey ar e p aid a c ap it at ion p ay m en t equ al t o 2 . 5 % of their
p at ien t p op u lat ion ’s av er age ex p ec t ed deman d m e an t t o f u n d a n RM T i n t er v en t ion w it h an
equivalent average cost. This RMT intervention will still result in a 5% average reduction in
patient risk as per the previous RMT scenarios. Providers can still manipulate demand, but as in
p re v ious RM T s c en ar ios t h ey man ip u lat e d ema n d f or t h eir p at ien t p op u lat ion ’s r egular n eed s, n ot
for RMT. Finally, as in the previous RMT scenarios, providers implementing RMT still reduce
their fixed cost resources per capita devoted to regular patient needs by the same amount that the
RMT reduces patient risk, 5%.
Providers under this payment system show the following change in the distribution of
average PID per year per capita compared to regular FFS without RMT:
Figure 4.12: Change in PID when adding paid RMT (2.5% cost) to FFS
110
I t is n ’t a h u ge red u c t ion in P I D a n d m ost of it is in t h e s h ort t er m , b u t giv en t h at P I D f or r egular F F S p rov ider s is n ’t ex t re m e t o begi n w it h it it ’s s t i ll a si g n if ic an t dr op in P I D.
4.3.2 - FFS plus paid RMT (3% cost)
FFS with paid RMT (3% cost) is identical to the paid RMT scenario in the previous section,
with the only difference being that the RMT intervention costs 3% of average expected patient
demand, and thus the RMT capitation payment is increased to match this higher cost. This level
of cost is chosen to provide an appropriate comparison to FFS with shared savings and losses plus
RMT, in which the RMT intervention also cost 3% of average expected patient demand.
Under this payment scenario, providers show the following change in the distribution of
average PID per year per capita compared to baseline FFS:
Figure 4.13: Change in PID when adding paid RMT (3% cost) to FFS
T h ough t h ey ar en ’t s t ric t l y c om p ar ab le, t h es e differential values for PID are not dissimilar from
the values for FFS plus paid RMT (2.5% cost) from the previous section.
4.3.3 - FFS plus paid RMT (5% cost)
In this scenario the RMT intervention is assumed to cost the same as the amount of
patient demand it mitigates on average, 5%. Thus, this scenario provides the fairest comparison to
paying for RMT through either kind of capitation, both of which can also support such RMT
interventions with a 1 to 1 ROI sustainably.
111
The change in the distribution of average PID per year per capita for providers under this
payment system in comparison to baseline FFS looks like the following:
Figure 4.14: Change in PID when adding paid RMT (5% cost) to FFS
These results again look similar to the paid RMT scenarios in the previous sections.
4.4 - Summary and Contextualization of Results
Imagine that you are a health care provider currently being paid regular FFS and you wish
to implement an RMT intervention. There are a variety of payment systems you could either
choose or, more likely, have imposed upon you by a payer, which could theoretically pay
sustainably for RMT. The following sections attempt to contextualize this decision by looking at
the differential provider financial risk involved in the choice between payment systems, as well as
the potential impact on payer costs (since payers would likely have a significant say in the choice
as well).
The main focus is on the difference between payment systems that pay indirectly for RMT
(simulation objective two) and directly for RMT (simulation objective three) for the same-cost
RMT intervention. Of particular interest is how financial risk changes for small providers (5,000
patients), since these smaller providers make up a significant proportion of the providers in the
U.S., as well as the difference in felt financial risk between small and large providers, as a large
difference might drive more provider mergers and their potential unintended consequences.
112
4.4.1 - Paying for RMT (2.5% Cost) with Shared Savings or with Direct Payment
Assuming there exists an RMT intervention that can produce at least a 2 to 1 ROI, e.g., it
costs no more than 2.5% of average expected patient demand in order to reduce the risk of new
patient medical needs by 5%, a provider could be paid for it indirectly by adding a shared savings
c om p o n en t t o t h e p rov ider ’s b as elin e F F S t h at h as a sh ar ed sa v in gs r at e of 50 % . A p rov ider c ould
also just be paid the cost of the RMT directly up front with its own capitation payment. The
graphs below show the resulting change in provider financial risk (in the form of PID) when
implementing this RMT intervention under these two payment systems (shared savings left,
direct pay right) over the first 3 years, starting from baseline FFS:
Figure 4.15: ΔPID from baseline FFS when implementing RMT (2.5% cost) under FFS w/ SS (left) or paid RMT
(right), first 3 years
We can also look at the difference in PID between these two payment systems directly, going
from direct pay RMT to indirect pay through shared savings:
113
Figure 4.16: ΔPID from FFS + paid RMT (2.5% cost) to FFS w/ SS + RMT, first 3 years
I t ’s c lear f rom t h es e gra p h s t h at p a yi n g f or t h is k in d of RM T t h rou gh s h ar ed sa v in gs r es u lt s in a
moderate increase in provider financial risk in comparison to baseline FFS, whereas paying for the
RMT through its own direct payment actually decreases financial risk. And altogether this adds
up to significantly more provider financial risk in the former payment system compared to the
latter.
When it comes t o f in a n c ial r is k, we ’r e es p ec ially w orri ed ab out t h e ef f ec t it w ill ha v e on small providers. The following table shows the same information from the figures above but in
terms of the magnitude of change in financial risk, instead of absolute change, experienced by the
median small provider (5,000 patients) as they make the switch to each payment system:
Payment System
Magnitude
ΔP ID
Shared Savings 1.81
Paid RMT 0.50
Difference
(SS / Paid RMT)
3.62
Table 4.3: Magnitude ΔPID from baseline FFS when implementing RMT (2.5% cost) under shared savings or
paid RMT for median 5,000 patient provider, first 3 years
The median small provider, starting from a baseline of regular FFS, experiences 1.81 times more
financial risk when implementing RMT (2.5% cost) and paying for it through shared savings.
Wher eas t h e s ame p rov ider ’s f in an c ial r is k dr op s by h alf w h en imp lem e nting the same RMT
intervention but having it paid for directly. And altogether this means that the median small
114
provider will experience 3.62 times more financial risk when trying to pay for RMT through
shared savings compared to direct pay.
I t ’s als o u seful to look at difference in the reduction in financial risk going from smaller to
larger providers between these two payment systems. Specifically, we can look at the ratio
between: A) the decrease in financial risk experienced by providers under shared savings as they
increase in size, and B) the same decrease experienced by providers under direct pay for RMT.
This ratio can be calculated by the following equation, where we are trying to calculate the
differential benefit (in the form of reduced financial risk) for providers who increase in size under
Payment A vs Payment B:
Differential Benefit Ratio = ([Payment A Smaller Provider PID] - [Payment A Larger
Provider PID]) / ([Payment B Smaller Provider PID] - [Payment B Larger Provider PID])
The following table shows this ratio for the median provider, where FFS with shared savings is
Payment A in the equation above and direct pay RMT is Payment B, for the possible provider size
increases in the simulation:
Patient Population
Size Increase
Differential
Benefit Ratio
5,000 to 10,000 2.31
10,000 to 25,000 2.26
25,000 to 50,000 1.23
Table 4.4: Ratios of reduction in PID from various provider size increases, [FFS w/ SS] / [paid RMT], for
median provider when implementing RMT (2.5% cost), first 3 years
This means that, for example, for the median provider trying to implement RMT under FFS with
shared savings, if they can increase their patient population size from 5,000 to 10,000 this reduces
their financial risk by 2.31 times as much as an equivalent provider trying to do the same under
direct pay RMT. Or in other words, the median small provider under FFS with shared savings has
2.31 times the incentive to increase their size compared to under direct pay RMT as a way of
mitigating financial risk. This is true to various degrees for all of the possible provider size
increases.
F in a ll y, it ’s als o us ef u l t o c om p ar e t h e d if f er en c e in t h e c ost t o p ay er s of t h es e t w o
payment systems since they would also likely play a major role in choosing between them. If this
RMT intervention were implemented it should decrease average payer costs per capita by about
2 . 5% i n eit h er s c en ar io , b ec au se n ew p at ien t medi c al n eed s w ill be r edu c ed by 5% a n d t h e p ay er ’s cost to reduce them, either in the form of a shared savings payment or a direct RMT payment, is
only 2.5% of average expected patient demand (hence -5% + 2.5% = -2.5% payer costs). Thus, we
can compare the actual payer costs in both scenarios to an ideal baseline FFS in which per capita
115
payer costs have been reduced by 2.5%. During the first 3 years, the left graph below shows payer
costs under FFS with shared savings relative to this ideal FFS baseline, and the right graph shows
payer costs under FFS plus direct pay for RMT relative to the same baseline:
Figure 4.17: Δ payer costs from ideal FFS when implementing RMT (2.5% cost) under FFS w/ SS (left) or paid
RMT (right), first 3 years
Bein g ge n er ous , t h er e d o es n ’t s eem t o be a c lear a dv an t age f or p ay er s be t w een t h es e p ay m en t options in the graphs above.
Thus, in summary, compared to paying for the RMT with its own direct capitation
payment, paying for it out of shared savings is associated with higher provider financial risk in
general, 3.62 times higher financial risk for the median small provider, a significantly increased
incentive to grow in size for the median provider (including a 2.31 times greater incentive for the
median small provider), and essentially no change in payer costs.
4.4.2 - Paying for RMT (3% Cost) with Shared Savings and Losses or with Direct Payment
Assuming there exists an RMT intervention that can produce at least a 5 to 3 ROI, e.g., it
costs 3% of average expected patient demand to produce a 5% reduction in patient risk of new
medical needs, it could be paid for either through a shared savings payment system with a savings
rate of 60% (even when this also includes the possibility of shared losses) or just by paying the 3%
cost directly up front. The graphs below show the difference in provider financial risk (in the form
of PID), again using baseline FFS (without RMT) as a reference, between these two payment
systems in the first 3 years.
116
Figure 4.18: ΔPID from baseline FFS when implementing RMT (3% cost) under FFS w/ SS/SL (left) or paid RMT
(right), first 3 years
And again we can also just look at the difference in PID between these two payment systems
directly:
Figure 4.19: ΔPID from FFS + paid RMT (3% cost) to FFS w/ SS/SL + RMT, first 3 years
Because the cost of the RMT intervention is different than in the previous section these graphs are
not strictly comparable. Yet the same pattern emerges; adopting RMT under shared savings and
losses is associated with a moderate increase in provider financial risk, while just paying for the
RMT in its own direct capitation payment is associated with a decrease in financial risk. And there
is a significant direct difference in absolute PID between the two payment systems.
117
Again we should worry about the effects of financial risk on small providers. The following
table shows the magnitude of change in financial risk (instead of absolute change) for the median
small provider (5,000 patients) going from baseline FFS to either of these two payment systems,
as well as the magnitude difference between them:
Payment System
Magnitude
ΔP ID
Shared Savings and
Losses
1.75
Paid RMT 0.48
Difference
(SS/SL / Paid RMT)
3.64
Table 4.5: Magnitude ΔPID from baseline FFS when implementing RMT (3% cost) under shared savings and
losses or paid RMT for median 5,000 patient provider, first 3 years
The median small provider seems to be faced with similar magnitudes of changes in financial risk
as in the previous section. Going from baseline FFS to paying for RMT (3% cost) through shared
savings and losses results in 1.75 times more financial risk (compared to 1.81 times more in the
previous section), while paying for it directly results in 0.48 times as much financial risk
(compared to 0.50 times as much in the previous section). The magnitude of difference in
financial risk between these payment systems directly is 3.64, very similar to the difference in
magnitude of 3.62 seen in the previous section.
I t ’s again also u se f u l t o l o ok at t h e magn it u de dif f er en c e in f in a n c ial r is k g oi n g f ro m smaller to larger providers under these two payment systems. The following table shows the
differential benefit ratio for increases in provider size between paying for RMT through shared
savings and losses (numerator) and direct payment (denominator) for the median provider:
Patient Population
Size Increase
Differential
Benefit Ratio
5,000 to 10,000 1.97
10,000 to 25,000 1.53
25,000 to 50,000 1.17
Table 4.6: Ratios of reduction in PID from various provider size increases, [FFS w/ SS/SL] / [paid RMT], for
median provider when implementing RMT (3% cost), first 3 years
Here the increased benefit from increases in provider size under shared savings and losses, in
comparison to direct pay for RMT, is not as significant as in the previous section. For instance, the
median small provider has 1.97 times the incentive to increase in size under the former payment
118
system compared to the latter, whereas a similar provider in the previous section faced 2.31 times
the incentive. Nevertheless the pattern is still the same.
Again we can compare the cost to payers between these systems. Given that patient risk of
new needs is reduced by 5% but the cost of making that reduction is approximately 3% of average
expected patient demand, payer costs per capita should ideally drop by 2% (-5% + 3% = -2%).
Thus, we can compare payer costs in the first 3 years under these two scenarios to an ideal FFS
reference in which payer costs have been reduced by 2%.
Figure 4.20: Δ payer costs from ideal FFS when implementing RMT (3% cost) under FFS w/ SS/SL (left) or paid
RMT (right), first 3 years
Once again, even though these graphs of per capita payer costs are not strictly comparable to the
ones in the previous section (because the RMT interventions cost different amounts), the basic
p at t er n is t h e s ame. T h er e d oesn ’t s eem t o be a c o n si st en t benefit in terms of payer costs between
these two payment systems.
Thus again, in summary, paying for RMT indirectly with shared savings and losses, in
comparison to paying for it directly with its own capitation payment, seems to produce more
provider financial risk generally, 3.64 times as much financial risk for the median small provider, a
higher incentive to increase in provider size, and no consistent advantage in payer costs.
4.4.3 - Paying for RMT (5% Cost) with Full Capitation or with Direct Payment
As long as there exists an RMT intervention that can produce at least a 1 to 1 ROI, e.g., it
costs 5% of average expected patient demand and produces a 5% reduction in the risk of new
patient medical needs on average, it can be paid for either with full capitation or with a direct
RMT-only capitation payment that covers the 5% cost. The graphs below show the difference in
119
provider financial risk (in the form of PID), again using baseline FFS as a reference point, between
these two options.
Figure 4.21: ΔPID from baseline FFS when implementing RMT (5% cost) under capitation (left) or paid RMT
(right), first 3 years
We can again also look at the difference in PID between these two payment systems directly:
Figure 4.22: ΔPID from FFS + paid RMT (5% cost) to capitation + RMT, first 3 years
Once again, these graphs are not strictly comparable to the graphs in the previous two sections
since the cost of the RMT intervention is different in all three. Nevertheless, they once again show
a similar pattern; paying for RMT through full capitation is associated with an increase in
120
financial risk from baseline FFS, while paying for RMT through its own direct capitation payment
(while continuing to pay for other care needs through FFS) is associated with a decrease in
financial risk. And again there is a significant difference in financial risk when you directly
compare the two payment systems.
The following table again examines the magnitude difference in PID (as opposed to the
absolute difference in the graphs above) for the median small (5,000 patient) provider:
Payment System
Magnitude
ΔP ID
Capitation 2.13
Paid RMT 0.59
Difference
(Cap / Paid RMT) 3.64
Table 4.7: Magnitude ΔPID from baseline FFS when implementing RMT (5% cost) under capitation or paid
RMT for median 5,000 patient provider, first 3 years
Here the results look similar to the previous two sections. The median small provider sees a 2.13
times increase in financial risk from baseline FFS when switching to capitation to pay for RMT
(5% c ost ) , w h ic h is h igher t h an t h at in t h e p re v iou s se c t ion . T h e s ame p rov i der does n ’t s ee qu it e
as much of a decrease in risk when paying for the RMT up front, however (0.59 factor decrease).
So the magnitude difference in financial risk between these two payment systems ends up being
about the same; paying for RMT through capitation is associated with 3.64 times more financial
risk for the median small provider than paying for it directly up front.
Once again we can also look at the difference between these two payment systems in their
incentive for providers to try to increase in size. The following table shows the ratio of differential
benefit from provider size increases for the median provider between capitation (numerator) and
direct pay RMT (denominator) over the first 3 years:
Patient Population
Size Increase
Differential
Benefit Ratio
5,000 to 10,000 2.48
10,000 to 25,000 3.93
25,000 to 50,000 2.07
Table 4.8: Ratios of reduction in PID from various provider size increases, [capitation] / [paid RMT], for
median provider when implementing RMT (5% cost), first 3 years
Here the pattern is significantly different than in the previous two sections. The median provider
has a significantly greater incentive to grow larger when paying for RMT through capitation
121
compared to direct pay (greater even than in the comparisons in the previous two sections), and
this greater incentive persists for all possible provider size increases. Thus, it seems like capitation
especially would encourage provider mergers and growth in response to financial risk when
implementing RMT.
We can again look at the difference is payer costs between these two payment systems,
and here is where we find the first true tradeoff in our results. Since the RMT intervention costs
5% of average expected patient demand for a 5% reduction in patient risk of new needs it is
budget neutral on average. Thus, we can use the payer costs in baseline FFS as a reference without
modification.
Figure 4.23: Δ payer costs from baseline FFS when implementing RMT (5% cost) under capitation (left) or paid
RMT (right), first 3 years
There seems to be a clear benefit to paying for RMT through full capitation instead of an RMT-
only capitation payment in terms of payer costs. Even though adding a paid RMT component to
ba se lin e F F S d oesn ’t s ig n i f ic an t ly c h an ge p ay er c ost s, it s t ill ca n ’t mat c h t h e ad v an t age of f u ll capitation in this regard.
Thus, paying for RMT with full capitation vs an RMT-only capitation payment plus FFS for
the rest presents a tradeoff between the effects on providers and payers. It causes more provider
financial risk generally, including 3.64 times as much financial risk for the median small provider,
and is associated with a significantly higher incentive for providers to grow larger going all the
way up to the highest provider size tested (50,000 patients). But paying for RMT through full
capitation is also significantly less expensive for payers. This tradeoff will be discussed in more
detail in the next chapter.
122
4.4.4 - Paying for RMT (5% Cost) with Risk Adjusted Capitation or with Direct Payment
The graphs below compare the PID produced when paying for an RMT intervention
identical to that in the previous section using risk adjusted capitation (with the values scaled
down by 1.457 to account for the lack of patient risk selection) versus paying for the RMT with its
own direct capitation payment during the first 3 years:
Figure 4.24: ΔPID from baseline FFS when implementing RMT (5% cost) under risk adjusted capitation
(scaled) (left) or paid RMT (right), first 3 years
And again we can also look at the direct difference in PID between these two payment systems:
Figure 4.25: ΔPID from FFS + paid RMT (5% cost) to risk adjusted capitation + RMT (scaled), first 3 years
123
Similar to the previous section, paying for RMT using risk adjusted capitation is associated with
an increase in provider financial risk from baseline FFS, only in this case the increase is
significantly larger than that for regular capitation (they are directly comparable in this case
because the cost of the RMT intervention is the same), while paying for the RMT directly is again
associated with a decrease in financial risk.
The following table again examines the magnitude difference in PID (as opposed to the
absolute difference) for the median small provider:
Payment System
Magnitude
ΔP ID
Risk Adjusted
Capitation
3.14
Paid RMT 0.59
Difference
(RA Cap / Paid RMT)
5.37
Table 4.9: Magnitude ΔPID from baseline FFS when implementing RMT (5% cost) under risk adjusted
capitation (scaled) or paid RMT for median 5,000 patient provider, first 3 years
Here the results are more extreme than in the previous section. The median small provider sees a
3.14 times increase in financial risk from baseline FFS when switching to risk adjusted capitation
to pay for RMT (5% cost), which is higher than the 2.13 times increase seen in the previous
section. And the magnitude difference in financial risk between paying for RMT through risk
adjusted capitation vs direct pay is much higher, a 5.37 times increase.
And once again we can look at the difference between these two payment systems in their
incentive for providers to try to increase in size. The following table shows the ratio of differential
benefit from provider size increases for the median provider between risk adjusted capitation
(scaled) (numerator) and direct pay RMT (denominator) over the first 3 years:
Patient Population
Size Increase
Differential
Benefit Ratio
5,000 to 10,000 2.33
10,000 to 25,000 5.24
25,000 to 50,000 5.88
Table 4.10: Ratios of reduction in PID from various provider size increases, [risk adjusted capitation (scaled)] /
[paid RMT], for median provider when implementing RMT (5% cost), first 3 years
The incentive for small providers to increase in size is similar to that in the previous section, but
124
the incentive for larger providers to increase in size is much greater. This version of risk adjusted
capitation seems like it would greatly encourage provider mergers and growth in response to
financial risk in comparison to direct pay for RMT.
Finally, looking again at the difference in payer costs between the two payment systems:
Figure 4.26: Δ payer costs from baseline FFS when implementing RMT (5% cost) under risk adjusted capitation
(left) or paid RMT (right), first 3 years
Once again, paying for RMT with risk adjusted capitation is associated with a significant decrease
in payer costs, whereas paying for RMT with a direct payment does not change payer costs
significantly. One small difference in this case is that, while the median decrease in payer costs for
risk adjusted capitation is the same as for regular capitation, the spread of the decrease is smaller
across all provider sizes.
Thus, paying for RMT with risk adjusted capitation also presents a similar tradeoff as in
the previous section. In comparison to paying for RMT directly, it involves significantly higher
provider financial risk generally, including 5.37 times the financial risk for the median small
provider, and comes with a very high incentive for providers to increase in size, even more so than
regular capitation. But it also significantly decreases payer costs in comparison. Furthermore, risk
adjusted capitation also has the advantage of discouraging providers from risk selecting for
h ealt h ier p at ien t s. As w e’ll see in a sp ec ial s c en ar i o at t h e end of t h is c h ap t er , h owev er , t h is tradeoff can change if providers are able to change some of their fixed costs into variable ones.
These possibilities will also be discussed further in the next chapter.
4.5 - Sensitivity Analysis
As a reminder, these were the baseline inputs used in the simulation:
125
Input Baseline Value(s) Notes
Patient population size 5000, 10000, 25000, and
50000 (all batched
separately)
Held constant over the
course of each simulation run
Prior year weight for existing
patient needs
0.45 Constant weight, actual
patient needs (expenditures)
vary
Patient turnover rate 0.35 Varies binomially
Fraction fixed costs 0.85 Constant
“ Ov er t im e” v ar iab le c ost fraction
0.30 Constant for any patient
demand above the threshold
“ Ov er t im e” in it ializa t ion threshold
2% above baseline average Constant
Max demand modification 0.10 Constant, per year max
Table 4.11: Baseline simulation input parameters
It is important to determine how much these baseline simulation input parameters play a role in
determining the PID values in the simulation. For a sensitivity analysis, these inputs (except for
patient population size) were varied up and down one by one by 20% and then 40% to determine
the sensitivity of PID to changes in them for both baseline FFS and capitation providers. Then, the
slope of a least squares regression line going through the PID outputs from the -40%, -20%,
baseline, +20%, and +40% changes to each of these input parameters was calculated. The
following graphs show how the slope of that line changes for the median provider depending on
the input being changed and on the patient population size. Positive values in the graphs indicate
a positive correlation between changes in the input parameter and provider financial risk, while
negative values indicate a negative correlation. Tables containing these slope values for both the
short term (3 years) and long term (10 years) can be found in the appendix, section A.7.
For the median provider under baseline FFS, the sensitivity analysis looks like the
following:
126
Figure 4.27: Sensitivity analysis, % change in average PID per year per capita for every 20% change in input
parameters, median FFS provider, full 10 years
The bars on the graph above show the percent change in average PID per year per capita for each
20% change in the listed input parameters. The financial risk of the median FFS provider does not
seem to be particularly sensitive to changes in most of the input parameters. There is a small
as soc iat ion b et w een i n c re as es in t h e “ ove rt im e” v ar iab le c ost f ra c tion and increases in financial
ris k. In c re as es in b oth t h e p at ien t t u rno v er r at e a n d t h e “ ove rt im e” in it ializa t ion t h re sh ol d a re associated with a small decrease in financial risk. And increases in the fraction of fixed costs seem
to have the largest impact; they are associated with a small to moderate decrease in financial risk.
For the median capitated provider, the sensitivity analysis looks quite different:
127
Figure 4.28: Sensitivity analysis, % change in average PID per year per capita for every 20% change in input
parameters, median capitation provider, full 10 years
The financial risk of the median capitated provider seems to be much more sensitive to changes
in the input parameters than the median FFS provider. Here, increases in the fraction of fixed
costs are associated with a fairly large increase in financial risk (as opposed to a moderate
dec re as e). T h e f in an c ial r is k of c ap it at ed p rov ider s is als o m u c h mo re s e n si t iv e t o t h e “ ove rt im e” v ar iab le c ost f ra c t ion and t h e “ ove rt im e” in it ializa t ion threshold compared to FFS providers,
signally an increased sensitivity to capacity limits. And finally, increases in the prior year weight
for existing patient needs seem to be significantly associated with a small reduction in financial
risk for capitate d p rov ide rs , w h er eas t h er e d oesn ’t s eem t o be mu c h of a li n k f or F F S p rov ider s.
The difference between FFS and capitated providers in the sensitivity of financial risk to
c h an ges in f ixed c ost s an d t h e “ ove rt im e” c ap ac i t y -related inputs is probably the most notable
finding in the sensitivity analysis. The last special scenario tested in the next section looks at how
both kinds of providers are affected in a situation where these two parameters are able to be
lowered.
4.6 - Special Scenarios
I want to look at the results for provider financial risk under a few special scenarios, including:
● Removing shared savings and losses thresholds: Do these thresholds themselves cause
some provider financial risk?
128
● Random effect RMT: Does provider financial risk change if the effect of RMT is random
with an average risk reduction of 5%, compared to the constant 5% used previously?
● Lowe r f ixed and “ ove rt im e” c ost s: H ow d oes p rov id er f in an c ial r is k c h an ge f o r p rov ider s
under various payment systems when a greater proportion of care costs are made variable
(an d t h u s, a lower p rop or t ion ar e f ixed and “ ove rt im e” ) u p f ron t ?
The raw PID values for each of these special scenarios can be found in the appendix, section A.8.
4.7.1 - FFS with Shared Savings and Losses Plus RMT, No Savings or Losses Thresholds
Given that much of the increase in provider financial risk (in the form of PID) when
adding an RMT intervention to a provider paid via FFS with shared savings and losses may come
from the thresholds used to determine savings and losses, it would be interesting to see how this
financial risk changes when these thresholds are removed. In other words, without thresholds
providers would share in 60% of all payer savings below the benchmark (the average expected
patient demand) even if patient demand was only slightly below the benchmark, rather than at
least 2% below it. Similarly, providers would also share in 40% of all payer losses above the
benchmark, even if patient demand was only slightly above it, rather than at least 2% above it.
Removing these thresholds makes it less likely that providers are denied a savings payment that
t h ey de se rv e, t h ought it a lso m ak es it m ore lik ely t h ey ar e c h ar ged w it h lo ss es t h at t h ey don ’t deserve.
Removing the savings and losses thresholds for providers engaging in RMT that costs 3%
of average expected patient demand results in the following changes in PID:
Figure 4.29: ΔPID, removing thresholds for FFS with shared savings and losses plus RMT (3% cost)
129
As you can see, most providers experience essentially no changes in financial risk when removing
these thresholds, short or long term, while a minority of large providers in the upper quantiles of
the distribution experience a moderate decrease in financial risk. Thus, some large providers
being paid under this system who decide to engage in RMT may see a benefit to choosing to
eliminate their shared savings and losses thresholds (which is an option for them under the
Medicare Shared Savings Program), but few other providers would likely benefit from such an
arrangement.
4.7.2 - Random Effect RMT
Thus far the RMT interventions tested have reduced patient risk of future needs by a
constant 5%. The actual needs that occur are still random within the modeled parameters from
the MEPS data, but the reduction in risk itself was assumed to be constant. An interesting
question is what happens if that reduction in risk is also subject to some variation. I tested to see
how provider financial risk (in the form of PID) changes from the baseline RMT intervention for
each payment system when the reduction in risk, instead of a constant 5%, is assumed to be a
beta random variable with a range of approximately 2 to 8% (but still an average of 5%).
It turns out that giving RMT a random effect produces nearly the same results as a
constant effect. For instance, the following graphs compare the change in PID from regular to
random RMT for capitation (left) and direct pay RMT (right) over the first 3 years:
Figure 4.30: ΔPID for capitation (left) and paid RMT (right) from regular to random RMT (5% cost)
All of the other payment systems show a similar pattern. There seems to be little to no increase in
financial risk when RMT has a random effect.
130
4.7.3 - Lower Fixed and “Overtime” Costs
Finally, it would be interesting to see how provider financial risk changes for providers
with a different fixed versus variable cost structure who decide to engage in RMT. The following
graphs show scenarios where, instead of a baseline 85% fraction of fixed costs, it is assumed that
providers have converted enough of those fixed costs to variable costs to reduce this fraction to
65%. They could accomplish this by, for instance, outsourcing a greater amount of care and/or
investing in more flex contracts for on demand nursing and technician staff. Accordingly, since it
is those kinds of staff that would typically need to receive actual overtime pay or be hired on a
t empora ry b as is in c as es w h en ad dit ion a l “ ove rt im e” v ar iab le c ost s kic k in in t h e s im u lat io n m odel , t h e “ ove rt im e” c ost f ra c t ion h as als o b een reduced by the same amount, from 30% to 10%.
This also makes sure that costs under overcapacity conditions are comparable in these scenarios
t o t h ose ou t lin ed p re v iou sly . Low er i n g f ixed and “ o v er t im e” c ost s ar e inte re s t in g bec au se , w h ile
they make fixed costs for providers lower, they make variable costs higher. Thus, provider costs
become much more responsive to patient demand, which depending on the payment system
might be desirable or undesirable.
For providers paying for some kind of RMT through shared savings, changing their cost
structure as described above seems to result in an increase in financial risk. The left graph below
shows this increase for shared savings, and the right graph shows it for shared savings and losses:
Figure 4.31: ΔPID, regular to fixed/overtime cost reduction, FFS w/ SS + RMT (2.5% cost) (left) and FFS w/
SS/SL + RMT (3% cost) (right), first 3 years
Thus, this kind of cost structure would likely not be good for providers paying for RMT through
shared savings; the baseline cost structure would be preferable.
On the other hand, providers paying for RMT through some kind of capitation see a
serious decrease in financial risk when adopting this cost structure. The following graphs show
131
this decrease for providers under regular capitation (left) and risk adjusted capitation (scaled
downward by a factor of 1.457) (right):
Figure 4.32: ΔPID, regular to fixed/overtime cost reduction, capitation + RMT (5% cost) (left) and risk
adjusted capitation + RMT (5% cost) (scaled) (right), first 3 years
The decrease in financial risk is especially prominent for providers under risk adjusted capitation.
These providers would seem to greatly benefit from changing their cost structure to be more
variable and less fixed.
Finally, we can look at the same cost structure change for providers being paid a direct
capitation payment to implement RMT, with FFS for all other care needs:
Figure 4.33: ΔPID, regular to fixed/overtime cost reduction, FFS + paid RMT (5% cost), first 3 years
132
There seems to be no obvious benefit or lack thereof to this cost structure for these providers.
Comparing the PID values for RMT (5% cost) paid via capitation (left) vs risk adjusted
capitation (right) under this new cost structure, we can see that this change has brought both
payment systems nearly to par with one another (using baseline FFS as a reference point):
Figure 4.34: ΔPID when implementing RMT (5% cost) from baseline FFS to reduced fixed/overtime costs
capitation (left) and risk adjusted capitation (right), first 3 years
Paying for RMT with its own direct payment is still less financially risky, however:
Figure 4.35: ΔPID when implementing RMT (5% cost) from baseline FFS to reduced fixed/overtime costs paid
RMT, first 3 years
133
It is interesting to see how well these three payment systems stack up against each other.
Red u c in g t h e f ixed and “ o v er t im e” c ost s in t h e s im u lat ion and i n c re as in g va ria ble c ost s h as brought the two full capitation payment systems closer to parity with direct pay RMT. The change
for risk adjusted capitation is especially notable; it is now only slightly more financially risky than
regular capitation. Given that risk adjusted capitation also discourages risk selecting for healthier
patients, this mild increase in financial risk might make it worth it over regular capitation.
However, paying for RMT directly still retains a significant advantage when it comes to
financial risk, especially for small providers. The following table shows the magnitude change in
PID for the median small (5,000 patient) provider with this new cost structure, from baseline FFS
to each of the three payment systems, during the first 3 years:
Payment System
Magnitude
ΔP ID
Capitation 1.49
Risk Adjusted
Capitation
1.95
Paid RMT 0.55
Difference
(Cap / Paid RMT)
2.72
Difference
(RA Cap / Paid RMT)
3.54
Table 4.12: Magnitude ΔPID from baseline FFS when implementing RMT (5% cost) under capitation, risk
adjusted capitation (scaled), or paid RMT for median 5,000 patient provider with reduced fixed/overtime costs,
first 3 years
Despite the change in cost structure that significantly lowers provider financial risk under
capitation and risk adjusted capitation, they are still 2.72 times and 3.54 times more financially
risky, respectively, for the median small provider than paying for RMT directly.
Looking at the differential benefit from increasing in provider size, paying for RMT
directly still seems to have an advantage there as well:
Patient Population
Size Increase
Differential
Benefit Ratio, Cap /
Paid RMT
Differential
Benefit Ratio, RA
Cap / Paid RMT
5,000 to 10,000 1.86 2.34
10,000 to 25,000 1.31 2.35
134
25,000 to 50,000 1.51 2.64
Table 4.13: Ratios of reduction in PID from various provider size increases, [risk adjusted capitation (scaled)]
or [capitation] / [paid RMT], for median provider with reduced fixed/overtime costs when implementing RMT
(5% cost), first 3 years
Paying for RMT through capitation or risk adjusted capitation, even with lower fixed and
“ ove rt im e” c os ts, still comes with a significantly greater incentive for the median provider to
merge or grow larger than when paying for RMT directly.
I n s u m m ar y , t h is r edu c ed f ixed and “ ove rt im e” c ost s t ru c t u re s eems t o sign if i c an t ly les se n financial risk for providers trying to implement RMT under both kinds of capitation, though it
increases financial risk for the same providers under shared savings. However, despite this, paying
for RMT through capitation is still associated with significantly more provider financial risk than
paying for it directly, especially for the median small provider, as well as a significantly greater
incentive to increase in size. Both kinds of capitation still retain the same advantage in terms of
payer costs, so there is still the same tradeoff as noted earlier in this chapter. The next chapter
will discuss this tradeoff in more detail.
135
Chapter 5 - Discussion
T h e p roblem w it h u si n g i n c en t iv es t o sh ap e p h ys ic ians’ b eh av ior is t h e b lun t n es s of t h e
method. Hillman has contrasted such in c en t iv es w it h w h at h e c alls “ r u les,” s u c h as guidelines for care, which specify clinical policies for doctors to follow. Rules attempt to
prescribe correct care; financial incentives leave that choice to the doctor. Incentives
thereby create an ethical quandary for the doctor that rule-based management does not.
With financial incentives, patients must depend on doctors to take the correct course even
when it is against their economic self-interest; rule-based management shifts that burden
to the person who makes the rules. (Berwick, 1996, p. 1228)
This study has taken a look at one of the major barriers to better prevention of medical
needs and patient harm in health care – how we currently pay health care providers – and offered
some theory with the aid of a simulation on ideal strategies for moving forward. The first chapter
examined the past and present major payment reform movements (whose goals have always been
myriad and not just limited to encouraging more prevention) and noted what is perhaps their
most common recurrent theme: the use of financial incentives to try to push health care provider
behavior in more desirable directions. This theme makes sense in light of the recurring criticism
of fee-for-service (FFS), the predominant provider payment system, that it incentivizes higher
volumes of care. Changing that financial incentive to favor value instead seems like a worthy goal
under this framework.
There are a few thorny problems to this approach, however. Increasing value in health
care is vital, but providers have more control over some aspects of value than others. They have a
lot of control over their attempts to increase value, as long as they are paid in a way that allows
them to conduct valuable interventions. But the actual outcomes of those interventions are less
certain. And if value is measured in the typical way, as outcomes divided by costs, then this means
that the value of medical interventions is also subject to some degree of uncertainty outside of
provider control. This is th e c en t ra l p re m is e of K e n n et h Ar row ’s (1963 ) f amo u s p ap er t h a t effectively launched the field of health care economics; that uncertainty in both patient medical
needs and the efficacy of their treatment drives all of the special economic problems in health
care. If financial incentives are contingent on providers achieving valuable medical outcomes then
any uncertainty in those outcomes will always create some degree of uncertainty in provider
finances.
There are several other concerns with using financial incentives to motivate provider
behavior that Kronick, Casilano, and Bindman (2015) neatly frame in the question of whether
providers should be paid more like apple pickers, where financial outcomes are strongly linked to
performance, or federal judges, where there is essentially no relation. But the main concern of this
study is how uncertainty in provider financial incentives creates provider financial risk, especially
for providers engaged in needs and harm prevention. Predominant FFS payment systems do not
136
especially encourage prevention and can in fact often create a barrier to it. Alternative payment
systems have the potential to get around this barrier, but if they create excessive financial risk for
providers who engage in prevention they may erect a n ew b ar rie r in it s p lac e. “ T yp e 2 er rors ” in shared savings, for instance, where savings are not paid out to providers who deserve them
because of randomly high patient demand for care outside of their control (Pope & Kautter, 2012)
could be incredibly detrimental to prevention if those providers were relying on those savings
payments to reimburse their preventive interventions. This is especially true if, as is likely, this
kind of financial risk is more likely to affect smaller providers, as such providers still make up a
fairly large proportion of the health care providers in the U.S.
Removing financial incentives from health care would be an impossible and unnecessary
task. There will always be some link between the actions of health care providers and their
financial performance. But as the quote by Berwick (1996) above suggests, and as Berenson and
Ric e (2 0 15 ) f u rt h er elab or at e, t h er e’s mo re t o en c o u ra gin g des ira ble p rov ide r be h av ior t h an j u st financial incentives. We also have to determine what that desirable behavior actually looks like;
i. e. , w h at t h e “ ru les” f or t h at b eh av ior ough t t o be. An d o n c e t h at is es t ab lis h ed, it is n ec es sa ry t o
make sure that providers are paid in a way that makes it possible for them to live up to those rules
to the best of their ability. Making sure that payment systems foster normative structures that
facilitate prevention may be just as if not more important than using financial incentives to
reward providers for successfully preventing, especially if the latter creates provider financial risk
that may undermine the former.
5.1 - Simulation Model Recap
The simulation model described in chapters three and four attempts to explore whether it
might be possible to change provider payment to encourage prevention without significantly
increasing provider financial risk. It is based on the theoretical conclusions of chapter two: that
while a major source of provider financial risk is variation and uncertainty in medical needs and
patient harm, there is significantly less variation and uncertainty in the medical risk of those
needs and harm itself, and that this distinction might in theory allow the design of payment
systems that specifically target this medical risk (i.e., specifically pay for prevention) without
creating significantly more provider financial risk.
There are other potential sources of provider financial risk that are not explored in the
simulation. One other major potential source is market based risk, i.e., the provider financial risk
created by the dynamics of competition and patient movement within the health care market.
The purpose the simulation model is not to perfectly recreate the complete dynamics of provider
financial risk in health care, but rather to generate some insights into the theoretical strength of
its most irreducible and ever-present contributing force – variation and uncertainty in patient
needs and harm – under various payment system scenarios and especially while providers are
engaging in preventive interventions.
137
5.1.1 - Shared Savings and Financial Risk
Shared savings has a lot of potential merit. The simulation results suggest that, when
added to a regular FFS provider who is not engaging in a costly risk mitigation treatment (RMT)
intervention, shared savings reduces financial risk for that provider (especially smaller ones). If
much of the undesirable volume-over-value behavior of FFS providers comes from their attempts
to mitigate the financial risk of lower than expected patient demand, then shared savings on its
own may moderate the incentive to produce wasteful volume. This in and of itself would be a
great thing; many patient safety and other health care experts have noted the potential for
wasteful volumes of care to result in preventable patient harms (Chassin, 2013), in addition to the
added cost of such care to payers and society. On the other hand, shared savings comes with a
cost to payers in the form of shared savings payments. The simulation model results showed no
actual reductions in payer costs under shared savings despite the lower financial risk. Adding
shared losses produced some reductions in payer costs, though it also increased financial risk for
some providers. Nevertheless, as a tool for discouraging wasteful volumes of care, a payment
system including shared savings and losses seems promising.
On the other hand, the results from the simulation suggest that shared savings might have
less merit as a mechanism for financing prevention. The first obvious limitation comes from the
fact that providers only share in a portion of payer savings. This means that providers under
shared savings payment systems will only ever be able to pay for preventive RMT interventions
that produce ROIs in line with their savings payments. Under the Medicare Shared Savings
P rogra m ’s most common 50% savings rate payment, for instance, providers will be limited to
RMT interventions that produce a 2 to 1 return or more. While RMT interventions with this kind
of potential ROI are certainly possible, they might also be a tall order at a time when such
interventions are in active development, with providers still figuring out how best to achieve
necessary efficiencies and economies of scale.
The second limitation is related to financial risk. Providers in the simulation under an FFS
plus shared savings payment system (with or without shared losses) who decide to engage in RMT
see a spike in financial risk, even though the RMT costs an amount that is sustainable under the
shared savings payment. This is similar to the conclusions of DeLia (2016), who determined that
sh ar ed sa v in gs ar ra n ge m en t s as c u rr en t ly c on st ru c t ed c om e w it h s ign if ic an t r is k of “ t y p e 2 er rors ” , whereby providers who do actually achieve savings nevertheless may not clear the savings
threshold needed to trigger shared savings payments because of randomly high variation in their
p at ien t s’ de m an d f or c ar e. I t is t h eoret ic ally p oss ib le t o red u c e t h is f in an c ial r is k by r edu c in g t h e
t h re sh ol d needed t o t rigg er s av in gs, b u t t h en t h is in c re as es t h e r is k of a “ t y p e 1 er ror” , w h er eb y
payers pay out savings payments to providers who did not in reality achieve savings. Because of
this, payers would almost certainly demand shared losses to guard against type 1 errors, and
adding these seems to increase financial risk even when the savings threshold is reduced.
Moreover, the results from the first special scenario at the end of the previous chapter show that
even if you remove the savings and losses thresholds entirely, so that savings and losses are
triggered for any deviations in patient demand from the benchmark (not just those at least 2%
138
away), this makes a negligible impact on provider financial risk for the vast majority of providers.
There are some echoes of these simulation results in the current Medicare shared savings
environment. Most shared savings experiments to date have determined savings based on
benchmarks that differ for each individual provider based on their historical patient utilization.
Many experts have noted potential problems with this methodology for determining savings,
in c lud in g t h at it in c en t iv izes p rov ider s t o ramp u p t h eir p at ien t s’ de m an d p rior t o join i n g t h e
shared savings program so that they might gain savings simply by letting demand fall to normal
lev els (D ouv en , M c G u ire , & M c Will iams , 2 0 1 5). I t ’s p oss i ble that the – thus far – low net Medicare
savings in these experiments could be related to this unintended incentive (McWilliams, Hatfield,
Chernew, Landon, & Schwartz, 2016).
As a result, Medicare has slowly begun moving its shared savings methodology toward
calculating provider benchmarks based on a regional average, instead of individually for each
provider. This should eliminate provider incentives to ramp up patient demand prior to joining a
shared savings program. On the other hand, it may also discourage inefficient providers from
signing up for the program in the first place. Given the high level of variation in the efficiency of
accountable care organizations (ACOs) participating in the Medicare Shared Savings Program,
experts have cautioned that these regional benchmarks should be implemented gradually (Rose,
Zaslavsky, & McWilliams, 2016), and indeed the Centers for Medicare and Medicaid Services
(CMS) has ruled that it will phase in these regional benchmarks only for providers who have
already participated for in the program for some time under the original benchmarks (CMS,
2016b).
The shared savings component in the simulation is analogous to a regional benchmark to
reflect this new direction for the Medicare Shared Savings Program. In fact, by design the
si m u lat ion ’s s h ar ed sa v in gs b en c h m ar k p er f ec t ly r ef lec t s t h e s im u lat ed p a t ien t p op u lat ion ’s p re -
savings expected average level of demand; i.e., it is a benchmark meant to be reflective of each
p rov ider ’s r eal c u rr en t lev el of ef f ic ien c y. T h is means that the increased financial risk seen by
providers in the simulation under shared savings who decide to engage in an RMT intervention
has nothing to do with those providers facing a savings benchmark that is beyond their current
capabilities to meet, as would be a concern with real-world providers today (Rose et al., 2016).
Rather, this increased risk is solely due to random variation in patient needs causing providers to
not see expected returns from their efforts to produce savings, something all providers would be
at risk for even with a completely appropriate regional benchmark. The fact that actual providers
w ould a lso be s u bj ec t t o regio n al s av in gs b en c h m a rk s t h at ar en ’t p er f ec t ly r ef lec t iv e of t h eir efficiency would likely add to the financial risk seen in the simulation.
There are a few real-world possible results of this increased financial risk when
implementing RMT as seen in the simulation. Even though the simulation measures financial risk
as the amount of provider manipulation of patient demand that would be needed to mitigate
an t ic ip at ed f in an c ial s h ort f alls, t h is does n ’t mea n t h at all p rov ider s w ould a c t u ally man ip u lat e
demand upward when engaging in RMT under shared savings. Rather, providers would have a few
options. They could ignore the financial risk and see what happens, which might put them in
financial jeopardy or, because of the nature of risk, it might not. Or they could respond to this
139
financial risk, not by increasing patient demand, but rather by scaling back their RMT related
interventions or even abandoning them entirely. If word spread among providers that a
significant minority of them was experiencing increased financial losses when trying to
im p lem e n t an RMT i n t er v en t ion b ec au se t h ey w er en ’t r ec eiv in g e n ough sh a re d s av ings to
compensate, this might push many providers into abandoning RMT, especially given the
uncertainty involved in running such an intervention in the first place.
To the extent that current provider care volumes are truly wasteful, shared savings has the
potential to encourage providers to become leaner in their care delivery and lower payer costs in
the process; assuming, of course, that providers are willing to accept reductions in their average
revenues per patient, as would be fiscally necessary for any payer to see net savings. But the
results of this simulation suggest that shared savings may be less useful as a tool for encouraging
providers to engage in genuine preventive efforts. Thus, the ability for shared savings to produce
net reductions in payer costs likely hinges on the proportion of current patient care that is truly
wasteful and could simply be eliminated versus that which could be prevented but only by
investing in interventions to reduce the risk of new patient needs and harm. Experts have
produced various estimates of the proportion of each of these in health care, but it is difficult to
say which is greater with any certainty (not to mention that there is significant disagreement
ab out w h at ac t u ally c on st it u t es “ w as t ef u l” c ar e). W h at I can say is that, as far as paying for
prevention is concerned, the results from the simulation suggest that some form of capitation
may offer a better path.
5.1.2 - Full Capitation vs FFS Plus RMT-Only Capitation
Full capitation has a significant advantage over shared savings when it comes to paying for
prevention. Since providers keep the whole capitation payment no matter what, they keep all of
the savings they generate (instead of only a portion) if they reduce medical needs and patient
harm. This means that they can sustainably pay for any RMT intervention that produces at least a
1 to 1 ROI on average, as opposed to the much higher ROIs required to sustain such interventions
under shared savings. Despite the fact that RMT-style interventions, like care coordination for
chronically ill patients, have been around for many years now, we are still in the early days of
RMT development. New research on preventive interventions for health care providers is
constantly ongoing, whether it involves new systems to reduce medical errors and health care
ac quir ed in f ec t ion s, n ew sy st ems of en ga gem e n t w it h c om m u n it y re sou rc es t o keep “ su p er u se rs ” of health care out of the hospital, or new programs for end of life care centered around
minimizing medical intervention and maximizing quality of life. Having a payment structure that
only asks for a 1 to 1 ROI at a minimum allows for a lot more experimentation with these
preventive interventions than would be possible with a more restrictive payment system like
shared savings, and thus may result in faster innovation in RMT.
The full capitation plus RMT scenario tested in the simulation has an RMT intervention
that costs more than the one tested in either shared savings scenario (5% of average expected
140
patient demand vs. 2.5% and 3% respectively). And yet, full capitation still produces an amount of
provider financial risk comparable to both shared savings payment systems in the short term (a
little more for smaller providers and about the same for large providers), and it produces less
financial risk for providers of all sizes in the long term. Furthermore, as the last special scenario at
the end of the previous chapter suggests , if it is p o ss ib le t o c on v er t s om e f ix ed an d “ ove rt im e” costs into normal variable costs, the financial risk associated with capitation might be
significantly reduced, making it clearly less risky than a shared savings system as a way to pay for
prevention. There are some important differences to consider with regards to provider behaviors
that might result from financial risk under capitation vs FFS-based payment systems; more on
that in a moment. But at least when it comes to raw financial risk while a provider is engaging in
RMT, capitation seems to result in a similar or lower amount of it than shared savings (risk
adjusted capitation is another matter; more on that in the next section).
Given that, for providers engaging in RMT, full capitation offers more flexibility in the
ROIs of RMT interventions, similar or lower financial risk, and is less costly to payers, the
simulation results suggest it ought to be the winner in a head to head comparison with shared
savings as a method of paying for the prevention of patient needs and harm. The only real
comparative downside is how providers might respond to financial risk. Providers under FFS-
based payment systems (including shared savings), in response to financial risk, typically have an
in c en t iv e t o bot h s eek ou t s ic ke r p at ien t s (a go od t h in g) a n d a rt if ic ially i n f la t e t h eir p at ien t s’ demand for care (not a good thing). Whereas providers under capitation-based systems, in
response to financial risk, have an incentive to both avoid sicker patients (a pretty bad thing) and
ar t if ic ially d ec re as e t h ei r p at ien t s’ de m an d f or c ar e (a go od or b ad t h in g dep e nding on your view
of the appropriateness of that demand for that care in the first place). Given an equal amount of
financial risk, it is arguable that the incentives for providers under FFS-based systems are not as
bad as those for providers under capitated systems. The incentive under capitation to avoid sicker
patients is especially bad, and is largely why there have been so many efforts to find good ways to
risk adjust the capitation payments so that this incentive might be lessened.
This concern is even more relevant when looking at the difference between paying for
RMT with full capitation vs paying for it with an RMT-only capitation payment on top of regular
FFS. When it comes to paying for prevention, this seems to be the main payment system dilemma
in the simulation results. FFS plus an RMT-only capitation payment is associated with lower
provider financial risk than full capitation plus RMT while allowing the same flexibility in RMT
ROIs. Taken along with the difference in incentives that encourages seeking out sicker patients
(as opposed to avoiding them), the FFS system seems to have a clear advantage here. The
financial risk associated with full capitation can be significantly reduced if it is possible for
convert some fixed costs to variable costs, however, as indicated by the special scenario in the last
chapter (though it is still higher than FFS plus paid RMT). And full capitation is significantly less
costly to payers than any of the FFS-based payment systems. Thus, there appears to be a tradeoff
between provider financial risk (which may be lessened through changes in the provider's cost
structure) and payer costs with these two payment systems.
There are some important caveats to consider, however, when evaluating this tradeoff.
141
F irs t , it ’s imp ort an t t o rev iew h ow p rov ider f in an c i al r is k is ac t u ally c alc u lat ed in t h e s im u lat io n . The traditional conception of financial risk is that of variation in net cash flows. The beginning of
the previous chapter looked at how this variation in net cash flows differs between regular FFS
and capitation when risk-averse provider manipulation of patient demand is turned off. Similar to
Gapenski and Langland-Orban (1996), who also created a Monte Carlo simulation comparing FFS
and capitation, my simulation found that variation in net cash flows was much lower for capitated
providers than FFS ones. They acknowledged, however, that FFS providers can potentially
mitigate financial risk more easily than capitated providers, as any action by FFS providers to
increase patient demand will increase both revenues and variable costs, while any action by
capitated providers to decrease patient demand will only decrease variable costs with no changes
to revenues. Thus, especially if variable costs make up a small proportion of total costs, capitated
providers will have to move patient demand much more than FFS ones for the same financial
outcome. Since providers have so much control over patient demand and, in theory, they could
exercise this control to mitigate the risk of poor financial performance, I decided to incorporate
this explicitly into the model by measuring financial risk as the amount of provider manipulation
of demand that would be necessary to offset anticipated financial shortfalls (i.e., measure financial
risk according to one of its potential outcomes, rather than directly).
I f y ou b eliev e t h at c a p it at ed p rov ider s’ r edu c ed p o w er t o p u ll t h emselv es out of a f in an c ial
hole compared to FFS providers is not an important driver of their behavior, and instead you
prefer the more standard conceptualization of financial risk as simply variation in net cash flows,
then you might be inclined to think that capitation is less financially risky than FFS. But Gapenski
and Langland-Orban (1996) cautioned against this conclusion, despite their simulation results
showing it to be so. First, they believed that provider ability to manipulate patient demand in
re sp on se t o ris k w as p roba bly imp ort an t , ev e n t h o u gh t h ey di dn ’t ac c oun t f or it in t h eir mo del . And second, this concern becomes even more important when there is uncertainty in the
appropriate level of the capitation payment in addition to uncertainty in patient needs.
Both Gapenski and Langland- Orb an ’s ( 1996) m ode l an d my m ode l exa m i n e objective
financial risk, which is uncertainty in financial outcomes but within a known distribution of
expected patient needs (in the case of my model this known distribution comes from the MEPS
dataset). In the real world, however, payment systems also involve subjective financial risk, which
concerns uncertainty in the distribution of expected patient needs itself. Under purely objective
risk, where the average expected patient demand is known with certainty and demand varies
normally around that average, capitation makes a lot of sense. Fixed costs tend to be high in
health care in the short term, and a properly set capitation payment ensures that those fixed costs
will be taken care of as long as the capitation payment accurately reflects average demand. But
high fixed costs also mean low variable costs, and because the capitation payment is fixed this
means that, for capitated providers, marginal changes in profit in response to changes in demand
are entirely dependent on changes in these low variable costs. This means that if the capitation
payment is not set at a proper level (e.g., set too low) because of uncertainty in the distribution of
patient needs itself, capitated providers can very quickly run into difficulty covering their fixed
costs with little recourse to ameliorate the situation by changing patient demand. FFS providers,
142
on the other hand, even with the same high fixed costs, are much more adaptable to
unanticipated levels of patient demand because their revenues also change when demand changes
(i.e., their marginal profit in response to changes in demand is much higher than capitated
providers when fixed costs are high).
This is why, when subjective financial risk is high (i.e., there is significant uncertainty in
the distribution of patient needs itself), payment systems with a higher marginal profit rate from
changes in patient demand tend to fare better as they are more adaptable to this uncertainty.
Ironically, FFS providers can increase this marginal profit rate by having lower variable costs (i.e.,
a higher fraction of fixed costs) while capitated providers can accomplish this by having higher
variable costs. This is the opposite of Gapenski and Langland- Orb a n ’s (199 6 ) f in ding s on obj ec t iv e
provider financial risk measured as variation in cash flows, whereby risk decreased for capitated
providers when their variable costs decreased while FFS providers saw the inverse. But they
surmised that high subjective financial risk could undermine or even negate their findings on
objective financial risk, and indeed my model suggests this is possible. Even though I use a known
distribution of patient needs to generate objective financial risk, which would normally be fine to
m eas u re as j u st ex p ec t ed v ar iat ion in c as h f lo w s, m eas u ring f in a n c ial r is k in s t ead as t h e p rov ider ’s actual response to try to mitigate cash flow variation is a way of measuring how providers might
react to subjective financial risk as well , ev en t h ou g h I c an ’t ac t u ally a c c oun t f or s u bj ec t i v e r is k in the simulation. In statistical parlance, my model essentially uses a nonparametric measure of
provider financial risk. Since a large component of provider risk-averse behavior in the simulation
is j u st a si m p le h eu ris t i c f or r es p on di n g t o a lev e l o f c as h on h an d b elo w a t a rget , it does n ’t m at t er h ow a c c u ra t ely p rov ider s h av e a ss es se d t h e d is t rib u t ion of t h eir p at ien t s’ n eed s. All t h at mat t er s is that providers know how they got to their current financial position (i.e., their historical patient
demand and cost structure) and with that information they can figure out how demand needs to
change in order to reach their financial target. A higher marginal profit rate from changes in
patient demand makes it easier to reach financial targets under uncertainty, and so it should
reduce the severity of both objective and subjective financial risk in the real world.
FFS has a higher marginal profit rate from changes in patient demand than capitation at
baseline in the simulation, though changing some fixed costs into variable costs narrows this gap.
This is why, even though the simulation only explicitly models objective financial risk (since it
c an ’t , b y def in it ion , m ode l su b jective risk), the lessons from the simulation should apply to
situations containing both objective and subjective financial risk. Under a significant amount of
either kind of risk, FFS-based payment systems would likely be preferable over capitation-based
ones to both providers and, if we assume providers would act to mitigate this financial risk by
manipulating demand, to the public as well. This is especially the case for smaller providers who
would be more vulnerable to outlier swings in patient demand. Thus, from the standpoint of
minimizing the role of financial risk, both objective and subjective, as a barrier to prevention, it
seems that paying for RMT via an FFS-based system with an RMT-only capitation payment might
be preferable to paying for RMT through full capitation. This is in alignment with Conrad (2015)
and other health care experts who have urged caution, given the financial risks involved, in paying
providers via full capitation, especially smaller providers. Though if it is possible to turn some
143
fixed costs into variable costs, this might make full capitation more attractive, especially given the
lower payer costs involved. For large providers who are capable of changing their cost structure in
this way, full capitation could be a great way to pay for RMT. For all other providers, the
simulation results would suggest that using FFS with an RMT-only capitation payment on top is
probably the better way to go.
5.1.3 - Risk Adjusted Capitation
Adding risk adjustment to capitation payments in the simulation model, which changes
the payments slightly to reflect differences in patient population medical risk, significantly
increases measured provider financial risk. This is largely due to the fact that providers in the
simulation are assumed to have fixed costs based on the average expected level of patient demand
in any give n y ear , a n d t h u s if t h ose f ixed c ost s ar e h igh t h ey w on ’t b e v er y f lexib le in de ali n g wit h short term changes in underlying patient medical risk (which occur in the model solely due to
patient turnover). Combined with the low marginal profit from changes in demand associated
with capitation, this increased likelihood of a mismatch between fixed costs and patient demand
greatly increases financial risk in comparison to regular capitation. The higher patient turnover is,
the more this would be a concern.
However, risk adjusted capitation, because it adjusts payments to reflect patient
population medical risk, significantly reduces the incentive to risk select for healthier patients
(and avoid sicker ones) compared to regular capitation, since any attempt to avoid sick patients
would in theory result in lower capitation payments. Thus, for a provider being paid via
capitation, it is worth considering how to make risk adjustment work better, as removing the
incentive to avoid sick patients goes a long way to making the financial risk in capitation more
palatable.
The special scenario at the end of the previous chapter showed that, by turning some fixed
costs into variable costs, thereby making costs more responsive to patient demand, the financial
risk associated with risk adjusted capitation can be greatly reduced. So much so, in fact, that the
modeled change in fixed costs resulted in a distribution of financial risk for providers under risk
adjusted capitation nearly on par with those under regular capitation. There would still be an
incentive for providers to stint on care (reduce demand) under risk adjusted capitation that might
be undesirable. And it is arguable that patients might find this incentive more upsetting than the
incentive to provide too much care under FFS, even if both are potentially harmful. But at least
f or lar ger c ap it at ed p rov i der s w h o are n ’t s u bj ec t t o as muc h f in an c ial r is k, as lo n g as t h ey c an increase the fraction of their costs that are variable, the simulation results suggest that risk
adjusted capitation may be a good way to pay these providers for prevention.
144
5.2 - Implications
There are more ways of paying health care providers for their services than just the
payment systems discussed in this study. Bundled payments, for instance, instead of paying for
each service provided as in FFS, pay a lump sum for one episode of care which is usually defined
by a certain duration of time. Such bundled payments usually define explicitly which kinds of
services are and are not included in the bundle so that providers know for which services they are
financially liable. In this way bundles sit somewhat in between FFS and capitation, though they
share elements of both. There are also more explicit hybrids of FFS and capitation, such as those
that use a full capitation payment equal to half of average expected patient demand and then add
on FFS payments equal to half their normal value under regular FFS. And then there are also
payment systems that carve out certain services under a capitation payment (that pays the full
average expected value of demand for those services) and pays for other services under some kind
of FFS – the FFS plus an RMT-only capitation payment system tested in the simulation is a variant
of this.
All of these payment systems are comparable to regular FFS and capitation, however,
especially in how they create provider financial risk. As represented in the simulation model,
financial risk is just a product of variation in patient needs as w ell as t h e p rov ider ’s margi n al p rof it in response to changes in patient demand, which is itself a result of both the payment system and
t h e p rov ider ’s in t er n al c o st s t ru c t u re . Hig h er v ar iat ion in p at ie n t needs leads to higher financial
risk, while a higher marginal profit from changes in patient demand means providers have an
easier time dealing with financial risk (and thus, the felt effect of the financial risk is lower),
though it also means that net cash flows will vary more with variations in patient demand.
T h er e like ly is n ’t a u n iv er sa l “ be st ” p ay m en t s ys t e m f or all s it u at ion s a n d h ealt h c ar e
se rv ic es . T h e id eal p ay m en t s ys t em f or a si t u at ion de p en ds on a p rov ider ’s c ost s t ru c t u re , including how flexible their costs are to changes in patient demand as well as how much their
c ost s in c re as e in o v er c ap ac it y sc en ar ios. I t als o dep en ds on h ow muc h t h e p rov ider ’s p at ien t population needs vary over time. However, this study has been concerned primarily with paying
for the treatment of patient risk of future morbidity (both needs and harm related risk). This
patient medical risk, unlike the resulting morbidities themselves, does not vary significantly for
the same patient population in the short term. It varies more at a single provider when patient
turnover is high but still not as much as the actual needs and harm that result from medical risk;
this is just a mathematical truism. Thus, the results of the simulation suggest that paying directly
to treat this medical risk with a targeted capitation payment does not increase provider financial
risk, and in fact seems to reduce it. In addition, paying for RMT with its own capitation payment
gives providers a great amount of flexibility in how they decide to treat medical risk, in that they
n eed n ’t b e li m it ed t o o n l y p rov iding s er v ic es f or w h ic h t h ey c an ea si ly a n d/or c ost ef f ec t i v ely b ill a
payer. How the care of other needs and harm should be paid for may vary depending on the types
of needs and the provider itself. But at least just for the treatment of medical risk, a specific RMT
capitation payment seems like a promising solution.
145
5.2.1 - Primary Care
There are a couple specific practical situations in which the implications of this study are
especially relevant, and primary care is one of them. Many of the activities envisioned in RMT
interventions, such as care coordination for chronically ill patients and end of life care
management, are practiced in some form predominantly by primary care providers. Primary care
p rov ider s also t yp ic ally e n ga ge in oth er t re at m en t ac t iv it ies t h at ar en ’t r elat ed t o t h e t re at m en t of m edic al r is k, su c h as t h e t re at m en t of ac u t e illn es s es , s o t h er e is n ’t p er f ec t ov er lap . Nev er t h eless , given their central role in the health care system, primary care physicians would almost certainly
be involved in most RMT-style interventions.
The scenarios modeled in the simulation involve providers who are responsible for all of
t h eir p at ien t s’ c ar e n eed s (o r at leas t all of t h e n eed s en c om p as s ed by the MEPS dataset). An
analogous provider in the real world would be an accountable care organization (ACO) that either
provides itself or outsources all primary and specialty outpatient care as well as all inpatient care
for its assigned patient population. Such ACOs typically employ or have in their network primary
care providers who are responsible for providing primary care services, often including services
related to RMT. The payment systems modeled in the simulation, however, governed how the
ACO as a whole is paid for both RMT and non-RMT patient care, not just individual members of
the ACO like primary care physicians. Thus, financial risk is assessed at the entire ACO level. If
the ACO is being paid FFS with shared savings and losses, it is assumed that the primary care
physicians practicing RMT affect financial risk for the ACO as a whole, not just for themselves.
There are some ACO arrangements, however, in which the ACO only consists of primary
care doctors and other related physicians. They are still assigned a patient population, but they
ar en ’t r es p on si ble f or al l of t h e n eed s of t h at p at ie n t p op u lat ion . T h u s, t h ey may r ec eiv e s h ar ed
sa v in gs p ay m e n t s ba se d on t h eir p at ien t p op u lat ion ’s t ota l dema n d f or h ealt h c ar e, ev en t h ough
they are only responsible for the primary care portion of it. In theory this could be a great way of
paying for RMT, since the primary care ACO itself would not suffer from reduced patient demand
as a result of the RMT (some other provider would have to deal with that). However, as
Mostashari, Sanghavi, and McClellan (2014) detail, there is a paradox to this way of paying
primary care providers. Primary care is only responsible for around 5% of health care spending in
the US. Imagine a primary care physician-run ACO that only receives shared savings, no shared
losses, at a 50% savings rate. In theory, such a payment system could pay for RMT at that ACO
that costs 2.5% of the average expected total demand of its patient population as long as the RMT
achieved at least a 5% reduction in medical risk on average. However, if the primary care
p h ys ic ians ar e on ly r es p o n si ble f or 5 % of t h eir p at ien t p op u lat ion ’s p ay er c o st s t o begin w it h , in v es t in g i n an i n t er v en t i on t h at c ost s 2 . 5 % of t h ei r p at ien t p op u lat ion ’s ex p ec t ed p ay er costs
would be a massive increase in expenses. Even if there is already significant overlap between the
p rimar y c ar e p h ys ic ians’ c u rr en t t re at m en t ac t iv it i es and t h e p rop osed RM T in t er v en t ion , t h er e
are still serious start up investments necessary to create the kind of practice infrastructure that is
capable of reducing patient medical risk by 5%. Making the RMT intervention dependent on
shared savings payments that have a decent chance of not materializing at all, even if the primary
146
care ACO is successful in treating medical risk, is a recipe for encouraging only tentative
investment in RMT. It is likely that primary care physicians who are seriously interested in RMT
would elect to join up with a larger ACO that can help finance it and shoulder the financial risk of
shared savings, which would lead to exactly the situation modeled in the simulation.
Nevertheless, there is some early but promising evidence that primary care physician-run
ACOs may be pressing on and achieving some savings anyway. McWilliams et al. (2016) note that
primary care physician-run ACOs showed significantly more gross savings in 2012 and 2013 than
hospital-run ACOs, though it is unclear how much net savings they created for Medicare, since as
a whole Medicare did not see any net savings from its Shared Savings Program over this time
period. This is at least in line with theory, however, which would suggest that primary care
physician-run ACOs are more likely to benefit from shared savings than hospital-run ACOs. If
there are enough low hanging fruit, high ROI RMT opportunities to be had, perhaps even small
investments in RMT by primary care physician ACOs could result in significant reductions in
patient medical risk that are sustainable via shared savings. At some point, however, as the
marginal cost to treat further amounts of patient medical risk increases, even primary care
physician-run ACOs will hit theoretical limits in their ability to practice RMT, as variation in
shared savings payments will simply create too much financial risk to sustain it. The results of the
simulation suggest that, eventually, in order to see more significant reductions in patient medical
risk, RMT provided by primary care providers will have to be financed through some form of
direct payment for it, not indirectly through shared savings alone.
Primary care- o n ly c ap it at ion p ay m e n t s ar e n ot a n ew c on c ep t , b u t t h ey h av en ’t ex ac t ly
taken off since their introduction. Berenson and Rich (2010) note that primary care-only
capitation has been generally plagued by too low of capitation payments to properly fund all
necessary primary care activities, inadequate risk adjustment, and the effects of the managed care
ba c klas h of t h e 199 0 s. It ’s als o unders t an da ble h ow , w h en c ap it at ion is u se d t o f u n d a n in -person
visit-based primary care workload that includes acute care patient visits, patients and providers
alike may come to dislike the incentive to avoid visits and services under capitation (as Berenson
and Rich also note). This is why, ideally, an RMT-only capitation payment would pay only for
services related to the treatment of medical risk, such as care coordination and ongoing chronic
disease management, while the morbid needs and harm that result from that risk, such as acute
illnesses, accidents, and complications, would be paid for through regular FFS payments. That
way, any time acute morbidity needs arose, patients could be confident that their primary care
p rov ider w ould b e c om p e n sa t ed f or t re at in g t h em i f t h ey de emed it n ec es sa ry and w ould n ’t h av e
an incentive to ignore them.
Right now the most promising reform that is testing something close to RMT-only
c ap it at ion p ay m en t s f or p rimar y c ar e is M edic ar e’s Co m p re h e n si v e P rimar y Car e P lus (C P C +)
initiative (Sessums, McHugh, & Rajkumar, 2016). This initiative is split into two tracks, a basic and
ad v an c ed on e . P rov ider s on t h e b as ic t ra c k w ill be p aid v ia n orm al F F S, b u t i n ad dit ion t h ey ’ll
receive a $15 per beneficiary per month (PBPM) payment, on average, to fund care management.
That is n ’t muc h ; c on si der in g t h e a v er age b e n ef ic iar y c ost s Me dic ar e a rou n d $ 10 , 0 0 0 p er y ear , t h is amo u n t s t o on ly a r oun d 1 . 8 % of av er age y ear ly b e n ef ic iar y exp en se s, bu t it ’s st ill som et h in g at
147
leas t . F or p rov ider s o n t h e a dv an c ed t ra c k, how ev er , t h ey ’ll rec ei ve on average a $27 PBPM
payment, which amounts to about 3.2% of average yearly beneficiary expenses (the payment is
risk adjusted, so providers taking care of the sickest patients will receive $100 PBPM). In addition,
instead of being paid via normal FFS for all services, providers will receive normal FFS for non-
evaluation and management services (i.e., acute care) and a hybrid capitation/FFS payment for
ev alu at ion and m an a ge m en t s er v ic es (i. e. , RM T ). I t ’s n ot a 1 0 0 % c ap it at ion p ay m en t f or RM T , b u t the idea is to tweak the fraction of FFS vs capitation so that the FFS payments are just about equal
to the marginal cost to the provider of a patient visit. That way, in theory, providers neither have
an incentive to provide nor avoid in-person patient visits ( t h ough it ’s u n c lear if , f or t h ose
providers who opt to provide more virtual care and less in-person care, the capitation portion of
the payment will be increased to compensate).
The only slightly questionable element of CPC+, given the results from the simulation, is
that there is an additional payment, of about $4 PBPM in the advanced track, which is given to
p rov ider s “ at r is k ” and is t ak en b ac k if t h ose p rov i der s f ail t o l ower p at ien t de m an d b e lo w a
benchmark and meet quality targets. This is essentially equivalent to adding a shared loss
component, which means there will be some financial risk associated with it; some providers may
have to pay it back even if they do genuinely reduce demand due to randomness in patient needs
(a “ t yp e 2 er ror” ). At on ly $ 4 P B P M it is n ’t a h u ge p ote n t ial f or s h ar ed lo ss , t h ough f or an av er age
primary care practice of 2,000 patients this is still equivalent to $96,000 per year, around the
salary of a nurse practitioner. Nevertheless, the CPC+ initiative is still an extremely promising
program that could potentially go a long way toward making RMT sustainable in primary care.
5.2.2 - Patient Safety
The other topic on which discussion of RMT and payment is especially relevant is patient
sa f et y. P re v en t ion a n d p a t ien t s af et y ar en ’t alw ay s t h ought of in t h e s ame li ght , b u t es p ec ially
when it comes to payment there is a lot of common ground. Berwick (1996) relays a story about a
h osp it al in T w in F a ll s, Id ah o wh ic h led a c om m u n it y c amp aign t o red u c e c h ildr en ’s in j u rie s f rom bicycle accidents. The result? Bicycle injuries went down, and the hospital had to eat a $150,000
decrease in its emergency room revenues. To me, that was as much an effort related to patient
safety as it was to prevention. And the payment implications are stark. When the predominant
F F S p ay m en t s ys t em does n ’t t en d t o p ay w ell f or p at ien t s af et y ef f ort s, and in addition it reduces
provider revenues when those safety efforts are successful anyway, to me that is an unsustainable
impediment to improvement in the safety and quality of the health care system. In terms of the
conceptual framework from chapter t h re e (r ep rodu c ed as F igur e 5. 1), it ’s as if w e’r e a sk in g
providers to invest in safety-related RMT that doesn’t itself result in much net profit, and when it
successfully reduces morbid patient harm it results in less treatment and even less net profit.
There are a few ways in which modern payment system reforms have touched on patient
sa f et y. T h e mo st p rom in e n t w as M edic ar e a n d ot h er p ay er s’ de c is ion s t ar t in g in 2 0 0 8 t o c eas e
reimbursing hospitals for the additional costs of certain hospital-acquired conditions that were
148
not present when the patient was admitted, including foreign objects left in the body after
surgery, air emboli, stage 3 or 4 pressure ulcers, and infusions of incompatible blood (Milstein,
2009; CMS, 2015a). Medicare also stopped paying any payments at all to both hospitals and
physicians in cases involving the most egregious medical errors: wrong site, wrong procedure, and
w ron g p at ien t s u rger ies . All of t h es e c an b e c on si d er ed “ n ev er ev en t s” in t h at t h ey s h ould s im p ly
never occur in a s af e h osp it al r egard less of t h at h osp it al’s c as e mix. T h u s, p at ien t s af et y ef f ort s
dir ec t ed at t h e se n ev er ev en t s don ’t r eally f all u n de r t h e u m br ella of RM T . R MT is ab out r edu c in g
the risk of future patient needs and harm. If a hospital can take actions to guarantee that
p ar t ic u lar m edic al er rors n ev er oc c u r t h en t h at is n ’t a m at t er of r edu c in g ris k bu t r at h er of creating certainty. It makes sense that, if hospitals can 100% guarantee avoidance of these medical
errors, then there is no reason to pay hospitals extra for when they occur.
Figure 5.1: A conceptual framework of morbidity risk, morbidity, and payment (created using Insight Maker)
T h e c on t rov er sy ar is es in M edic ar e’s n on p ay m en t f or t h e a dd it ion al c ost s of oth er hospital-acquired conditions, such as catheter associated infections and surgical site infections,
which – unlike never events – are only potentially preventable rather than 100% preventable
(Lembitz & Clarke, 2009). These health care associated infections (HAIs) are incredibly harmful,
costly, and may be between 55 and 70% preventable using current evidence-based guidelines
(Umscheid et al., 2011; Anderson et al., 2014). Providers ought to implement the best evidence-
based strategies to reduce the risk of these HAIs occurring. But due to variations in patient risk as
w ell as r an do m c h an c e, t h ey w on ’t alw ay s su c c eed in p re v en t in g t h em. Whi c h mean s t h at providers face the possibility that, even if they are doing everything right when it comes to the
prevention of HAIs, they will still be financially penalized (via nonpayment for the additional
costs of treating them) when they occur.
149
The prevention of HAIs could be considered a form of RMT because it is a matter of
reducing the risk of HAIs occurring rather than taking actions guaranteed to eliminate them.
Given this, nonpayment for the additional costs associated with HAIs is somewhat analogous to
the shared savings and losses scenario in the simulation. Even though the payment mechanism is
different, it still involves providers investing in the prevention of future medical needs and patient
harm with the possibility t h at t h ey ’ll su s t ain f in an c ial loss es , even if that prevention works as
intended, due to random chance. Because of this, the results of the simulation suggest that this
nonpayment for the additional costs of HAIs likely increases provider financial risk.
Providers could react to this increased financial risk in a number of ways. It is doubtful
that providers would scale back HAI reduction efforts that they have already begun. However, the
increase in financial risk involved in such efforts may dissuade other providers from seriously
investing in them in the first place. For instance, there has been a very mild decrease in the rate of
these HAIs since Medicare implemented its nonpayment policy, but there is some evidence that
this may reflect changes in hospital coding rather than genuine investment in interventions to
prevent HAIs (Kawai et al., 2015). After all, why bill for additional costs related to HAIs if you
know that Medi c ar e w on ’ t r eimbu rs e y ou f or t h em an yw ay ?
But even if, eventually, professional norms and organizations like the Joint Commission
succeed in pushing more providers to adopt substantive, evidence-based efforts to reduce HAIs
despite the increased financial risk caused by nonpayment for them, there are still two potential
unintended consequences that remain. First, providers may react to this financial risk by
in c re as in g p at ien t de m an d f or ot h er c ar e t o c om p en sa t e, c o n t rib u t in g t o t h e “ ove ru se ” of medic al
services that is itself an area of concern for patient safety advocates due to the potential harm
associated with it (Chassin, 2013). And second, providers may attempt to avoid high risk patients,
such as those with multiple comorbidities who are more likely to contract HAIs, making it more
difficult for such patients to access care.
HAI prevention is just one example in patient safety where nonpayment or reduced
payment for undesirable but not-completely-preventable patient outcomes has gained in
popularity. Another example relates to potentially preventable hospital readmissions. Medicare,
through its Readmissions Reduction Program, has been reducing payments to hospitals deemed
to have excessive readmissions within 30 days post-discharge (CMS, 2016e). Medicare has also
implemented bundled payments for some surgical procedures, such as hip and knee replacement,
that cover a period of 90 days post-discharge (CMS, 2016a). Both of these policies are intended to
penalize hospitals which have readmissions within a certain timeframe after patients are
discharged and encourage them to adopt readmission-prevention interventions. But because
these readmissions are not completely preventable, the same fears are applicable as in HAI
prevention. This nonpayment or reduced payment for potentially preventable readmissions may
increase provider financial risk and encourage providers to either increase patient demand, lower
care quality, or potentially even avoid complex patients who are more likely to be readmitted.
In effect, these nonpayment mechanisms for HAIs and readmissions are a form of blunt
financial incentives when, in accordance with the Berwick (1996) quote at the beginning of this
chapter, an emphasis on rules would probably be more effective. If indeed current evidence-based
150
practice can, on average, reduce HAIs and readmissions, why not just mandate such practice
through policy? And since high quality HAI and readmissions prevention typically involves a
certain amount of costs to health care providers, it would make some sense for payers to pay for
this prevention up front, similar to the up-front payments for RMT in the simulation. Since
hospitals are typically paid for surgical episodes in standardized DRG payments, payers could
designate a portion of those payments as conditional on the provision of HAI prevention as long
as that portion is sufficient to cover the costs. Paying for HAI prevention up front and getting rid
of the policy of nonpayment when HAIs and readmissions occur could help create a rules-based
culture, rather than an incentives based one, that involves less financial risk to providers (and its
unintended consequences) and more actual compliance with high quality evidence-based care,
which is ultimately all these payment reforms are trying t o ac h iev e. I sn’t t h is t h e p oin t of comparative effectiveness research after all; to discover the most effective interventions,
det er m in e w h ic h on es w e’r e w illin g t o p ay f or, and t h en s im p ly p ay f or t h em?
Ultimately, I fear that this focus on fostering safety through financial incentives based on
patient outcomes can distract from the ways in which payment can be used to create a health care
system that is more vertically aligned toward prevention at all levels. The story of the hospital that
helped reduce c h ildr en ’s b ic yc le ac c ident s an d s u f f er ed f in an c ially f or it is n o isolat ed exa m p le . There are many such prevention paradoxes in our current health care provider payment systems.
It is certainly possible and necessary for the health care system to try to reduce, say, the number
of HAIs t h at oc c u r p er p a t ien t en c oun t er . But it ’s p oss ib le an d n ec es sa ry f or t h e h ealt h c ar e
system to try to reduce the number of patient encounters too. Many patient safety advocates have
ac kno w ledg ed h ow “ ove r u se ” c an lead t o m o re p at ien t h ar m , b u t s o c an j u st “ u se ” . Red u c in g bot h would also reduce patient safety-related harm, with the bonus of preventing many patients from
be in g p u t at r is k of t h at h ar m in t h e f irs t p lac e. I t ’s im p er at iv e t h at w e c on si der h ow t o red es ign provider payment systems to encourage the holistic prevention of patient harm in this way.
5.3 - Limitations
5.3.1 - Simulation Model Limitations
The simulation model examines provider financial risk that is due solely to variation in
patient needs over time. As noted previously, another potential source of financial risk that is not
examined is market-based or competitive risk. This kind of risk arises due to the complex
interactions between a p r ovid er ’s mar ket share, perceived quality, and the flow of patients from
one provider to another in a health care market. This kind of risk would also differ significantly
from one health care market to another depending on the number of providers and payers, as well
as idiosyncratic factors such as regional provider culture. Nevertheless, it is likely that this source
of provider financial risk would exist on top of the baseline amount of financial risk that stems
from variation in patient needs. Thus, the simulation model is examining the minimum likely
151
difference, from one payment system to another, in their influence on provider financial risk, a
difference that would probably exist regardless of the particular market situation.
In addition, though the simulation uses risk-averse provider manipulation of patient
demand as a proxy for measuring financial risk, it does not consider the potential role of profit
seeking behavior in also generating this behavior. Once again though, any profit seeking motive
that results in provider manipulation of patient demand would likely be additive to similar
behaviors in response to financial risk. So the model provides an idea of the baseline difference in
provider financial risk between different payment systems that should be expected regardless of
market conditions and provider altruism (or lack thereof).
F u rt h er m ore, al l o f t h e p rov ider ’s p at ien t s in t h e s im u lat ion ar e c on si der ed t o be u n der the same payment system during the course of each run. In other words, when the simulation is
testing a 5,000 patient population for a provider paid via capitation, it is assumed that the
provider is receiving capitated payments for all 5,000 patients. In real life, providers typically
interact with many payers who might pay differently for each of their patients. This is especially
the case with Medicare; a provider might receive shared savings payments from Medicare for its
Medicare ACO patients, while at the same time it receives normal FFS patients for its private
payer patients. Experts have noted potential problems when providers have these mixtures of
different payments for different patients. It may be the case, for instance, that in order for
alt er n at iv e p ay m en t s ys t ems f or p rimar y c ar e p rov ider s t o en c our age t h eir de si re d new “ n or m s”
for care, the providers would need to receive these alternative payments for some critical
minimum fraction of their patients (Landon, 2017). This is where there is another promising
aspect of the CPC+ initiative mentioned earlier. This initiative attempts to bring on board other
private payers, in addition to Medicare, who will agree to use the same alternative payment
structure so that the care of a greater proportion of the patients of participating primary care
providers is being paid for in this way. Nevertheless, this is still a persistent problem with
payment reform initiatives, especially in regions with many different payers, and it would
certainly be an impediment for a provider attempting to implement (and be reimbursed for) an
RMT intervention.
Finally, the simulation uses demand manipulation in response to anticipated financial
shortfalls as a proxy for financial risk, though as the last special scenario in the previous chapter
makes clear, providers also have the option of changing their cost structure to reduce risk. It is
difficult to say how much providers would do of one or the other in response to perceived
financial risk, and there are almost certainly limitations to both. Nevertheless, the special scenario
showed – for providers implementing a specific reduction in fixed costs (and a corresponding
increase in variable costs) – how their incentive to engage in risk-averse manipulation of patient
demand would change. If capitated providers were able to they would probably change their cost
structure in this way, although they would never be able to completely eliminate the incentive to
manipulate patient demand in response to financial risk. FFS providers, on the other hand, would
probably not bother with this change.
152
5.3.2 - General Limitations / Caveats
The simulation results suggest that paying for RMT up front might be a better way to pay
for prevention than paying for RMT out of the savings it produces (which are variable and may
not always cover its costs). But in order to pay for RMT up front there would ideally need to be
some consensus about what constitutes good RMT. There is growing consensus about the types of
treatment and care delivery structures that are valuable in chronic care management, for
instance, but even this oldest of RMT-related care activities is still going through a lot of
innovation and change. In order to eventually have a system where we can pay specifically for
RMT, it would be helpful to have a better idea of the spectrum of RMT interventions that are
currently out there, what their main components are, and how much they appear to reduce
patient risk of future needs and harm, on average. It should be noted how much these
in t er v en t ion s c ost as w ell, b u t at t h is p oin t t h at in f orm at io n is n ’t as u se f u l as it ma y seem. All new
products and services generally cost more early in their development than they do when they
h av e mat u re d. I t ’s u su ally imp ort an t t o f irs t f in d p r odu c t s an d s er v ic es t h a t w ork, an d t h e n af t er that innovators can figure out how to improve economies of scale in order to deliver them more
cost effectively. Focusing on the early cost of these interventions might create unnecessary
p es si m is m t owa rd t h em. I n st ead of “ in t er v en t ion r edu c es medic al r is k bu t t h e c ost r es u lt s in n o
n et p ay er s av in gs” it sh ould b e “ in t er v en t ion r edu c es r is k, m ak es p at ien t s h ealt h ier , all w it h out in c re as in g t h e c ost t o p at ien t s an d s ocie t y” . Righ t n ow , b u ildin g t h is c on se n su s on w h at k in ds of RMT interventions there are and which ones are most effective at reducing patient medical risk
(n ot n ec es sa rily w h ic h on es ar e mo st “ v alu ab le” ) s h ould b e a maj or go al t o h elp mo v e t owa rd a
scenario where we can eventually be confident in paying for RMT directly.
Even once there is greater consensus on the standards for good RMT practice, however,
there will still likely need to be a push at the policy level to mandate delivery of it. The simulation
model assumes that the RMT interventions tested are fast acting, so as to come up with a best
case scenario for the amount of provider financial risk associated with implementing them.
However, it is likely that some RMT interventions will take longer to reduce patient medical risk
than others. And no matter how long it takes, there is always the risk that a provider will start
RMT for a patient only for that patient to leave for a different provider while the RMT is still
ongoing. If only a fraction of providers in a region are actually practicing RMT, then this reduces
the effectiveness of RMT overall since providers will more often than not have to start from
sc ra t c h in r edu c in g t h eir n ew p at ien t s’ medic al r is k. W h er eas if m ost p rov ider s ar e p ra c t i c in g
RM T , t h en p at ien t t u rno v er is less of a p roblem bec au se a p at ien t ’s n ew p rov ider c an j u st p ic k u p where the old provider left off.
As a corollary to mandating delivery of RMT, there will also likely need to be a policy push
to mandate payment for RMT and to protect RMT funding from being used for non-RMT
ac t iv it ies . M an y h ealt h s y st em exp er t s h av e n ote d h ow t h e “ t yr an n y of t h e u rgen t ” c an r es ult in
the health care system constantly needing to devote resources to pressing acute care needs while
neglecting the preventive care that could help avoid them in the first place (Berenson and Rich,
2010). One group of system analysts even created a simulation model showing how easy it is for
153
curative services to crowd out preventive ones (Bishai, Paina, Li, Peters, & Hyder, 2014). The
solut ion ? T h er e n eed s t o be a “ ring f en c e” p lac ed ar oun d f u n din g f or RM T i n t er v en t ion s s o t h at , even in times of high acute care demand, these RMT interventions are still delivered at a normal
level, similar to how we try to protect funding for education even during a recession.
Despite these current general limitations, there is still a lot that individual researchers,
providers, and payers can continue to do now, together, to advance RMT even in the absence of
more systemic policy geared toward prevention. Research must continue on the effectiveness of
RMT interventions. Given the nature of risk, it will either take longer or require larger studies in
order to determine the effectiveness of risk-reducing interventions, but these studies must
continue regardless. It is probably fair to expect that an RMT intervention deliver at least a 1 to 1
ROI (i.e., it is cost neutral), but beyond that I believe it is still premature to be evaluating these
interventions on their cost effectiveness. After all, even if an RMT intervention is only cost
n eu t ra l, is n ’t it s t ill a grea t t h in g t h at p at ien t r is k h as b een r edu c ed w it h n o i n c re as e in costs to
p ay er s an d p at ien t s? On c e RM T in t er v en t ion s h av e b ec om e m ore s t an da rd i zed and w e’v e gai n ed
a better idea of their effectiveness at reducing patient medical risk, only then will it make more
sense to ramp up investing in improving the value and cost effectiveness of them. Certainly we
can start improving cost effectiveness now as well, but given the early stage of these interventions
patience will be a virtue.
5.4 - Future Research
Probably the best immediate next step for this research would be to quantify the amount
of financial risk when implementing RMT due to competitive/market forces, in addition to the
financial risk from variation in patient needs quantified in the current simulation model. This
would involve expanding the model to c re at e a t ru e “ c lo se d loo p ” s ys t em, w h er eb y p at ien t s,
instead of disappearing into the ether and appearing out of nowhere when there is patient
turnover, actually move from one provider to another within a closed system. Patients could still
be risk selected in such a system, though there would be limitations as to how easily FFS
providers could accumulate the sickest patients, as it would be impossible, for instance, for all
providers in a closed system to increase the amount of sick patients in their population at the
same time. It would be easier, however, for a lot of providers to risk select downward (i.e., avoid
sicker patients) as this simply involves reducing access to these patients, some of whom would
presumably have difficulty finding treatment with any provider even in a closed system. Providers
in this closed system might also be able to reduce their costs and quality, in addition to or
perhaps instead of manipulating patient demand for care, in response to financial risk. Though
because of the in f orm at ion as ym m et rie s in h ealt h c ar e it ’s u n c lear h ow muc h a dec re as e in q u alit y
w ould r es u lt in c h an ges i n p at ien t de m a n d f or a p rov ider ’s s er v ic es . Som e a s su m p t ion s w ould
have to be made about which behaviors would be more likely to produce poor market outcomes
for a provider so that they could choose the least bad of those options. Given that reducing
quality could also potentially lead to more patient needs in the future, this expanded simulation
154
model could also look at the potentially detrimental dynamics of market competition in that
respect.
It may also be useful to add a mix of profit-seeking and more altruistic providers to the
model in the future. However, this may not be as relevant to the main question at hand – how to
reduce barriers to RMT – unless we have reason to believe that profit-seeking providers are more
or less likely to pursue RMT. If they are less likely, experimenting with payment for RMT to
ensure that more profit-oriented providers have a desire to provide it could be a useful area for
future simulation research. In the end, though, I suspect that broad implementation of RMT
interventions will likely require policy mandates as much as it will require sustainable payments,
so making sure that profit-oriented providers in particular are incentivized to provide RMT might
be an unnecessary effort.
Finally, the current simulation model assumes a linear relationship between expected
financial shortfalls and resulting provider risk- av er se b eh av ior. I t ’s p roba ble t h ough, bas ed o n behavioral economics and psychology research that has gained traction over the past few decades,
that higher expected financial shortfalls may produce a greater than expected (according to a
linear relationship) provider reaction. It would be interesting to create such a utility-based model
of provider risk-averse behavior. The tricky part would be in determining the utility curve for
p rov ider r is k av er si on , an d I ’m n ot su re if t h at h as be en de t er m i n ed ye t in t h e h ealt h s er v ic es literature.
5.5 - Conclusion
This study presents a body of theory aided by simulation that suggests that payment
systems for health care providers: 1) can be a direct impediment to the prevention of medical
needs and patient harm in their own right; 2) can be an indirect impediment to this prevention
through the creation of provider financial risk; and 3) can be specifically designed alleviate both of
these impediments. Changing the payment system in this way would not be a panacea. There is
still a lot of work to be done to discover both the most effective preventive interventions as well
as how to actually implement them across the health care system. But we still need payment
p u sh in g in t h e r ight di re c t ion . “ H ealt h c ar e a s if h e alt h mat t er ed” is n ec es sa r y an d a c h iev ab le
(Frieden & Mostashari, 2008). Changing provider payment systems to fully align with that mantra
would be a great step forward.
155
References
Acheson, R. M. (1978). The Definition and Identification of Need for Health Care. Journal of
Epidemiology and Community Health, 32(1), 10 –15.
Agency for Healthcare Research and Quality [AHRQ]. (2015). Medical Expenditure Panel Survey,
Household Component, 2013. Retrieved from http://meps.ahrq.gov
Agency for Healthcare Research and Quality [AHRQ]. (2016a). Never Events. Retrieved October 3,
2016, from https://psnet.ahrq.gov/primers/primer/3/never-events
Agency for Healthcare Research and Quality [AHRQ]. (2016b). National Healthcare Quality and
Disparities Report: Chartbook on Patient Safety. Retrieved from
http://www.ahrq.gov/research/findings/nhqrdr/chartbooks/patientsafety/index.html
Anderson, D. J., Podgorny, K., Berríos-Torres, S. I., Bratzler, D. W., Patchen Dellinger, E., Greene,
L. , … K ay e, K . S . ( 2 0 14 ) . St ra t egies t o P re v en t Su rgic al Sit e I n f ec t ion s in Ac u t e Car e
Hospitals: 2014 Update. Infection Control & Hospital Epidemiology, 35(6), 605 –627.
http://doi.org/10.1086/676022
An der so n , G . F . , Rei n h ar dt , U . E . , Hus se y, P . S . , & P et rosy an , V . (2 0 0 3) . I t ’s t h e P ric es , St u p id:
Why the United States Is So Different from Other Countries. Health Affairs, 22(3), 89 –105.
doi:10.1377/hlthaff.22.3.89
Armour, B. S., Pitts, M. M., Maclean, R., Cangialose, C., Kishel, M., Imai, H., & Etchason, J. (2001).
The Effect of Explicit Financial Incentives on Physician Behavior. Archives of Internal
Medicine, 161, 1261 –1266.
Arrow, K. J. (1963). Uncertainty and the Welfare Economics of Medical Care. The American
Economic Review, 53(5), 941 –973.
Aslakson, R., Cheng, J., Vollenweider, D., Galusca, D., Smith, T. J., & Pronovost, P. J. (2014).
Evidence-Based Palliative Care in the Intensive Care Unit: A Systematic Review of
Interventions. Journal of Palliative Medicine, 17(2), 219 –235.
http://doi.org/10.1089/jpm.2013.0409
Awad, S., Varadhan, K. K., Ljungqvist, O., & Lobo, D. N. (2013). A Meta-Analysis of Randomised
Controlled Trials on Preoperative Oral Carbohydrate Treatment in Elective Surgery.
Clinical Nutrition, 32(1), 34 –44. http://doi.org/10.1016/j.clnu.2012.10.011
Bader, A., & Hepner, D. L. (2009). The Role of the Preoperative Clinic in Perioperative Risk
Reduction. International Anesthesiology Clinics, 47(4), 151 –160.
http://doi.org/10.1097/AIA.0b013e3181b44311
Berenson, R. A., & Burton, R. A. (2012). Accountable Care Organizations in Medicare and the
Private Sector: A Status Update.
Berenson, R. A., & Rice, T. (2015). Beyond Measurement and Reward: Methods of Motivating
Quality Improvement and Accountability. Health Services Research, 50(S2), 2155 –2186.
http://doi.org/10.1111/1475-6773.12413
Berenson, R. A., & Rich, E. C. (2010). US Approaches to Physician Payment: The Deconstruction of
156
Primary Care. Journal of General Internal Medicine, 25(6), 613 –618. doi:10.1007/s11606-010-
1295-z
Berwick, D. M. (1996). Payment by Capitation and the Quality of Care. New England Journal of
Medicine, 335(16), 1227 –1231.
Berwick, D. M., & Hackbarth, A. D. (2012). Eliminating waste in US health care. Journal of the
American Medical Association, 307(14), 1513 –1516. doi:10.1001/jama.2012.362
Bishai, D., Paina, L., Li, Q., Peters, D. H., & Hyder, A. A. (2014). Advancing the Application of
Systems Thinking in Health: Why Cure Crowds Out Prevention. Health Research Policy
and Systems, 12(28), 1 –12. http://doi.org/10.1186/1478-4505-12-28
Bodenheimer, T., & Pham, H. H. (2010). Primary Care: Current Problems and Proposed Solutions.
Health Affairs, 29(5), 799 –805. doi:10.1377/hlthaff.2010.0026
Bodenheimer, T., Berenson, R. A., & Rudolf, P. (2007). The Primary Care –Specialty Income Gap:
Why It Matters. Annals of Internal Medicine, 146(4), 301 –306.
Bodenheimer, T., Wagner, E. H., & Grumbach, K. (2002). Improving Primary Care for Patients
with Chronic Illness. Journal of the American Medical Association, 288(14), 1775 –1779.
Briggs, A., Nixon, R., Dixon, S., & Thompson, S. (2005). Parametric Modelling of Cost Data: Some
Simulation Evidence. Health Economics, 14(4), 421 –428. http://doi.org/10.1002/hec.941
Brosig-Koch, J., Hennig-Schmidt, H., Kairies-Schwarz, N., & Wiesen, D. (2015). Using Artefactual
Field and Lab Experiments to Investigate How Fee-for-Service and Capitation Affect
Medical Service Provision. Journal of Economic Behavior and Organization.
http://doi.org/10.1016/j.jebo.2015.04.011
Brown, J., Duggan, M., Kuziemko, I., & Woolston, W. (2011). How Does Risk Selection Respond to
Risk Adjustment? Evidence From the Medicare Advantage Program (No. 16977). NBER
Working Paper.
Brown, R. S., Peikes, D., Peterson, G., Schore, J., & Razafindrakoto, C. M. (2012). Six Features of
Medicare Coordinated Care Demonstration Programs that Cut Hospital Admissions of
High-Risk Patients. Health Affairs, 31(6), 1156 –1166. http://doi.org/10.1377/hlthaff.2012.0393
Casalino, L. P. (1992). Balancing Incentives: How Should Physicians Be Reimbursed? Journal of the
American Medical Association, 267(3), 403 –405. doi:10.1001/jama.1992.03480030081042
Centers for Disease Control and Prevention [CDC]. (2016a). Chronic Disease Overview. Retrieved
March 30, 2016, from http://www.cdc.gov/chronicdisease/overview/index.htm
Centers for Disease Control and Prevention [CDC]. (2016b). FastStats - Life Expectancy. Retrieved
March 30, 2016, from http://www.cdc.gov/nchs/fastats/life-expectancy.htm
Centers for Medicare & Medicaid Services [CMS]. (2015a). Hospital-Acquired Conditions.
Retrieved April 14, 2017, from https://www.cms.gov/Medicare/Medicare-Fee-for-Service-
Payment/HospitalAcqCond/Hospital-Acquired_Conditions.html
Centers for Medicare & Medicaid Services [CMS]. (2015b). National Health Expenditure Data:
Historical. Retrieved from https://www.cms.gov/research-statistics-data-and-
systems/statistics-trends-and-
reports/nationalhealthexpenddata/nationalhealthaccountshistorical.html
Centers for Medicare & Medicaid Services [CMS]. (2015c). Shared Savings and Losses and
157
Assignment Methodology. Retrieved from https://www.cms.gov/Medicare/Medicare-Fee-
for-Service-Payment/sharedsavingsprogram/Downloads/Shared-Savings-Losses-
Assignment-Spec-V4.pdf
Centers for Medicare & Medicaid Services [CMS]. (2016a). Comprehensive Care for Joint
Replacement Model. Retrieved June 13, 2016, from
https://innovation.cms.gov/initiatives/cjr
Centers for Medicare & Medicaid Services [CMS]. (2016b). Final Medicare Shared Savings Program
Rule (CMS-1644-F). Retrieved April 10, 2017, from
https://www.cms.gov/Newsroom/MediaReleaseDatabase/Fact-sheets/2016-Fact-sheets-
items/2016-06-06.html
Centers for Medicare & Medicaid Services [CMS]. (2016c). Frequently Asked Questions about
Billing the Physician Fee Schedule for Advance Care Planning Services. Retrieved October
3, 2016, from https://www.cms.gov/Medicare/Medicare-Fee-for-Service-
Payment/PhysicianFeeSched/Downloads/FAQ-Advance-Care-Planning.pdf
Centers for Medicare & Medicaid Services [CMS]. (2016d). Medicare Program; Advancing Care
Coordination Through Episode Payment Models (EPMs); Cardiac Rehabilitation Incentive
Payment Model; and Changes to the Comprehensive Care for Joint Replacement Model
(CJR). Retrieved from https://innovation.cms.gov/Files/x/advancing-care-coordination-
nprm.pdf
Centers for Medicare & Medicaid Services [CMS]. (2016e). Readmissions Reduction Program
(HRRP). Retrieved October 3, 2016, from https://www.cms.gov/medicare/medicare-fee-
for-service-payment/acuteinpatientpps/readmissions-reduction-program.html
Cha ss in , M . R. (2 0 13) . I m p rov in g Th e Qu alit y of H ealt h Car e: Wh at ’s T ak in g So L on g? Health
Affairs, 32(10), 1761 –1765. doi:10.1377/hlthaff.2013.0809
Chassin, M. R., & Loeb, J. M. (2013). High-Reliability Health Care: Getting There from Here. The
Milbank Quarterly, 91(3), 459 –490. http://doi.org/10.1111/1468-0009.12023
Chernew, M., McGuire, T., & McWilliams, J. M. (2014). Refining the ACO Program: Issues and
Options.
Cohen, S. B. (2015). The Concentration and Persistence in the Level of Health Expenditures over
Time: Estimates for the U.S. Population, 2012-2013. Agency for Healthcare Research and
Quality, Rockville, MD. Retrieved from
http://meps.ahrq.gov/mepsweb/data_files/publications/st481/stat481.pdf
Conrad, D. A. (2015). The Theory of Value-Based Payment Incentives and Their Application to
Health Care. Health Services Research, 50, 2057 –2089. http://doi.org/10.1111/1475-
6773.12408
Cuc c iar e, M . A. , & O’D on ohu e, W . ( 2 0 0 6) . P re dic t in g F u t u re Healt h c ar e Cost s: H ow W ell D oes
Risk-Adjustment Work? Journal of Health Organization and Management, 20(2), 150 –162.
doi:10.1108/14777260610661547
Dartmouth Atlas of Health Care. (2016). Retrieved February 21, 2016, from
http://www.dartmouthatlas.org/
De Jonge, K. E., Jamshed, N., Gilden, D., Kubisiak, J., Bruce, S. R., & Taler, G. (2014). Effects of
158
Home-Based Primary Care on Medicare Costs in High-Risk Elders. Journal of the American
Geriatrics Society, 62(10), 1825 –1831. http://doi.org/10.1111/jgs.12974
DeLia, D. (2016). Monte Carlo Analysis of Payer and Provider Risks in Shared Savings
Arrangements. Medical Care Research and Review, 73(5), 511 –531.
http://doi.org/10.1177/1077558715618320
DeLia, D., Hoover, D., & Cantor, J. C. (2012). Statistical Uncertainty in the Medicare Shared
Savings Program. Medicare & Medicaid Research Review, 2(4), 1 –16.
http://doi.org/10.5600/mmrr.002.04.a04
Diehr, P., Yanez, D., Ash, A., Hornbrook, M., & Lin, D. Y. (1999). Methods for Analyzing Health
Care Utilization and Costs. Annual Review of Public Health, 20, 125 –144.
Donabedian, A. (1966). Evaluating the Quality of Medical Care. The Milbank Memorial Fund
Quarterly, 44(3), 166 –203. Retrieved from http://www.jstor.org/stable/10.2307/3348969
Donabedian, A. (1974). Aspects of Medical Care Administration. Cambridge, MA: Harvard
University Press.
Douven, R., McGuire, T. G., & McWilliams, J. M. (2015). Avoiding Unintended Incentives in ACO
Payment Models. Health Affairs, 34(1), 143 –149. http://doi.org/10.1377/hlthaff.2014.0444
Dranove, D., Simon, C. J., & White, W. D. (2002). Is Managed Care Leading to Consolidation in
Health-care Markets? Health Services Research, 37(3), 573 –594.
Ducel, G., Fabry, J., & Nicolle, L. (2002). Prevention of Hospital-Acquired Infections: A Practical
Guide. World Health Organization. doi:WHO/CDS/CSR/EPH/2002.12
Duncan, I., Loginov, M., & Ludkovski, M. (2016). Testing Alternative Regression Frameworks for
Predictive Modeling of Health Care Costs. North American Actuarial Journal, 20(1), 65 –87.
http://doi.org/10.1080/10920277.2015.1110491
Ellis, R. P., Fiebig, D. G., Johar, M. J., Jones, G., & Savage, E. (2013). Explaining Health Care
Expenditure Variation: Large-Sample Evidence Using Linked Survey and Health
Administrative Data. Health Economics, 22(9), 1093 –1110.
Ellwood, P. M., Anderson, N. N., Billings, J. E., Carlson, R. J., Hoagberg, E. J., & McClure, W. (1971).
Health Maintenance Strategy. Medical Care, 9(3), 291 –298.
Farley, T. A., Dalal, M. A., Mostashari, F., & Frieden, T. R. (2010). Deaths Preventable in the U.S.
by Improvements in Use of Clinical Preventive Services. American Journal of Preventive
Medicine, 38(6), 600 –609. doi:10.1016/j.amepre.2010.02.016
Fisher, E. S., McClellan, M. B., Bertko, J., Lieberman, S. M., Lee, J. J., Lewis, J. L., & Skinner, J. S.
(2009). Fostering Accountable Health Care: Moving Forward in Medicare. Health Affairs,
28(2), w219 –w231. doi:10.1377/hlthaff.28.2.w219
Fortmann-Roe, S. (2014). Insight Maker: A General-Purpose Tool for Web-Based Modeling &
Simulation. Simulation Modelling Practice and Theory, 47, 28 –45.
http://doi.org/10.1016/j.simpat.2014.03.013
Frakt, A. B., & Mayes, R. (2012). Beyond Capitation: How New Payment Experiments Seek To find
t h e “ Swe et Sp ot ” in Amo u n t of Ris k P rov ider s an d P ay er s Bear . Health Affairs, 31(9), 1951 –
1958. doi:10.1377/hlthaff.2012.0344
Freund, T., Campbell, S. M., Geissler, S., Kunz, C. U., Mahler, C., Peters-Klimm, F., & Szecsenyi, J.
159
(2013). Strategies for Reducing Potentially Avoidable Hospitalizations for Ambulatory
Care-Sensitive Conditions. Annals of Family Medicine, 11(4), 363 –370. doi:10.1370/afm.1498
Friedberg, M. W., Buendia, A. M., Lauderdale, K. E., & Hussey, P. S. (2013). Option Pricing: A
Flexible Tool to Disseminate Shared Savings Contracts. The American Journal of Managed
Care, 19(8), e285 –e292.
Frieden, T. R., & Mostashari, F. (2008). Health Care As If Health Mattered. Journal of the American
Medical Association, 299(8), 950 –952. http://doi.org/10.1001/jama.299.8.950
Fries, J. F. (1980). Aging, Natural Death, and the Compression of Morbidity. The New England
Journal of Medicine, 303(3), 130 –135.
Fries, J. F., Bruce, B., & Chakravarty, E. (2011). Compression of Morbidity 1980-2011: A Focused
Review of Paradigms and Progress. Journal of Aging Research, 2011. doi:10.4061/2011/261702
F rie s, J . F . , K oo p , C. E . , Be ad le, C . E . , Co op er , P . P . , E n gl a n d, M . J . , G re av es , R . F . , … T h e Hea lt h Project Consortium. (1993). Reducing Health Care Costs by Reducing the Need and
Demand for Medical Services. The New England Journal of Medicine, 329(5), 321 –325.
G allagher , T . H. , St . P et er , R. F . , Ch es n ey , M . , & L o, B. ( 2 0 0 1) . P at ien t s’ At t it u des T owa rd Cost Control Bonuses for Managed Care Physicians. Health Affairs, 20(2), 186 –192.
doi:10.1377/hlthaff.20.2.186
Gapenski, L. C., & Langland-Orban, B. (1996). Impact of Capitation Contracts on Financial Risk: A
Monte Carlo Simulation. Health Services Management Research, 9, 183 –191.
Gapenski, L. C., & Langland-Orban, B. (1998). The Financial Risk to Hospitals Inherent in DRG,
Per Diem, and Capitation Reimbursement Methodologies. Journal of Healthcare
Management, 43(4), 323 –338.
Gawande, A. (2009, June). The Cost Conundrum. The New Yorker. Retrieved from
http://www.newyorker.com/magazine/2009/06/01/the-cost-conundrum
Gawande, A. (2011, January). The Hot Spotters. The New Yorker, 41 –51. Retrieved from
http://www.newyorker.com/magazine/2011/01/24/the-hot-spotters
Gawande, A. (2014). Being Mortal. New York, NY: Metropolitan Books.
Gaziano, J. M. (2010). Fifth Phase of the Epidemiologic Transition: The Age of Obesity and
Inactivity. Journal of the American Medicine Association, 303(3), 275.
doi:10.1001/jama.2009.2025
Ghaferi, A. A., Birkmeyer, J. D., & Dimick, J. B. (2009). Variation in Hospital Mortality Associated
with Inpatient Surgery. The New England Journal of Medicine, 361(14), 1368 –1375.
Ginsburg, P. B. (2008). High and Rising Health Care Costs: Demystifying US Health Care
Spending. Retrieved from http://www.rwjf.org/en/research-publications/find-rwjf-
research/2008/10/high-and-rising-health-care-costs.html
Ginsburg, P. B. (2013). Achieving Health Care Cost Containment Through Provider Payment
Reform That Engages Patients And Providers. Health Affairs, 32(5), 929 –934.
doi:10.1377/hlthaff.2012.1007
G in sb u rg, P . B . , & Bere n s on , R. A . ( 2 0 0 7 ). Rev is in g Me dic ar e’s P h ys ic ian F ee Sc h edu le: M u c h Activity, Little Change. New England Journal of Medicine, 356(12), 1201 –1203.
Goldman, D. P., Cutler, D., Rowe, J. W., Michaud, P.-C., Sullivan, J., Peneva, D., & Olshansky, S. J.
160
(2013). Substantial Health and Economic Returns from Delayed Aging May Warrant a New
Focus for Medical Research. Health Affairs, 32(10), 1698 –1705. doi:10.1377/hlthaff.2013.0052
Goldman, D. P., Gaudette, É., & Cheng, W.-H. (2016). Competing Risks: Investing in Sickness
Rather Than Health. American Journal of Preventive Medicine, 50(5), S45 –S50.
doi:10.1016/j.amepre.2015.12.005
Gruenberg, E. M. (1977). The Failures of Success. The Milbank Memorial Fund Quarterly. Health
and Society, 55(1), 3 –24.
Harrison, J. D., Young, J. M., Butow, P. N., & Solomon, M. J. (2013). Needs in Health Care: What
Beast is That? International Journal of Health Services, 43(3), 567 –585.
doi:http://dx.doi.org/10.2190/HS.43.3.I
Hellinger, F. J. (1996). The Impact of Financial Incentives on Physician Behavior in Managed Care
Plans: A Review of the Evidence. Medical Care Research and Review, 53(3), 294 –314.
doi:10.1177/107755879605300305
Hillman, A. L. (1990). Health Maintenance Organization s, F in a n c ial I n c en t i v es , and P h ys ic ians’
Judgements. Annals of Internal Medicine, 112(12), 891 –893.
Hillman, A. L. (1991). Managing the Physician: Rules Versus Incentives. Health Affairs, 10(4), 138 –
146. doi:10.1377/hlthaff.10.4.138
Hillman, A. L., Pauly, M. V., & Kerstein, J. J. (1989). How Do Financial Incentives Affect
P h ys ic ians’ Cli n ic al Dec is ion s a n d t h e Fin a n c ial P e rf orm an c e of Hea lt h M ainte n a n c e
Organizations? The New England Journal of Medicine, 321(2), 86 –92.
Himmelstein, D. U., & Woolhandler, S. (2014). Global Amnesia: Embracing Fee-For-Non-Service —
Again. Journal of General Internal Medicine, 29(5), 693 –695. http://doi.org/10.1007/s11606-
013-2745-1
Himmelstein, D. U., Ariely, D., & Woolhandler, S. (2014). Pay-for-Performance: Toxic to Quality?
Insights from Behavioral Economics. International Journal of Health Services, 44(2), 203 –
214. http://doi.org/10.2190/HS.44.2.a
Hsu , J . , P ric e, M . , Sp irt , J. , Vo ge li, C. , Bra n d , R . , Ch e rnew, M . E . , … F er ris , T . ( 2 0 16 ) . P at ie n t Population Loss at a Large Pioneer Accountable Care Organization and Implications for
Refining the Program. Health Affairs, 35(3), 422 –430.
http://doi.org/10.1377/hlthaff.2015.0805
Institute of Medicine [IOM]. (2015). Dying in America: Improving Quality and Honoring Individual
Preferences Near the End of Life. Washington, D.C.: National Academies Press.
http://doi.org/10.17226/18748
James, J. T. (2013). A New, Evidence-Based Estimate of Patient Harms Associated with Hospital
Care. Journal of Patient Safety, 9(3), 122 –128. http://doi.org/10.1097/PTS.0b013e3182948a69
Jones, D. S., Podolsky, S. H., & Greene, J. a. (2012). The Burden of Disease and the Changing Task
of Medicine. The New England Journal of Medicine, 366(25), 2333 –2338.
doi:10.1056/NEJMp1113569. Retrieved from
http://www.nejm.org/doi/full/10.1056/NEJMp1113569
Joynt, K. E., Gawande, A. A., Orav, E. J., & Jha, A. K. (2013). Contribution of preventable acute care
spending to total spending for high-cost Medicare patients. Journal of the American
161
Medical Association, 309(24), 2572 –2578. http://doi.org/10.1001/jama.2013.7103
K ais er F a m ily F oun da t io n [K F F ] . ( 2 0 16 ) . 1 0 F AQs: Me dic ar e’s Role in E n d -of-Life Care. Retrieved
October 3, 2016, from http://kff.org/medicare/fact-sheet/10-faqs-medicares-role-in-end-
of-life-care/
Kao, A. C. (2015). Driven to Care: Aligning External Motivators with Intrinsic Motivation. Health
Services Research, 50(S2), 2216 –2222. doi:10.1111/1475-6773.12422
Kash, B. A., Zhang, Y., Cline, K. M., Menser, T., & Miller, T. R. (2014). The Perioperative Surgical
Home (PSH): A Comprehensive Review of US and Non-US Studies Shows Predominantly
Positive Quality and Cost Outcomes. The Milbank Quarterly, 92(4), 796 –821.
http://doi.org/10.1111/1468-0009.12093
Kawai, A. T., Calderwood, M. S., Jin, R., Soumerai, S. B., Vaz, L. E., Goldmann, D., & Lee, G. M.
(2015). Impact of the Centers for Medicare and Medicaid Services Hospital-Acquired
Conditions Policy on Billing Rates for 2 Targeted Healthcare-Associated Infections.
Infection Control & Hospital Epidemiology, 36(8), 871 –877.
http://doi.org/10.1017/ice.2015.86
Khandelwal, N., Kross, E. K., Engelberg, R. A., Coe, N. B., Long, A. C., & Curtis, J. R. (2015).
Estimating the Effect of Palliative Care Interventions and Advance Care Planning on ICU
Utilization: A Systematic Review. Critical Care Medicine, 43(5), 1102 –1111.
http://doi.org/10.1097/CCM.0000000000000852.Estimating
Kirkwood, T. B. L. (1979). Geometric Means and Measures of Dispersion. Biometrics, 35(4), 908 –
909.
Kronick, R., Casalino, L. P., & Bindman, A. B. (2015). Apple Pickers or Federal Judges: Strong
versus Weak Incentives in Physician Payment. Health Services Research, 50(S2), 2049 –
2056. http://doi.org/10.1111/1475-6773.12424
L&M Policy Research LLC. (2015). Evaluation of CMMI Accountable Care Organization Initiatives:
Pioneer ACO Evaluation Findings from Performance Years One and Two. Retrieved from
https://innovation.cms.gov/Files/reports/PioneerACOEvalRpt2.pdf
Lake, T. K., Rich, E. C., Valenzano, C. S., & Maxfield, M. M. (2013). Paying More Wisely: Effects of
Payment Reforms on Evidence-Based Clinical Decision-Making. Journal of Comparative
Effectiveness Research, 2(3), 249.
Landon, B. E. (2015). Structuring Payments to Patient-Centered Medical Homes. Journal of the
American Medical Association, 312(16), 1633 –1634.
Landon, B. E. (2017). Tipping the Scale — The Norms Hypothesis and Primary Care Physician
Behavior. New England Journal of Medicine, 376(9), 810 –811.
http://doi.org/10.1056/NEJMp1510923
Lee, P. R., Grumbach, K., & Jameson, W. J. (1990). Physician Payment in the 1990s: Factors that
Will Shape the Future. Annual Review of Public Health, 11, 297 –318.
Lembit z, A . , & C lark e, T . J. ( 2 0 0 9) . Clarif yi n g “Nev er E v en t s” and I n t rodu c in g “Alw ay s E v en t s.”
Patient Safety in Surgery, 3(1), 26. http://doi.org/10.1186/1754-9493-3-26
Liss, P.-E. (1993). Health Care Need: Meaning and Measurement. Aldershot: Avebury.
Ljungqvist, O. (2014). ERAS - Enhanced Recovery After Surgery: Moving Evidence-Based
162
Perioperative Care to Practice. Journal of Parenteral and Enteral Nutrition, 38(5), 559 –566.
http://doi.org/10.1177/0148607114523451
Makary, M. A., & Daniel, M. (2016). Medical Error —The Third Leading Cause of Death in the US.
BMJ, 353(i2139), 1 –5. doi:10.1136/bmj.i2139
Marmor, T., & Oberlander, J. (2012). From HMOs to ACOs: The Quest for the Holy Grail in U.S.
Health Policy. Journal of General Internal Medicine, 27(9), 1215 –1218. doi:10.1007/s11606-012-
2024-6
Mays, N. (2011). Reducing Unwarranted Variations in Healthcare in the English NHS. BMJ,
342(March), 665. doi:10.1136/bmj.d1849
McCall, N., Cromwell, J., & Urato, C. (2010). Evaluation of Medicare Care Management for High
Cost Beneficiaries (CMHCB) Demonstration: Massachusetts General Hospital and
Massachusetts General Physicians Organization (MGH).
McWilliams, J. M., Hatfield, L. A., Chernew, M. E., Landon, B. E., & Schwartz, A. L. (2016). Early
Performance of Accountable Care Organizations in Medicare. New England Journal of
Medicine, 374(24), 2357 –2366. http://doi.org/10.1056/NEJMsa1600142
Meadows, D. H. (2008). Thinking in Systems: A Primer. (D. Wright, Ed.). White River Junction:
Chelsea Green.
Mechanic, D. (2004). The Rise and Fall of Managed Care. Journal of Health and Social Behavior,
45(Extra Issue), 76 –86. doi:10.2307/3653825
Meyer, H. (2011). Changing the Conversation in California About Care Near the End of Life. Health
Affairs, 30(3), 390 –393. http://doi.org/10.1377/hlthaff.2011.0082
Mi c h ael, Y . L. , Wh it l ock , E . P . , Li n , J. S. , F u , R. , O’C on n or , E . A . , & G o ld, R . ( 2 0 10 ) . P rimar y Car e –
Relevant Interventions to Prevent Falling in Older Adults: A Systematic Evidence Review
for the U.S. Preventive Services Task Force. Annals of Internal Medicine, 153(12), 815 –825.
Mi lst ein , A . ( 2 0 0 9) . E n di n g E xt ra P ay m en t f or “ Ne v er E v en t s” –St ron ger I n c en t iv es f or P at ien t s’ Safety. The New England Journal of Medicine, 360(23), 2388 –2390.
http://doi.org/10.1056/NEJMp0809125
Moody’s I n v es t ors Serv ic e. (2 0 1 2 ) . Ra t in g Met h odo lo gy : Not -for-Profit Healthcare Rating
Methodology. U.S. Public Finance. Retrieved from http://www.calhospital.org/cha-news-
article/moodys-changes-credit-rating-methodology-nonprofit-hospitals-systems
Morrisey, M. A. (2013). History of Health Insurance in the United States. In Health Insurance,
Second Edition. Health Administration Press.
Mossialos, E., Wenzl, M., Osborn, R., & Sarnak, D. (2015). International Profiles of Health Care
Systems 2015. Retrieved from http://www.commonwealthfund.org/publications/fund-
reports/2016/jan/international-profiles-2015
Mostashari, F., Sanghavi, D., & McClellan, M. (2014). Health Reform and Physician-Led
Accountable Care: The Paradox of Primary Care Physician Leadership. Journal of the
American Medical Association, 311(18), 1855 –1856. http://doi.org/10.1001/jama.2014.4086
Mueller, M. S. (1974). Health Maintenance Organization Act of 1973. Retrieved from
https://www.ssa.gov/policy/docs/ssb/v37n3/v37n3p35.pdf
Nelso n , L . ( 2 0 12 ) . Le ss o n s f rom M edic ar e’s Demo n s t ra t ion P roj ec t s on D is eas e M an age m e n t and
163
Care Coordination (Working Paper Series) (Vol. 1). Washington, D.C. Retrieved from
http://www.cbo.gov/sites/default/files/WP2012-
01_Nelson_Medicare_DMCC_Demonstrations.pdf
Newhouse, J. P., Manning, W. G., Keeler, E. B., & Sloss, E. M. (1989). Adjusting Capitation Rates
Using Objective Health Measures and Prior Utilization. Health Care Financing Review,
10(3), 41 –54.
Nikolich- Zu gich , J. , G ol d m an , D. P . , Co h e n , P . R . , Co rt es e, D . , F o n t an a , L. , K en n edy , B. K . , … F ai n , M. J. (2015). Preparing for an Aging World: Engaging Biogerontologists, Geriatricians, and
the Society. Journals of Gerontology: Biological Sciences, 00(00), 1 –10.
doi:10.1093/gerona/glv164
Ogden, L. L., Richards, C. L., & Shenson, D. (2012). Clinical Preventive Services for Older Adults:
The Interface Between Personal Health Care and Public Health Services. American Journal
of Public Health, 102(3), 419 –425. doi:10.2105/AJPH.2011.300353
Olshansky, S. J., & Ault, A. B. (1986). The Fourth Stage of the Epidemiologic Transition: The Age
of Delayed Degenerative Diseases. The Milbank Quarterly, 64(3), 355 –391.
Omran, A. (1971). The Epidemiologic Transition: A Theory of the Epidemiology of Population
Change. The Milbank Memorial Fund Quarterly, 49(1), 509 –538. doi:10.1007/s13398-014-
0173-7.2
Peasah, S. K., McKay, N. L., Harman, J. S., Al-Amin, M., & Cook, R. L. (2013). Medicare Non-
Payment of Hospital-Acquired Infections: Infection Rates Three Years Post
Implementation. Medicare & Medicaid Research Review, 3(3), E1 –E13.
http://doi.org/10.5600/mmrr.003.03.a08
Pope, G. C., & Kautter, J. (2012). Minimum Savings Requirements in Shared Savings Provider
Payment. Health Economics, 21(11), 1336 –1347. http://doi.org/10.1002/hec.1793
Pozen, A., & Cutler, D. M. (2010). Medical Spending Differences in the United States and Canada:
The Role of Prices, Procedures, and Administrative Expenses. Inquiry, 47(2), 124 –134.
doi:10.5034/inquiryjrnl_47.02.124
Pronovost, P. J., & Bo-Linn, G. W. (2012). Preventing Patient Harms Through Systems of Care.
Journal of the American Medical Association, 308(8), 769 –770.
Pronovost, P. J., Marsteller, J. A., & Goeschel, C. A. (2011). Preventing Bloodstream Infections: A
Measurable National Success Story in Quality Improvement. Health Affairs, 30(4), 628 –
634. doi:10.1377/hlthaff.2011.0047
Public Health England. (2015). The NHS Atlas of Variation in Healthcare: Reducing Unwarranted
Variation to Increase Value and Improve Quality. Retrieved from
http://www.rightcare.nhs.uk/index.php/nhs-atlas/
Rab e, K . F . , Hur d, S . , An z u et o, A. , Bar n es , P . J. , Bu i st , S. A . , Ca lv er ley, P . , … Zielinski, J. (2007).
Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive
Pulmonary Disease: GOLD Executive Summary. American Journal of Respiratory and
Critical Care Medicine, 176(6), 532 –555. doi:10.1164/rccm.200703-456SO
Rich, E. C., Lake, T. K., Valenzano, C. S., & Maxfield, M. M. (2013). Paying the Doctor: Evidence-
Based Decisions at the Point-of-Care and the Role of Fee-for-Service Incentives. Journal of
164
Comparative Effectiveness Research, 2(3), 235.
Rizzo, J. A., & Zeckhauser, R. J. (2003). Reference Incomes, Loss Aversion, and Physician Behavior.
The Review of Economics and Statistics, 85(4), 909 –922.
Rober t s, R. R . , F ru t os, P . W. , Ciav ar ella , G . G . , G u ss ow, L. M . , M en sa h , E . K . , K amp e, L. M . , … Rydman, R. J. (1999). Distribution of Variable vs Fixed Costs of Hospital Care. Journal of
the American Medical Association, 281(7), 644 –649. http://doi.org/10.1001/jama.281.7.644
Robinson, J. C. (2001). Theory and Practice In the Design of Physician Payment Incentives. The
Milbank Quarterly, 79(2), 149 –177. doi:10.1111/1468-0009.00202
Rose, S., Zaslavsky, A. M., & McWilliams, J. M. (2016). Variation in Accountable Care Organization
Spending and Sensitivity to Risk Adjustment: Implications for Benchmarking. Health
Affairs, 35(3), 440 –448. http://doi.org/10.1377/hlthaff.2015.1026
Rosenthal, T. (2012). Geographic Variation in Health Care. Annual Review of Medicine, 63(1), 493 –
509. doi:10.1146/annurev-med-050710-134438
Saver, B. G., Ritzwoller, D. P., Maciosek, M., Goodman, M. J., Co n ra d , D. A . , F in ke lst ein , E . , … Cain, K. C. (2004). Does Payment Drive Procedures? Payment for Specialty Services and
Procedure Rate Variations in 3 HMOs. American Journal of Managed Care, 10(3), 229 –237.
Sessums, L. L., McHugh, S. J., & Rajkumar, R. (20 1 6 ). M edic ar e’s Visio n f or A dv an c ed P rimar y
Care: New Directions for Care Delivery and Payment. Journal of the American Medical
Association, E1 –E2. http://doi.org/10.1001/jama.2016.4472
Shafrin, J. (2010). Operating on Commission: Analyzing How Physician Financial Incentives Affect
Surgery Rates. Health Economics, 19, 562 –580.
Shaw, G. B. (1911). The Doctor’s Dilemma . New Y or k: Bren t a n o’s .
Shi, L. (2012). The Impact of Primary Care: A Focused Review. Scientifica, 2012.
doi:10.6064/2012/432892
Showstack, J. A., Blumberg, B. D., Schwartz, J., & Schroeder, S. A. (1979). Fee-for-Service Physician
Payment: Analysis of Current Methods and Their Development. Inquiry, 16(3), 230 –246.
Simpson, J. C., Moonesinghe, S. R., Grocott, M. P. W., Kuper, M., McMeeking, A., Ol iv er , C . M . , … Mythen, M. G. (2015). Enhanced Recovery from Surgery in the UK: An Audit of the
Enhanced Recovery Partnership Programme 2009-2012. British Journal of Anaesthesia,
115(4), 560 –568. http://doi.org/10.1093/bja/aev105
Spettell, C. M., Rawlins, W . S . , K ra ka u er , R . , F er n a n des , J . , Bre t o n , M . E . S . , G owd y, W . , … Brennan, T. A. (2009). A Comprehensive Case Management Program To Improve
Palliative Care. Journal of Palliative Medicine, 12(9), 827 –832.
Squires, D., & Anderson, C. (2015). U.S. Health Care from a Global Perspective: Spending, Use of
Services, Prices, and Health in 13 Countries. Commonwealth Fund. Retrieved from
http://www.commonwealthfund.org/publications/issue-briefs/2015/oct/us-health-care-
from-a-global-perspective
Stoddard, J. J., Reed, M. , & Hadley , J. ( 2 0 0 3) . F in a n c ial I n c en t iv es and P h ys ic ians’ P er c ep t ion s of Conflict of Interest and Ability to Arrange Medically Necessary Services. Journal of
Ambulatory Care Management, 26(1), 39 –50.
Sullivan, K. (2015). Patient Turnover in ACOs Destroys Accountability. Retrieved October 12, 2016,
165
from http://pnhp.org/blog/2015/06/23/patient-turnover-in-acos-destroys-accountability/
Teno, J. M., Freedman, V. A., Kasper, J. D., Gozalo, P., & Mor, V. (2015). Is Care for the Dying
Improving in the United States? Journal of Palliative Medicine, 18(8), 662 –666.
http://doi.org/10.1089/jpm.2015.0039
T en o , J. M . , G ozalo , P . L . , Byn u m , J . P . W . , L elan d , N. E . , M iller , S . C. , M orde n , N. E . , … M or , V . (2013). Change in End-of-Life Care for Medicare Beneficiaries. Journal of the American
Medical Association, 309(5), 470 –477.
The National Commission on Physician Payment Reform. (2013). Report of the National
Commission on Physician Payment Reform. Retrieved from
http://physicianpaymentcommission.org/report/
Thomasson, M. (2003). Health Insurance in the United States. EH.Net Encyclopedia. Retrieved
April 19, 2016, from https://eh.net/encyclopedia/health-insurance-in-the-united-states/
U.S. Department of Transportation, Research and Innovation Technology Administration (RITA),
Bureau of Transportation Statistics. (2014). US Air Carrier Safety Data. Retrieved May 4,
2016, from
http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/national_transportati
on_statistics/html/table_02_09.html
Umscheid, C. A., Mitchell, M. D., Doshi, J. A., Agarwal, R., Williams, K., & Brennan, P. J. (2011).
Estimating the Proportion of Healthcare-Associated Infections That Are Reasonably
Preventable and the Related Mortality and Costs. Infection Control & Hospital
Epidemiology, 32(2), 101 –114. http://doi.org/10.1086/657912
Urato, C., McCall, N., Cromwell, J., Lenfestey, N., Smith, K., & Raeder, D. (2013). Evaluation of the
Extended Medicare Care Management for High Cost Beneficiaries (CMHCB)
Demonstration: Massachusetts General Hospital (MGH).
van Baal, P. H. M., Polder, J. J., de Wit, G. A., Hoogenveen, R. T., Feenstra, T. L., Boshuizen, H. C.,
… Brou w er , W . B . F . (2 0 0 8 ). Li f et im e M edic al C ost s of Ob es it y: P re v en t io n No Cu re f or
Increasing Health Expenditure. PLoS Medicine, 5(2), e29. doi:10.1371/journal.pmed.0050029
Van Herck, P., De Smedt, D., Annemans, L., Remmen, R., Rosenthal, M. B., & Sermeus, W. (2010).
Systematic Review: Effects, Design Choices, and Context of Pay-for-performance in Health
Care. BMC Health Services Research, 10(247). doi:10.1186/1472-6963-10-247
Weil, T. P. (2002). Multispecialty Physician Practices: Fixed and Variable Costs, and Economies of
Scale. The Journal of Ambulatory Care Management, 25(3), 70 –77.
Wennberg, J. E. (2010). Tracking Medicine: A Researcher’s Quest to Understand Health Care. New
York: Oxford University Press.
Wennberg, J. E., Fisher, E. S., & Skinner, J. S. (2002). Geography and the Debate Over Medicare
Reform. Health Affairs, w96 –w114. http://doi.org/10.1377/hlthaff.w2.96
Wennberg, J. E., Fisher, E. S., Goodman, D. C., & Skinner, J. S. (2008). Tracking the Care of
Patients with Severe Chronic Illness: The Dartmouth Atlas of Healthcare 2008. Retrieved
from http://www.dartmouthatlas.org/publications/reports.aspx
Wick, E. C., Galante, D. J . , Ho bs on , D . B. , Be n so n , A. R. , Le e , K . H . K . , Bere n h ol t z, S . M . , … Wu , C. L. (2015). Organizational Culture Changes Result in Improvement in Patient-Centered
166
Outcomes: Implementation of an Integrated Recovery Pathway for Surgical Patients.
Journal of the American College of Surgeons, 221(3), 669 –677.
http://doi.org/10.1016/j.jamcollsurg.2015.05.008
Wijeysundera, D. N., & Sweitzer, B.-J. (2015). Preoperative Evaluation. In Miller’s Anesthesia (8th
ed., pp. 1085 –1155). Elsevier Inc. http://doi.org/10.1016/B978-0-7020-5283-5.00038-2
Yarnall, K. S. H., Pollak, K. I., Østbye, T., Krause, K. M., & Michener, J. L. (2003). Primary Care: Is
There Enough Time for Prevention? American Journal of Public Health, 93(4), 635 –641.
doi:10.2105/AJPH.93.4.635
167
Appendix
A.1 - Batch Repetitions Full Tables
Because this is a Monte Carlo simulation, the quantile results listed throughout chapter
four are the products of some amount of random variation. Since they are quantiles from
simulation batches containing 5,000 ru n s, t h e v ar ia t ion s h ould b e m in i m al . Nev er t h eless , it ’s useful to see how much these quantile results vary from batch to batch to have a better idea of
how significant the changes in results are from one payment system to another.
The following tables show a set of PID values from 15 repetitions of a 5,000 patient
population in the baseline FFS scenario. Starting with the short term (3 year average PID per year
per capita) values:
Repetition # 10th 25th Median 75th 90th
1 $0.00 $3.94 $41.98 $108.02 $169.59
2 $0.00 $3.72 $39.20 $108.74 $169.89
3 $0.00 $4.13 $39.92 $107.89 $167.80
4 $0.00 $4.53 $41.21 $108.33 $167.59
5 $0.00 $4.30 $44.13 $111.81 $170.89
6 $0.00 $3.20 $38.20 $105.86 $168.75
7 $0.00 $4.97 $41.94 $111.49 $173.33
8 $0.00 $4.62 $40.81 $107.70 $168.05
9 $0.00 $3.78 $38.51 $105.91 $166.30
10 $0.00 $3.82 $39.71 $108.38 $169.09
11 $0.00 $3.42 $36.66 $104.96 $165.49
12 $0.00 $4.09 $39.15 $109.68 $173.32
13 $0.00 $3.23 $36.49 $105.29 $164.74
14 $0.00 $3.65 $41.30 $109.23 $169.91
15 $0.00 $4.15 $42.80 $109.04 $168.43
Mean $0.00 $3.97 $40.14 $108.16 $168.88
Std Dev $0.00 $0.51 $2.18 $2.04 $2.47
CV NaN 12.73% 5.43% 1.88% 1.46%
Table A.1: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 5000 Patients, First 3 Years
168
Because this is the smallest patient population tested and these are short term values, we should
expect that these values would have the most variation from batch to batch. Nevertheless, they
are still fairly consistent, with the median having a mean PID value of $40.14 and a coefficient of
variation (CV) of 5.43%. The values for the long term (10 year average PID per year per capita) are
even more consistent:
Repetition # 10th 25th Median 75th 90th
1 $6.50 $14.09 $37.43 $66.44 $97.77
2 $6.46 $13.72 $37.23 $67.71 $97.51
3 $6.64 $14.44 $36.83 $66.03 $96.64
4 $6.75 $14.10 $37.18 $66.91 $97.95
5 $6.48 $14.42 $38.77 $66.73 $99.08
6 $6.24 $13.40 $35.52 $66.46 $95.85
7 $6.46 $14.14 $38.20 $67.72 $97.99
8 $6.58 $14.37 $37.40 $67.52 $97.18
9 $6.33 $13.82 $36.53 $67.33 $98.70
10 $6.48 $13.59 $36.34 $66.38 $96.03
11 $6.37 $13.36 $36.39 $66.76 $97.12
12 $6.56 $14.52 $36.51 $67.48 $99.68
13 $6.42 $13.35 $35.51 $64.89 $94.83
14 $6.29 $13.77 $36.72 $66.79 $95.72
15 $6.95 $14.57 $37.11 $67.75 $96.45
Mean $6.50 $13.98 $36.91 $66.86 $97.23
Std Dev $0.18 $0.44 $0.87 $0.78 $1.35
CV 2.79% 3.12% 2.37% 1.16% 1.39%
Table A.2: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 5000 Patients, Full 10 Years
Over the long term, the median small FFS provider has a mean value of $36.91 of PID with a CV of
2.37%, less than half the amount of variation in the short term.
Looking at a large provider (50,000 patients) as well, we find similar variability in these
values in percentage terms, but because the quantile PID values are lower for larger providers
there is a lower amount of absolute variation in PID. Starting with the short term results:
169
Repetition # 10th 25th Median 75th 90th
1 $0.00 $1.34 $11.97 $33.09 $53.58
2 $0.00 $1.29 $11.98 $32.88 $53.26
3 $0.00 $1.22 $11.87 $32.48 $52.20
4 $0.00 $1.29 $13.01 $33.87 $53.05
5 $0.00 $1.13 $12.43 $33.64 $54.21
6 $0.00 $1.32 $11.87 $32.48 $53.96
7 $0.00 $1.33 $11.34 $31.96 $52.22
8 $0.00 $1.30 $11.52 $32.48 $51.51
9 $0.00 $0.96 $11.01 $31.69 $52.92
10 $0.00 $1.14 $12.15 $32.74 $52.07
11 $0.00 $1.02 $12.18 $32.32 $52.69
12 $0.00 $1.06 $11.89 $32.28 $53.05
13 $0.00 $1.31 $12.41 $32.21 $52.52
14 $0.00 $1.05 $12.27 $33.59 $54.17
15 $0.00 $1.20 $11.78 $32.48 $53.17
Mean $0.00 $1.20 $11.98 $32.68 $52.97
Std Dev $0.00 $0.13 $0.48 $0.63 $0.80
CV NaN 10.71% 4.02% 1.93% 1.50%
Table A.3: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 50000 Patients, First 3 Years
Here we see that the median large FFS provider has an average of $11.98 of PID in the short term
with a CV of 4.02%. And then looking at the long term results:
Repetition # 10th 25th Median 75th 90th
1 $2.04 $4.24 $10.57 $19.52 $28.97
2 $2.11 $4.19 $10.35 $19.16 $29.05
3 $1.99 $4.23 $10.45 $19.54 $28.37
4 $2.07 $4.34 $10.77 $19.83 $28.28
5 $2.01 $4.18 $10.57 $19.91 $28.77
6 $2.02 $4.21 $10.46 $19.73 $28.82
7 $1.97 $4.18 $10.10 $19.49 $28.56
170
8 $2.05 $4.16 $10.17 $18.94 $28.21
9 $2.00 $4.04 $9.91 $19.47 $29.11
10 $1.90 $4.13 $10.26 $19.50 $28.66
11 $2.04 $4.16 $10.18 $18.86 $28.73
12 $2.04 $4.24 $10.49 $19.72 $28.64
13 $2.06 $4.23 $10.34 $19.35 $28.93
14 $1.96 $4.15 $10.33 $19.78 $28.77
15 $2.02 $4.05 $10.06 $19.23 $29.38
Mean $2.02 $4.18 $10.33 $19.47 $28.75
Std Dev $0.05 $0.07 $0.23 $0.31 $0.32
CV 2.47% 1.79% 2.20% 1.61% 1.11%
Table A.4: Repetitions, Average PID Per Year Per Capita, Baseline FFS, 50000 Patients, Full 10 Years
Over the long term the median large FFS provider has an average of $10.33 of PID with a CV of
2.20%, again about half the variation compared to the short term.
I t ’s als o im p ort an t t o l o o k at r ep et it ion s f or c ap it a t ed p rov ider s. Bec au se t h ey h av e a different profit structure, capitated providers produce PID values a bit differently than FFS
p rov ider s, so it ’s imp ort an t t o c h ec k if t h eir P I D v a lues als o v ar y by di f f er en t amo u n t s . Nevertheless, the PID values for capitated providers seem to show a similar amount of variation as
those of FFS providers. Starting with a 5,000 patient population in the short term:
Repetition # 10th 25th Median 75th 90th
1 $0.00 -$3.91 -$63.51 -$154.07 -$235.37
2 $0.00 -$3.62 -$60.47 -$149.16 -$239.50
3 $0.00 -$4.28 -$62.29 -$154.41 -$244.53
4 $0.00 -$3.26 -$57.87 -$149.16 -$240.88
5 $0.00 -$3.61 -$56.39 -$146.16 -$237.91
6 $0.00 -$3.66 -$64.24 -$157.11 -$245.33
7 $0.00 -$3.65 -$59.36 -$149.72 -$235.64
8 $0.00 -$3.83 -$64.11 -$157.37 -$239.84
9 $0.00 -$3.52 -$55.30 -$150.53 -$236.34
10 $0.00 -$4.01 -$61.89 -$159.80 -$246.25
11 $0.00 -$4.13 -$61.02 -$155.07 -$250.12
171
12 $0.00 -$3.76 -$61.34 -$155.49 -$241.80
13 $0.00 -$3.08 -$61.78 -$148.81 -$244.40
14 $0.00 -$3.43 -$57.16 -$148.39 -$231.58
15 $0.00 -$3.89 -$59.83 -$143.17 -$235.27
Mean $0.00 -$3.71 -$60.44 -$151.90 -$240.32
Std Dev $0.00 $0.32 $2.77 $4.67 $5.10
CV NaN 8.58% 4.59% 3.08% 2.12%
Table A.5: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 5000 Patients, First 3 Years
We can see that, in the short term, the median small capitated provider produces -$60.44 of PID
on average with a CV of 4.59%, which is a higher amount of PID but actually less variation
compared to a similar FFS provider. And over the long term:
Repetition # 10th 25th Median 75th 90th
1 -$6.95 -$24.63 -$58.76 -$98.68 -$137.23
2 -$6.57 -$22.63 -$57.81 -$97.38 -$134.07
3 -$6.41 -$22.39 -$59.15 -$99.38 -$138.70
4 -$6.18 -$21.74 -$58.29 -$98.09 -$135.32
5 -$6.49 -$21.45 -$56.98 -$96.15 -$136.27
6 -$6.56 -$23.66 -$58.42 -$99.19 -$139.42
7 -$6.27 -$21.44 -$57.57 -$98.02 -$134.45
8 -$6.84 -$23.25 -$58.81 -$99.74 -$137.69
9 -$6.21 -$21.40 -$56.18 -$97.83 -$136.91
10 -$6.80 -$23.10 -$60.00 -$100.07 -$137.92
11 -$6.65 -$21.36 -$57.93 -$98.92 -$139.61
12 -$6.47 -$20.85 -$57.83 -$99.31 -$137.56
13 -$6.42 -$21.31 -$56.06 -$95.52 -$135.43
14 -$6.15 -$21.07 -$56.61 -$96.84 -$134.66
15 -$6.42 -$22.39 -$57.56 -$97.79 -$135.67
Mean -$6.49 -$22.18 -$57.86 -$98.19 -$136.73
Std Dev $0.24 $1.09 $1.10 $1.32 $1.78
CV 3.75% 4.94% 1.91% 1.35% 1.30%
172
Table A.6: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 5000 Patients, Full 10 Years
The median small capitated provider produces an average of -$57.86 of PID over the long term
with a CV of 1.91%, which is again more absolute PID but a bit less variation than a similar FFS
provider.
Again there is a similar amount of variation for large capitated providers (50,000 patients).
Over the short term:
Repetition # 10th 25th Median 75th 90th
1 $0.00 -$1.34 -$12.20 -$34.16 -$62.01
2 $0.00 -$1.31 -$12.54 -$34.99 -$63.64
3 $0.00 -$1.23 -$12.06 -$34.16 -$61.65
4 $0.00 -$1.23 -$12.26 -$35.33 -$63.08
5 $0.00 -$1.30 -$12.25 -$34.90 -$61.09
6 $0.00 -$1.15 -$11.80 -$35.54 -$63.62
7 $0.00 -$1.26 -$11.57 -$33.16 -$62.39
8 $0.00 -$1.08 -$12.01 -$33.83 -$65.11
9 $0.00 -$1.02 -$11.42 -$33.68 -$63.59
10 $0.00 -$1.26 -$12.19 -$33.42 -$61.58
11 $0.00 -$1.01 -$11.84 -$34.20 -$62.21
12 $0.00 -$1.19 -$11.74 -$34.30 -$63.36
13 $0.00 -$1.16 -$12.18 -$34.51 -$62.91
14 $0.00 -$0.90 -$11.55 -$33.14 -$60.31
15 $0.00 -$1.35 -$12.33 -$34.72 -$62.16
Mean $0.00 -$1.19 -$12.00 -$34.27 -$62.58
Std Dev $0.00 $0.13 $0.33 $0.74 $1.21
CV NaN 11.25% 2.73% 2.17% 1.93%
Table A.7: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 50000 Patients, First 3 Years
The median large capitated provider produces on average -$12.00 of PID over the short term with
a CV of 2.73%, which is a bit more absolute PID but a bit less variation than an equivalent FFS
provider. And over the long term:
Repetition # 10th 25th Median 75th 90th
173
1 -$2.03 -$4.25 -$10.86 -$21.15 -$32.32
2 -$1.98 -$4.29 -$11.34 -$21.53 -$32.11
3 -$1.97 -$4.37 -$11.21 -$21.22 -$32.02
4 -$1.95 -$4.33 -$10.92 -$21.56 -$32.55
5 -$1.93 -$4.09 -$11.01 -$21.44 -$31.97
6 -$1.97 -$4.11 -$10.87 -$21.76 -$32.73
7 -$1.99 -$4.11 -$10.57 -$21.00 -$32.52
8 -$2.02 -$4.14 -$10.88 -$21.50 -$32.58
9 -$2.00 -$4.08 -$10.57 -$21.58 -$32.79
10 -$1.97 -$4.28 -$10.76 -$21.26 -$32.00
11 -$2.04 -$4.20 -$10.90 -$21.55 -$31.98
12 -$1.96 -$4.24 -$10.92 -$21.98 -$32.75
13 -$1.99 -$4.35 -$10.99 -$21.45 -$32.57
14 -$1.91 -$3.95 -$10.64 -$21.38 -$32.82
15 -$2.05 -$4.34 -$11.57 -$22.47 -$33.80
Mean -$1.98 -$4.21 -$10.93 -$21.52 -$32.50
Std Dev $0.04 $0.12 $0.27 $0.35 $0.48
CV 2.03% 2.94% 2.50% 1.65% 1.48%
Table A.8: Repetitions, Average PID Per Year Per Capita, Baseline Capitation, 50000 Patients, Full 10 Years
The median large capitated provider produces on average -$10.93 of PID over the long term with a
CV of 2.50%, which is again a bit more absolute PID but about the same amount of variation
compared to an equivalent FFS provider.
In summary, because the simulation contains random elements, the reported quantile
values in chapter four will change from batch to batch, even with the exact same input
parameters. But because these quantile values are generated from 5,000 simulation runs, a
repetition analysis shows that there is minimal variation in these values. In fact, it seems that as
long as there is at least a 10-20% difference in the quantile values of PID from one batch to
another, depending on whether they are short or long term values, then that difference is almost
certainly due to significant differences between the batches in their input parameters and
payment systems, with a minimal role from random variation. Since most of the important
comparisons in chapter four involve changes in PID values of 150% and more, I can say with a
high degree of confidence that these changes are significant.
174
A.2 - Comparative Financial Performance of FFS and Capitation without Risk Aversion
T o st ar t , let ’s lo ok at t h e d is t rib u t ion of t h e n u m b er of y ear s in w h ic h p rov ider s w h o are n ’t risk averse experience a cash flow loss of any size. Data will generally be shown for both the short
term (first 3 years) and long term (full 10 years), for providers of the four different patient
population sizes tested, and for the median, quartile, and decile providers in the distribution of
results. Starting with FFS providers, the distribution of the number of years with a loss is as
follows:
# Patients 10th 25th Median 75th 90th
5000 0 1 2 2 3
10000 0 1 2 2 3
25000 0 1 2 2 3
50000 0 1 1 2 3
Table A.9: Baseline FFS (no risk aversion), number of years with a loss, first 3 years
# Patients 10th 25th Median 75th 90th
5000 3 4 5 6 7
10000 2 4 5 6 7
25000 3 4 5 6 8
50000 3 4 5 6 7
Table A.10: Baseline FFS (no risk aversion), number of years with a loss, full 10 years
The distribution is similar for capitated providers:
# Patients 10th 25th Median 75th 90th
5000 0 1 2 2 3
10000 0 1 1 2 3
25000 0 1 1 2 3
50000 0 1 1 2 3
Table A.11: Baseline capitation (no risk aversion), number of years with a loss, first 3 years
# Patients 10th 25th Median 75th 90th
5000 3 4 5 6 8
10000 2 4 5 6 7
175
25000 3 4 5 6 7
50000 2 4 5 6 7
Table A.12: Baseline capitation (no risk aversion), number of years with a loss, full 10 years
As expected, because net profit is set to zero at the average expected level of patient needs
for care, the median health care provider of any size experiences 5 years of loss over a 10 year
period (with similar short term results), and there is a nearly symmetrical spread of outcomes in
the quantiles on either side of the median. This is the case for every payment system tested; while
FFS, capitation, and other tested alternatives may produce different average losses, they do not
generally influence the actual chance of a provider having a loss.
When p rov ider s h av e loss es it ’s als o im p ort an t t o n ote t h eir mag n it u de. Th o u gh FF S an d
capitated providers experience about the same number of years with a loss, both over the short
and long term, the actual magnitude of those losses is quite different, with capitated providers
experiencing far smaller average losses per capita across the board than FFS providers. This is
seen clearly in a graphical comparison (in all of the graphs in this section, the black square on the
graph represents the median value, while the blue, red, green, and purple points represent the
10th, 25th, 75th, and 90th percentiles, respectively):
176
Figure A.1: Baseline FFS (left) vs. capitation (right), no risk aversion, average loss per capita
This discrepancy between the financial performance of FFS and capitated providers also
sh ows u p w h en w e loo k a t t h e p rov ider s’ w orst lo ss es ov er t h e s h ort and lo n g t er m . Here is a
graphical comparison of the worst per capita losses experienced by FFS and capitated providers at
baseline (again without any provider risk aversion):
177
Figure A.2: Baseline FFS (left) vs. capitation (right), no risk aversion, max loss per capita
Once again, the worst losses of capitated providers are significantly lower across the board than
the worst losses of FFS providers.
In addition to differences in cash flow losses between FFS and capitated providers, we can
also look at differences in cash on hand. Providers in each simulation run start with 90 days cash
on hand, and show the following distribution of average days cash on hand over the course of the
run (again with no provider risk averse behavior):
178
Figure A.3: Baseline FFS (left) vs. capitation (right), no risk aversion, average days cash on hand
Here again it is clear that all FFS providers, regardless of size, experience a wider range of average
days cash on hand over the course of the simulation than capitated providers.
And finally, we can also look at the difference between FFS and capitated providers (again
without risk aversion) in the lowest days cash on hand they experience during the simulation:
179
Figure A.4: Baseline FFS (left) vs. capitation (right), no risk aversion, lowest days cash on hand
Once again, FFS providers are subjected to a much wider range of minimum days cash on hand
than capitated providers.
A.3 - Comparative Financial Performance of FFS and Capitation when Adding Risk Aversion
This section looks at how the financial performance measures from the previous section
change when turning provider risk aversion on. For example, enabling risk aversion slightly
reduces the number of years in which providers experience a loss. Here are the results for baseline
FFS with provider risk aversion, with the change from the corresponding previous table (without
provider risk aversion) in parentheses:
# Patients 10th 25th Median 75th 90th
5000 0 (0) 0 (-1) 1 (-1) 2 (0) 2 (-1)
10000 0 (0) 0 (-1) 1 (-1) 2 (0) 2 (-1)
25000 0 (0) 0 (-1) 1 (-1) 2 (0) 2 (-1)
50000 0 (0) 0 (-1) 1 (0) 2 (0) 2 (-1)
Table A.13: Baseline FFS (with risk aversion), number of years with loss, first 3 years
# Patients 10th 25th Median 75th 90th
5000 2 (-1) 3 (-1) 4 (-1) 5 (-1) 6 (-1)
10000 2 (0) 3 (-1) 4 (-1) 5 (-1) 6 (-1)
180
25000 2 (-1) 3 (-1) 4 (-1) 5 (-1) 6 (-2)
50000 2 (-1) 3 (-1) 4 (-1) 5 (-1) 6 (-1)
Table A.14: Baseline FFS (with risk aversion), number of years with loss, full 10 years
And here are the results for capitated providers practicing risk aversion:
# Patients 10th 25th Median 75th 90th
5000 0 (0) 0 (-1) 1 (-1) 1 (-1) 2 (-1)
10000 0 (0) 0 (-1) 1 (0) 1 (-1) 2 (-1)
25000 0 (0) 0 (-1) 1 (0) 2 (0) 2 (-1)
50000 0 (0) 0 (-1) 1 (0) 2 (0) 2 (-1)
Table A.15: Baseline capitation (with risk aversion), number of years with loss, first 3 years
# Patients 10th 25th Median 75th 90th
5000 2 (-1) 2 (-2) 3 (-2) 4 (-2) 5 (-3)
10000 2 (0) 3 (-1) 4 (-1) 4 (-2) 5 (-2)
25000 2 (-1) 3 (-1) 4 (-1) 5 (-1) 6 (-1)
50000 2 (0) 3 (-1) 4 (-1) 5 (-1) 6 (-1)
Table A.16: Baseline capitation (with risk aversion), number of years with loss, full 10 years
Risk averse behavior reduces the number of years that both FFS and capitated providers
experience a cash flow loss by about 1 year, both in the short and long term, except for some
slightly greater reductions for smaller capitated providers in the upper quantiles.
Provider risk averse behavior also reduces the size of those cash flow losses. Here are the
values for average losses per capita for baseline FFS providers after taking into account risk averse
behavior, again with the change from the previous values (without risk aversion) in parentheses:
# Patients 10th 25th Median 75th 90th
5000 $0 ($0) $0 (-$33) $64 (-$32) $131 (-$31) $201 (-$26)
10000 $0 ($0) $0 (-$21) $45 (-$22) $92 (-$21) $143 (-$12)
25000 $0 ($0) $0 (-$13) $28 (-$14) $60 (-$12) $93 (-$7)
50000 $0 ($0) $0 (-$8) $21 (-$9) $43 (-$8) $66 (-$5)
Table A.17: Baseline FFS (with risk aversion), average loss per capita, first 3 years
# Patients 10th 25th Median 75th 90th
181
5000 $57 (-$9) $83 (-$10) $113 (-$13) $147 (-$12) $185 (-$9)
10000 $40 (-$7) $59 (-$7) $79 (-$9) $105 (-$9) $130 (-$9)
25000 $25 (-$4) $36 (-$5) $51 (-$6) $66 (-$6) $81 (-$6)
50000 $19 (-$2) $26 (-$3) $36 (-$3) $47 (-$4) $58 (-$4)
Table A.18: Baseline FFS (with risk aversion), average loss per capita, full 10 years
And here are the results for average losses per capita for capitated providers after risk aversion:
# Patients 10th 25th Median 75th 90th
5000 $0 ($0) $0 (-$6) $9 (-$19) $40 (-$20) $81 (-$11)
10000 $0 ($0) $0 (-$3) $7 (-$6) $24 (-$12) $52 (-$8)
25000 $0 ($0) $0 (-$2) $5 (-$3) $11 (-$6) $23 (-$6)
50000 $0 ($0) $0 (-$2) $4 (-$2) $7 (-$2) $12 (-$3)
Table A.19: Baseline capitation (with risk aversion), average loss per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $11 (-$6) $23 (-$6) $37 (-$7) $57 (-$5) $78 (-$1)
10000 $7 (-$2) $13 (-$3) $23 (-$4) $34 (-$5) $46 (-$4)
25000 $5 (-$1) $7 (-$1) $10 (-$2) $16 (-$2) $23 (-$2)
50000 $3 ($0) $5 (-$1) $6 (-$1) $9 (-$1) $13 (-$1)
Table A.20: Baseline capitation (with risk aversion), average loss per capita, full 10 years
As you can see, risk aversion reduces average losses per capita for both FFS and capitation
providers by a good amount in the short term, though the long term average reduction is smaller
for both. The effect of risk aversion is also much less pronounced at the upper quantiles of
providers, indicating some natural limits to its effectiveness. And as before, average losses per
capita are still smaller for capitated providers than for FFS ones, though it is noteworthy that even
though capitated providers are engaging in more risk averse behavior, they are seeing less
absolute reductions in averages losses from those efforts.
The greater reduction to average losses in the short term after provider risk aversion in
comparison to the long term is likely due to its effect on mitigating extreme losses. This shows up
in the results for maximum losses under risk aversion, starting with FFS providers (again with the
change from no risk aversion in parentheses):
# Patients 10th 25th Median 75th 90th
182
5000 $0 ($0) $0 (-$37) $73 (-$49) $158 (-$58) $237 (-$62)
10000 $0 ($0) $0 (-$24) $52 (-$33) $110 (-$40) $164 (-$49)
25000 $0 ($0) $0 (-$14) $33 (-$22) $71 (-$23) $107 (-$27)
50000 $0 ($0) $0 (-$9) $24 (-$13) $51 (-$15) $77 (-$17)
Table A.21: Baseline FFS (with risk aversion), max loss per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $94 (-$20) $143 (-$30) $207 (-$33) $273 (-$38) $343 (-$37)
10000 $67 (-$14) $102 (-$20) $145 (-$24) $190 (-$30) $234 (-$35)
25000 $42 (-$8) $66 (-$11) $91 (-$15) $121 (-$18) $150 (-$21)
50000 $30 (-$6) $47 (-$8) $66 (-$10) $85 (-$13) $106 (-$14)
Table A.22: Baseline FFS (with risk aversion), max loss per capita, full 10 years
And here are the results for capitated providers:
# Patients 10th 25th Median 75th 90th
5000 $0 ($0) $0 (-$7) $10 (-$29) $50 (-$38) $93 (-$40)
10000 $0 ($0) $0 (-$3) $8 (-$10) $30 (-$25) $61 (-$26)
25000 $0 ($0) $0 (-$2) $6 (-$4) $12 (-$12) $29 (-$16)
50000 $0 ($0) $0 (-$2) $4 (-$3) $9 (-$3) $13 (-$10)
Table A.23: Baseline capitation (with risk aversion), max loss per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $19 (-$16) $47 (-$19) $78 (-$23) $112 (-$26) $147 (-$29)
10000 $11 (-$5) $26 (-$11) $49 (-$15) $72 (-$19) $97 (-$20)
25000 $7 (-$2) $11 (-$2) $22 (-$8) $38 (-$10) $53 (-$10)
50000 $5 (-$1) $8 (-$1) $11 (-$2) $18 (-$7) $29 (-$8)
Table A.24: Baseline capitation (with risk aversion), max loss per capita, full 10 years
Risk aversion seems to be pretty effective at reducing maximum cash flow losses over the whole 10
year simulation period, though there are obviously still limits to its ability to do so, as shown by
the still-high maximum losses in the upper quantiles of providers. Again it is notable that, while
capitated providers still have lower maximum per capita losses than FFS ones, their risk averse
behavior has less of an absolute effect on these maximum losses as well.
The effects and limits of provider risk averse behavior show up most notably in the results
183
for days cash on hand. For FFS providers, they show the following average days cash on hand with
risk averse behavior (again with the change from no risk aversion shown in parentheses):
# Patients 10th 25th Median 75th 90th
5000 80 (+30) 87 (+18) 96 (+8) 108 (+3) 121 (+2)
10000 84 (+22) 88 (+13) 94 (+5) 104 (+2) 114 (+1)
25000 87 (+15) 90 (+9) 94 (+4) 100 (+1) 106 (+0)
50000 88 (+10) 90 (+6) 93 (+3) 97 (+1) 102 (+0)
Table A.25: Baseline FFS (with risk aversion), average days cash on hand, first 3 years
# Patients 10th 25th Median 75th 90th
5000 87 (+52) 93 (+33) 103 (+19) 119 (+11) 134 (+6)
10000 89 (+38) 93 (+26) 101 (+14) 112 (+7) 125 (+5)
25000 90 (+25) 93 (+16) 98 (+9) 106 (+5) 113 (+2)
50000 90 (+17) 92 (+12) 97 (+7) 102 (+4) 108 (+2)
Table A.26: Baseline FFS (with risk aversion), average days cash on hand, full 10 years
Whereas the results for capitated providers are somewhat different:
# Patients 10th 25th Median 75th 90th
5000 86 (+12) 88 (+7) 91 (+3) 94 (+1) 97 (+1)
10000 88 (+9) 89 (+5) 91 (+2) 93 (+0) 95 (+0)
25000 89 (+5) 90 (+2) 91 (+1) 92 (+0) 93 (+0)
50000 90 (+2) 90 (+1) 91 (+1) 91 (+0) 92 (+0)
Table A.27: Baseline capitation (with risk aversion), average days cash on hand, first 3 years
# Patients 10th 25th Median 75th 90th
5000 88 (+23) 90 (+15) 92 (+8) 95 (+4) 99 (+2)
10000 89 (+14) 90 (+9) 92 (+5) 94 (+2) 97 (+1)
25000 90 (+7) 90 (+5) 92 (+2) 93 (+1) 95 (+1)
50000 90 (+4) 90 (+2) 91 (+1) 92 (+1) 93 (+0)
Table A.28: Baseline capitation (with risk aversion), average days cash on hand, full 10 years
Remember that providers in the simulation prefer to have at least 90 days cash on hand. Above
184
this level they will not be risk averse from their level of cash on hand, only from anticipated cash
flow losses. Below this level they will seek to return to having at least 90 days cash on hand. Now
note that, despite the fact that capitated providers at baseline are more risk averse, they end up
with lower average days cash on hand than FFS providers across the board. This goes to show why
capitated providers end up being more risk averse; they have to be just to consistently reach their
desired level of cash, whereas FFS providers reach this level far more easily with less effort to
manipulate demand.
This is also borne out in the results for the lowest days cash on hand experienced by
providers. For risk averse FFS providers, the values for lowest days cash on hand experienced are:
# Patients 10th 25th Median 75th 90th
5000 69 (+29) 77 (+17) 88 (+7) 100 (+2) 112 (+1)
10000 76 (+22) 81 (+13) 88 (+5) 98 (+2) 107 (+0)
25000 82 (+15) 85 (+9) 90 (+4) 95 (+0) 101 (+0)
50000 84 (+10) 87 (+6) 90 (+3) 94 (+1) 99 (+0)
Table A.29: Baseline FFS (with risk aversion), lowest days cash on hand, first 3 years
# Patients 10th 25th Median 75th 90th
5000 64 (+63) 71 (+40) 80 (+19) 92 (+7) 106 (+3)
10000 72 (+46) 77 (+29) 83 (+13) 93 (+5) 103 (+3)
25000 79 (+30) 82 (+18) 86 (+9) 93 (+3) 99 (+1)
50000 83 (+21) 85 (+13) 88 (+6) 92 (+2) 97 (+1)
Table A.30: Baseline FFS (with risk aversion), lowest days cash on hand, full 10 years
Whereas risk averse capitated providers experience the following lowest days cash on hand:
# Patients 10th 25th Median 75th 90th
5000 81 (+12) 85 (+7) 89 (+3) 92 (+1) 95 (+1)
10000 85 (+8) 87 (+5) 90 (+2) 92 (+0) 94 (+0)
25000 88 (+5) 89 (+3) 90 (+1) 91 (+0) 92 (+0)
50000 89 (+3) 89 (+1) 90 (+1) 91 (+0) 92 (+0)
Table A.31: Baseline capitation (with risk aversion), lowest days cash on hand, first 3 years
# Patients 10th 25th Median 75th 90th
5000 79 (+27) 82 (+18) 85 (+10) 89 (+4) 93 (+1)
185
10000 83 (+17) 85 (+11) 88 (+6) 90 (+2) 92 (+0)
25000 87 (+8) 88 (+5) 89 (+2) 90 (+1) 92 (+0)
50000 88 (+5) 89 (+3) 90 (+1) 90 (+1) 91 (+0)
Table A.32: Baseline capitation (with risk aversion), lowest days cash on hand, full 10 years
Note that risk aversion is much more impactful for increasing the minimum days cash on hand
experienced by FFS providers than it is for capitated ones, especially for smaller providers in the
more extreme quantiles.
A.4 - Simulation Objective 1: PID Values for Baseline Payment Systems without RMT
A.4.1 - Baseline Fee-for-service (FFS) PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $4.18 $39.63 $106.81 $169.78
10000 $0.00 $2.83 $28.26 $76.52 $123.20
25000 $0.00 $1.69 $15.85 $45.35 $74.94
50000 $0.00 $1.16 $10.73 $31.19 $51.63
Table A.33: Baseline FFS, average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $6.47 $14.10 $36.69 $67.75 $98.37
10000 $4.61 $9.64 $25.58 $47.02 $68.09
25000 $2.93 $6.09 $14.67 $27.58 $40.85
50000 $2.01 $4.15 $10.18 $18.84 $28.35
Table A.34: Baseline FFS, average PID per year per capita, full 10 years
A.4.2 - Baseline Capitation PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 -$3.89 -$60.26 -$156.53 -$245.49
10000 $0.00 -$2.31 -$30.98 -$102.83 -$172.58
25000 $0.00 -$1.61 -$17.09 -$57.11 -$102.91
186
50000 $0.00 -$1.22 -$11.93 -$35.28 -$62.02
Table A.35: Baseline capitation (with risk aversion), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$6.42 -$23.19 -$58.37 -$98.70 -$141.36
10000 -$4.36 -$10.94 -$35.79 -$63.84 -$90.30
25000 -$2.93 -$6.19 -$18.31 -$35.25 -$52.54
50000 -$1.97 -$4.10 -$10.67 -$21.87 -$32.22
Table A.36: Baseline capitation (with risk aversion), average PID per year per capita, full 10 years
A.4.3 - Baseline FFS with Shared Savings PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $4.03 $32.22 $95.92 $160.59
10000 $0.00 $2.79 $22.55 $65.85 $114.04
25000 $0.00 $1.66 $15.69 $43.01 $70.77
50000 $0.00 $1.08 $11.46 $31.27 $51.73
Table A.37: Baseline FFS with shared savings, average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $6.51 $13.08 $27.79 $52.53 $79.06
10000 $4.58 $9.05 $19.30 $36.61 $55.74
25000 $2.85 $5.71 $12.61 $24.43 $36.47
50000 $2.06 $4.21 $9.57 $18.02 $26.51
Table A.38: Baseline FFS with shared savings, average PID per year per capita, full 10 years
A.4.4 - Baseline FFS with Shared Savings and Losses PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $3.65 $27.10 $120.15 $214.97
10000 $0.00 $2.72 $19.16 $80.78 $152.48
25000 $0.00 $1.50 $12.60 $46.07 $86.73
187
50000 $0.00 $1.14 $10.60 $33.92 $59.46
Table A.39: Baseline FFS with shared savings and losses, average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $6.15 $11.96 $26.57 $66.06 $110.98
10000 $4.40 $8.77 $18.70 $44.91 $75.90
25000 $2.76 $5.67 $12.59 $27.15 $43.65
50000 $2.03 $4.14 $9.72 $19.77 $30.70
Table A.40: Baseline FFS with shared savings and losses, average PID per year per capita, full 10 years
A.4.5 - Determining the Scaling Factor for PID under Capitation with Risk Adjustment
In order to make a proper comparison between regular capitation and capitation with risk
ad j u st m en t , it ’s f irs t u se f u l t o l oo k at t h e d at a f or c ap it at ed p rov ider s, wi t h out r is k ad j u st m en t , under a scenario where they are also not able to risk select. In the absence of the ability to risk
select for healthier patients, we would anticipate that capitated providers would perform more
provider induced demand (downward), and this is indeed the case. Here are the values for
average PID per year per capita for regular capitated providers, without risk adjustment, and with
patient risk selection turned off:
# Patients 10th 25th Median 75th 90th
5000 $0.00 -$4.42 -$80.83 -$217.28 -$343.91
10000 $0.00 -$3.67 -$44.38 -$148.49 -$253.41
25000 $0.00 -$2.55 -$24.37 -$80.60 -$144.79
50000 $0.00 -$1.61 -$15.65 -$48.26 -$88.19
Table A.41: Regular capitation, no patient risk selection, average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$8.48 -$31.05 -$84.27 -$143.72 -$201.18
10000 -$6.32 -$16.22 -$52.71 -$93.83 -$136.87
25000 -$3.96 -$8.93 -$26.89 -$51.83 -$77.36
50000 -$2.59 -$5.81 -$15.37 -$31.09 -$47.60
Table A.42: Regular capitation, no patient risk selection, average PID per year per capita, full 10 years
188
Because there is no concurrent patient risk selection occurring, however, these values still
represent an equivalent amount of financial risk compared to the PID values for regular capitated
providers noted earlier, despite the fact that they are greater in magnitude.
Based on the math outlined in the previous chapter, we would expect that these values
would be greater in magnitude by a factor of around 1.375 compared to the baseline PID values for
regular capitated providers. In fact, because of the emergent properties of the simulation, these
values are generally slightly more than 1.375 times as big as the baseline capitation PID values,
though due to random chance there is also some variation in the increase. The following tables
show the actual magnitude increase in PID for capitated providers when patient risk selection is
turned off, compared to the PID values for regular capitated providers shown previously:
# Patients 10th 25th Median 75th 90th
5000 NaN 1.137 1.341 1.388 1.401
10000 NaN 1.587 1.432 1.444 1.468
25000 NaN 1.576 1.426 1.411 1.407
50000 NaN 1.318 1.311 1.368 1.422
Table A.43: Magnitude increase in PID when turning off risk selection for capitated providers, first 3 years
# Patients 10th 25th Median 75th 90th
5000 1.322 1.339 1.444 1.456 1.423
10000 1.449 1.483 1.473 1.470 1.516
25000 1.350 1.443 1.469 1.470 1.472
50000 1.319 1.418 1.441 1.421 1.477
Table A.44: Magnitude increase in PID when turning off risk selection for capitated providers, full 10 years
The increase in PID is a little more variable both in the short term and for smaller providers, as
well as for providers in the lower quantiles of PID where the values are generally smaller in the
absolute. Nevertheless, the factor increase in median 10 year PID averaged across all provider sizes
in the above table is 1.457, and that should work as a reasonable scaling factor for use in making a
f air c om p ar is on of t h e f in an c ial r is k of r is k ad j u st e d c ap it at ion p rov ider s, wh o don ’t p er f orm patient risk selection, with that of all other providers who do.
A.4.6 - Baseline Capitation with Risk Adjustment PID Values (Scaled)
# Patients 10th 25th Median 75th 90th
5000 -$28.39 -$75.28 -$113.20 -$200.90 -$262.96
189
10000 -$16.78 -$46.46 -$100.86 -$152.87 -$223.55
25000 -$10.86 -$23.70 -$58.37 -$104.54 -$162.35
50000 -$6.98 -$15.05 -$37.23 -$77.13 -$114.88
Table A.45: Capitation with risk adjustment, average PID per year per capita (scaled), first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$29.94 -$51.01 -$95.99 -$164.06 -$237.45
10000 -$19.15 -$36.88 -$70.33 -$125.30 -$187.72
25000 -$10.93 -$23.43 -$45.39 -$83.61 -$129.18
50000 -$6.96 -$15.82 -$31.21 -$58.43 -$89.54
Table A.46: Capitation with risk adjustment, average PID per year per capita (scaled), full 10 years
A.5 - Simulation Objective 2: PID Values for Payment Systems Paying Indirectly for RMT
A.5.1 - FFS with Shared Savings Plus RMT (2.5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $17.00 $38.55 $71.66 $141.79 $225.10
10000 $0.00 $10.48 $47.46 $106.16 $169.01
25000 $0.00 $0.00 $35.53 $73.31 $123.26
50000 $0.00 $0.00 $32.68 $43.88 $101.67
Table A.47: FFS with shared savings plus RMT (2.5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $41.52 $54.77 $73.17 $104.17 $140.20
10000 $26.10 $38.43 $55.94 $79.90 $105.93
25000 $21.19 $31.84 $43.92 $60.70 $77.96
50000 $19.81 $23.00 $35.30 $51.29 $65.16
Table A.48: FFS with shared savings plus RMT (2.5% cost), average PID per year per capita, full 10 years
A.5.2 - FFS with Shared Savings and Losses Plus RMT (3% Cost) PID Values
190
# Patients 10th 25th Median 75th 90th
5000 $0.00 $14.11 $69.18 $182.11 $274.53
10000 $0.00 $6.59 $51.43 $146.74 $225.28
25000 $0.00 $0.00 $43.01 $117.17 $168.45
50000 $0.00 $0.00 $40.02 $83.53 $142.96
Table A.49: FFS with SS/SL plus RMT (3% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $31.75 $50.10 $75.69 $128.86 $188.87
10000 $27.40 $42.93 $67.07 $102.37 $150.82
25000 $24.99 $37.69 $53.33 $84.51 $121.31
50000 $24.03 $35.67 $49.04 $73.83 $108.65
Table A.50: FFS with SS/SL plus RMT (3% cost), average PID per year per capita, full 10 years
A.5.3 - Capitation Plus RMT (5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 -$2.22 -$12.90 -$84.49 -$171.54 -$259.66
10000 -$4.67 -$13.30 -$60.53 -$125.23 -$195.75
25000 -$8.15 -$14.40 -$36.76 -$80.28 -$122.16
50000 -$10.57 -$15.06 -$29.70 -$53.58 -$81.72
Table A.51: Capitation plus RMT (5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$8.58 -$28.36 -$63.77 -$104.29 -$141.85
10000 -$6.92 -$19.23 -$44.01 -$72.39 -$100.39
25000 -$5.60 -$10.77 -$24.85 -$42.68 -$59.98
50000 -$5.43 -$8.50 -$16.91 -$27.81 -$38.50
Table A.52: Capitation plus RMT (5% cost), average PID per year per capita, full 10 years
A.5.4 - Capitation with Risk Adjustment Plus RMT (5% Cost) PID Values
191
# Patients 10th 25th Median 75th 90th
5000 -$39.94 -$86.40 -$124.51 -$203.95 -$265.17
10000 -$27.01 -$58.10 -$102.05 -$168.55 -$228.43
25000 -$17.00 -$34.14 -$70.35 -$118.92 -$179.52
50000 -$13.95 -$24.32 -$50.29 -$90.08 -$133.72
Table A.53: Risk adjusted capitation plus RMT (5% cost), average PID per year per capita (scaled), first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$31.07 -$54.19 -$100.69 -$171.48 -$237.05
10000 -$21.58 -$39.53 -$75.15 -$131.57 -$194.91
25000 -$13.81 -$26.36 -$49.73 -$88.23 -$134.86
50000 -$9.30 -$18.60 -$36.13 -$64.93 -$98.43
Table A.54: Risk adjusted capitation plus RMT (5% cost), average PID per year per capita (scaled), full 10 years
A.6 - Simulation Objective 3: PID Values for Payment Systems Paying Directly for RMT
A.6.1 - FFS Plus Paid RMT (2.5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $0.02 $19.81 $80.02 $139.25
10000 $0.00 $0.00 $9.33 $46.95 $96.24
25000 $0.00 $0.00 $4.05 $22.11 $48.12
50000 $0.00 $0.00 $1.74 $10.29 $25.50
Table A.55: FFS plus paid RMT (2.5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $5.29 $10.90 $28.78 $58.70 $89.44
10000 $3.67 $7.29 $16.87 $37.42 $59.08
25000 $2.05 $4.01 $8.13 $19.45 $32.45
50000 $1.20 $2.56 $4.89 $10.80 $20.80
Table A.56: FFS plus paid RMT (2.5% cost), average PID per year per capita, full 10 years
192
A.6.2 - FFS Plus Paid RMT (3% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $0.50 $19.02 $80.50 $138.07
10000 $0.00 $0.00 $10.02 $45.59 $92.60
25000 $0.00 $0.00 $4.53 $23.97 $49.75
50000 $0.00 $0.00 $1.98 $12.19 $27.31
Table A.57: FFS plus paid RMT (3% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $5.31 $10.87 $28.02 $57.34 $88.55
10000 $3.56 $7.27 $16.51 $38.14 $60.23
25000 $1.99 $4.03 $8.31 $19.66 $33.32
50000 $1.23 $2.54 $4.94 $10.94 $19.59
Table A.58: FFS plus paid RMT (3% cost), average PID per year per capita, full 10 years
A.6.3 - FFS Plus Paid RMT (5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $1.22 $23.19 $84.45 $145.64
10000 $0.00 $0.13 $13.54 $52.79 $98.45
25000 $0.00 $0.00 $7.49 $29.26 $55.22
50000 $0.00 $0.00 $4.08 $17.06 $33.82
Table A.59: FFS plus paid RMT (5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $5.48 $11.06 $29.79 $59.36 $90.00
10000 $3.84 $7.61 $18.76 $40.25 $61.11
25000 $2.27 $4.44 $9.93 $21.94 $34.79
50000 $1.50 $2.90 $5.73 $13.00 $22.15
Table A.60: FFS plus paid RMT (5% cost), average PID per year per capita, full 10 years
193
A.7 - Sensitivity Analysis
As a reminder, these were the baseline input parameters used in the simulation:
Input Baseline Value(s) Notes
Patient population size 5000, 10000, 25000, and
50000 (all batched
separately)
Held constant over the
course of each simulation run
Prior year weight for existing
patient needs
0.45 Constant weight, actual
patient needs (expenditures)
vary
Patient turnover rate 0.35 Varies binomially
Fraction fixed costs 0.85 Constant
“ Ov er t im e” v ar iab le c ost fraction
0.30 Constant for any patient
demand above the threshold
“ Ov er t im e” in it ializa t ion threshold
2% above baseline average Constant
Max demand modification 0.10 Constant, per year max
Table A.61: Baseline simulation input parameters
The following tables show the results of a sensitivity analysis for each of the input parameters in
the simulation. The values in the tables reflect the percent change in provider induced demand
(PID) per 20% change in the relevant input parameter from baseline for the median provider of
various sizes (positive values imply a positive correlation, while negative values imply a negative
correlation). The sensitivity analysis was performed for both FFS and capitated providers at
baseline (without RMT).
A.7.1 - Prior Year Weight for Existing Patient Needs
# Patients FFS Capitation
5000 -0.23% -2.11%
10000 -0.13% -1.97%
25000 -0.89% -0.80%
50000 -0.28% -0.25%
194
Table A.62: % change in PID per 20% change in prior year weight, median provider, first 3 years
# Patients FFS Capitation
5000 -0.28% -1.56%
10000 -0.24% -1.38%
25000 -0.18% -1.40%
50000 0.20% -0.67%
Table A.63: % change in PID per 20% change in prior year weight, median provider, full 10 years
A.7.2 - Patient Turnover Rate
# Patients FFS Capitation
5000 -0.54% -0.16%
10000 -0.93% 0.36%
25000 -1.07% -0.30%
50000 -0.57% -1.00%
Table A.64: % change in PID per 20% change in patient turnover rate, median provider, first 3 years
# Patients FFS Capitation
5000 -0.75% -0.59%
10000 -1.10% -0.55%
25000 -1.05% -0.32%
50000 -1.01% -0.96%
Table A.65: % change in PID per 20% change in patient turnover rate, median provider, full 10 years
A.7.3 - Fraction Fixed Costs
# Patients FFS Capitation
5000 -0.23% 3.18%
10000 -1.10% 0.62%
25000 -1.98% 0.35%
50000 -0.86% 0.39%
195
Table A.66: % change in PID per 20% change in fraction fixed costs, median provider, first 3 years
# Patients FFS Capitation
5000 -1.93% 4.00%
10000 -1.60% 2.92%
25000 -0.92% 2.00%
50000 -0.31% 0.56%
Table A.67: % change in PID per 20% change in fraction fixed costs, median provider, full 10 years
A.7.4 - “Overtime” Variable Cost Fraction
# Patients FFS Capitation
5000 0.89% 2.17%
10000 -0.09% 1.75%
25000 -0.05% -0.49%
50000 -0.32% 0.65%
Table A.68: % change in PID per 20% change in “overtime” cost fraction, median provider, first 3 years
# Patients FFS Capitation
5000 0.80% 2.37%
10000 0.59% 1.81%
25000 -0.04% 0.99%
50000 0.08% 0.71%
Table A.69: % change in PID per 20% change in “overtime” cost fraction, median provider, full 10 years
A.7.5 - “Overtime” Initialization Threshold
# Patients FFS Capitation
5000 -0.36% -3.79%
10000 -0.70% -3.84%
25000 -0.18% -1.61%
50000 -0.65% -0.78%
196
Table A.70: % change in PID per 20% change in “overtime” initialization threshold, median provider, first 3
years
# Patients FFS Capitation
5000 -0.81% -1.94%
10000 -0.76% -2.62%
25000 -0.61% -3.10%
50000 -0.45% -2.34%
Table A.71: % change in PID per 20% change in “overtime” initialization threshold, median provider, full 10
years
A.7.6 - Max Demand Modification
# Patients Median Median
5000 -0.03% 0.01%
10000 -0.17% -0.56%
25000 0.68% -0.05%
50000 0.19% 0.22%
Table A.72: % change in PID per 20% change in max demand modification, median provider, first 3 years
# Patients Median Median
5000 0.13% 0.55%
10000 0.01% 0.16%
25000 -0.03% 0.05%
50000 0.15% 0.11%
Table A.73: % change in PID per 20% change in max demand modification, median provider, full 10 years
A.8 - Special Scenarios: PID Values
A.8.1 - FFS with Shared Savings and Losses Plus RMT, No Savings/Losses Thresholds, PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $31.03 $73.30 $187.13 $282.89
197
10000 $0.00 $0.00 $49.88 $149.20 $227.59
25000 $0.00 $0.00 $42.15 $100.22 $162.65
50000 $0.00 $0.00 $39.77 $73.56 $137.59
Table A.74: FFS with shared savings and losses, no thresholds, average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $31.11 $49.43 $75.46 $125.10 $193.77
10000 $27.10 $41.66 $63.80 $97.37 $145.16
25000 $24.87 $37.14 $51.08 $72.72 $99.76
50000 $23.65 $26.58 $39.91 $61.02 $81.12
Table A.75: FFS with shared savings and losses, no thresholds, average PID per year per capita, full 10 years
A.8.2 - FFS with Shared Savings Plus Random Effect RMT (2.5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $17.00 $34.52 $39.43 $45.87 $64.50
10000 $0.00 $7.68 $24.91 $40.30 $54.97
25000 $0.00 -$1.66 $19.83 $30.30 $52.49
50000 $0.00 -$1.08 $21.22 $12.61 $49.93
Table A.76: FFS with shared savings plus random RMT (2.5% cost), average PID per year per capita, first 3
years
# Patients 10th 25th Median 75th 90th
5000 $35.01 $41.70 $45.38 $51.64 $61.14
10000 $21.52 $29.37 $36.64 $43.29 $50.19
25000 $18.34 $26.13 $31.31 $36.26 $41.50
50000 $17.75 $18.79 $25.73 $33.26 $38.65
Table A.77: FFS with shared savings plus random RMT (2.5% cost), average PID per year per capita, full 10
years
A.8.3 - FFS with Shared Savings and Losses Plus Random Effect RMT (3% Cost) PID Values
198
# Patients 10th 25th Median 75th 90th
5000 $0.00 $35.11 $74.23 $190.55 $280.72
10000 $0.00 $9.61 $53.65 $150.40 $224.30
25000 $0.00 $0.00 $44.33 $115.08 $170.02
50000 $0.00 $0.00 $40.87 $84.37 $141.80
Table A.78: FFS with SS/SL plus random RMT (3% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $32.95 $51.86 $77.52 $132.29 $195.19
10000 $27.91 $43.92 $68.28 $107.00 $153.42
25000 $25.46 $38.68 $56.72 $86.13 $122.10
50000 $24.35 $36.68 $50.98 $75.35 $104.66
Table A.79: FFS with SS/SL plus random RMT (3% cost), average PID per year per capita, full 10 years
A.8.4 - Capitation Plus Random Effect RMT (5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 -$2.90 -$13.89 -$92.42 -$177.73 -$261.06
10000 -$4.78 -$14.03 -$60.93 -$123.92 -$197.34
25000 -$8.49 -$14.75 -$41.81 -$85.70 -$128.12
50000 -$10.70 -$15.90 -$33.84 -$61.65 -$95.01
Table A.80: Capitation plus random RMT (5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$9.42 -$31.98 -$66.92 -$107.76 -$146.20
10000 -$7.37 -$20.80 -$45.47 -$74.01 -$102.99
25000 -$6.62 -$13.93 -$29.08 -$47.77 -$67.52
50000 -$6.28 -$10.80 -$21.19 -$35.03 -$49.11
Table A.81: Capitation plus random RMT (5% cost), average PID per year per capita, full 10 years
A.8.5 - Risk Adjusted Capitation Plus Random Effect RMT (5% Cost) (Scaled) PID Values
199
# Patients 10th 25th Median 75th 90th
5000 -$37.52 -$85.86 -$125.81 -$204.02 -$267.37
10000 -$28.56 -$63.68 -$104.00 -$171.18 -$230.70
25000 -$17.66 -$36.11 -$72.27 -$120.33 -$179.02
50000 -$14.52 -$26.79 -$54.64 -$94.21 -$135.37
Table A.82: Risk adjusted capitation plus random RMT (5% cost) (scaled), average PID per year per capita,
first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$31.04 -$55.67 -$101.32 -$171.66 -$235.74
10000 -$25.53 -$42.88 -$79.29 -$137.04 -$193.28
25000 -$14.11 -$27.58 -$52.00 -$91.10 -$135.51
50000 -$10.32 -$20.17 -$38.06 -$68.31 -$104.63
Table A.83: Risk adjusted capitation plus random RMT (5% cost) (scaled), average PID per year per capita, full
10 years
A.8.6 - FFS Plus Random Effect Paid RMT (5% Cost) PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $1.19 $24.34 $89.26 $150.01
10000 $0.00 $0.00 $14.85 $54.61 $103.18
25000 $0.00 $0.00 $9.23 $33.24 $62.57
50000 $0.00 $0.00 $5.99 $21.52 $41.62
Table A.84: FFS plus random paid RMT (5% cost), average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $5.62 $11.85 $31.94 $61.44 $93.19
10000 $3.87 $7.94 $19.85 $42.07 $65.38
25000 $2.49 $5.22 $11.92 $26.10 $40.85
50000 $1.97 $3.95 $8.29 $18.37 $29.79
Table A.85: FFS plus random paid RMT (5% cost), average PID per year per capita, full 10 years
200
A.8.7 - FFS with Shared Savings Plus RMT (2.5% Cost) Lower Fixed and “Overtime” Costs PID
Values
# Patients 10th 25th Median 75th 90th
5000 $33.33 $59.23 $94.07 $157.77 $231.15
10000 $7.79 $29.87 $62.18 $114.81 $185.05
25000 $0.00 $0.00 $45.54 $89.17 $140.84
50000 $0.00 $0.00 $42.84 $53.51 $119.65
Table A.86: FFS with shared savings plus RMT (2.5% cost), reduced fixed and “overtime” costs, average PID
per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $58.11 $73.40 $92.58 $117.20 $147.63
10000 $34.57 $49.74 $68.68 $91.57 $118.60
25000 $26.92 $40.75 $55.21 $73.01 $96.53
50000 $25.79 $29.37 $45.44 $66.21 $84.29
Table A.87: FFS with shared savings plus RMT (2.5% cost), reduced fixed and “overtime” costs, average PID per
year per capita, full 10 years
A.8.8 - FFS with Shared Savings and Losses Plus RMT (3% Cost) Lower Fixed and “Overtime”
Costs PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $49.75 $91.75 $219.73 $295.66
10000 $0.00 $20.03 $70.31 $184.19 $253.79
25000 $0.00 $0.00 $55.89 $155.17 $203.71
50000 $0.00 $0.00 $51.98 $110.69 $176.61
Table A.88: FFS with shared savings and losses plus RMT (3% cost), reduced fixed and “overtime” costs,
average PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 $47.72 $68.07 $97.61 $161.37 $228.95
10000 $39.16 $62.57 $88.33 $144.47 $200.78
201
25000 $34.12 $50.72 $80.67 $129.12 $175.12
50000 $31.66 $47.38 $66.55 $114.67 $157.97
Table A.89: FFS with shared savings and losses plus RMT (3% cost), reduced fixed and “overtime” costs,
average PID per year per capita, full 10 years
A.8.9 - Capitation Plus RMT (5% Cost) Lower Fixed and “Overtime” Costs PID Values
# Patients 10th 25th Median 75th 90th
5000 -$2.00 -$12.39 -$59.19 -$123.37 -$193.37
10000 -$4.20 -$12.65 -$43.73 -$96.32 -$143.11
25000 -$8.22 -$14.13 -$34.48 -$65.03 -$95.13
50000 -$10.59 -$15.11 -$29.93 -$52.56 -$73.21
Table A.90: Capitation plus RMT (5% cost), reduced fixed and “overtime” costs, average PID per year per
capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$8.06 -$17.46 -$41.17 -$71.60 -$101.88
10000 -$6.40 -$12.75 -$30.14 -$51.92 -$74.12
25000 -$5.77 -$9.71 -$20.69 -$33.95 -$47.74
50000 -$5.48 -$8.61 -$16.18 -$25.55 -$34.75
Table A.91: Capitation plus RMT (5% cost), reduced fixed and “overtime” costs, average PID per year per
capita, full 10 years
A.8.10 - Risk Adjusted Capitation Plus RMT (5% Cost) (Scaled) Lower Fixed and “Overtime” Costs
PID Values
# Patients 10th 25th Median 75th 90th
5000 -$26.60 -$46.10 -$77.12 -$106.27 -$146.07
10000 -$22.84 -$35.49 -$57.70 -$83.67 -$109.50
25000 -$20.65 -$27.48 -$41.14 -$58.47 -$76.39
50000 -$18.95 -$23.91 -$33.20 -$46.33 -$58.71
Table A.92: Risk adjusted capitation plus RMT (5% cost) (scaled), reduced fixed and “overtime” costs, average
202
PID per year per capita, first 3 years
# Patients 10th 25th Median 75th 90th
5000 -$16.58 -$28.08 -$45.88 -$75.09 -$108.26
10000 -$12.76 -$21.32 -$35.30 -$55.49 -$78.81
25000 -$9.57 -$14.62 -$23.41 -$35.85 -$52.48
50000 -$7.97 -$11.67 -$17.77 -$26.99 -$37.70
Table A.93: Risk adjusted capitation plus RMT (5% cost) (scaled), reduced fixed and “overtime” costs, average
PID per year per capita, full 10 years
A.8.11 - FFS Plus Paid RMT (5% Cost) Lower Fixed and “Overtime” Costs PID Values
# Patients 10th 25th Median 75th 90th
5000 $0.00 $1.12 $21.98 $83.55 $137.03
10000 $0.00 $0.24 $13.69 $52.84 $95.54
25000 $0.00 $0.00 $6.64 $27.02 $51.84
50000 $0.00 $0.00 $3.63 $17.24 $33.43
Table A.94: FFS plus paid RMT (5% cost), reduced fixed and “overtime” costs, average PID per year per capita,
first 3 years
# Patients 10th 25th Median 75th 90th
5000 $5.55 $11.14 $27.48 $53.67 $82.94
10000 $3.66 $7.54 $17.67 $36.71 $56.94
25000 $2.22 $4.38 $9.17 $19.90 $33.07
50000 $1.48 $2.79 $5.86 $12.89 $21.43
Table A.95: FFS plus paid RMT (5% cost), reduced fixed and “overtime” costs, average PID per year per capita,
full 10 years
Abstract (if available)
Abstract
It is difficult to pay for prevention in health care. Most payment system alternatives to fee-for-service (FFS) try to encourage prevention by creating financial incentives that reward providers for reducing the medical needs and patient harm that lead to demand for care and health expenditures. But due to uncertainty in medical treatment and health outcomes, providers (especially smaller ones) might invest in preventive interventions only to fail to see expected returns. The chance for financial incentives not to cover the cost to providers of prevention creates financial risk for those providers who invest in it, and this financial risk could be a barrier that makes prevention hard to sustain. I created a simulation model to test the theoretical contribution of variation in patient needs to the financial risk of providers under various payment system alternatives who are attempting to practice prevention (what I call “risk mitigation treatment” or RMT for the purposes of the model). The results of the model suggest that paying for RMT directly with its own up-front payment while paying for all other care needs with FFS is associated with less provider financial risk than alternative payment systems that pay for RMT indirectly out of payer or provider savings. FFS plus a direct payment for RMT was unambiguously superior to paying for RMT through ACO shared savings (with or without shared losses). Paying for RMT through full capitation was associated with more provider financial risk than FFS plus direct payment for RMT but also less payer costs. Choosing between these two payment systems to fund RMT interventions will require consideration of provider size, provider cost structure, and the amount of variation in medical needs and patient harm in the provider’s patient population as well as concerns about the unintended financial incentives of FFS vs capitation.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Investigation of health system performance: effects of integrated triple element method of high reliability, patient safety, and care coordination
PDF
Developing an agent-based simulation model to evaluate competition in private health care markets with an assessment of accountable care organizations
PDF
Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy
PDF
An agent-based model to study accountable care organizations
PDF
Using a human factors engineering perspective to design and evaluate communication and information technology tools to support depression care and physical activity behavior change among low-inco...
PDF
Total systems engineering evaluation of invasive pediatric medical therapies conducted in non-clinical environments
PDF
A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system
PDF
A system framework for evidence based implementations in a health care organization
PDF
BRIM: A performance-based Bayesian model to identify use-error risk levels in medical devices
PDF
Clinical prediction models to forecast depression in patients with diabetes and applications in depression screening policymaking
PDF
A framework for examining relationships among electronic health record (EHR) system design, implementation, physicians’ work impact
PDF
Human and organizational factors of PTC integration in railroad system and developing HRO-centric methodology for aligning technological and organizational change
PDF
A risk analysis methodology to address human and organizational factors in offshore drilling safety: with an emphasis on negative pressure test
PDF
Vehicle routing and resource allocation for health care under uncertainty
PDF
Essays on access to prescription drugs
PDF
Optimizing chronic disease screening frequencies considering multiple risk factors and patient behavior
PDF
Systems engineering and mission design of a lunar South Pole rover mission: a novel approach to the multidisciplinary design problem within a spacecraft systems engineering paradigm
PDF
Essays in health economics and provider behavior
PDF
Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance
PDF
Inter professional education and practice in the health care setting: an innovative model using human simulation learning
Asset Metadata
Creator
Franklin, John
(author)
Core Title
Designing health care provider payment systems to reduce potentially preventable medical needs and patient harm: a simulation study
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Industrial and Systems Engineering
Publication Date
07/19/2017
Defense Date
04/25/2017
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
financial risk,health care,incentives,OAI-PMH Harvest,payment,Prevention,simulation,systems
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Meshkati, Najmedin (
committee chair
), Wu, Shinyi (
committee chair
), Capron, Alexander (
committee member
)
Creator Email
franklin.johnp@gmail.com,jpfrankl@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-403459
Unique identifier
UC11264069
Identifier
etd-FranklinJo-5545.pdf (filename),usctheses-c40-403459 (legacy record id)
Legacy Identifier
etd-FranklinJo-5545.pdf
Dmrecord
403459
Document Type
Dissertation
Rights
Franklin, John
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
financial risk
incentives
payment
simulation
systems