University of Southern California Dissertations and Theses

Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

(USC Thesis Other)

Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ECONOMIC, CLINICAL, AND BEHAVIORAL OUTCOMES FROM
MEDICAL AND PHARMACEUTICAL TREATMENTS

by
Ning Ning, M.S.

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(HEALTH ECONOMICS)

December 2019

Copyright 2019 Ning Ning
ii
Table of Contents

Acknowledgements ...................................................................................................................... iv
Chapter 1: Introduction ............................................................................................................... 1
References .................................................................................................................................. 5
Chapter 2: Long-Term Trends in the Quality and Cost of Surgical Care, 2002–2014 .......... 7
Introduction ............................................................................................................................... 7
Methods .................................................................................................................................... 10
Data ....................................................................................................................................... 10
Surgeries ............................................................................................................................... 10
Sample ................................................................................................................................... 11
Analysis................................................................................................................................. 11
Innovation ............................................................................................................................. 13
Results ...................................................................................................................................... 13
Discussion ................................................................................................................................ 17
Conclusion ............................................................................................................................... 21
Tables and & Figures ............................................................................................................. 22
Appendix .................................................................................................................................. 26
References ................................................................................................................................ 31
Chapter 3: Long-Term Trends in the Productivity of Surgical Care, 2002–2014 ................ 34
Introduction ............................................................................................................................. 34
Methods .................................................................................................................................... 37
Data Sources ......................................................................................................................... 37
Sample selection ................................................................................................................... 37
Analysis................................................................................................................................. 38
Results ...................................................................................................................................... 40
Summary Statistics................................................................................................................ 40
Surgical Cost and Quality ..................................................................................................... 40
Surgical Productivity ............................................................................................................ 41
Discussion ................................................................................................................................ 42
iii
Conclusion ............................................................................................................................... 48
Tables and Figures .................................................................................................................. 50
References ................................................................................................................................ 55
Chapter 4: Variation in Demand Response to FDA Black Box Warnings ............................ 58
Introduction ............................................................................................................................. 58
Methods .................................................................................................................................... 60
Data Sources ......................................................................................................................... 60
Adjusting Warning Dates ...................................................................................................... 60
Selecting Sample Drugs ........................................................................................................ 61
Identifying Demand Response in Aggregated Days of Supply ............................................ 61
Study Population and Drugs for Statistical Analysis ............................................................ 62
Quantifying Cessation Among Current Users ...................................................................... 62
Quantifying Initiation in New Users ..................................................................................... 64
Disparities ............................................................................................................................. 64
Results ...................................................................................................................................... 64
Impact of Safety Warnings on Total Aggregated Days of Supply ....................................... 64
Impact of Safety Warnings on Current Users ....................................................................... 65
Disparities in Cessation Rates of Current Users ................................................................... 66
Impact of Safety Warnings on New Users ............................................................................ 67
Discussion ................................................................................................................................ 68
Conclusion ............................................................................................................................... 72
Tables and Figures .................................................................................................................. 73
Appendix .................................................................................................................................. 75
References ................................................................................................................................ 77
Chapter 5: Summary and Future Research Directions ........................................................... 80

iv
Acknowledgements

First and foremost, I would like to thank my advisor, Professor John Romley, for his
immeasurable guidance and support. He is not only a passionate and excellent researcher, but
also a wonderful mentor. I am very grateful for all he taught me. Not only did I learnt how to
properly conduct and present a scientific study, but also, I learnt how to arrange my time and life
from him. He cares about students, and is always ready to lead them with his wisdom and
experience.

It has also been an honor for me to work with 2 other wonderful professors during my PhD
study. I want to thank Dr. Geoffrey Joyce, he kindly accepted my proposal to mentor my third
project regarding the Black Box Warning for my dissertation. He is kind, passionate, and
responsible. He always thinks one step forward for project and for his students. I am truly
grateful to work for Dr. Erin Trish as a Research Assistant. She is kind and supportive whenever
I have doubts and questions both about my project or my life. I am always amazed about how
tender and strong she could be at the same time. I also want to thank the other 2 professors on
my proposal defense committee, Dr. Rebecca Myerson and Dr. Michael Nichol.

There are many others that I thank for. I want to thank Dr. Jeff McCombs. He always welcomes
students with an open door and treats us with his kindness and stories. I then want to thank Laura
Gascue. She taught me how to be a rigorous analyst along the way when we worked on projects
together for many years. Next, I want to thank our collaborators Dr. Yang Lu and Dr. Alex
Haynes for patiently providing professional and clinical information whenever we needed. Thank
Tommy Chiou for putting a lot of effort into writing paper for our project. Thank Yuchen Ding
v
for our time together exploring many possibilities regarding BBWs during the first year of our
PhD study.

I am grateful to have gotten to know a lot of great friends. I would especially thank Meng Li and
Yingying Zhu for they encouraged and inspired me a lot. When I was preparing my dissertation
defense, I was really touched that they kindly spent more than 5 hours in one afternoon listening
to my presentation and asking questions. I also want to thank Drs. Wendy Cheng, Cynthia Gong,
Yifan Xu, and Xiayu Jiao for sharing their experience with me. I like to thank Liming Xie, Dai
Hong, and Vicky Yang for the pressure they shared and encouragement they provided me, even
though they are not majoring in Health Economics. I am grateful to have you with me, and am
lucky to call you my friends.

Finally, I have been very lucky to have a wonderful and big family. In memory of my and my
husband’s grandmothers, they were the best grandmothers that I can dream of. They raised us up
with full of their love and I know they would always support us unconditionally. I sincerely wish
that they could see our marriage and graduations, and I hope to show them that we will not let
them down. My deepest gratitude goes to my parents. They will always be there for me, no
matter how busy they are. Their great achievements in their own fields motivate me to fight on. I
am always very proud of them, and feel really blessed to be their daughter. I also want to thank
my husband’s parents, their kindness makes me feel like home, and their passion in their lives
and work inspires me a lot. I want to thank my dear Aunt Jolin and Uncle Bobby for loving me
as their own. In addition, I want to express my best wishes and love to the rising stars of our
family, Yuze, Jiachen, and Ziyang. Wish you all the best. Finally, I need to tell my husband,
vi
George, how truly grateful I am for his unconditional love, encouragement, and support. Thank
you for being with me in sickness and health, in joy and in sorrow.

In memory of the people we love
In memory of all that we gained and lost in 2019
1
Chapter 1: Introduction

Healthcare spending in the United States increased by 4.8% in 2016, and 3.9% in 2017,
according to the estimates released by the National Health Expenditure Accounts. These
numbers were roughly double to triple the inflation amounts in those same years. By the end of
2017, the total national health expenditures reached $3.5 trillion, roughly $11,000 per person.
Healthcare spending accounted for 17.9% of the nation’s gross domestic product (GDP) in 2017
in the United States (Medicare, 2018). Among them, 32.7% of the spending came from hospitals,
with about 4.6% increase in 2017. The total prescription drug expenditure took about 9.5% of the
total national health expenditures, with a decreasing growth rate (0.4%) in 2017 comparing to the
previous year’s (Medicare, 2018). However, increasing health care expenditures did not
necessarily correlate with an improvement in health outcomes. For example, as reported by the
Centers for Disease Control and Prevention (CDC), death rates increased by 0.4% from 2016 to
2017, after adjusting for age. CDC also reported increasing inpatient hospital deaths from 2000
to 2010. These numbers signal an increasing need to improve quality and safety of healthcare,
and at the same time decrease cost of US health treatments.

Health outcomes research can provide guidance on a wide range of interventions, decisions, and
treatments, as well as provide quality controls for the safety, effectiveness, equity, efficiency,
responsiveness of the healthcare system. It is a methodology used to measure the correlation
between treatments or interventions provided and the outcomes achieved (Ellis, 2019). Outcomes
which are commonly studied can be grouped into economic, clinical, and behavioral outcomes.
Commonly seen economic measures include costs, charges, and payments of specific treatments
or medications. Commonly used clinical measures are mortality/survival rates, complications,
2
biomarker levels, and readmission rates from clinical trials or claims data sets. Behavioral
outcome studies focus on behaviors (decisions, cessation, compliance, initiation, etc.) of patients
and physicians in order to reveal, predict, prevent, and manage health-related issues. Common
treatments for health outcomes research include the furnishing of medicine, medical or surgical
treatment, nursing, hospital service, etc (Oregon Legislature, 2017). In this dissertation, we aim
to evaluate a wide range of treatment methods, including both inpatient surgical care and
prescription medications, and examine the changes in quality, safety and cost of these health
services using economic, clinical, and behavioral outcomes.

Our research uses data from Medicare, which covers certain people under 65 with disabilities
and most of US citizens aged 65 and older (Resdac, 2019). We received access to a 20% random
sample of Medicare Part A, B, and D data, which contain fee-for-service claims submitted by
inpatient hospital providers, professional providers, or Medicare prescription drug plan,
respectively. Medicare data sets contain extensive patient information, including demographic
and covered service data, including age, race, birth date, death date, service dates, diagnosis,
procedures, payments, and service charges (Virnig and Parsons, 2018). Medicare data is also one
of the richest data sources in the country, with an estimated > 45 million beneficiaries,
accounting for approximately 98% of adults aged 65+, which allows us to perform detailed
subgroup analysis for inpatient surgical care and prescription medications.

There are deep concerns regarding whether quality of surgical care is improved while cost of
surgical care is well controlled. It is a major point of interest for policy makers, healthcare
providers, and patients alike. Previous studies failed to provide evidences of long-term changes
3
in quality and cost over a wide range of inpatient surgical care. To address this question, Chapter
2 used data from Medicare Part A and B, and focused on the 11 surgical groups which contained
hundreds of surgical procedures, which were well defined in the previous literature (Weiser et
al., 2011). Compared to previous studies, we introduced an improved metric of surgical quality,
incorporating both inpatient mortality (inside or outside hospitals) and unplanned readmissions,
as well as a better cost metric by incorporating both inpatient costs and non-institutional charges.
We also explored innovations in surgical care that had been implemented from 2002– 2014.

To move one step forward from evaluating cost and quality of surgical care, we applied the
quality-adjusted productivity metric – the output that hospitals produce from inputs such as labor
and capital – to inpatient surgical cares in Chapter 3. Productivity growth reflects whether there
were improvements in efficiency and technology in inpatient surgical care. We assessed long-
term trend in productivity growth among U.S. hospitals from 2002 to 2014. In this way, we can
examine whether productivity growth of inpatient surgical care had lagged behind that of other
industries from a societal perspective (Romley et al., 2015; Gu et al., 2019).

In Chapter 4, we focused on Medicare Part D prescription data, and explored behavioral
outcomes in response to warnings about unsafe adverse effects. We aimed to study a wide range
of commonly used oral or transdermal drugs that received Black Box Warnings (BBWs), the
most severe class of warnings issued by FDA, from 2007 to 2013, and to track their
corresponding behavioral response in drug use among the elderly US population after issuance of
warnings. Moreover, we examined behavioral outcomes of subgroups, such as cessation among
current users, who had been on the drug for some time, and initiation among new users, who
4
sought to start a new medication. We also explored disparities in behavioral outcomes (cessation
or initiation) across subgroups of different patient or plan characteristics among current users or
new users.

Finally, we summarized our findings in Chapter 5, and provided some directions for future
research in these areas.

5
References

Anon. (2019). Different Black Box Warning Labeling for Same-Class Drugs. Available from:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3101972/ [Accessed 19 September 2019].
Anon. 2005. Warning Labels on High-risk Drugs Inconsistently Heeded by Doctors. Harvard
Gazette. Available from: https://news.harvard.edu/gazette/story/2005/11/warning-labels-on-high-
risk-drugs-inconsistently-heeded-by-doctors/ [Accessed 11 September 2019].
Ashton, Del Junco, Souchek, Wray, Mansyur. 1997. The Association Between the Quality of
Inpatient Care and Early Readmission: A Meta-Analysis of the Evidence. Medical Care 35(10):
1044.
Berwick DM, Hackbarth AD. 2012. Eliminating Waste in US Health Care. JAMA 307(14):
1513–1516.
Boulware LE, Cooper LA, Ratner LE, LaVeist TA, Powe NR. 2003. Race and Trust in the
Health Care System. Public Health Reports (1974-) 118(4): 358–365.
Cheng CM, Shin J, Guglielmo BJ. 2014. Trends in Boxed Warnings and Withdrawals for Novel
Therapeutic Drugs, 1996 Through 2012. JAMA Internal Medicine 174(10): 1704.
DiNicolantonio JJ, Serebruany VL. 2013. Challenging the FDA Black Box Warning for High
Aspirin Dose with Ticagrelor in Patients with Diabetes. Diabetes 62(3): 669–671.
Dorsey ER. 2010. Impact of FDA Black Box Advisory on Antipsychotic Medication Use.
Archives of Internal Medicine 170(1): 96.
Dusetzina SB, Caleb Alexander G. 2011. Drug vs Class-Specific Black Box Warnings: Does
One Bad Drug Spoil the Bunch? Journal of General Internal Medicine 26(6): 570–572.
Ellis L. 2019. Using Health Outcomes Research to Improve Quality of Care | Executive and
Continuing Professional Education | Harvard T.H. Chan School of Public Health. Available
from: https://www.hsph.harvard.edu/ecpe/using-health-outcomes-research-to-improve-quality-
of-care/ [Accessed 19 October 2019].
Fleming ST. 1991. The Relationship Between Quality and Cost: Pure and Simple? Inquiry 28(1):
29–38.
Gallini A, Andrieu S, Donohue JM, Oumouhou N, Lapeyre-Mestre M, Gardette V. 2014. Trends
in Use of Antipsychotics in Elderly Patients with Dementia: Impact of National Safety Warnings.
European Neuropsychopharmacology 24(1): 95–104.
Ghali WA, Ash AS, Hall RE, Moskowitz MA. 1997. Statewide Quality Improvement Initiatives
and Mortality After Cardiac Surgery. JAMA 277(5): 379–382.
6
Lasser KE, Seger DL, Yu DT, Karson AS, Fiskio JM, Seger AC, Shah NR, Gandhi TK,
Rothschild JM, Bates DW. 2006. Adherence to Black Box Warnings for Prescription
Medications in Outpatients. Archives of Internal Medicine 166(3): 338–344.
Lester P, Kohen I, Stefanacci RG, Feuerman M. 2011. Antipsychotic Drug Use Since the FDA
Black Box Warning: Survey of Nursing Home Policies. Journal of the American Medical
Directors Association 12(8): 573–577.
Medicare. 2018. Historical National Health Accounts by Service Type and Funding Source.
Available from: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-
and-Reports/NationalHealthExpendData/NationalHealthAccountsHistorical.html [Accessed 19
October 2019].
Oregon Legislature. 2017. Definition of Health Care Services - Oregon Legal Glossary.
Available from: https://www.oregonlaws.org/glossary/definition/health_care_services [Accessed
19 October 2019].
Qato DM, Trivedi AN, Mor V, Dore DD. 2016. Disparities in Discontinuing Rosiglitazone
Following the 2007 FDA Safety Alert. Health Affairs (Project Hope) 31(1): 188–198.
Shi L, Zhao Y, Szymanski K, Yau L, Fonseca V. 2011. Impact of Thiazolidinedione Safety
Warnings on Medication Use Patterns and Glycemic Control Among Veterans with Diabetes
Mellitus. Journal of Diabetes and Its Complications 25(3): 143–150.
Van Leeuwen E, Petrovic M, van Driel ML, De Sutter AI, Stichele RV, Declercq T, Christiaens
T. 2018. Discontinuation of Long-Term Antipsychotic Drug Use for Behavioral and
Psychological Symptoms in Older Adults Aged 65 Years and Older with Dementia. Journal of
the American Medical Directors Association 19(11): 1009–1014.
Virnig B, Parsons H. 2018. Strengths and Limitations of CMS Administrative Data in Research |
ResDAC. Available from: https://www.resdac.org/articles/strengths-and-limitations-cms-
administrative-data-research [Accessed 18 October 2019].

7
Chapter 2: Long-Term Trends in the Quality and Cost of Surgical
Care, 2002–2014
Ning Ning, Alex Haynes, and John Romley

Introduction
Controlling healthcare spending while simultaneously improving the quality of service has been
an ongoing effort in the US. Healthcare spending in the US was reported to increase by 5.3% in
2014 (Martin et al., 2016), following relatively lower growth in healthcare spending in 2013
(3.5%) and 2012 (3.7%; Hartman et al., 2015; Martin et al., 2014). By the end of 2014,
healthcare spending in the US reached $3.0 trillion (Martin et al., 2016). Some have suggested,
however, that a substantial proportion of healthcare spending brings little or no value (Altman
and Mechanic, 2018). The Institute of Medicine recently stated that “the only sensible way to
restrain costs is to enhance the value of the healthcare system, thus extracting more benefit from
the dollars spent.”

Costs and mortality rates have long been used to evaluate surgeries. In addition, attention has
been increasingly paid to readmission rates following surgery. In 1983, the Centers for Medicare
and Medicaid Services (CMS) introduced the Prospective Payment System to reimburse
hospitals for fixed amounts that were predetermined on the basis of a classification system for
the service rendered. Through this method, the CMS was able to reduce hospital admissions and
lengths of stay. In 2012, the CMS introduced the Hospital Value Based Purchasing Program to
improve the safety and quality of inpatient care. Also in 2012, the Hospital Readmissions
8
Reduction Program (HRRP), based on the Affordable Care Act (ACA), was established to
further control readmissions and unnecessary costs by linking reimbursements for surgeries
made through the Inpatient Prospective Payment System (IPPS) to readmission rates (Medicare,
cms.gov). These policy changes were made to simultaneously control hospitalization costs and
decrease unnecessary readmissions following surgery.

Apart from the policies that were implemented to improve healthcare quality and reduce costs,
new technological innovations and guidelines were introduced to improve the safety and
outcomes of surgery (Skinner et al. 2011; Cutler et al. 2007). The policy changes and innovations
appeared to make surgeries safer, but they also had the potential to increase the use of surgery as
a treatment option for patients in more critical health conditions, which might lead to a decrease
in the rate of high-quality outcomes. Although policy changes and new technologies might lead
to cost control, they also have the potential to increase total surgical cost by driving up the prices
of surgical services. Hence, the real long-term changes in healthcare quality and cost remain
unknown.

There are many published studies regarding the quality and cost of surgeries. Mortality rates are
commonly used to quantify surgical quality. Regarding readmission rates, most studies focus
primarily on a limited number of treatments, hospitals, or reforms and use decreases in short-
term readmission rates as an indicator of improved quality (Kulaylat et al., 2018; Henke et al.,
2017; Gonzalez, Shih, et al., 2014; Gonzalez, Girotti, et al., 2014; Salerno et al., 2017). A few
studies have looked at underlying factors that affect readmission rates, such as specific
characteristics of patients or hospitals (Glebova et al., 2016; Tsai et al., 2013). A lot of studies
9
regarding surgical cost are cost-effectiveness analysis (Cheng et al., 2019; Fitzsimmons et al.,
2014).

There have been a few studies of long-term changes in the quality (mortality and readmission
rates) and cost of healthcare services. One study that focused on the trends in mortality rates
during 1996–2006 for 11 surgery types (Weiser et al., 2011) found that mortality rates following
surgery had improved significantly for only 1 out of 11 surgical groups over those years. That
study did not look at surgical costs or readmissions, however. Another study examined
productivity, defined on the basis of mortality and readmission rates, in the treatment of heart
attack, heart failure, and pneumonia among patients on Medicare during 2002–2011 (Romley et
al., 2015). That study did not examine long-term survival rates, unplanned readmission rates, and
cost separately, nor did it look at inpatient surgical care. To our knowledge, there are no
published studies of long-term patterns in the cost and quality, incorporating mortality rates and
unplanned readmission rates, of a wide range of inpatient surgeries at the national level in the
US.

In this study, we sought to answer three questions: What are the long-term patterns of surgery
quality and cost? Is the variation in the quality and cost of surgical care fully explained by
changes in patient demographics and illness severity? What innovations in surgical care have
been implemented from 2002–2014? We hope that by answering those questions and studying
the long-term trends in survival rates, readmission rates, and cost, we can provide some
meaningful insight on the effectiveness of recent policy changes.

10
Methods
Data
Our primary data source was the Medicare 20% Part A and Part B claims data from 2002–2014.
The claims data include patient diagnosis codes, inpatient service procedure codes, the non-
institutional service Healthcare Common Procedure Coding System (HCPCS) codes, dates of
service, hospital discharge dates, provider identification for inpatient stays at short-term
hospitals, and service charges. The research-identifiable version of the data also reports patient
demographic information, zip codes, and death dates. In addition, we integrated information
from the US Census to add community characteristics to the data.

Surgeries
We included specific categories of surgeries on the basis of International Classification of
Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes that were defined and
selected by five independent surgeons in a previous study (Weiser et al. 2011). The included
surgeries comprised non-operating room and operating room procedures, involving “incision,
excision, suturing, and manipulation of tissue, usually with regional or general anesthesia or
profound sedation to control pain” (Weiser et al. 2011). Similar interventions were grouped into
11 categories based on the Clinical Classification System (CCS) of the Agency for Healthcare
Research and Quality (AHRQ; Weiser et al., 2011). The 11 CCS categories included in our study
are listed in Table 1.

11
Sample
We identified patients who underwent at least one of the selected inpatient surgical procedures
between 2002 and 2014. We used inpatient admission types as one of the proxies for patient
illness severity, and excluded patients whose inpatient admission types were missing. Finally, we
limited the cohorts to patients who enrolled in fee-for-service (FFS) Medicare for the study
period.

Analysis
To quantify surgical quality, we first created a binary variable indicating whether or not a
hospitalization case is a high-quality stay. A high-quality stay was one in which a patient
survived for at least 30 days after his/her initial hospital admission and was not readmitted again
unplanned within 30 days after hospital discharge. The methodology to identify unplanned
readmissions was provided by the CMS (CMS, 2013; CMS, 2016).

To quantify surgical cost, we summed the costs of inpatient stays and physician services based
on the Medicare Part A and Part B claims. For the cost of inpatient stays, we followed the
methodology described by Romley et al. (2011). We first used the cost-to-charge ratio, which
hospitals report to the CMS, to estimate inpatient costs based on total Medicare hospital charges
(Romley et al., 2015; Medicare, 2012). We then adjusted the labor-related portion of the base
payment rate in the IPPS using the hospital wage index to account for geographic differences in
labor prices (Romley et al., 2015). Next, we adjusted for inflation by converting the costs to 2014
US dollars using the CMS inpatient market basket of the prospective payment system. In order to
incorporate non-institutional costs, we selected all Part B claims billed by physicians during each
12
inpatient stay. Finally, we converted the costs of non-institutional claims to 2014 US dollars
using the Medicare economic index and economy-wide private nonfarm business multifactor
productivity from the Physician Fee Schedule.

We used logistic regression to evaluate the changes in surgical quality during the study period.
We used multivariate linear models to evaluate the changes in cost during the same period. We
first regressed the outcomes on a group of year indicators using the whole sample from 2002 to
2014 in order to estimate the year-by-year changes in quality and cost. We then used only the
data from 2002 and 2014 to regress the outcomes on the year indicator and calculate the
annualized growth rates of quality and cost. In that way, we could identify cumulative changes
that took place from 2002 to 2014. All regressions were carried out at beneficiary level. The
standard errors were clustered at hospital level to account for correlations in the residuals.

To address the potential confounding effect of changes in the patients’ illness severity, we
adjusted for admission types, from the Medicare claims, and Charlson comorbidities, which
indicated in the hospital discharge records (Romley et al., 2015). In addition, we adjusted for
patient demographics and socio-demographic information, including age, gender, race/ethnicity,
area poverty rate, percentage of elderly residents with disabilities, percentage of residents with a
high level of education, and other factors. We also defined and controlled for the teaching status
of the hospitals according to a previously described methodology (Romley et al., 2015).
Specifically, we defined a teaching hospital as a hospital with a resident-to-bed ratio greater than
0.6. We then ran regressions with these control variables.

13
Since CMS define unplanned readmissions based on whether patients got readmitted by hospitals
within 30 days after hospitalization discharges, so there is a possibility that some beneficiaries
did not have readmissions because they died within this 30-day period. In order to address this
possibility, we redefined the quality measure in a sensitivity analysis. In that analysis, patients
were considered to have a high-quality stay only if they survived for 30 days after initial hospital
discharge without any unplanned readmissions to the hospital. We also redefined the cost
measure in the sensitivity analysis as the mean cost per day during hospitalization. Finally, for
patients with heart attack only, we adjusted for the specific locations of the heart attack and
predicted likelihood of death during the hospital stay together with the control variables we
mentioned above.

Innovation
We defined surgical innovations as clinically distinct procedures that were nonexistent or only in
limited use in 2002. Because the International Classification of Diseases, Tenth Revision,
Clinical Modification (ICD-10-CM) did not go into effect until October 2015, we identified
innovations using their five-digit ICD-9-CM procedure codes. We then calculated the
frequencies of novel procedures within the different surgery categories for each calendar year.

Results
The patient samples for the 11 surgery categories from 2002 to 2014 ranged in size from 5,256
patients to 506,508 patients (Table 1). Nine of the 11 surgery categories were associated with a
sample size >40,000 over the 13-year study period. Summary statistics of key variables were
reported in Table 2.
14
The unadjusted year-by-year changes in surgical quality and cost are presented in Figures 1 and
2. Although quality fluctuated during the study period, the quality improved for all surgery
categories except PTCA (CCS45; see Appendix). The trends in cost were more varied. The cost
of tracheostomy (CCS34) decreased dramatically starting in 2002 and kept decreasing through
2012. The costs associated with the other surgery categories increased in the early years of the
study period and then decreased in the later years. By the end of 2014, the unadjusted costs of
coronary artery bypass grafting (CABG; CCS44), percutaneous transluminal coronary
angioplasty (PTCA; CCS45), gastrostomy (CCS71), and exploratory laparotomy (CCS89) were
higher than in 2002, whereas the unadjusted costs of the other seven surgery categories were
lower in 2014 than in 2002 (See Appendix). However, the changes in costs of other surgeries
were much smaller than the changes in cost of tracheostomy (CCS34; Figure 2).

When only the data from 2002 and 2014 were considered, logistic regressions of the unadjusted
quality measures revealed a significant decrease in quality for PTCA (CCS45) and significant
increases in quality for tracheostomy (CCS34), heart valve procedures (CCS43),
CABG(CCS44), gastrostomy (CCS71), ileostomy and other enterostomy (CCS73), colorectal
resection (CCS78), and debridement of wound (CCS169; p < 0.05 for each category). There
were no significant changes in unadjusted quality for small bowel resection (CCS75),
exploratory laparotomy (CCS89), or excision, lysis peritoneal adhesions (CCS90). The
unadjusted annualized quality growth rates ranged from -0.39% for PTCA (CCS45) to 0.83% for
ileostomy and other enterostomy (CCS73).

15
The unadjusted growth in cost based on only the data from 2002 and 2014 was negative for
tracheostomy (CCS34), heart valve procedures (CCS43), small bowel resection (CCS75),
colorectal resection (CCS78), and debridement of wound (CCS169; p < 0.05 for each category),
indicating significant reductions in cost for those procedures over the entire study period. There
were increases in unadjusted costs over entire study period for CABG (CCS44), PTCA (CCS45),
and gastrostomy (CCS71; p < 0.05 for each category). The unadjusted costs for excision, lysis
peritoneal adhesions (CCS90), ileostomy and other enterostomy (CCS73), and exploratory
laparotomy (CCS89) did not change significantly over the study period (p>0.05). Among the 11
surgery categories, tracheostomy (CCS34) had the largest decrease in unadjusted cost (-2.74%),
and PTCA (CCS45) had the largest increase in unadjusted cost (1.65%).

When we adjusted for patient demographics, illness severity, hospital characteristics, and social
characteristics (Figure 3a), the rate of quality growth increased for all surgery categories except
heart valve procedures (CCS43). The adjusted quality increased over the study period for all of
the surgery categories (p < 0.05 for each category). The surgery category with the largest
increase in adjusted quality was ileostomy and other enterostomy (CCS73), whereas that with the
smallest increase in adjusted quality was PTCA (CCS45).

When we adjusted the cost for the same factors used to adjust quality, the cost decreased as a
result of the adjustment for all surgery categories except tracheostomy (CCS34) and heart valve
procedures (CCS43; Figure 3b). The adjusted costs decreased significantly over the study period
for tracheostomy (CCS34), heart valve procedures (CCS43), small bowel resection (CCS75),
colorectal resection (CCS78), excision, lysis peritoneal adhesions (CCS90), and debridement of
16
wound (CCS169), whereas they increased significantly for CABG (CCS44) and PTCA (CCS45;
p < 0.05 for each category). There was no significant change in adjusted cost over the study
period for gastrostomy (CCS71), ileostomy and other enterostomy (CCS73) and exploratory
laparotomy (CCS89; p>0.05 for each category).

After adjustment, tracheostomy (CCS34), heart valve procedures (CCS43), small bowel
resection (CCS75), colorectal resection (CCS78), excision, lysis peritoneal adhesions (CCS90),
and debridement of wound (CCS169) had a significant increase in quality and a significant
decrease in cost, whereas gastrostomy (CCS71), exploratory laparotomy (CCS89), and ileostomy
and other enterostomy (CCS73) had a significant increase in quality and no significant change in
cost. Adjusted quality and adjusted cost both increased significantly for CABG (CCS44) and
PTCA (CCS45; Figure 3c).

The results were similar in the sensitivity analyses in which we also required 30-day survival
after discharges. The growth rates of daily hospitalization costs increased overall for all the CCS
surgery categories except tracheostomy (CCS34), although there was a decline in daily
hospitalization costs for some categories around 2012 (See Appendix). For patients with heart
attack, when we controlled for the locations of the heart attacks and the likelihood of inpatient
mortality, the trends in quality over the study period were consistent with our initial results.

Among the 11 surgery categories, only heart valve procedures (CCS43) and colorectal resection
(CCS78) had growth in the use of novel procedures. For the former, endovascular replacement of
aortic valve (ICD-9-CM 3505) was first used in 2011 and increased in frequency to 21.21% by
17
the end of 2014. Transapical replacement of aortic valve (ICD-9-CM 3506) was first used in
2012 and increased in frequency to 4.30% by the end of 2014. Percutaneous mitral valve repair
with implant (ICD-9-CM 3597) was first introduced in 2010 and had a frequency of 1.77% in
2014. For colorectal resection (CCS78), six novel procedures were introduced between 2008 and
2009: laparoscopic cecectomy (ICD-9-CM 1732), laparoscopic right hemicolectomy (ICD-9-CM
1733), laparoscopic left hemicolectomy (ICD-9-CM 1735), laparoscopic sigmoidectomy (ICD-9-
CM 1736), open abdominoperineal resection of the rectum (ICD-9-CM 4852), and open total
intra-abdominal colectomy (ICD-9-CM 4582). The frequencies of laparoscopic cecectomy (ICD-
9-CM 1732), laparoscopic right hemicolectomy (ICD-9-CM 1733), laparoscopic left
hemicolectomy (ICD-9-CM 1735), and laparoscopic sigmoidectomy (ICD-9-CM 1736)
increased in frequency from 2008 to 2012 and then decreased after 2012. Among those,
laparoscopic right hemicolectomy (ICD-9-CM 1733) had the highest frequency at 15.46% in
2014. The frequencies of open abdominoperineal resection of the rectum (ICD-9-CM 4852) and
open total intra-abdominal colectomy (ICD-9-CM 4582) only increased from 2008 to 2009,
reaching frequencies of 2.07% and 1.42%, respectively, in 2014.

Discussion
We assessed the trends in the cost and quality of 11 surgery categories involving inpatient
hospital stays from 2002 to 2014. Adjustment of the data for health-related covariates resulted in
higher quality metrics for all 11 of the surgery categories. The adjustment for disease severity
and patient outcomes revealed significant trends in quality for four surgery categories that were
not apparent in the unadjusted data. In particular, the growth in quality for PTCA (CCS45)
changed from significantly negative to significantly positive after the adjustment. Adjustment
18
also impacted the trends in cost to some extent. Compared with the unadjusted cost metrics, the
adjusted metrics were lower for all of the surgery categories except tracheostomy (CCS34) and
heart valve procedures (CCS43), and the significance of the cost trends changed for gastrostomy
(CCS71) and excision, lysis peritoneal adhesions (CCS90). Those results indicate that health-
related covariates are important drivers of changes in quality and cost metrics and have the
potential to mask real changes in surgical quality and costs. One possible reason for this is that
the patient demographics changed over the 13 years of the study period. Physician skill and
technologies might have also changed, allowing more seriously ill patients to undergo certain
surgeries. Furthermore, patients with less severe conditions were increasingly moved to
outpatient sectors during the study period. Therefore, the patients with inpatient stays in the later
years of the study might be, on average, more unwell than those at the beginning of the study.
These changes may have led to worse apparent quality outcomes and higher surgical costs per
stay.

After the data were adjusted for illness severity, the surgical quality improved consistently over
the study period for all 11 surgery categories. We also found that the surgical costs declined for
eight of the surgery categories, except for CABG (CCS44) and PTCA (CCS45). The
improvement in quality might have been caused by changes in physician experience, improved
quality assurance, and technological innovations (Weiser et al., 2011). Surgeons might have
gained experience in preforming surgeries and making decisions based on patients’ conditions
over the 13 years of the study period. Improved quality assurances such as improved antibiotic
timing and glucose control, prevention of hypothermia, and reduction of unnecessary
transfusions would help to control complications, readmissions, and deaths at early stages
19
(Weiser et al., 2011). Innovations may have effectively controlled costs and improved quality for
certain surgeries. Among the 11 surgery categories, only heart valve procedures (CCS43) and
colorectal resections (CCS78) had novel procedures come into use during the study period. For
heart valve procedures (CCS43), there was a relative increase in use of aortic valve surgery
versus mitral valve surgery and also an increase in use of less invasive procedures. For colorectal
resection (CCS78), there was a major shift from open procedures to laparoscopic procedures.
The newer and less invasive procedures require only small incisions, which minimizes the
burden on both patients and hospitals, leading to quicker recoveries and lower costs.

The adjusted costs increased significantly for two of the 11 surgery categories: CABG (CCS44)
and PTCA (CCS45). Although quality improved significantly for both of those surgery
categories over the study period, the costs increased at a faster rate than the quality, so it is hard
to judge whether the increases in cost was worth the improvements in quality. The use of
productivity metrics to further assess the changes in the values of the surgeries might help to
determine if the changes implemented over the study period were beneficial or not.

Our findings are similar to those of previous studies that found quality improvements, such as
Weiser et al. (2011) who found a significant decrease in mortality rates for only one
(tracheostomy) out of 11 types of surgery, however. Romley et al. (2015) focused on
productivity (incorporating quality and cost) improvement in the treatment of disease conditions
and did not directly analyze changes in the quality and cost of surgical care. Romley et al. (2015)
found significant improvement in the productivity of services related to heart attacks. Although
we found that the cost increases for CABG (CCS44) and PTCA (CCS45) outpaced the increases
20
in quality of care, those two surgery categories are not the only medical procedures used to treat
heart attacks. Hence, the productivity increase reported by Romley et al. may have been driven
by other services. Our study complements previous studies to provide a more complete picture of
the long-term changes in trends for a wide range of surgical procedures. Furthermore, our study
provides additional information regarding surgical quality and cost. Compared with the metrics
used in previous studies, our quality metric is less noisy and more accurate. We used national
Medicare data, which are clean, well formatted, and complete for FFS beneficiaries, so we could
capture mortality both inside and outside of hospitals. We also included readmission information
based on the CMS algorithm. To determine costs, we incorporated not only inpatient costs,
which were often used in previous analyses, but also costs of non-institutional claims billed
during inpatient stays. Our results were robust across several sensitivity analyses.

Our analysis has some limitations. First, our quality metrics did not account for quality of life,
functional status, life expectancy, complications, time to return to the community, or clinical
biomarkers, all of which can be used to quantify quality of care. For example, one previous study
used the return of patients to the community to evaluate the productivity of skilled nursing
facilities (Gu et al., 2019). Nevertheless, survival without unplanned readmission is a meaningful
quality metric that has received increasing attention from patients, physicians, and policy makers
(Romley et al., 2015). A second limitation of our study is that there was a decrease in the sample
size from 2002 to 2014. A possible explanation for that is that surgeons might have become less
inclined to perform certain surgeries. Also, some of the surgeries moved from the inpatient sector
to the outpatient sector and therefore were no longer captured by our data. The increasing
utilization of outpatient services means that patients with less severe conditions would not be
21
admitted to hospitals. That might lead us to underestimate the reduction in cost and increase in
quality of surgical care, as outpatient care is generally associated with lower costs and lower
rates of readmission and mortality. Nevertheless, it would be worthwhile to perform a sensitivity
analysis to include surgical services from outpatient sectors in the future. Lastly, while we
descriptively characterized innovations from 2002 to 2014, we did not explore the causal
relationships among quality, cost, and innovations, which would be another good direction for
future research.

Conclusion
Our results suggest that there was significant quality improvement for all 11 surgery categories
over the study period. Costs decreased significantly for 5 of the 11 surgery categories, although
they increased significantly for 2 categories. Future work can further explore the value of
inpatient surgeries, especially when surgical cost and quality increase simultaneously.

22
Tables and & Figures
Table 1. The 11 Surgery Categories and Corresponding Sample Sizes, 2002–2014
CCS
Category
Surgery
Type
Number of
Cases
Number of
Hospitals
34 Tracheostomy 45,944 2,700
43 Heart Valve Procedures 102,116 1,266
44 Coronary Artery Bypass Grafting (CABG) 187,275 1,299
45 Percutaneous Transluminal Coronary Angioplasty (PTCA) 506,508 1,878
71 Gastrostomy 98,413 3,393
73 Ileostomy and Other Enterostomy 5,256 1,785
75 Small Bowel Resection 43,982 3,119
78 Colorectal Resection 220,428 3,537
89 Exploratory Laparotomy 7,109 2,114
90 Excision, Lysis Peritoneal Adhesions 48,988 3,192
169 Debridement of Wound 106,782 3,568
Total 1,372,801 27,851

23

Variables
34
Tracheostomy
43
Heart Valve
Procedures
44
CABG
45
PTCA
71
Gastrostomy
73
Ileostomy and
Other Enterostomy
75
Small Bowel
Resection
78
Colorectal Resection
89
Exploratory
Laparotomy
90
Excision, Lysis
Peritoneal Adhesions
169
Debridement of
Wound
Inpatient stay case number, n 45,944 102,116 187,275 506,508 98,413 5,256 43,982 220,428 7,109 48,988 106,782
Hospital number, n 2,700 1,266 1,299 1,878 3,393 1,785 3,119 3,537 2,114 3,194 3,568
30-day survival with no unplanned readmissions 82.1% 92.6% 92.5% 93.4% 77.1% 79.9% 86.6% 92.4% 62.1% 93.3% 89.4%
Inpatient costs (2014 $) 84,302.9 52,725.8 40,010.7 18,820.9 17,112.2 24,303.6 32,064.6 24,748.5 24,922.6 23,102.7 15,686.6
Part B non-institutional charges (2014 $) 10,374.4 7,669.9 6,274.4 1,900.4 2,587.6 3,494.3 4,578.8 4,118.4 3,548.3 3,273.1 2,308.2
Total costs (2014 $) 94,689.6 60,408.8 46,290.6 20,735.3 19,690.2 27,895.2 36,653.1 28,871.1 28,540.8 26,468.4 17,993.5
Inpatient stay days, n 30.6 12.2 10.7 4.2 12.2 13.3 14.1 11.4 10.4 11.7 10.7
Teaching hospital 77.1% 81.8% 68.3% 66.7% 54.9% 64.1% 55.9% 53.7% 64.6% 52.2% 55.3%
Patient Characteristics
Age (Mean ± sd) 76.2 ± 7.2 76.7± 6.6 74.0± 5.8 75.4 ± 6.7 81.7 ± 8.0 77.4 ± 7.8 77.8 ± 7.6 76.8 ± 7.3 76.8 ± 7.3 77.0 ± 7.6 78.1 ± 8.2
Female 48.3% 44.9% 32.3% 42.1% 55.8% 45.6% 61.7% 58.3% 57.3% 64.1% 57.5%
Race : African American 17.5% 3.8% 5.2% 5.9% 21.7% 12.3% 8.9% 7.9% 8.8% 9.5% 15.7%
Race : Hispanic 2.6% 1.1% 1.4% 1.4% 3.5% 2.1% 1.5% 1.2% 1.8% 1.5% 2.5%
Race : White 75.7% 93.1% 90.7% 90.1% 70.7% 81.9% 87.3% 88.6% 86.5% 87.1% 79.4%
Race : Other 4.2% 2.0% 2.7% 2.6% 4.1% 3.7% 2.3% 2.3% 2.9% 1.9% 2.4%
Social Characteristics
Median household income (2014 $) 41,706 ± 16,058 44,853 ± 16,931 42,334 ± 15,299 42,725 ± 15,566 41,631 ± 16,277 43,259 ± 16,101 43,614 ± 16,215 43,624 ± 16,162 41,834 ± 15,152 43,558± 16,211 41,744 ± 15,801
Social Security income (2014 $) 11,111 ± 1,572 11,556 ± 1,492 11,322 ± 1,461 11,359 ± 1,468 11,040 ± 1,580 11,285 ± 1,504 11,389 ± 1,500 11,410 ± 1,484 11,238 ± 1,487 11,384 ± 1,476 11,154 ± 1,547
Poor 13.9% 10.9% 11.8% 11.7% 14.1% 12.4% 11.7% 11.5% 12.4% 11.7% 13.3%
Employed 93.4% 94.7% 94.5% 94.4% 93.4% 94.0% 94.4% 94.5% 94.1% 94.4% 93.7%
Less than high school education 22.0% 18.3% 19.9% 19.7% 22.3% 19.9% 19.1% 19.0% 20.4% 19.2% 21.4%
Urban 76.8% 70.3% 67.2% 69.0% 77.7% 75.0% 72.2% 71.4% 70.0% 72.2% 73.6%
Hispanic 11.6% 8.1% 8.1% 8.2% 12.4% 10.1% 8.8% 8.3% 8.8% 8.8% 10.5%
Single 44.1% 40.7% 40.6% 41.1% 44.1% 42.7% 41.7% 41.5% 42.0% 41.6% 43.2%
Elderly in an institution 5.5% 5.4% 5.4% 5.5% 5.9% 5.7% 5.5% 5.6% 5.5% 5.6% 5.7%
Non-institutionalized elderly with physical disability 30.2% 28.2% 29.4% 29.3% 30.4% 29.3% 29.0% 28.9% 29.5% 29.1% 30.0%
Sensory disability among elderly 14.6% 14.2% 14.7% 14.6% 14.6% 14.4% 14.4% 14.4% 14.7% 14.5% 14.6%
Mental disability 11.7% 10.4% 10.9% 10.9% 12.0% 11.1% 10.8% 10.7% 11.2% 10.8% 11.5%
Self-care disability 10.4% 9.2% 9.6% 9.6% 10.6% 9.8% 9.5% 9.5% 9.8% 9.6% 10.2%
Difficulty going-outside-the-home disability 21.8% 19.6% 20.3% 20.3% 22.0% 20.7% 20.2% 20.1% 20.6% 20.3% 21.4%
Patient Disease Severity
Number of Charlson-Deyo comorbidities 1.56 1.29 1.40 1.36 1.54 1.46 0.98 1.36 1.37 0.88 1.31
1 Charlson-Deyo comorbidity 33.6% 36.5% 35.9% 36.6% 32.0% 34.5% 34.6% 33.7% 33.2% 34.5% 35.3%
2 Charlson-Deyo comorbidities 30.2% 23.1% 25.7% 24.2% 27.8% 29.3% 18.7% 26.3% 27.2% 16.4% 24.4%
3 Charlson-Deyo comorbidities 13.5% 9.5% 11.3% 10.7% 13.9% 11.4% 6.1% 11.1% 10.7% 5.0% 10.3%
4 Charlson-Deyo comorbidities 3.9% 3.1% 3.4% 3.4% 4.5% 3.7% 1.5% 3.2% 3.3% 1.2% 2.9%
5+ Charlson-Deyo comorbidities 1.2% 1.0% 1.0% 1.0% 1.4% 0.8% 0.4% 0.8% 0.7% 0.3% 0.8%
Transferred from other hospital 9.5% 8.7% 15.2% 13.9% 3.5% 4.3% 4.2% 1.9% 5.8% 3.3% 2.5%
Emergency Inpatient Admissions 64.3% 14.5% 24.6% 38.5% 70.2% 47.9% 59.7% 28.6% 52.3% 54.6% 53.1%
Urgent Inpatient Admissions 18.1% 17.5% 26.3% 27.8% 19.6% 21.3% 18.9% 15.2% 18.7% 18.7% 25.6%
Elective Inpatient Admissions 16.1% 67.8% 48.9% 33.4% 9.9% 30.5% 21.1% 56.0% 27.9% 26.5% 20.7%
CABG Related Risk Factor Adjustors ( CCS 44)
AHRQ predicted inpatient mortality 2.7%
NO of cases with AHRQ predicted inpatient mortality 41,339
Location of heart attack: Anterolateral (410.0x) 75.8%
Location of heart attack: Other Anterior Wall (410.1x) 2.4%
Location of heart attack: Inferolateral Wall (410.2x) 0.8%
Location of heart attack: Inferoposterior Wall (410.3x) 0.5%
Location of heart attack: Other Inferior Wall (410.4x) 2.5%
Location of heart attack: Other Lateral Wall (410.5x) 0.3%
Location of heart attack: True Posterior Wall (410.6x) 0.1%
Location of heart attack: Sub-Endocardial (410.7x) 15.9%
Location of heart attack: Other Specified Sites (410.8x) 0.4%
Location of heart attack: Unspecified site (410.9x) 1.3%
Surgeries Case Number Hospital Number
34 Tracheostomy 45,944 2,700
43 Hearth Valve Procedures 102,116 1,266
44 CABG 187,275 1,299
45 PTCA 506,508 1,878
71 Gastrostomy 98,413 3,393
73 Ileostomy and Other Enterostomy 5,256 1,785
75 Small Bowel Resection 43,982 3,119
78 Colorectal Resection 220,428 3,537
89 Exloratory Laparotomy 7,109 2,114
90 Excision, Lysis Peritoneal Adhesions 48,988 3,192
169 Debridement of Wound 106,782 3,568
Total 1,372,801 27,851
Table 1. Demographic Information for hospitalized patients who underwent a surgical intervention
24
Figure 1 ．Unadjusted Year-By-Year Changes in Surgical Quality for 11 CCS Categories, 2002–2014

Notes: This figure shows the predicated surgical quality, which is the predicted rates of high-quality stays, from multivariate logistic
regressions. A high-quality stay was one in which the patient survived 30 days after inpatient admissions without unplanned
readmissions after discharges. The unadjusted models are regressed on year indicators at the individual level. Rates in 2014 were
significantly different from those in 2002 for all surgery categories (p < 0.05) for all surgeries except for exploratory laparotomy (CCS
code 89) and excision (CCS code 90).

Figure 2 ．Unadjusted Year-By-Year Changes in Surgical Cost for 11 CCS Categories, 2002–2014

Notes: This figure shows predicted surgical costs from ordinary least squares regressions. The unadjusted models are regressed on
year indicators at the individual level. Costs in 2014 were significantly different from the costs in 2002 for all surgery categories
(p<0.001) except for CCS 73, 89 and 90.

25

Figure 3. Unadjusted and Adjusted Annualized Growth Rates of Quality and Cost, 2002–2014
Figure 3a. Unadjusted Annualized Growth Rates of Quality, 2002–2014 Figure 3b. Unadjusted Annualized Growth Rates of Cost, 2002–2014

Figure 3c. Adjusted Annualized Growth Rates of Cost and Quality, 2002–2014

Figure 4. Shares of New ICD-9-CM Procedures in 2014, Compared with 2002

26
Appendix
Appendix Exhibit 1. Unadjusted and Adjusted Year-By-Year Quality Growth for Each CCS Category,
2002-2014

Notes: This figure shows predicted rates of "high-quality stays" and associated confidence intervals from multivariate logit
regressions in which indicators for 30-day survival without readmissions for patients are the dependent variables. The unadjusted
and adjusted models are regressed on year indicators at the individual level. The adjusted regressions are controlled for age,
gender, race/ethnicity, the Charlson comorbidities, area sociodemographic, AHRQ inpatient mortality risk for CABG, teaching
hospital indicators, and inpatient stay characteristics like diagnosis codes and inpatient admission types.

Figure 1. Quality Changes By CCS Categories, 2002 - 2014
86%
87%
88%
89%
90%
91%
92%
93%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
90: Excision, lysis Peritoneal Adhesions
Unadjusted
#REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
50%
55%
60%
65%
70%
75%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
80%
82%
84%
86%
88%
90%
92%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
65%
70%
75%
80%
85%
90%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
73%
74%
75%
76%
77%
78%
79%
80%
81%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
88%
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
78%
79%
80%
81%
82%
83%
84%
85%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
27
Appendix Exhibit 2. Unadjusted and Adjusted Year-By-Year Cost Growth for Each CCS Category, 2002-
2014

Notes: This figure shows predicted costs and associated standard errors from ordinary least squares regressions in which sum of
inpatients and non-institutional costs are the dependent variables. The unadjusted and adjusted models are regressed on year
indicators at the individual level. The adjusted regressions are controlled for age, gender, race/ethnicity, the Charlson
comorbidities, area sociodemographic, AHRQ inpatient mortality risk for CABG, teaching hospital indicators, and inpatient stay
characteristics like diagnosis codes and inpatient admission types.

Figure 1. Cost Changes By CCS Categories, 2002 - 2014
14.0
15.0
16.0
17.0
18.0
19.0
20.0
21.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
23.0
24.0
25.0
26.0
27.0
28.0
29.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
90: Excision, Lysis Peritoneal Adhesions
Unadjusted #REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
20.0
22.0
24.0
26.0
28.0
30.0
32.0
34.0
36.0
38.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
25.0
26.0
27.0
28.0
29.0
30.0
31.0
32.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
30.0
32.0
34.0
36.0
38.0
40.0
42.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
20.0
22.0
24.0
26.0
28.0
30.0
32.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
17.5
18.0
18.5
19.0
19.5
20.0
20.5
21.0
21.5
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
17.0
18.0
19.0
20.0
21.0
22.0
23.0
24.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
42.0
43.0
44.0
45.0
46.0
47.0
48.0
49.0
50.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
52.0
54.0
56.0
58.0
60.0
62.0
64.0
66.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
60.0
70.0
80.0
90.0
100.0
110.0
120.0
130.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
28

Appendix Exhibit 3. Sensitivity Analysis: Adjusted Models with Redefined Quality Metric (Surviving 30
days after discharge without readmissions)

Figure 1. Quality Changes By CCS Categories, 2002 - 2014
82%
83%
84%
85%
86%
87%
88%
89%
90%
91%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
88%
89%
90%
91%
92%
93%
94%
95%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
90: Excision, lysis Peritoneal Adhesions
Unadjusted #REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
50%
55%
60%
65%
70%
75%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
87%
88%
89%
90%
91%
92%
93%
94%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
78%
80%
82%
84%
86%
88%
90%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
61%
66%
71%
76%
81%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
66%
67%
68%
69%
70%
71%
72%
73%
74%
75%
76%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
88%
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
88%
89%
90%
91%
92%
93%
94%
95%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
50%
55%
60%
65%
70%
75%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
29

Appendix Exhibit 4. Sensitivity Analysis: Adjusted Models with Redefined Cost Metric (Cost per Day in
Hospital)

Figure 1. Cost Changes By CCS Categories, 2002 - 2014
2.0
2.1
2.2
2.3
2.4
2.5
2.6
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
2.4
2.5
2.6
2.7
2.8
2.9
3.0
3.1
3.2
3.3
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
90: Excision, Lysis Peritoneal Adhesions
Unadjusted #REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
3.0
3.2
3.4
3.6
3.8
4.0
4.2
4.4
4.6
4.8
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
2.7
2.9
3.1
3.3
3.5
3.7
3.9
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
2.7
2.8
2.9
3.0
3.1
3.2
3.3
3.4
3.5
3.6
3.7
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
1.9
2.0
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
1.8
1.9
1.9
2.0
2.0
2.1
2.1
2.2
2.2
2.3
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
5.5
6.0
6.5
7.0
7.5
8.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
4.8
4.9
5.0
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
5.5
6.0
6.5
7.0
7.5
8.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
3.6
3.8
4.0
4.2
4.4
4.6
4.8
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
30

Appendix Exhibit 5: Sensitivity Analysis: Redefined Cost and Quality of CCS 44 with Additional Risk
Factors

Figure f. Growth Rates of Redefined Quality with Same Health
Adjustors, Locations of Heart Attack and AHRQ Predicted
Figure c. Growth Rates of Daily Hospitalization Costs with Same Health
Adjustors, Locations of Heart Attack and AHRQ Predicted Mortality
Appendix: Cost and Quality of CCS 44 with Risk Factors
Figure b. Growth Rates of Daily Hospitalization Costs with Same Health
Adjustors and Locations of Heart Attack
Figure a. Growth Rates of Daily Hospitalization Costs with Same Health
Adjustors as Previous Models
Figure d. Growth Rates of Redefined Quality with Same Health
Adjustors as Previous Models
Figure e. Growth Rates of Redefined Quality with Same Health
Adjustors and Locations of Heart Attack
70%
72%
74%
76%
78%
80%
82%
84%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
43
44
45
46
47
48
49
50
51
52
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
70%
72%
74%
76%
78%
80%
82%
84%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
70%
72%
74%
76%
78%
80%
82%
84%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
43
44
45
46
47
48
49
50
51
52
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
43
44
45
46
47
48
49
50
51
52
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
31
References

Altman, S., and Mechanic, R. (2018). Health Care Cost Control: Where Do We Go from Here? |
Health Affairs. 10.1377/hblog20180705.24704.

Boulware, L.E., Cooper, L.A., Ratner, L.E., LaVeist, T.A., and Powe, N.R. (2003). Race and
Trust in the Health Care System. Public Health Reports (1974-) 118, 358–365.

Bradshaw, P.J., Jamrozik, K., Le, M., Gilfillan, I., and Thompson, P.L. (2002). Mortality and
Recurrent Cardiac Events After Coronary Artery Bypass Graft: Long Term Outcomes in a
Population Study. Heart 88, 488–494.

Cheng, C.M., Shin, J., and Guglielmo, B.J. (2014). Trends in Boxed Warnings and Withdrawals
for Novel Therapeutic Drugs, 1996 Through 2012. JAMA Intern Med 174, 1704.

CMS (2013). 2013 Measures Updates and Specifications: Acute Myocardial Infarction, Heart
Failure, and Pneumonia 30-Day Risk-Standardized Mortality Measure (Version 7.0). 55.
https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/
HospitalQualityInits/Mortality_AMI-HF-PN_Measures_Updates_Report_FINAL_06-13-
2013.pdf

CMS (2016). 2016 Measure Information About the 30-Day All-Cause Hospital Readmission
Measure. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/
PhysicianFeedbackProgram/Downloads/2015-ACR-MIF.pdf

DiNicolantonio, J.J., and Serebruany, V.L. (2013). Challenging the FDA Black Box Warning for
High Aspirin Dose with Ticagrelor in Patients with Diabetes. Diabetes 62, 669–671.

Dorsey, E.R. (2010). Impact of FDA Black Box Advisory on Antipsychotic Medication Use.
Arch Intern Med 170, 96.

Dusetzina, S.B., and Caleb Alexander, G. (2011). Drug vs Class-Specific Black Box Warnings:
Does One Bad Drug Spoil the Bunch? J Gen Intern Med 26, 570–572.

Ellis, L. (2019). Using Health Outcomes Research to Improve Quality of Care | Executive and
Continuing Professional Education | Harvard T.H. Chan School of Public Health.
https://www.hsph.harvard.edu/ecpe/using-health-outcomes-research-to-improve-quality-of-care/

Fitzsimmons, D., Phillips, C.J., Bennett, H., Jones, M., Williams, N., Lewis, R., Sutton, A.,
Matar, H.E., Din, N., Burton, K., et al. (2014). Cost-effectiveness of Different Strategies to
Manage Patients with Sciatica. PAIN® 155, 1318–1327.

Hartman, M., Martin, A.B., Lassman, D., Catlin, A., and National Health Expenditure Accounts
Team (2015). National Health Spending in 2013: Growth Slows, Remains in Step with the
Overall Economy. Health Affairs 34, 150–160.
32
Lasser, K.E., Seger, D.L., Yu, D.T., Karson, A.S., Fiskio, J.M., Seger, A.C., Shah, N.R., Gandhi,
T.K., Rothschild, J.M., and Bates, D.W. (2006). Adherence to Black Box Warnings for
Prescription Medications in Outpatients. Arch Intern Med 166, 338–344.

Lester, P., Kohen, I., Stefanacci, R.G., and Feuerman, M. (2011). Antipsychotic Drug Use Since
the FDA Black Box Warning: Survey of Nursing Home Policies. Journal of the American
Medical Directors Association 12, 573–577.

Martin, A.B., Hartman, M., Whittle, L., and Catlin, A. (2014). National Health Spending In
2012: Rate of Health Spending Growth Remained Low for The Fourth Consecutive Year. Health
Affairs 33, 67–77.

Martin, A.B., Hartman, M., Benson, J., and Catlin, A. (2016). National Health Spending In 2014:
Faster Growth Driven by Coverage Expansion and Prescription Drug Spending. Health Affairs
35, 150–160.

Medicare, CMS. for, Baltimore, M.S. 7500 S.B., and Usa, M. (2012). Historical Impact Files for
FY 1994 through Present. https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/
AcuteInpatientPPS/Historical-Impact-Files-for-FY-1994-through-Present

Medicare, CMS. for, Baltimore, M.S. 7500 S.B., and Usa, M. (2018). National Health Accounts
Historical. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-
Reports/NationalHealthExpendData/NationalHealthAccountsHistorical

Oregon Legislature (2017). Definition of Health Care Services.
https://www.oregonlaws.org/glossary/definition/health_care_services

Orszag, P.R., and Emanuel, E.J. (2010). Health Care Reform and Cost Control. New England
Journal of Medicine 363, 601–603.

Qato, D.M., Trivedi, A.N., Mor, V., and Dore, D.D. (2016). Disparities in Discontinuing
Rosiglitazone Following the 2007 FDA Safety Alert: Medical Care 54, 406–413.

Resdac Getting Started with CMS Data | ResDAC. https://www.resdac.org/

Robinson, J.C., and Smith, M.D. (2008). Cost-Reducing Innovation in Health Care. Health
Affairs 27, 1353–1356.

Ross, J.S., Jackevicius, C., Krumholz, H.M., Ridgeway, J., Montori, V.M., Alexander, G.C.,
Zerzan, J., Fan, J., and Shah, N.D. (2012). State Medicaid Programs Did Not Make Use of Prior
Authorization to Promote Safer Prescribing After Rosiglitazone Warning. Health Affairs
(Millwood) 31, 188–198.

Shi, L., Zhao, Y., Szymanski, K., Yau, L., and Fonseca, V. (2011). Impact of Thiazolidinedione
Safety Warnings on Medication Use Patterns and Glycemic Control Among Veterans with
Diabetes Mellitus. J. Diabetes Complicat. 25, 143–150.
33
Virnig, B., and Parsons, H. (2018). Strengths and Limitations of CMS Administrative Data in
Research | ResDAC. http://resdac.org/articles/strengths-and-limitations-cms-administrative-data-
research.

Weiser, T.G., Semel, M.E., Simon, A.E., Lipsitz, S.R., Haynes, A.B., Funk, L.M., Berry, W.R.,
and Gawande, A.A. (2011). In-hospital Death Following Inpatient Surgical Procedures in the
United States, 1996–2006. World J Surg 35, 1950–1956.

History of the Affordable Care Act (ACA). http://affordablehealthca.com/history-affordable-
care-act/

Boards of Trustees. (2012) Annual Report of the Boards of Trustees of the Federal Hospital
Insurance and Federal Supplementary Medical Insurance Trust Funds. CMS.gov
https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-
Reports/ReportsTrustFunds/Downloads/TR2012.pdf

34
Chapter 3: Long-Term Trends in the Productivity of Surgical Care,
2002–2014
Tommy Chiou, Ning Ning, Alex Haynes, and John Romley

Introduction
Cost containment and quality improvement are two of the top priorities of the American
healthcare system today. The attainment of both goals is particularly challenging because cost
reduction, if implemented indiscriminately, could come at the expense of quality of care, while
inefficient efforts to improve quality could unnecessarily inflate spending (Burke and Ryan,
2014; Newhouse et al., 2013). Consistent with these ideals, numerous value-based
reimbursement programs sponsored by both the Centers for Medicare and Medicaid Services
(CMS) and private institutions have been developed and carried out from 2012 in an effort to
incentivize better hospital care at lower costs (Chee et al., 2016). Meanwhile, according to the
reports released by Medicare, hospital cost is still a big part of total health care spending
(32.7%), with about 4.6% increase in 2017 (Medicare, 2018). Among all hospital services,
approximately 45 million surgical procedures, accounting for more than half a trillion dollars in
annual expenditures, are performed in the United States each year. There is need to study long-
term trends in surgical cost and quality, and more importantly, there is need to evaluate whether
money (resources) was spent efficiently to provide desirable level of services.

In 2013, the Institute of Medicine released a report endorsing value-based care as the “only
sensible way” to control costs without compromising quality (Newhouse et al., 2013).
35
Now, value becomes an increasingly recognized goal, there is a growing need to operationalize
and measure value of healthcare. Given the significance of surgical care from both a healthcare
delivery and expenditure standpoint, as well as the upward projections in both cost and quality of
surgical care, it is important to analyze whether the value of surgical care has improved in the
United States. This has become an increasingly popular proposition in healthcare.
,

Multifactor productivity – “output per unit of labor, capital, and other measurable inputs”
(Harper et al., 2010) – is often used as a proxy for value outside of the healthcare sector. In the
context of hospital care, the calculation of productivity has traditionally involved either the
derivation of implicit output and input quantities based on a hospital’s clinical revenues and
expenses, reference to available data on input quantities, or the use of specific output definitions,
such as number of hospital stays (Ashby et al., 2000; Cylus and Dickensheets, 2007). These
estimates may not yield accurate portrayals of a hospital’s productivity, however, as they do not
account for the heterogeneity of outputs, which may vary by quality, and inputs, which may vary
based on a patient’s pre-admissions clinical profile (Cylus and Dickensheets, 2007; Sheiner and
Hutchins, 2016). For instance, productivity estimates will be biased downward if the quality of
healthcare services provided has improved over time, or if the patients being treated are on
average sicker.

One approach of tackling this problem is to further adjust outputs for service quality outcomes
and inputs for patient characteristics, thereby controlling for the effects of changes in patient
case-mix or the quality of healthcare interventions on productivity (Gu et al., 2019; Romley et
al., 2015). Using this method, Romley et al. (2015) estimated productivity growth in hospital
36
care for heart attack, heart failure, and pneumonia patients from 2002 to 2011 (Romley et al.,
2015). They found that the quality- and patient health-adjusted productivity estimates were
significantly higher compared to the unadjusted values, suggesting the importance of such
adjustments in generating more accurate estimates. It would be important to look at long-term
changes in productivity, with this quality and patient illness severity adjustment, by disease
conditions, but also by inpatient surgical services in hospitals.

There is evidence to suggest steady declines in mortality after inpatient surgery in U.S. hospitals
from 1996 to 2006, even though the declines in mortality were identified as a significant change
for only 1 out of 11 defined surgical groups (Weiser et al., 2011). In addition, they did not study
changes in surgical cost metric alongside to the changes in quality. Moreover, the quality metric
used in this paper only accounted for in-hospital mortality only when captured by survey data.
So, we wanted to build upon their work, and improve quality metric by incorporating unplanned
readmission rates, as well as mortality inside and outside hospitals. We also wanted to introduce
cost metric in order to calculate surgical productivity to evaluate the value in hospital surgical
care.

In this study, we used Medicare fee-for-service claims to assess quality-adjusted productivity
trends across 11 classes of surgical procedures among U.S. hospitals from 2002 to 2014. We
built up the prior work of Weiser et al (2011), and analyzed the productivity growth in a wide
range of surgery groups, which contain hundreds of inpatient surgical procedures.

37
Methods
Data Sources
A random 20% sample of the Medicare inpatient claims (Part A) was queried for selected
surgical procedures performed for Medicare fee-for-service beneficiaries during inpatient
hospital stays from 2002 to 2014. Medicare Part B data was linked to Part A data providing cost
of non-institutional claims billed by surgeons during inpatient stays. We also linked Medicare
data with US Census data to add area characteristics to our data.

Sample selection
The inclusion criteria for the surgical procedures has previously been described (Weiser et al.,
2011). Briefly, five surgeon reviewers independently evaluated ICD-9-CM procedure codes
(excluding ophthalmologic and dental procedures) to select for “operating room and non-
operating room procedures that involve incision, excision, manipulation, or suturing of tissue,
and that usually require regional or general anesthesia, or profound sedation to control pain”
(Weiser et al., 2011). The qualifying procedures were then grouped into clinical meaningful
categories based on the Agency for Healthcare Research and Quality (AHRQ) Clinical
Classification System (CCS).

A total of 11 classes of surgery were analyzed in this study. These include tracheostomy
(temporary and permanent), heart valve procedures, coronary artery bypass graft (CABG),
percutaneous transluminal coronary angioplasty (PTCA), gastrostomy (temporary and
permanent), ileostomy and other enterostomy, small bowel resection, colorectal resection,
exploratory laparotomy, excision (lysis peritoneal adhesions), and debridement of wound for
38
infection or burn (Weiser et al., 2011). We selected patients who experienced at least one of the
qualifying inpatient surgical procedures between 2002 and 2014. Beneficiaries who transferred
to other hospitals during their stays were excluded from our patient sample. We further restricted
our patient sample to elderly beneficiaries who enrolled in fee-for-service Medicare.

Analysis
The main outcome of interest in this study is the surgical productivity of hospital inpatient care,
which was measured as the ratio of hospital output to inputs. To measure hospital inputs, we
obtained surgical costs by converting hospital charges covered by Medicare into costs via the
cost-to-charge ratios submitted by each hospital to CMS (CMS, 2019; Medicare, 2012). These
costs were then adjusted for geographic differences in labor prices using the hospital wage index,
and for inflation to 2014 dollars (CMS, 2019). We also selected all non-institutional claims billed
by physicians during each inpatient stay. We converted the costs from non-institutional claims to
2014 US dollars using the Medicare economic index and the Physician Fee Schedule.

To quantify hospital output, we first measured the number of inpatient stays at each hospital each
year. We then identified the number of “high-quality stays”, which were defined as the hospital
stays with 30-day survival beyond inpatient admission and with no unplanned readmissions
within thirty days of their discharge, for each hospital in each year. We identified unplanned
readmissions using the methodology developed by CMS (CMS, 2016; Romley et al., 2015). Rate
of high-quality stays in each hospital was also a measure of quality for each surgery group.

39
Finally, for each surgery group, hospital productivity was measured as the ratio of hospital
output to inputs (hospital costs) using two different outputs we mentioned above: 1) the number
of hospital stays, and 2) the number of high-quality stays.

To obtain productivity growth rates, ordinary least squares (OLS) regressions were performed
with the logarithm of productivity as the dependent variable and a trend variable for year of
admission as the independent variable. We weighted each hospital-year observation by the
number of hospital stays in our analysis sample. To deal with possible correlation in the
residuals, standard errors were clustered at the hospital level. To obtain changes in hospital
inputs and output separately, we also applied the same regressions to cost and quality for each
surgery group.

To control for potential confounding due to changes in patient characteristics, such as illness
severity. We adjusted for patient demographics (age, sex, race/ethnicity), community
demographics (education level, social security income, the median household income, poverty
rate in the patients’ zip codes, etc.), patient health status (the proportion of patients with Charlson
comorbidities in a hospital’s discharge records, admission types, the most common CCS
diagnoses, etc.) (AHRQ, 2010; Romley et al., 2015). Finally, we used a threshold of 0.6 or more
residents per bed to define teaching hospitals and adjusted for hospital teaching status (Volpp et
al., 2007).

40
Results
Summary Statistics
A total of 1,414,232 inpatient hospital stays corresponding to one of the 11 surgical classes took
place at 170,253 hospitals between 2002 and 2014 in our study sample (Table 1). Compared to
inpatient hospital stays in 2002, hospital stays in 2014 had higher costs ($31,911 vs $30,908,
p=0.001), lower readmissions rates (13.9% vs 16.6%, p<0.001), and slightly higher rates of high-
quality stays (91.4% vs 91.1%, p=0.031). Comparing our sample of patients between 2002 and
2014, patients in 2014 were slightly older (mean age 76.3 vs 75.9, p<0.001), less likely to be
White (86.7% vs 88.4%, p<0.001), and had a higher median household income ($4,373 vs
$4,254, p<0.001). Patients in 2014 were also more likely to have more than one Charlson
comorbidity (54.2% vs 34.7%, p<0.001), and had a higher percentage of emergency inpatient
admissions (46.6% vs 34.6%, p<0.001) compared to patients in our sample in 2002.

Surgical Cost and Quality
Hospital cost trends adjusted for patient characteristics for each of the 11 classes of surgery are
displayed in Figure 1 (unadjusted trends included in Appendix). Average adjusted costs for the
surgical procedures ranged from a low of $15,400 for debridement of wound in 2013 to a high of
$122,000 for heart valve procedures in 2002. From 2002 to 2014, adjusted hospital costs
declined among all procedures except for CABG procedures, which increased from $44,273 to
$45,083. Adjusted surgical care quality, as measured by percent of hospital stays with 30-day
survival without unplanned readmissions, for each of the 11 classes of surgery are displayed in
Figure 2 (unadjusted trends included in Appendix). Between 2002 and 2014, adjusted surgical
41
quality improved among all 11 classes of surgery, with the biggest improvements concentrated in
exploratory laparotomy (82.7% to 90.1%) and tracheostomy (81.7% to 88%) procedures.

Surgical Productivity
We first quantified hospital output as the number of patient stays, and regressed this unadjusted
productivity only on a trend indicator without adjustment for patient or plan characteristics. The
unadjusted hospital productivity for the 11 classes of surgery increased by an average of 0.53%
annually between 2002 and 2014 (Figure 3). The surgical procedures with positive productivity
growth included debridement of wound (1.69%; p<0.001), colorectal resection (0.42%;
p<0.001), small bowel resection (0.60%; p=0.002), heart valve procedures (0.51%; p<0.001),
and tracheostomy (4.24%; p<0.001; Figure 1). In contrast, unadjusted productivity decreased for
gastrostomy (-0.54%; p<0.001), PTCA (-1.57%; p<0.001), and CABG (-0.62%; p<0.001), while
unadjusted productivity in excision, lysis peritoneal adhesions (-0.12%; p=0.52), exploratory
laparotomy (-0.48%; P=0.40), and ileostomy and other enterostomy (0.53%; p=0.39) did not
change significantly between 2002 and 2014.

After adjustment for patient characteristics and illness severity, the average increase in
annualized productivity growth rate across the 11 classes of surgery was 1.14%. Specifically,
adjusting for patient health increased productivity growth rates for gastrostomy (-0.54% p<0.001
to 0.29% p=0.06), CABG (-0.62% p<0.001 to -0.12% p=0.44), excision, lysis peritoneal
adhesion (-0.12% p=0.517 to 0.91% p<0.001), and PTCA (-1.57% p<0.001 to 0.43% p=0.044).
In contrast, productivity estimates for tracheostomy decreased from 4.24% (p<0.001) to 3.28%
(p<0.001) after adjustment for patient severity.
42
Then we further adjusted productivity by surgical care quality, and used the number of high-
quality stays as hospital output to calculate productivity. The average annualized productivity
growth rate increased by 1.57%. The quality and illness severity-adjusted productivity estimates
were greater than the illness severity-adjusted estimates for all 11 classes of surgery, and the
estimates increased from nonsignificant or marginal significant to positive for exploratory
laparotomy (0.75% p=0.268 to 1.68% p=0.025), gastrostomy (0.29% p=0.060 to 0.75%
p<0.001), and PTCA (0.43% p=0.044 to 0.64% p=0.004). Only the productivity estimates for
CABG (0.24%, p=0.13) and ileostomy, enterostomy (1.03%, p=0.164) remained statistically
insignificant after the quality and illness severity adjustment.

Discussion
Surgical care is a significant component of the American healthcare system. Approximately 45
million surgical procedures are performed each year, representing up to 40% of all hospital and
physician spending (Birkmeyer et al., 2010). While the quality and cost of surgical care has been
the focus of much research attention, past studies have not examined surgical quality, cost and
productivity trends at a national level, across a wide range of surgeries, and in relation to each
other (Finks et al., 2011; Lawson et al., 2014). In this study, we applied a quality and patient
health-adjusted productivity metric, incorporating both surgical cost and quality, to estimate the
annualized productivity growth of 11 classes of surgeries performed on Medicare fee-for-service
beneficiaries from 2002 to 2014.

Overall, the surgeries examined in this study showed positive productivity growth between 2002
and 2014. However, this unadjusted metric – which calculated productivity using number of
43
hospital stays as hospital output– does not account for any changes that may have occurred in the
patient case-mix between 2002 and 2014. Upon further adjustment for patient characteristics and
illness severity, annualized productivity growth rates increased. This upward shift after
adjustment suggests that the patients examined in this study in 2014 were on average sicker upon
admission compared to those in 2002. Indeed, summary statistics showed that patients in 2014
were much more likely to have been admitted on an emergent basis compared to 2002, and a
higher proportion of them had more than one Charleson comorbidity at the time of admission
(Table 1). Adjustment for quality (using high-quality stays as hospital output in calculation for
productivity) further increased the productivity growth rates, suggesting that patients in 2014
also received hospital care with higher quality compared to those in 2002. These patterns of
increased productivity growth rates after adjustment are consistent with a previous study
employing similar productivity metrics for hospital care for pneumonia, heart failure and heart
attack, and highlight the importance of controlling for heterogeneity in patients’ baseline health
and outcomes achieved when estimating productivity (Romley et al., 2015).

There are several possible factors that can explain the productivity growth observed among these
11 classes of surgery. One potential driver of productivity is innovation in surgical technique.
Tracheostomy, which involves the establishment of airways, has traditionally been performed by
surgically creating a 2-3cm hole through the neck (Rashid and Islam, 2017). However, a
percutaneous approach requiring much smaller incisions has also become an option of late.
Compared to the surgical approach, percutaneous tracheostomy can be performed faster, and has
also been found to lead to lower stoma inflammation and wound infection rates among patients
(Brass et al., 20; Putensen et al., 2014). Furthermore, one study estimated the cost of a
44
percutaneous tracheostomy to be at $1,569, compared to the $3,172 estimated for a surgical
tracheostomy (Rashid and Islam, 2017). While it is unknown to what extent utilization trends of
percutaneous vs surgical tracheostomy has changed between 2002 and 2014, a shift in in favor of
the percutaneous approach may help explain the 4.03% annualized productivity growth of
tracheostomy.

Improvements in surgical technique might similarly explain the productivity growth observed for
debridement of wound procedures. The removal of necrotic tissue has historically involved
surgical removal via a scalpel or curette (Bekara et al., 2018). Alternative approaches such as the
biodebridement (e.g. sterile maggots) or enzymatic debridement have also since been developed.
More recently, the use of hydrosurgery, ultrasound, and plasma-mediated bipolar radio-
frequency ablation have been explored as more precise tools for wound cleaning (Bekara et al.,
2018). While relatively new, the majority of limited literature comparing the effectiveness of
these techniques have found them to enable faster wound debridement as well as improved
wound healing time (Bekara et al., 2018). Although it is unclear to what extent these newer
techniques are utilized, these innovations may at least partially be responsible for the observed
2.91% annualized productivity growth in wound debridement surgeries.

Besides innovations in surgical technology, another potential factor driving productivity in
surgical care is increased adoption of evidence-based practice guidelines and quality
improvement initiatives. Over the past decade, there has been an increased emphasis among
professional medical societies on the creation and dissemination of best practices/standards of
care. One such example is the Enhanced Recovery After Surgery initiative, which produces
45
recommendations for optimal perioperative care (Ljungqvist et al., 2017). Currently guidelines
exist for cardiac surgery, including heart valve procedures, PTCA and CABG; colorectal
resections; as well as gastrointestinal procedures, including gastrostomy, laparotomy, ileostomy,
other enterotomy, and small bowel resection (Engelman et al., 2019; Gustafsson et al., 2019;
Scott et al., 2015). Compliance with ERAS protocols has been found to reduce post-operative
complications by up to 50% and shorten length of stay, with one study focused on ERAS for
colorectal resection reporting a 42% reduction in mortality (Ljungqvist et al., 2017). Increased
adoption of these guidelines, as well as a rise in other quality improvement collaboratives, may
all play a role in enhancing the productivity of the surgeries examined in this study (Thompson et
al., 2017).

Finally, the implementation of value-based reimbursement programs throughout the past decade
may also have contributed to growth in surgical care productivity in the latter part of our analysis
window. Since the enactment of the Affordable Care Act, in 2010, the Center for Medicare and
Medicaid have proposed and implemented various value-based financing programs, including
the Accountable Care Organizations, Hospital Value-based Purchasing, and Episode-based
Payment Initiatives (CMS.gov). These programs, as well as similar initiatives organized among
private payers, all seek to structure reimbursements in ways that encourage cost-effective care,
e.g. by tying payment to the achievement of certain quality metrics, or by bundling payments
prospectively to encourage prudent resource use (CMS.gov). Further study of the productivity
impacts of these reforms is clearly needed.

46
Tracking the growth in surgical care productivity has important implications for Medicare
reimbursement. To adequately compensate healthcare providers, the Center for Medicare and
Medicaid has historically adjusted payments in accordance with the rate of healthcare cost
inflation (Medicare Payment Advisory Commission). However, this arrangement did not
account for potential gains in productivity, which would enable providers to deliver higher
quality or quantity of care. To ensure that Medicare could also “benefit from productivity gains
in the economy at large”, the Affordable Care Act reduced annual updates on Medicare
reimbursements by the rate of multifactor productivity growth in the private nonfarm economy
(Medicare Payment Advisory Commission). Given this adjustment, it is therefore important to
track productivity in the provision of healthcare services and assess their growth rate relative to
that of the general economy. If productivity growth for certain surgeries were less than that of
the economy at large, for instance, then the slower growth in reimbursements may not allow
providers to maintain the same quality or quantity of care.

Based on Medicare’s 2012 Board of Trustees Report, future productivity growth of the general
economy was forecast at 1.1% per year, while the productivity growth of the healthcare sector
was forecast at 0.4%. Comparing these values to the productivity growth rates estimated in this
study, we find that all 11 classes of surgeries except for CABG experienced higher productivity
growth compared to the healthcare sector in general. Additionally, the annualized productivity
growth of debridement of wound, excision, lysis peritoneal adhesions, exploratory laparotomy,
small bowel resection, and tracheostomy outpaced that of the general economy. These
comparisons suggest that Medicare’s benchmarking of reimbursement against economy-wide
productivity growth may not negatively affect U.S. hospitals’ ability to continue providing high
47
quality/quantity surgical care to patients (Romley et al., 2015). The high rates of productivity
observed among all but one of the surgery classes examined also suggest that the phenomenon of
cost disease, in which “in which a heavy reliance on labor limits opportunities for cost
efficiencies stemming from technological improvement,” may not be a salient concern in the
context of surgical care (Boards of Trustees, 2012).

To our knowledge, this study is the first to apply a quality-adjusted productivity and assess the
quality, cost, and productivity trends of a broad range of surgeries across a national sample of
U.S. hospitals. While the cost and quality of surgical care have been examined, the relationships
between cost and quality over time have at most only been analyzed at the level of simple
correlations and in the context of specific surgical procedures (Finks et al., 2011; Lawson et al.,
2014; Weiser et al., 2011). The productivity metric employed in this study is also unique in two
ways. First, we modelled an extensive set of covariates on patients’ socio-demographics and
illness severity in order to control for potential changes in quality/cost of surgical care caused by
differences in patient case mix over time. This adjustment for the heterogeneity of inputs could
help produce more accurate estimates compared to other methodologies previously used in the
productivity literature.
8
Second, our productivity metric measured output not only in terms of the
number of hospital stays “produced” by hospitals, but also by the patient outcomes achieved,
both in terms of survival and readmission. The quality metric used – 30-day survival without
unplanned readmissions within 30 days of discharge – has in particular gained attention as an
important marker for quality-of-care (Kassin et al., 2012).

48
This paper has several limitations. First, our findings may not be generalizable beyond the
specific surgeries and populations studied. Although the procedures examined span the
categories of general, cardiac, and abdominal/gastrointestinal surgery, there remain other
surgeries, such as those involving the brain, that were not included. Similarly, surgical care for
Medicare beneficiaries may differ in significant ways from care received by commercially
insured / younger patients. Second, our productivity estimates are sensitive to the quality metric
used. While 30-day survival without readmissions is an important quality metric, it does not
capture other important patient outcomes, such long-term mortality, quality of life, or functional
status. Finally, the study only analyzed surgeries performed in inpatient settings.

Conclusion
Despite the recent slowdown in the growth of healthcare costs, spending on health care continues
to represent a substantial and potentially unsustainable portion of total national GDP (CMS.gov).
Achieving cost reduction while maintaining quality of care would therefore require an
improvement in the value – or in other words, productivity – of healthcare delivery. Using
Medicare claims for fee-for-service beneficiaries, we assessed trends in quality- and patient
illness severity-adjusted productivity of 11 classes of surgeries between 2002 and 2014. Adjusted
annualized productivity growth rates were larger than the productivity growth of the healthcare
sector for 10 out of 11 surgery groups, suggesting that their quality of care delivered has been
outpacing the hospital resources used. Innovations in surgical technology, increased adoption of
evidence-based guidelines, as well as the implementation of value-based reimbursement
programs may all have contributed to this improvement in productivity. Future studies should
49
consider applying alternate quality definitions to measure productivity, as well as examining
surgical care productivity in the outpatient/ambulatory care setting.

50
Variables
All 2002 2014 P-value
Inpatient stay case number, n 1,414,232 136,535 63,764
Hospital number
1
, n 170,253 14,202 10,782
30-day survival with no unplanned readmissions 91.5% 91.1% 91.4% 0.031
Readmission rates 16.5% 16.6% 13.9% <0.001
Inpatient costs (2014 $) 27,263.9 26,872.2 28,186.4 <0.001
Part B non-institutional charges (2014 $) 3,762.7 4,035.5 3,724.5 <0.001
Total costs (2014 $) 31,026.6 30,907.7 31,910.9 0.001
Teaching hospital 65.7% 67.9% 67.8% 0.912
CCS Surgeries
34 Tracheostomy 7.5% 3.4% 2.2% <0.001
43 Heart Valve Procedures 6.9% 5.5% 13.5%
44 CABG 6.7% 17.4% 12.7%
45 PTCA 10.7% 37.4% 31.8%
71 Gastrostomy 12.8% 7.2% 7.0%
73 Ileostomy and Other Enterostomy 1.6% 0.2% 0.4%
75 Small Bowel Resection 9.4% 2.2% 3.2%
78 Colorectal Resection 17.8% 15.7% 18.7%
89 Exploratory Laparotomy 1.7% 0.3% 0.3%
90 Excision, Lysis Peritoneal Adhesions 10.5% 2.6% 3.5%
169 Debridement of Wound 14.4% 8.1% 6.8%
Patient Characteristics
Age (Mean ± sd) 76.1 ± 3.3 75.9 ± 3.3 76.3 ± 3.7 <0.001
Female 47.2% 47.6% 46.7% 0.001
Race : White 87.5% 88.4% 86.7% <0.001
Race : African American 8.3% 8.0% 8.1% 0.689
Race : Hispanic 1.6% 1.6% 1.5% 0.158
Race : Other 2.6% 1.9% 3.7% <0.001
Social Characteristics
Median household income (2014 $) 4,289.1 ± 1,035.9 4,253.8 ± 982.4 4,373.1 ± 1,156.1 <0.001
Social Security income (2014 $) 1,133.8 ± 95.4 1,133.8 ± 92.8 1,136.6 ± 101.1 0.033
Poor 12.0% 12.1% 11.8% <0.001
Employed 94.3% 94.3% 94.4% <0.001
Less than high school education 19.8% 19.9% 19.4% <0.001
Urban 71.0% 71.6% 70.5% <0.001
Hispanic 9.0% 8.8% 8.8% 0.934
Single 41.6% 41.9% 41.2% <0.001
Elderly in an institution 5.6% 5.7% 5.4% <0.001
Non-institutionalized elderly with physical disability 29.3% 29.3% 29.1% 0.018
Sensory disability among elderly 14.5% 14.5% 14.5% 0.991
Mental disability 11.0% 10.9% 10.9% 0.832
Self-care disability 9.7% 9.7% 9.6% 0.015
Difficulty going-outside-the-home disability 20.5% 20.5% 20.3% <0.001
Patient Disease Severity
1 Charlson-Deyo comorbidity 35.4% 37.0% 29.5% <0.001
2 Charlson-Deyo comorbidities 24.7% 23.8% 26.0%
3 Charlson-Deyo comorbidities 10.7% 8.8% 16.4%
4 Charlson-Deyo comorbidities 3.3% 1.9% 8.1%
5+ Charlson-Deyo comorbidities 1.0% 0.2% 3.7%
Emergency Inpatient Admissions 38.6% 34.6% 46.6% <0.001
Urgent Inpatient Admissions 23.2% 26.2% 17.7% <0.001
Elective Inpatient Admissions 38.2% 39.2% 35.7% <0.001
CABG Related Risk Factor Adjustors ( CCS 44)
AHRQ predicted inpatient mortality 2.9% 1.9% 3.4% <0.001
NO of cases with AHRQ predicted inpatient mortality 11,419 915 779 <0.001
Location of heart attack: Anterolateral (410.0x) 70.7% 75.1% 64.9% <0.001
Location of heart attack: Other Anterior Wall (410.1x) 2.5% 2.9% 2.2% <0.001
Location of heart attack: Inferolateral Wall (410.2x) 0.9% 0.8% 1.0% 0.009
Location of heart attack: Inferoposterior Wall (410.3x) 0.6% 0.6% 0.6% 0.853
Location of heart attack: Other Inferior Wall (410.4x) 2.7% 3.3% 2.3% <0.001
Location of heart attack: Other Lateral Wall (410.5x) 0.3% 0.4% 0.3% 0.249
Location of heart attack: True Posterior Wall (410.6x) 0.1% 0.2% 0.1% 0.271
Location of heart attack: Sub-Endocardial (410.7x) 20.1% 14.7% 26.8% <0.001
Location of heart attack: Other Specified Sites (410.8x) 0.5% 0.6% 0.4% 0.275
* Mean values, distributions and tests were weighted by number cases of each hospital.
* Data are presented as percent, or mean ± SD.
* p-values were calculated from Welch's t-test for continuous variables or chi-square test for categorical variables, comparing between 2012 and 2014.
1
Hospital number reported in "2002" or "2014" columns is the number of hospitals in each year. The number reported in "All" column is the sum of
of numbers of hospitals from 2002 to 2014.
Table 1. Demographic Information for hospitalized patients who underwent a surgical intervention
Tables and Figures

51
Figure 1: Adjusted Year-by-Year Changes in Surgical Cost for 11 Surgery Groups, 2002-2014.

Note: Surgical cost trends were adjusted for patient illness severity. Costs are adjusted to 2014 USD.

Figure 2: Adjusted Year-by-Year Changes in Surgical Quality for 11 Surgery Groups, 2002-2014.

Note: Surgical cost trends were adjusted for patient illness severity. Surgical care quality was measured as the percent of hospital
stays for which the patient had 30-day survival beyond admissions and no unplanned readmissions within 30 days of discharges.

52
Figure 3: Annualized Productivity Growth Rates by CCS Categories

Note: Unadjusted growth rate calculated via OLS regression with the logarithm of productivity as the dependent
variable and binary variables for year of admission as the independent variable. Adjusted regression includes
controls for patient health, admission type, and common Dx CCS. Values highlighted in yellow are statistically
significantly different from 0.

53
Appendix

Appendix Exhibit 1. Unadjusted and Adjusted Year-By-Year Quality Growth for Each CCS
Category, 2002-2014

Notes: This figure shows predicted rates of "high-quality stays" and associated confidence intervals from multivariate logit
regressions in which indicators for 30-day survival without readmissions for patients are the dependent variables. The unadjusted
and adjusted models are regressed on year indicators at the individual level. The adjusted regressions are controlled for age,
gender, race/ethnicity, the Charlson comorbidities, area sociodemographic, AHRQ inpatient mortality risk for CABG, teaching
hospital indicators, and inpatient stay characteristics like diagnosis codes and inpatient admission types.

Figure 1. Quality Changes By CCS Categories, 2002 - 2014
86%
87%
88%
89%
90%
91%
92%
93%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
90: Excision, lysis Peritoneal Adhesions
Unadjusted
#REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
50%
55%
60%
65%
70%
75%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
80%
82%
84%
86%
88%
90%
92%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
65%
70%
75%
80%
85%
90%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
73%
74%
75%
76%
77%
78%
79%
80%
81%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
88%
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
89%
90%
91%
92%
93%
94%
95%
96%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
78%
79%
80%
81%
82%
83%
84%
85%
86%
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Quality
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval

54
Appendix Exhibit 2. Unadjusted and Adjusted Year-By-Year Cost Growth for Each CCS Category, 2002-
2014

Notes: This figure shows predicted costs and associated standard errors from ordinary least squares regressions in which sum of
inpatients and non-institutional costs are the dependent variables. The unadjusted and adjusted models are regressed on year
indicators at the individual level. The adjusted regressions are controlled for age, gender, race/ethnicity, the Charlson
comorbidities, area sociodemographic, AHRQ inpatient mortality risk for CABG, teaching hospital indicators, and inpatient stay
characteristics like diagnosis codes and inpatient admission types.

Figure 1. Cost Changes By CCS Categories, 2002 - 2014
14.0
15.0
16.0
17.0
18.0
19.0
20.0
21.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
169: Debridement of Wound
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
23.0
24.0
25.0
26.0
27.0
28.0
29.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
90: Excision, Lysis Peritoneal Adhesions
Unadjusted #REF!
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
20.0
22.0
24.0
26.0
28.0
30.0
32.0
34.0
36.0
38.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
89: Exploratory Laparotomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
25.0
26.0
27.0
28.0
29.0
30.0
31.0
32.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
78: Colorectal Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
30.0
32.0
34.0
36.0
38.0
40.0
42.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
75: Small Bowel Resection
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
20.0
22.0
24.0
26.0
28.0
30.0
32.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
73: Ileostomy and Other Enterostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
17.5
18.0
18.5
19.0
19.5
20.0
20.5
21.0
21.5
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
71: Gastrostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
17.0
18.0
19.0
20.0
21.0
22.0
23.0
24.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
45: PTCA
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
42.0
43.0
44.0
45.0
46.0
47.0
48.0
49.0
50.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
44: CABG
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
52.0
54.0
56.0
58.0
60.0
62.0
64.0
66.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
43: Heart Valve Procedures
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval
60.0
70.0
80.0
90.0
100.0
110.0
120.0
130.0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Cost
Year
34: Tracheostomy
Unadjusted Adjusted
Unadjusted Model 95% Confidence Interval Adjusted Model 95% Confidence Interval

55
References

AHRQ (2010). Agency for Healthcare Research and Quality. https://www.ahrq.gov/

Ashby, J., Guterman, S., and Greene, T. (2000). An Analysis of Hospital Productivity and
Product Change. Health Affairs 19, 197–205.

Bekara, F., Vitse, J., Fluieraru, S., Masson, R., Runz, A.D., Georgescu, V., Bressy, G., Labbé,
J.L., Chaput, B., and Herlin, C. (2018). New Techniques for Wound Management: A Systematic
Review of Their Role in the Management of Chronic Wounds. Archives of Plastic Surgery 45,
102–110.

Birkmeyer, J.D., Gust, C., Baser, O., Dimick, J.B., Sutherland, J.M., and Skinner, J.S. (2010).
Medicare Payments for Common Inpatient Procedures: Implications for Episode-Based Payment
Bundling. Health Services Research 45, 1783–1795.

Boards of Trustees. (2012) Annual Report of the Boards of Trustees of the Federal Hospital
Insurance and Federal Supplementary Medical Insurance Trust Funds. CMS.gov
https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-
Reports/ReportsTrustFunds/Downloads/TR2012.pdf

Brass, P., Hellmich, M., Ladra, A., Ladra, J., and Wrzosek, A. (20). Percutaneous Techniques
Versus Surgical Techniques for Tracheostomy. The Cochrane Database of Systematic Reviews
7, CD008045–CD008045.

Burke, L.A., and Ryan, A.M. (2014). The Complex Relationship between Cost and Quality in
US Health Care. AMA Journal of Ethics 16, 124–130.

Chee, T.T.M., Ryan, A.M., Wasfy, J.H.M., and Borden, W.B. (2016). Current State of Value-
Based Purchasing Programs. [Miscellaneous Article]. Circulation 133, 2197–2205.

CMS (2014). FY 2014 IPPS final rule. https://www.cms.gov/Medicare/Medicare-Fee-for-
Service-Payment/AcuteInpatientPPS/FY-2014-IPPS-Final-Rule-Home-Page-Items/FY-2014-
IPPS-Final-Rule-CMS-1599-F-Data-Files

CMS. Innovation Models | Center for Medicare & Medicaid Innovation.
https://innovation.cms.gov/initiatives/

Cylus, J.D., and Dickensheets, B.A. (2007). Hospital Multifactor Productivity: A Presentation
and Analysis of Two Methodologies. Health Care Financing Review; Washington 29, 49–64.

56
Engelman, D.T., Ben Ali, W., Williams, J.B., Perrault, L.P., Reddy, V.S., Arora, R.C., Roselli,
E.E., Khoynezhad, A., Gerdisch, M., Levy, J.H., et al. (2019). Guidelines for Perioperative Care
in Cardiac Surgery: Enhanced Recovery After Surgery Society Recommendations. JAMA
Surgery 154, 755–766.

Finks, J.F., Osborne, N.H., and Birkmeyer, J.D. (2011). Trends in Hospital Volume and
Operative Mortality for High-Risk Surgery. New England Journal of Medicine 364, 2128–2137.

Gustafsson, U., Scott, M., Hubner, M., Nygren, J., Demartines, N., Francis, N., Rockall, T.,
Young-Fadok, T., Hill, A., Soop, M., et al. (2019). Guidelines for Perioperative Care in Elective
Colorectal Surgery: Enhanced Recovery After Surgery (ERAS ®) Society Recommendations:
2018. World Journal of Surgery 43, 659–695.

Harper, M.J., Khandrika, B., Kinoshita, R., and Rosenthal, S. (2010). Nonmanufacturing
Industry Contributions to Multifactor Productivity 1987-2006. Monthly Labor Review: MLR;
Washington 133, 16–31.

Kassin, M.T., Owen, R.M., Perez, S.D., Leeds, I., Cox, J.C., Schnier, K., Sadiraj, V., and
Sweeney, J.F. (2012). Risk Factors for 30-Day Hospital Readmission among General Surgery
Patients. Journal of the American College of Surgeons 215, 322–330.

Lawson, E.H., Zingmond, D.S., Stey, A.M., Hall, B.L., and Ko, C.Y. (2014). Measuring Risk-
Adjusted Value Using Medicare and ACS-NSQIP: Is High-Quality, Low-Cost Surgical Care
Achievable Everywhere? Annals of Surgery 260, 668–679.

Ljungqvist, O., Scott, M., and Fearon, K.C. (2017). Enhanced Recovery After Surgery: A
Review. JAMA Surgery 152, 292–298.

Medicare (2012). Historical Impact Files for FY 1994 through Present.
https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/AcuteInpatientPPS/
Historical-Impact-Files-for-FY-1994-through-Present

Newhouse, J.P., Garber, A.M., Graham, R.P., McCoy, M.A., Mancher, M., Kibria, A., Variation
in Health Care Spending: Target Decision Making, Not Geography (Washington, D.C., UNITED
STATES: National Academies Press).

Putensen, C., Theuerkauf, N., Guenther, U., Vargas, M., and Pelosi, P. (2014). Percutaneous and
Surgical Tracheostomy in Critically Ill Adult Patients: A Meta-analysis. Crit Care. 2014 Dec
19;18(6):544.

Rashid, A.O., and Islam, S. (2017). Percutaneous Tracheostomy: A Comprehensive Review.
Journal of Thoracic Disease 9, S1128–S1138.

Romley, J., Goldman, D., and Sood, N. (2015). US Hospitals Experienced Substantial
Productivity Growth During 2002–11 | Health Affairs 34: 511–518

57
Scott, M.J., Baldini, G., Fearon, K.C.H., Feldheiser, A., Feldman, L.S., Gan, T.J., Ljungqvist, O.,
Lobo, D.N., Rockall, T.A., Schricker, T., et al. (2015). Enhanced Recovery After Surgery
(ERAS) for Gastrointestinal Surgery, part 1: Pathophysiological Considerations. Acta
Anaesthesiologica Scandinavica 59, 1212–1231.

Sheiner, L., and Hutchins, A.M. (2016). Measuring Productivity in Healthcare: a Analysis of the
Literature. https://www.semanticscholar.org/paper/MEASURING-PRODUCTIVITY-IN-
HEALTHCARE-%3A-AN-ANALYSIS-Sheiner-Hutchins/
13429271e467091768467f327a3355d93e207318

Thompson, R.E., Pfeifer, K., Grant, P.J., Taylor, C., Slawski, B., Whinney, C., Wellikson, L.,
and Jaffer, A.K. (2017). Hospital Medicine and Perioperative Care: A Framework for High-
Quality, High-Value Collaborative Care. Journal of Hospital Medicine 12, 277–282.

Volpp, K.G., Rosen, A.K., Rosenbaum, P.R., Romano, P.S., Even-Shoshan, O., Wang, Y.,
Bellini, L., Behringer, T., and Silber, J.H. (2007). Mortality Among Hospitalized Medicare
Beneficiaries in the First 2 Years Following ACGME Resident Duty Hour Reform. JAMA 298,
975–983.

Weiser, T.G., Semel, M.E., Simon, A.E., Lipsitz, S.R., Haynes, A.B., Funk, L.M., Berry, W.R.,
and Gawande, A.A. (2011). In-hospital Death Following Inpatient Surgical Procedures in the
United States, 1996–2006. World J Surg 35, 1950–1956.

58
Chapter 4: Variation in Demand Response to FDA Black Box
Warnings
Ning Ning, Yang Lu, Laura Gascue, and Geoffrey Joyce

Introduction
Many drugs have known side-effects when they are approved for sale by the FDA, and
occasionally new adverse effects are also found post-market approval. In dealing with drugs that
have potentially severe adverse effects, the FDA must decide how to provide corresponding risk
information to patients and providers. These communications typically include “Public Health
Advisories,” “Safety Alerts,” “Dear Healthcare Provider” letters, and boxed warnings (Dusetzina
et al., 2012). Boxed warnings, which are also commonly referred to as black box warnings (BBWs),
are the most severe warnings that can be issued by the FDA (Cheng et al., 2014; Maggs and
Kesselheim, 2014.; Dusetzina and Caleb Alexander, 2011; Panagiotou et al., 2011). BBWs can be
issued as a requirement for approval (pre-market warning) or after a drug has entered the market
(post-market warning). Post-market BBWs are usually issued in response to reports of adverse
effects in the general population after these events are previously unobserved during clinical trials,
perhaps due to smaller sample sizes. Because of the severity of BBWs, evaluation of their impact
on drug use is important for public safety and for informing the effectiveness of future risk
communication efforts. In particular, a decrease in drug use following the issuance of a BBW can
reduce patients’ exposure to severe adverse effects.

59
Although BBWs are the most serious class of FDA-issued warnings, they are issued more
frequently than one may expect. Among pharmacological products approved between 1996 and
2012, 30% received BBWs. Among all pharmacological products that received BBWs, 42%
received post-market BBWs (Cheng et al., 2014). Despite their prevalence, empirical evidence on
the effectiveness of BBWs in reducing risks of adverse effects is uncertain. Studies of BBWs on
antipsychotic medications found decreased use in outpatient and office-based settings, but no
change in use in nursing homes (Desai et al., 2012; Dorsey, 2010; Lester et al., 2011). While
several studies have found that minorities have significantly lower rates of adoption of new
technologies and novel drug therapies, there is little evidence that their response to safety warnings
is also delayed (Glied and Lleras-Muney, 2008; Groeneveld et al., 2005; Wang et al., 2006). The
only evidence comes from an analysis of rosiglitazone (Avandia) following its widely publicized
safety alert, where low-income patients and minorities were slower to discontinue use in response
to the BBW (Qato et al., 2016; Shi et al., 2011). However, these studies were limited in several
ways. First, they focused on a specific drug or drug class. Second, none of the prior studies
compared post-warning cessation rates to pre-warning rates, making it hard to identify the causal
effect of the warning from other factors which affect patient and provider decisions to discontinue
a medication. Finally, none of the studies examined the response of new versus current users of a
BBW drug. Patients and their providers may be more reluctant to discontinue use of a drug that
has been well tolerated and effective prior to warning, whereas they be more willing to start new
patients on alternative therapies in response to a safety warning.

In this study, we use longitudinal data from a nationally representative sample of Medicare
beneficiaries to assess the demand response to a large number of BBWs issued between March

60
2007 and December 2013. We examined the response to BBWs among current and new users, and
the extent to which response varied across patient, plan and area characteristics.

Methods
Data Sources
Our primary data source was a 20% random sample of Medicare beneficiaries enrolled in Part D
from 2006-2014. The Part D claims include National Drug Codes (NDC), dates of services, times
of refills, and days of supply for each claim. The research-identifiable version of the Medicare
Master Beneficiary Summary File (MBSF) reports patient demographic information, including
date of birth, gender, race/ethnicity, enrollment status, dual-eligibility status, and date of death.
The race/ethnicity variables used were externally validated by the Research Triangle Institute
(RTI), which incorporated information on beneficiary geography, first names and last names to
improve accuracy (Eicheldinger and Bonito, 2008). We used the Medicare Chronic Conditions
Data Warehouse (CCW) to identify patients’ chronic comorbid conditions. In addition, we
augmented Medicare claims with neighborhood characteristics from the American Community
Survey (ACS) to capture zip-level measures of education and median household income.

Adjusting Warning Dates
While BBWs are issued for drugs to emphasize the severity of certain adverse effects, the contents
of these warnings may have been communicated to patients and physicians via other types of
warnings issued by the FDA prior to escalating to BBWs. In order to capture the right timing that
physicians and patients are exposed to the content of BBWs, we used the FDA MedWatch data to
identify all forms of safety communications that had the same contents as BBWs. If the changes

61
in drug use happened around the alert or letter dates prior to the BBW dates, we instead used those
alert or letter dates as warning dates for accuracy. Other detailed drug information was obtained
from First Databank, which provided ingredient names, the NDC codes, and routes of drug
administration.

Selecting Sample Drugs
We selected our drugs of interest using multiple criteria, as summarized in Figure 1. We first
selected all post-market warnings and restricted our samples to those BBWs issued for oral or
transdermal medications between March 2007 and December 2013. We excluded BBWs which
applied to adolescents (increasing suicides in adolescents) or pregnant women (causing injury and
death to the developing fetus). The remaining 126 BBWs would affect the elderly population in
the United States. Since multiple warnings may apply to the same drug, we then combined the
drugs based on their active ingredients (N=80). We selected all Part D claims of these 80 drugs
between years 2006 and 2014, and aggregated days of supply prescribed in Part D claims for each
month. We restricted our final sample to widely used drugs, with aggregated days of supply greater
than 6000 at the warning month, in the Medicare population (N=57).

Identifying Demand Response in Aggregated Days of Supply
For all drugs with aggregated days of supply >6000, we looked at the changes in monthly
aggregated days of supply, and identified whether there was a demand response for each drug post-
warning. We defined demand response as 1) a significant decrease in aggregated days of supply
in Month 1-6 post-warning compared to Month 1-6 pre-warning; or 2) a statistically significant
break in trend of aggregated days of supply in Month 1-6 post-warning period.

62
We then used Medicare Part D data to measure average days’ supply of each medication in order
to identify short- vs. long-term medications, depending on whether the average days’ supply was
less or greater than 180 days, respectively. For the drugs with significant demand response, we
then looked at the impact of the corresponding BBWs on drug use for current users and new users
separately, given that physicians may be reluctant to discontinue a long-term medication in
response to a safety warning if the patient has been responded well to the drug for an extended
period of time.

Study Population and Drugs for Statistical Analysis
We analyzed the cessation in response to a safety warnings among current users of that drug, as
well as the initiation of use among new users. We defined “current users” as those in possession
of a BBW drug within 30 days prior to its warning date (Figure 2). We restricted our study sample
to those who did not switch Part D plan (PDP and MAPD) or beneficiary status (Dual, LIS, Non-
LIS) for at least 12 months after the warning month. We also focused on chronic medications for
current user analysis.

We defined “new users” as those who were enrolled in Medicare Part D without possession of a
BBW drug for at least 120 days prior to their first prescriptions of this BBW drug. Analyses of
“new users” examined both short- and long-term medications.

Quantifying Cessation Among Current Users
The primary dependent variable was discontinuation of use among current users at the time of the
warning. We defined a beneficiary as stopping therapy, i.e. “cessation” if they were not in

63
possession of the warning drug at any time during the period 7 to 12 months after the warning
(Figure 2). We chose this conservative window to allow sufficient time for patients and their
providers to incorporate this new information into their clinical practice.

Given that beneficiaries are likely to stop taking a medication for reasons other than a newly
issued safety warning, it is important to measure the change in cessation rates in the periods
before and after a warning. In order to calculate a baseline cessation rate as close to the warning
date as possible, we measured the cessation rate in a 6-month period immediately prior to the
warning date, calculated against a baseline index date set 13 months prior to the warning date,
such that the baseline cessation rate is defined as Month 7-12 after the baseline index date,
paralleling the cessation rate and warning date chronology. Whereas before “current user” was
defined as the users with possession of BBW drugs within 30 days prior to the warning date, the
“current user” used to calculate baseline cessation rate is defined as the users with possession of
BBW drugs within 30 days preceding the baseline index date (Figure 2).

We used logistic regression to examine the impact of a BBW on drug cessation, and predicted
unadjusted and adjusted cessation rates pre- and post-BBW. We controlled for a detailed set of
patient, plan and area characteristics, including patient demographics (age, gender, race,
beneficiary type), plan type (PDP vs MAPD), and comorbidities captured in the Chronic
Conditions Warehouse (CCW). We controlled for area characteristics, including medium
household income and average education level, at the ZIP Code level for each year based on data
from the American Community Survey. We presented both unadjusted and fully interacted models
for each drug, as well as pooled models of BBW drugs.

64
Quantifying Initiation in New Users
For new users, we looked at the number of initiation events among those who started the
medication in the period 6 months pre- and 12 months post-warning. We reported absolute
numbers and relative changes (%) in mean number of new users between specific time intervals.

Disparities
We also examined whether there was heterogeneous response in drug cessation and initiation
across groups with different patient/plan characteristics. For current users, we predicted cessation
rates of patient demographic or plan variables and also applied interaction terms between warning
and these variables to check two questions. First, whether there were significant disparities in
absolute cessation levels across different patient groups, including race, socioeconomics statues,
and plan types. Second, whether there were significant disparities in changes of cessation rates
pre- and post-warning across different groups. For new users, we preformed Chi-square tests to
explore the changes in distributions across subgroups, including age, race/ethnicity, gender, Part
D plan type, socioeconomic status, and region.

Results
Impact of Safety Warnings on Total Aggregated Days of Supply
Among the 57 drugs which met our inclusion/exclusion criteria, only 12 (21%) showed some
significant demand response in aggregated days of supply (Table 1). Ten out of 12 drugs had a
statistically significant decrease in aggregated days’ supply in pre- and post-warning periods, while
the other 2 drugs, ziprasidone and paliperidone, showed a statistically significant break in trend

65
post-warning. Summary statistics of key variables for current and new users were reported in
Appendix a.

The remaining 45 drugs did not exhibit a statistically significant demand response, as measured
by aggregate days of supply. The majority of these 45 drugs belong to the following major
therapeutic areas: 4 antibiotics, 8 narcotics, 4 HIV antivirals, 2 antipsychotics, 3
immunosuppressive drugs, 3 anticonvulsants, 2 antiarrhythmics, 2 blood thinners, 2 smoking
cessation aids, and etc. Their corresponding warnings were related to severe adverse conditions,
including liver injury, cardiac effects, hypersensitivity reaction, etc.

Impact of Safety Warnings on Current Users
We examined cessation rates of current users before and after the BBWs for 7 long-term
medications out of the 12 drugs with a demand response. Six of them were antipsychotics drugs
(Table 1). We ran pooled models on all antipsychotics drugs, and drug-specific models for each
antipsychotic drug and rosiglitazone.

The unadjusted baseline cessation rate varied by antipsychotic drugs, ranging from 4.5% for
clozapine to 38.5% for paliperidone. The averaged unadjusted cessation rate of all 6 antipsychotic
drugs was 16% before BBW. This indicated that current users of antipsychotic drugs dropped the
medication for reasons other than BBWs. The unadjusted cessation rate post-warning, which
ranged from 4.4% to 32.3%, was generally smaller than their unadjusted baseline cessation rate.
The average unadjusted cessation rate of all 6 antipsychotic drugs was 15.4% after BBW. It
indicated that fewer patients dropped the warning drugs post-warning than before BBWs. The

66
results were substantively unchanged after controlling for patient, community, and plan
characteristics. The adjusted cessation rate ranged from 5.5% to 37.1%, and averaged cessation
rate of all antipsychotics after warnings was 15.2%. This means that even though we found BBW
led to significant demand response in aggregated days’ supply for 6 antipsychotic drugs post-
warning, it did not have big impacts on drug cessation among their current users (Figure 3a).

Rosiglitazone is a special case, whose usage was significantly affected by BBW (Figure 3b). The
unadjusted baseline cessation rate of rosiglitazone was 13.7%, which is much lower than both the
unadjusted and adjusted cessation rates post-warning (57.7% and 57.5%, respectively).
This may largely result from its widely received media attention.

Disparities in Cessation Rates of Current Users
Despite negligible changes in overall cessation rates pre- and post-BBW for some drugs, we found
disparities across different patient and plan characteristics (Figure 4). The results were same for
antipsychotics and rosiglitazone. African Americans, Hispanics, and Asian/Pacific Islanders were
more likely to discontinue a medication in comparison to Whites (ORs>1, p<0.001). Medicare
beneficiaries with fully subsidized Part D coverage or eligibility for the low-income subsidies were
less likely to stop a medication (ORs<1, p<0.001) in comparison to individuals who were not
eligible for subsidized Part D coverage. In addition, beneficiaries enrolled in Medicare Advantage
plan (MAPD) were more likely to stop medications than those who enrolled in prescription drug
plan (ORs>1, p<0.001).

67
Except for different levels of cessation, we also observed differential responses in change of
cessation rates by patient and plan characteristics (Figure 4). The odds ratios of changes in
cessation rates pre- and post-BBW were observed using interaction terms in the adjusted
regressions. For rosiglitazone, African Americans and Hispanics were more likely to be affected
by BBWs, and their post-warning cessation rates increased more than the Whites (OR>1, p<0.001).
In addition, patients with DUAL eligibility or MAPD plans were less likely to be affected by
BBWs (OR<1, p<0.001). For antipsychotics, there is no such disparity in changes of cessation
rates pre- and post-warning across groups (p>0.05). BBWs did not associate with significantly
different changes in cessation rates of antipsychotics among any racial, SES or plan subgroups
(p>0.05).

Impact of Safety Warnings on New Users
Even though current users showed some differential responses to BBWs across subgroups, they
did not show significant increase in overall cessation rates for most of the drugs post-warning.
This means that the demand response, quantified using aggregated days of supply, may come from
the reduction in the number of new users. Physicians may have changed their prescribing behaviors
for naïve patients, which would lead to less new users (and less prescribed days of supply) post-
warning. New user analysis focused on the initiation of medications, so it was applied to all 12
drugs with significant demand response post-warning (Table 1).

Our results suggest that BBWs indeed may play a larger role in influencing prescribers’ decision
to initiate medication (Table 3). We found all drugs experienced decreases in numbers of new
users immediately within Month 1-6 post-warning. The percentage decrease ranges from 2% for

68
thiothixene to 85% for rosiglitazone. The monthly mean number of new users went down for 11
out of 12 drugs (except for thiothixene) within 1 year after warnings. There were no disparities or
major changes in distributions of demographic features for new users. The distributions and
significant levels were reported in Appendix b.

Discussion
A black box warning is the strongest safety alert that the FDA issues, reserved for medications
with potentially severe adverse effects. Between 1996-2012, 12.6% of FDA approved
medications received a BBW after being introduced to the market (Cheng et al., 2014). We found
that only 21% of drugs receiving a BBW between March 2007 and Dec 2013 exhibited a demand
response in the year following the warning, and almost all of response was due to lower rates of
new starts rather than more discontinuations in treatment of current users. Despite modest
changes in overall demand in response to a BBW, low income beneficiaries and their physicians
were far less likely to heed these warnings.

Apart from the 12 drugs with a demand response, most of the drugs in our study sample (N=45,
79%) do not exhibit a demand response. We characterized the warnings of these 45 drugs, and the
top 5 commonly ignored BBWs were regarding 1) liver injury/hepatotoxicity; 2) a risk evaluation
and mitigation strategy (REMS) for extended-release and long-acting opioid medications; 3)
cardiac effects; 4) hypersensitivity reaction/immunosuppression; and 5) tendinitis and tendon
rupture. It means patients are exposed to risks of theses severe adverse effects, which are ignored
by physicians.

69
Physicians and patients may have potentially ignored these warnings for the following reasons.
First, physicians may overlook BBWs, or do not comply with BBWs (Lasser et al., 2006; Wagner
et al., 2006), because they do not believe what were suggested in BBWs. For example, high non-
compliance with potassium level and liver enzyme monitoring was found, because it is unclear
whether liver function monitoring prevents liver toxicity (Wagner et al., 2006). Second, BBWs
warn against commonly seen risks, e.g. liver toxicity, thus physicians worry less about them. Third,
a BBW may be ignored when the perceived benefits of a drug outweigh the warned risk, even if
that risk has increased. For example, moxifloxacin is an antibiotic with a BBW warning of an
increased risk of tendinitis and tendon rupture, but it is still widely used.

In addition, we found different response patterns in current and new users for anti-psychotics,
further supporting the idea that physicians are primarily responsible for determining responses to
issuance of BBWs. If patients had been on a medication and proven the medication was effective
and relatively safe, physicians may be reluctant to switch these patients off the drug in response to
a BBW. Switching patients to a new drug may bring them more risks and uncertainties. In contrast,
for patients who are new to the drug, physicians may compare all available drugs side by side and
decide which one to prescribe. In this case, a BBW would stand out with life-threatening adverse
effects, which makes a drug less appealing to prescribe.

Previous studies have examined disparities among people of different race/ethnicity in their
response rate to novel developments in medical technology or drugs, and in general have found
minority populations to be slower in adopting to change. This might result from having slower
access to new information regarding advances in the medical field, a factor which might cause a

70
similar slower response in adapting to BBWs. However, contrary to this hypothesis, we found that
in general, minorities had higher cessation rates than Whites for antipsychotics. The higher
cessation rates of minorities may be due to their lower rates of medication adherence, because their
baseline cessation rates were also higher than Whites. We also found no significant differences in
changes of cessation rates across racial groups in the post-warning period for antipsychotics.
Similar results were also found within different socioeconomic groups and plan types. The
cessation rates of DUAL and LIS groups were smaller for antipsychotics compared to their
counterparts. Because socioeconomic status is often a good proxy for education, this may also
suggest that individuals with less education may be less likely to adhere to directions on their
medications. When comparing pre- and post-warning periods, there were no disparities in BBW’s
impact on changes in cessation rates of DUAL and LIS groups. For patients enrolled in MAPD,
their absolute cessation rates were higher than that of patients in PDP, and the changes between
pre- and post-warning cessation rates were not significantly different for PDP and MAPD groups.
In conclusion, disparities in cessation rates exist for some patient subgroups, but there were no
observed disparities in changes between pre- and post-warning cessation rates across different
groups for antipsychotics.

Apart from antipsychotics, rosiglitazone was also included in analyses of current and new users.
It is special in all cases. Large increases in cessation and reductions in initiation were found for
rosiglitazone post-warning. Disparities in impact of BBWs on cessation changes across subgroups
were also detected. All these may be because rosiglitazone and its severe cardiovascular adverse
effects were published in the New England Journey of Medicine (Nissen and Wolski, 2007), and
it has received wide range of media coverage and attention since then. Thus, a much larger amount

71
of information is available and for patients highlighting the risks of rosiglitazone presented in its
BBW, which could amplify the disparity in rates of access to new information among racial/ethnic
groups. This effect could explain why we found rosiglitazone alone to be an outlier in our results.
This hypothesis could be indirectly proven by examining changes in use of pioglitazone, a different
antidiabetic with similar adverse effects as rosiglitazone. Pioglitazone did not exhibit a demand
response in use, although it received the same BBW as rosiglitazone around the same time, but did
not receive the same degree of media attention.

Our study had some limitations. First, demand response may be limited by the types of warning
being issued. For example, a BBW may warn about potential risks in a small subgroup of patients,
and it may have large impact on drug use within this patient subgroup who have specific diagnosis
of certain diseases. In this case, we may underestimate the impact of BBW by looking at the
demand response averaged across all patients associated with this warning drug. Nonetheless, if
the demand changes were not captured in our study, it means the response was small enough in
general. Second, for those current users who dropped warning drug early and new users who were
supposed to be put on the warning drug, one concern is whether they found a proper substitution
treatment. If they used no other drugs, or switched to drugs with other severe health issues, then
the impact of BBW is less desirable. This might be a good direction for future studies. Third, we
could have delved deeper into all 57 drugs and come up with a specific study plan for each drug.
Nevertheless, this study is a good first step. We started with the big picture regarding BBWs and
can dive into the details in future studies.

72
Conclusion
Among 57 commonly used oral or transdermal drugs that received BBWs, the response in use is
minimal among the elderly US population after issuance of warnings. For the 7 long-term drugs
with an overall decrease in use, we found only modest response to BBWs among current users,
except for rosiglitazone. New patients, however, were less likely to initiate a medication after
issuance of BBW. For current users of long-term medications, we found higher pre- and post-
warning cessation rates among minorities, MAPD enrollees, and patients with DUAL and LIS
eligibility. However, the disparities in changes of cessation associated with BBW were detected
only for rosiglitazone. No disparities were identified between any population stratifications for
new users.

73
Tables and Figures
Figure 1. A Flowchart of the Drug Selection and Analysis Process

Figure 2. Study Design of Current User Analysis

Table 1. FDA Safety Warnings of 12 Medications with A Demand Response

74

Figure 3. Adjusted and Unadjusted Cessation Rates of Warning Drugs for Long-term Use
Figure 3a. Antipsychotic Drugs Figure 3b. Rosiglitazone

Figure 4. Odds Ratio of Patient Demographic Variables, Plan Characteristics, and Socioeconomic Status
Figure 4a. Odds Ratio of Cessation of Pooled Model for Antipsychotic Drugs Figure 4b. Odds Ratio of Cessation of Model for Rosiglitazone

Table 2. Percentage Change of Numbers of New Users for 12 Drugs with Demand Response to BBWs

75
Appendix
Appendix a. Summary Statistics of Key Variables for Current User Analysis*

PDP = Prescription Drug Plan;
MA-PD = Medicare Advantage Prescription Drug Plan.
*FDA safety warnings include safety alerts and black box warnings issued by the Food and Drug Administration.
† p <0.05 compared with the baseline group, calculated from two-sided t test for continuous variables or chi-square
test for categorical variables.

76

Appendix b. Summary Statisics of Key Variables for Current User Analysis*
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
1-6 mon
pre
warning
1-6 mon
post
warning
Total Current Users (N) 632 548 2,849 2,515 41,083 34,449 9,208 8,124 5,339 3,496 2,704 1,925 334 167 15,932 15,156 22,027 3,391 438 431 29,913 15,354 4,355 3,902
Race/ethnicity
White 73.3% 73.5% 66.1% 62.5%*** 67.9% 67.7% * 75.9% 74.4% * 87.9% 87.3% 69.2% 67.0% 82.9% 76.7% 73.1% 73.5% 55.3% 53.9%*** 67.8% 67.1% 87.1% 84.3%*** 72.5% 72.2%
African American 15.7% 16.4% 11.4% 11.1% 13.3% 13.0% 11.1% 12.5% 5.3% 5.3% 19.1% 20.2% 10.2% 13.2% 14.4% 14.2% 16.1% 14.8% 14.4% 15.1% 8.1% 9.8% 15.5% 15.4%
Asian/Pacific Islander 2.1% 3.3% 2.8% 2.0% 3.0% 3.4% 2.1% 2.3% 1.4% 1.8% 1.3% 1.0% <4% <7% 1.9% 1.8% 4.6% 4.1% 3.0% 2.6% 0.5% 0.6% 1.3% 1.1%
Hispanic 6.8% 5.8% 18.4% 22.8% 14.4% 14.7% 9.5% 9.6% 4.4% 4.5% 9.0% 10.7% 5.4% 7.8% 9.6% 9.2% 22.4% 25.8% 14.2% 13.5% 3.2% 4.0% 9.1% 9.7%
Socioeconomic Status
Dual-eligible 77.4% 79.2% 34.4% 31.7% 44.7% 44.8%* 61.8% 64.3%*** 15.0% 15.3% 82.1% 83.5% 34.7% 37.7% 62.1% 61.0%* 45.9% 43.3%** 66.0% 62.7% 55.3% 55.2% 75.9% 76.0%
Low-income subsidy 7.9% 6.8% 3.7% 3.4% 5.6% 5.1% 4.3% 4.8% 4.4% 4.8% 5.8% 5.6% 9.9% 7.2% 4.1% 3.8% 7.1% 7.0% 5.0% 5.6% 8.2% 8.2% 5.8% 6.3%
Other 14.7% 14.1% 61.9% 64.9% 49.7% 50.0% 33.8% 30.9% 80.6% 80.0% 12.1% 10.9% 55.4% 55.1% 33.8% 35.2% 46.9% 49.7% 29.0% 31.8% 36.5% 36.6% 18.3% 17.7%
Plan Type
PDP 90.8% 91.2% 57.2% 58.5% 69.8% 68.2%*** 79.4% 79.7% 66.5% 66.56%*** 88.2% 88.5% 70.1% 65.3% 78.6% 76.3%*** 69.6% 66.7%*** 80.4% 77.7% 79.5% 75.9%*** 84.1% 83.5%
MA-PD 8.7% 8.2% 38.9% 37.9% 28.5% 30.0% 19.6% 19.3% 31.2% 29.6% 11.1% 10.7% 27.0% 29.9% 20.5% 22.7% 29.6% 31.7% 18.7% 20.9% 19.5% 22.1% 14.9% 15.6%
Gender
Male 56.5% 58.2% 51.0% 50.2% 33.7% 33.5% 38.6% 39.8% 35.7% 37.4% 49.3% 49.5% 24.9% 21.6% 38.0% 38.4% 43.4% 47.6%*** 50.0% 44.3% 41.4% 43.1%*** 40.0% 38.9%
Female 43.5% 41.8% 49.0% 49.8% 66.3% 66.5% 61.4% 60.2% 64.3% 62.6% 50.7% 50.5% 75.2% 78.4% 62.0% 61.6% 56.7% 52.4% 50.0% 55.7% 58.6% 56.9% 60.0% 61.2%
Age
<65 82.9% 82.7% 30.3% 28.19%* 25.8% 26.1% 38.6% 39.3% 15.3% 15.8% 86.7% 86.4% 30.2% 37.1% 33.0% 33.8% 23.6% 25.86%*** 67.6% 75.17%** 55.8% 56.1% 72.8% 73.3%
65 -74 9.0% 10.4% 41.5% 44.3% 37.2% 37.6% 17.0% 16.9% 59.2% 59.2% 7.7% 7.5% 37.4% 33.5% 15.4% 15.5% 43.7% 44.8% 14.2% 12.1% 36.2% 36.1% 10.8% 11.0%
75 -84 5.7% 5.3% 21.5% 19.5% 25.7% 25.6% 23.8% 23.3% 22.8% 22.2% 3.7% 4.4% 26.4% 21.6% 25.8% 25.1% 26.1% 23.4% 8.7% 9.3% 7.4% 7.2% 9.2% 9.4%
>=85 2.4% 1.6% 6.8% 8.0% 11.4% 10.8% 20.7% 20.5% 2.8% 2.9% 1.9% 1.8% 6.0% 7.8% 25.8% 25.6% 6.6% 5.9% 9.6% 3.5% 0.6% 0.6% 7.2% 6.3%
Region
Northeast 27.2% 27.2% 12.8% 12.0%* 15.4% 16.0%*** 20.0% 19.9% 16.3% 14.13%* 14.7% 17.09%** 4.8% <7% 21.0% 20.9% 19.2% 13.7%*** 20.1% 19.0% 18.7% 18.4%*** 15.7% 15.5%
Midwest 35.4% 34.1% 15.8% 16.2% 19.5% 20.3% 22.2% 21.7% 18.1% 18.7% 28.0% 26.8% 12.0% <7% 22.0% 21.3% 16.2% 17.4% 16.9% 21.1% 25.8% 26.1% 20.1% 21.3%
South 21.8% 24.1% 42.6% 41.4% 41.5% 40.2% 35.5% 35.9% 50.1% 50.0% 43.1% 40.9% 79.6% 86.2% 36.9% 36.8% 39.5% 40.3% 42.9% 39.4% 41.2% 40.8% 46.8% 46.4%
West 15.4% 14.6% 20.0% 19.0% 20.2% 20.1% 20.7% 20.6% 14.7% 16.0% 13.5% 13.4% <4% <7% 18.8% 19.7% 18.9% 17.8% 14.6% 14.6% 13.6% 14.4% 16.2% 15.7%
Ketoconazole Osmoprep Metoclopramide Pentazocine Clozapine Ziprasidone Varenicline Paliperidone Thiothixene Olanzapine Risperidone Rosiglitazone
PDP = Prescription Drug Plan;
MA-PD = Medicare Advantage Prescription Drug Plan.
*FDA safety warnings include safety alerts and black box warnings issued by the Food and Drug Administration.
†P<0.05 compared with the baseline group, calculated from two-sided t test for continuous variables or chi-
square test for categorical variables.

77
References

Amanda Butler (2015). Listen to Reason: Patient Reasons for Medication Noncompliance.
https://orbisbio.com/listen-to-reason-patient-reasons-for-medication-noncompliance/
Bates, D.W., Cullen, D.J., Laird, N., Petersen, L.A., Small, S.D., Servi, D., Laffel, G., Sweitzer,
B.J., Shea, B.F., Hallisey, R., et al. (1995). Incidence of Adverse Drug Events and Potential
Adverse Drug Events: Implications for Prevention. JAMA 274, 29–34.
Bates, D.W., Spell, N., Cullen, D.J., Burdick, E., Laird, N., Petersen, L.A., Small, S.D., Sweitzer,
B.J., and Leape, L.L. (1997). The Costs of Adverse Drug Events in Hospitalized Patients. JAMA
277, 307–311.
Begosh, A., Goldsmith, J., Hass, E., Lutter, R.W., Nardinelli, C., and Vernon, J.A. (2006). Black
Box Warnings and Drug Safety: Examining the Determinants and Timing of FDA Warning Labels.
National Bureau of Economic Research. https://www.nber.org/papers/w12803
Boulware, L.E., Cooper, L.A., Ratner, L.E., LaVeist, T.A., and Powe, N.R. (2003). Race and Trust
in the Health Care System. Public Health Reports (1974-) 118, 358–365.
Budnitz, D.S., Lovegrove, M.C., Shehab, N., and Richards, C.L. (2011). Emergency
Hospitalizations for Adverse Drug Events in Older Americans. New England Journal of Medicine
365, 2002–2012.
Cheng, C. (2014). The Rising Tide of Boxed Warning Drugs.
https://www.fdbhealth.com/insights/articles/2014-09-23-are-you-prepared-to-meet-the-rising-
tide-of-boxed-warning-drugs
Cheng, C.M., Shin, J., and Guglielmo, B.J. (2014). Trends in Boxed Warnings and Withdrawals
for Novel Therapeutic Drugs, 1996 Through 2012. JAMA Intern Med 174, 1704–1705.
DePetris, A.E., and Cook, B.L. (2013). Differences in Diffusion of FDA Antidepressant Risk
Warnings Across Racial-ethnic Groups. Psychiatr Serv 64, 466–471, 471.e1-4.
Desai, V.C.A., Heaton, P.C., and Kelton, C.M.L. (2012). Impact of the Food and Drug
Administration’s Antipsychotic Black Box Warning on Psychotropic Drug Prescribing in Elderly
Patients with Dementia in Outpatient and Office-based Settings. Alzheimer’s & Dementia 8, 453–
457.
DiNicolantonio, J.J., and Serebruany, V.L. (2013). Challenging the FDA Black Box Warning for
High Aspirin Dose with Ticagrelor in Patients with Diabetes. Diabetes 62, 669–671.
Dorsey, E.R. (2010). Impact of FDA Black Box Advisory on Antipsychotic Medication Use. Arch
Intern Med 170, 96.

78
Downing, N.S., Shah, N.D., Aminawung, J.A., Pease, A.M., Zeitoun, J.-D., Krumholz, H.M., and
Ross, J.S. (2017). Post-market Safety Events Among Novel Therapeutics Approved by the US
Food and Drug Administration Between 2001 and 2010. JAMA 317, 1854–1863.
Dusetzina, S.B., and Caleb Alexander, G. (2011). Drug vs Class-Specific Black Box Warnings:
Does One Bad Drug Spoil the Bunch. J Gen Intern Med 26, 570–572.
Dusetzina, S.B., Higashi, A.S., Dorsey, E.R., Conti, R., Huskamp, H.A., Zhu, S., Garfield, C.F.,
and Alexander, G.C. (2012). Impact of FDA Drug Risk Communications on Health Care
Utilization and Health Behaviors: A Systematic Review. Medical Care 50, 466–478.
Eicheldinger, C., and Bonito, A. (2008). More Accurate Racial and Ethnic Codes for Medicare
Administrative Data. Health Care Financing Review; Washington 29, 27–42.
Gallini, A., Andrieu, S., Donohue, J.M., Oumouhou, N., Lapeyre-Mestre, M., and Gardette, V.
(2014). Trends in Use of Antipsychotics in Elderly Patients with Dementia: Impact of National
Safety Warnings. European Neuropsychopharmacology 24, 95–104.
Glied, S., and Lleras-Muney, A. (2008). Technological Innovation and Inequality in Health.
Demography 45, 741–761.
Groeneveld, P.W., Laufer, S.B., and Garber, A.M. (2005). Technology Diffusion, Hospital
Variation, and Racial Disparities among Elderly Medicare Beneficiaries 1989-2000. Medical Care
43, 320–329.
Halloran, K., and Barash, P.G. (2010). Inside the Black Box: Current Policies and Concerns with
the United States Food and Drug Administration’s Highest Drug Safety Warning System. Current
Opinion in Anesthesiology 23, 423.
Hurren, K.M., Taylor, T.N., and Jaber, L.A. (2011). Antidiabetic Prescribing Trends and
Predictors of Thiazolidinedione Discontinuation following the 2007 Rosiglitazone Safety Alert.
Diabetes Res. Clin. Pract. 93, 49–55.
Lasser, K.E., Allen, P.D., Woolhandler, S.J., Himmelstein, D.U., Wolfe, S.M., and Bor, D.H.
(2002). Timing of New Black Box Warnings and Withdrawals for Prescription Medications.
JAMA 287, 2215–2220.
Lasser, K.E., Seger, D.L., Yu, D.T., Karson, A.S., Fiskio, J.M., Seger, A.C., Shah, N.R., Gandhi,
T.K., Rothschild, J.M., and Bates, D.W. (2006). Adherence to Black Box Warnings for
Prescription Medications in Outpatients. Arch Intern Med 166, 338–344.
Lester, P., Kohen, I., Stefanacci, R.G., and Feuerman, M. (2011). Antipsychotic Drug Use Since
the FDA Black Box Warning: Survey of Nursing Home Policies. Journal of the American Medical
Directors Association 12, 573–577.

79
Libby, A.M., Brent, D.A., Morrato, E.H., Orton, H.D., Allen, R., and Valuck, R.J. (2007). Decline
in Treatment of Pediatric Depression After FDA Advisory on Risk of Suicidality with SSRIs. AJP
164, 884–891.
Maggs, L., and Kesselheim, A. The Role of Black Box Warnings in Safe Prescribing Practices |
Health Affairs. https://www.healthaffairs.org/do/10.1377/hblog20140820.040875/full/.
Nissen, S.E., and Wolski, K. (2007). Effect of Rosiglitazone on the Risk of Myocardial Infarction
and Death from Cardiovascular Causes. New England Journal of Medicine 356, 2457–2471.
Panagiotou, O.A., Contopoulos-Ioannidis, D.G., Papanikolaou, P.N., Ntzani, E.E., and Ioannidis,
J.P.A. (2011). Different Black Box Warning Labeling for Same-Class Drugs. J Gen Intern Med
26, 603–610.
Papanikolaou, K., Richardson, C., Pehlivanidis, A., and Papadopoulou-Daifoti, Z. (2006). Efficacy
of Antidepressants in Child and Adolescent Depression: A Meta-analytic Study. J Neural Transm
113, 399–415.
Qato, D.M., Trivedi, A.N., Mor, V., and Dore, D.D. (2016). Disparities in Discontinuing
Rosiglitazone Following the 2007 FDA Safety Alert: Medical Care 54, 406–413.
Ross, J.S., Jackevicius, C., Krumholz, H.M., Ridgeway, J., Montori, V.M., Alexander, G.C.,
Zerzan, J., Fan, J., and Shah, N.D. (2012). State Medicaid Programs Did Not Make Use of Prior
Authorization to Promote Safer Prescribing After Rosiglitazone Warning. Health Affairs
(Millwood) 31, 188–198.
Shi, L., Zhao, Y., Szymanski, K., Yau, L., and Fonseca, V. (2011). Impact of Thiazolidinedione
Safety Warnings on Medication Use Patterns and Glycemic Control Among Veterans with
Diabetes Mellitus. J. Diabetes Complicat. 25, 143–150.
Solotke, M.T., Dhruva, S.S., Downing, N.S., Shah, N.D., and Ross, J.S. (2018). New and
Incremental FDA Black Box Warnings from 2008 to 2015. Expert Opin Drug Saf 17, 117–123.
Van Leeuwen, E., Petrovic, M., van Driel, M.L., De Sutter, A.I., Stichele, R.V., Declercq, T., and
Christiaens, T. (2018). Discontinuation of Long-Term Antipsychotic Drug Use for Behavioral and
Psychological Symptoms in Older Adults Aged 65 Years and Older with Dementia. Journal of the
American Medical Directors Association 19, 1009–1014.
Wagner, A.K., Chan, K.A., Dashevsky, I., Raebel, M.A., Andrade, S.E., Lafata, J.E., Davis, R.L.,
Gurwitz, J.H., Soumerai, S.B., and Platt, R. (2006). FDA Drug Prescribing Warnings: Is the Black
Box Half Empty or Half Full? Pharmacoepidemiology and Drug Safety 15, 369–386.
Wang, J., Noel, J.M., Zuckerman, I.H., Miller, N.A., Shaya, F.T., and Mullins, C.D. (2006).
Disparities in Access to Essential New Prescription Drugs between Non-Hispanic Whites, Non-
Hispanic Blacks, and Hispanic Whites. Med Care Res Rev 63, 742–763.

80
Chapter 5: Summary and Future Research Directions

This dissertation described our efforts to analyze the economic, clinical and behavioral outcomes
of two important types of treatments: inpatient surgical care and prescription medications. For
inpatient surgical care, we investigated the potential contribution of surgical care, documented
trends in clinical outcome (quality) and economic outcomes (cost and productivity), and explored
innovations in treatments among Medicare beneficiaries from 2002 to 2014. For medications with
BBWs, we focused on changes in 3 behavior outcomes, demand response in total prescriptions
(aggregated days’ supply) for all users, cessation among current users, and initiation among new
users, after the issuance of BBWs from 2007 to 2013. These studies focused on important issues
that received a lot of attention from policy makers, physicians, and patients, but were not
sufficiently examined in the previous literature.

Based on our studies on inpatient surgical care, we believe that quality has improved significantly
for all 11 surgery categories from 2002 to 2014, while trends in cost varied by surgeries. We also
found 10 out of 11 surgery groups had productivity growth above expected productivity growth of
healthcare in general, indicating that these 10 surgeries had not suffered from what has been termed
cost disease. Future work in this area should integrate outpatient/ambulatory care settings, and may
consider using alternate quality definitions, such as quality of life or return to community.

For our study on commonly used medications with BBWs, the response in use is minimal among
the elderly US population after issuance of warnings. We found only modest response to BBWs
among current users, except for rosiglitazone. The disparities in changes of cessation associated

81
with BBWs were detected only for rosiglitazone. New patients only exhibited a behavioral
response in reducing initiation of medications after issuance of BBWs, with no disparities across
any population stratifications. Future work could focus on further exploring the other drugs which
did not exhibit a demand response, and identify changes in use of their substitution treatments.

Abstract (if available)

Abstract OBJECTIVES: Health spending in the United States has continued to increase in the recent years, without guarantee that health outcomes of medical and pharmaceutical treatments have improved simultaneously. This has resulted in wide recognition of the need for analyzing economic, clinical, and behavioral outcomes of US health treatments. In this dissertation, we aim to evaluate a wide range of treatment methods, including both inpatient surgical care and prescription medications, and examine the changes in quality, safety and cost of these health services using economic, clinical, and behavioral outcomes. This dissertation documents trends in risk-adjusted quality, cost, and productivity for a variety of surgical procedures, and explores innovation in treatment among Medicare beneficiaries from 2002 through 2013 in Chapter 2 and 3. In Chapter 4, we aim to assess the demand response to a large number of Black Box Warnings issued between March 2007 and December 2013. We want to examine the response to Black Box Warnings (BBWs) among current and new users, and the extent to which response varied across patient, plan and area characteristics. ❧ METHODS: For medical treatments, we focus on 11 classes of inpatient surgery, defined by the Agency for Health Research and Quality’s (AHRQ’s) Clinical Classification System

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The impact of treatment decisions and adherence on outcomes in small hereditary disease populations

PDF

Three essays in health economics

PDF

Thesis on Medicare Part D

PDF

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

PDF

Evaluating cancer treatments with health economics methods

PDF

Value in oncology care and opportunities for improvement

PDF

Health economics and outcomes research for informed decision making in rapidly evolving therapeutic areas

PDF

New approaches using probabilistic graphical models in health economics and outcomes research

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Understanding primary nonadherence to medications and its associated healthcare outcomes: a retrospective analysis of electronic medical records in an integrated healthcare setting

PDF

Essays in health economics and provider behavior

PDF

Outcomes of antibiotic use among children with acute respiratory tract infections

PDF

Characterization of health outcomes in patients with hemophilia A and B: Findings from psychometric and health economic analyses

PDF

Documenting the impact of under-treatment for chronic disorders

PDF

Value in health in the era of vertical integration

PDF

Three essays on comparative economic and clinical outcomes of antipsychotics using retrospective claims data

PDF

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

The impact of healthcare interventions using electronic health records: an evaluation within an integrated healthcare system

PDF

Economic aspects of obesity

Asset Metadata

Creator Ning, Ning (author)

Core Title Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Health Economics

Publication Date 12/16/2019

Defense Date 09/25/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Black Box Warning,cessation,Cost,demand response,Health Economics,initiation,inpatient surgery,mortality,OAI-PMH Harvest,outcomes research,productivity,quality,treatments,unplanned re-admissions

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Romley, John (committee chair), Joyce, Geoffrey (committee member), Trish, Erin (committee member)

Creator Email ningn@usc.edu,ningningusc@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-255083

Unique identifier UC11673459

Identifier etd-NingNing-8073.pdf (filename),usctheses-c89-255083 (legacy record id)

Legacy Identifier etd-NingNing-8073.pdf

Dmrecord 255083

Document Type Dissertation

Rights Ning, Ning

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