Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Developing an agent-based simulation model to evaluate competition in private health care markets with an assessment of accountable care organizations
(USC Thesis Other)
Developing an agent-based simulation model to evaluate competition in private health care markets with an assessment of accountable care organizations
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
DEVELOPING AN AGENT-BASED
SIMULATION MODEL TO
EVALUATE COMPETITION IN
PRIVATE HEALTH CARE MARKETS
WITH AN ASSESSMENT OF ACCOUNTABLE
CARE ORGANIZATIONS
By
ABDULLAH ALIBRAHIM
A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF DOCTOR IN PHILOSOPHY TO
FACULTY OF THE USC GRADUATE SCHOOL
December 2017
Committee Chair: Shinyi Wu, PhD
Daniel J. Epstein Department of Industrial and Systems Engineering
University of Southern California
2
Acknowledgements
“In the name of God, most Gracious, most Merciful”
In my journey towards completing my PhD Dissertation, I was graced with the guidance and
support of many individuals who made this the precious experience it is.
First, I would like to thank my mentor, Dr. Shinyi Wu, to whom I am immensely grateful for
fostering a stimulating environment and consistently pushing me to realize my potential. Since the
beginning of my journey, her guidance, insights, and perspective continuously improved the way
I think. I would like to thank her for her patience and helping me find my best interest. I am
privileged to have an inspiring advisor and a great friend.
I would also like to thank the esteemed members of my PhD dissertation committee, Dr. John
Romely and Dr. Detlof von Winterfeldt, for their time, comments, and constructive criticism. Also,
I am grateful for comments and insights by Dr. Gregory Stevens, Dr. Phebe Vayanos, and Dr.
Ketan Savla during the proposal stage of this dissertation.
I would like to thank the Industrial & Management Systems Engineering Department in the
College of Engineering & Petroleum of Kuwait University for their sponsorship and support during
my PhD studies.
I would like to express a special gratitude for my parents, Nuhad and Ibrahim, for providing the
love and safety throughout the years from which I was able to complete this journey. A special
thanks to my sisters, Hanan and Muneera, and my brother, Faisal, for their invaluable love and
support. Lastly, I would like to thank my brothers, Hamad and Salem, for alleviating the pressure
and sharing my frustrations during this eventful journey.
3
Abstract
Health market evolution necessitates continuous reevaluation of initiatives, competitive
regulations, and antitrust policies. The complexity of health care markets and competition make it
especially challenging to understand the relationships between ongoing initiatives that accumulate
market power, market player behaviors, and, ultimately, outcomes. For example, recent health care
reform efforts that leverage purchasing power through state-level health insurance exchanges, such
as Covered California, and spread collaborative, performance-based delivery models, such as
Accountable Care Organizations (ACOs), may tip the delicate competitive balance in health care
markets. Market conditions play a critical role in determining the extent to which delivery and
payment models such as the ACOs may foster anti-competitive behavior [1].
This dissertation compares features of competition in health care markets to those of complex
adaptive systems (CAS) and establishes CAS modeling as a suitable theoretical approach to
construct health system-level behaviors and patterns as a method of explanation to emergent
competitive dynamics in the markets. Studying health care markets as CAS, an agent-based model
(ABM) is created to simulate private health care markets aimed at exploring tensions and trade-
offs between competition and collaboration in health care provision under different market
conditions. A conceptual framework is developed to characterize a private health care market by
managed competition where insurers sell semi-standardized plans with access to a network of
providers on an insurance exchange and individuals are required to purchase coverage. Active
market players are modeled as 3 agents that represent health care plans, providers, and patients
engaging in local interactions and decisions. These interactions collectively create system-level
behaviors and outcomes. The model defines a market as an insurance rating area (IRA), which is
a geographic area where insurers rate communities risk as part of premium setting, typically
corresponding to a Metropolitan Statistical Area. In a hypothetical IRA, consumers purchase
insurance plans from a marketplace, and insurance plans and health providers reach price
agreements that are determined, in part, by their corresponding market influence. Bargaining
between insurers and providers is guided by solving a stylized Nash bargaining problem.
Consumers health care needs and willingness-to-pay (WTP) to access a provider govern their
choice of insurer during enrollment period and choice of provider for a treatment episode. This is
to reconstruct empirically established relationships between market features and outcomes.
Simulated health care markets varied in consumer population, number of providers, number of
insurers, and IRA size. Insurer premiums, provider quality, health outcomes, and expenditures
were observed for 3 years, a modeling time set to mitigate the impact of not capturing demographic
etiologic, and epidemiological changes in the model. Empirical relationships drawn from
systematically selected econometrics literature is used to establish the face-validity of the observed
relationships from the model’s simulation results. Upon validation, the simulation model
formalizes the quantitative associations between market conditions and observable outcomes.
4
This model is then used to implement a case study on the impact of ACO formations under
different market conditions on market outcomes, prices, and provider behaviors. Under various
simulated market configurations, two providers form an ACO, and the impact of this formation is
observed and compared to identical baseline market outcomes with no ACO formations. ACO
formations allow participating provider to bargain for higher price due to the combined market
share and WTP. While increasing provider variable costs, ACO formations tend to reduce
mortality and readmissions rates due to care coordination and provider investments.
Illustrative results show that in more competitive markets, ACO providers have the most market
power and price to gain from forming ACOs. In such markets, because provider prices are kept
due to balanced insurer/provider market power, ACO formations tip this balance in favor for the
ACO providers. In contrast, in less competitive markets with high baseline provider prices ACO
formations induce a smaller marginal provider price increases. The model reveals insights relating
to the synergistic impact of consumer quality preferences and ACO risk-based agreements on
sustaining strong value-based provider pricing compared to baseline markets with weaker
indications of value-based pricing.
In comparing competition in health care markets to CAS, this dissertation utilizes CAS simulation
approaches to investigate elusive aspects of competition typically overlooked in conventional
econometric research. The artificial health care markets reconstruct empirical relationships in a
bottom-up fashion, uncover system leverage points, and can serve as fertile testbed for policy
interventions. The case study on the competitive implications ACO formations demonstrates the
flexibility and utility of this approach. The findings challenge existing notions on the effectiveness
of ACO formations in competitive markets and quantify the effect of patient preferences on
increasing value-based competition in ACO markets.
5
Table of Contents
Acknowledgements ......................................................................................................................... 2
Abstract ........................................................................................................................................... 3
List of Figures ................................................................................................................................. 7
List of Tables .................................................................................................................................. 9
1. Introduction ........................................................................................................................... 10
2. Literature Review.................................................................................................................. 12
2.1. Health Care Market and Competition ........................................................................... 12
2.1.1. Health Care Market Composition ......................................................................... 12
2.1.2. Measuring Competition ........................................................................................ 13
2.1.3. Health Market Industrial Organization ................................................................. 14
2.1.4. Health Care Markets ............................................................................................. 16
2.2. ACOs and Potential Market Effects .............................................................................. 17
2.2.1. ACO Cost and Quality Improvements .................................................................. 17
2.2.2. Accountable Care Organizations and Anti-Competitive Concerns ...................... 18
2.2.3. ACO Concerns ...................................................................................................... 19
2.3. Complex Adaptive Systems and Health Care Market Competition ............................. 20
2.3.1. Complexity & Economics ..................................................................................... 21
2.3.2. Agent-Based Modeling ......................................................................................... 22
2.3.3. Uses of ABM ........................................................................................................ 23
2.3.4. ABM for understanding the competitive implications of ACOs .......................... 24
3. Structural Framework ........................................................................................................... 26
3.1. Health Care Market Model ........................................................................................... 26
3.1.1. Agent Structures & Components .......................................................................... 27
3.1.2. Conditions and Environment of the Private Health Care Market ......................... 28
3.2. Cooperation and Competition in the Health Care Market Model ................................. 30
3.2.1. Competition........................................................................................................... 30
3.2.2. Cooperation ........................................................................................................... 32
3.3. Delimitations of the Model Design ............................................................................... 36
3.3.1. Nature of Consumer Choices ................................................................................ 36
3.3.2. External Competitive Influences........................................................................... 36
3.3.3. Differentiated Products and Market Exits ............................................................. 37
4. Methods................................................................................................................................. 37
4.1. Components of the Health Care Market Competition ABM Simulation ...................... 38
4.1.1. Initial Agents and Key Attributes ......................................................................... 38
4.1.2. Initial Distributions and Agent Parameters ........................................................... 40
4.1.3. Initial Environment and System Parameters ......................................................... 41
4.2. Agents and Modules ..................................................................................................... 42
4.2.1. Consumers............................................................................................................. 42
4.2.2. Health Care Providers ........................................................................................... 47
4.2.3. Health Plans .......................................................................................................... 53
4.2.4. Health care Provider and Plan Agreements .......................................................... 53
4.3. Yearly Interaction and Event Sequence ........................................................................ 60
4.4. Observable Outcomes ................................................................................................... 61
4.5. Base Model Verification and Validation Methods ....................................................... 62
4.5.1. Verification Methods ............................................................................................ 62
6
4.5.2. Validation Methods ............................................................................................... 63
4.6. Base Model Run Settings .............................................................................................. 65
4.7. Case Study: Accountable Care Organizations .............................................................. 69
4.7.1. ACO Structure and Arrangements ........................................................................ 69
4.7.2. ACO Formation Verification Methods ................................................................. 71
4.7.3. ACO Experiment Settings..................................................................................... 71
4.7.4. ACO Sensitivity Analysis ..................................................................................... 72
5. Results ................................................................................................................................... 74
5.1. Model Implementation Overview ................................................................................. 74
5.2. Base Model Results....................................................................................................... 77
5.2.1. Base Model Verification ....................................................................................... 78
5.2.2. Base Model Face-Validity Assessment ................................................................ 82
5.3. ACO Case Study Results .............................................................................................. 86
5.3.1. ACO Formation Verification ................................................................................ 86
5.3.2. ACO Market Effects ............................................................................................. 87
5.3.3. ACO Provider Results - ACO vs. Non-ACO providers ....................................... 91
5.3.4. ACO Sensitivity Analysis ..................................................................................... 94
6. Discussion ........................................................................................................................... 101
6.1. Studying Health Care Markets as Complex Adaptive Systems .................................. 102
6.2. ABM of Private Health Care Markets ........................................................................ 102
6.3. ACO Model Findings .................................................................................................. 104
6.3.1. ACOs & Market power ....................................................................................... 104
6.3.2. ACOs & Quality Outcomes ................................................................................ 106
6.3.3. ACOs & Value .................................................................................................... 108
6.4. Policy & Research Implications.................................................................................. 111
6.5. Research limitations & Future work ........................................................................... 113
7. Conclusion .......................................................................................................................... 115
References ................................................................................................................................... 118
Appendices .................................................................................................................................. 135
Appendix A – Model Initialization Settings ........................................................................... 135
Appendix B – Base Model Steady State Assessment ............................................................. 136
Appendix C – Base Model Verification .................................................................................. 138
Appendix D – Base Model Face-Validity Assessment ........................................................... 143
Appendix E – ACO Model Verification ................................................................................. 149
Appendix F – ACO Model Sensitivity Analysis..................................................................... 150
7
List of Figures
Figure 2-1. An illustration of health care markets studied using a reductionist approach............ 25
Figure 2-2. An illustration of the natural components of an abstraction of health care markets
when utilizing a CAS approach. ................................................................................................... 26
Figure 2-3. Structural framework of the ABM adapted from Managed Competition market
framework. .................................................................................................................................... 29
Figure 2-4. Representation of an insurance rating area with a defined market structure for
providers and plans. ...................................................................................................................... 30
Figure 2-5. Illustration of a potential network configuration of agreements in a market a market
with 3 insurance plans and 4 hospitals, 2 of which form an ACO. ............................................... 33
Figure 2-6. ACO effects and pathways for altering health expenditures in the ABM that studies
competition in health care markets. .............................................................................................. 35
Figure 4-1. The organization of modules outlined in this model that allows each agent to play the
designated role in the agent-based model. Diamond shaped modules correspond to decision
modules. ........................................................................................................................................ 38
Figure 4-2. A flowchart representing the relationships between agents’ modules in the health care
market ABM simulation. .............................................................................................................. 40
Figure 4-3. Components that make up the intention for the provider agent for each effort level
according to the Theory of Planned Behavior. ............................................................................. 49
Figure 4-4. The relationship between types of episodes, health care need, and health outcomes. 51
Figure 4-5. Plot of Effort Multiplier to probability of adverse outcome, namely probability of
mortality or readmissions. ............................................................................................................. 52
Figure 4-6. A plot of an example insurer’s per enrollee profits with and without a provider j. ... 55
Figure 4-7. Plot of optimal prices in reference to the $1 per unit of intensity baseline to illustrate
the effect of provider market share on equilibrium prices. ........................................................... 56
Figure 4-8. Plot of insurer utility for a provider's mortality rate depicting the shape of the insurer
utility function. .............................................................................................................................. 58
Figure 4-9. Simulation sequence illustrated in a flow chart. Each loop constitutes a year, and the
model is observed for 5 years in total, observing only the last 3 years. ....................................... 68
Figure 5-1. Snapshot of the outcomes dashboard of the simulation model built on Anylogic
v.7.1.2............................................................................................................................................ 75
Figure 5-2. Snapshot of the virtual IRA as generated by the model using Anylogic v. 7.1.2. ..... 76
Figure 5-3. Snapshot of the market status component of the model's visual dashboard depicting
market-related measures over the simulation period as generated by Anylogic 7.1.2. ................ 76
Figure 5-4. Snapshot of the summary of health exchange related outcomes on insurer premiums
and network size. .......................................................................................................................... 77
Figure 5-5. Snapshot of the end-of-run observable outcomes averaged over the last 3 years of the
model as generated by an illustrative model run in Anylogic 7.1.2. ............................................ 77
Figure 5-6. Verification of the expected relationship between average patient distance per
episode. ......................................................................................................................................... 79
Figure 5-7. Verification of the quadratic relationship between provider effort and costs. ........... 80
Figure 5-8. Verification of the diminishing relationship between provider effort and mortality
rates. .............................................................................................................................................. 80
Figure 5-9. Scatter plot depicting the direct relationship between average proportion of providers
included in insurance plan networks and plan premiums. ............................................................ 81
8
Figure 5-10. Box plot of insurer premiums for artificially inflated consumer utilization rates
(UtilizationMultiplier factor). ....................................................................................................... 82
Figure 5-11. The base model experiment's direct relationship between provider concentration
(HHI) and provider service prices (p-value<0.001). ..................................................................... 84
Figure 5-12. Base model experiment results depicting the “mixed” association between provider
concentration (HHI) and average provider effort. ........................................................................ 85
Figure 5-13. Base model experiment results on the association between provider concentration
(HHI) and provider effort separated by market area representing different directions of
associations. .................................................................................................................................. 85
Figure 5-14. Verification of continuity of care benefits which are realized when a higher
proportion of the care is delivered in an ACO. ............................................................................. 87
Figure 5-15. Correlation between average provider price and effort in ACO and baseline markets
separated by market attributes (area & population size). .............................................................. 88
Figure 5-16. The association between provider concentration and provider prices in baseline and
ACO markets. ............................................................................................................................... 89
Figure 5-17. Average proportion of market providers included in insurer networks in baseline
and ACO markets. ......................................................................................................................... 90
Figure 5-18. A comparison of average provider prices in baseline markets with ACO and non-
ACO prices in ACO markets. ....................................................................................................... 92
Figure 5-19. Relationship between ACO prices and ACO savings per ACO enrollee................. 93
Figure 5-20. Sensitivity of market effort to ACO effort increase multiplier. ............................... 95
Figure 5-21. ACO savings for higher CareContinuity multiplier levels are higher but not positive.
....................................................................................................................................................... 96
Figure 5-22. The correlation between provider quality and price for different degrees of
consumer preferences of provider quality..................................................................................... 97
Figure 5-23. ACO savings increase for higher expected patient utilization rates. ....................... 98
Figure 5-24. Box plot of average provider effort in baseline and ACO markets for different
provider orientations. .................................................................................................................... 99
Figure 5-25. Inclusion of market provider in insurer networks for different settings of insurer
quality orientation. ...................................................................................................................... 100
Figure 5-26. Insurer quality orientation and effects on provider prices in ACO and baseline
markets. ....................................................................................................................................... 101
Figure 6-1. ACO provider prices compared to non-ACO provider prices grouped by number of
providers in market. .................................................................................................................... 105
Figure 6-2. Marginal cost-benefit assessment of providers in ACO markets. ............................ 108
Figure 6-3. Consumer quality preference & value-based competition in ACO and baseline
markets. ....................................................................................................................................... 110
9
List of Tables
Table 4-1. A summary of the active agents, attributes and behaviors in the health care market
ABM. ............................................................................................................................................ 38
Table 4-2. Table of distributions and parameters involved in the model and sources.................. 41
Table 4-3. List of individual agent's attributes along with source variables sampled from
NHANES 2013-2014. ................................................................................................................... 42
Table 4-4. Coefficients from Keeler et al (1988) [110] used in negative binomial regressions
equations for predicting episode rates. .......................................................................................... 43
Table 4-5. Coefficients from Keeler et al (1988) [110] used to generate log(episode costs). ...... 44
Table 4-6. Parameters used to estimate patient utility of received care at a given provider. ....... 46
Table 4-7. Systematic search on primary variable and observable model outcomes to establish
model face-validity. ...................................................................................................................... 64
Table 4-8. Summary of sources of model heterogeneity associated with model development and
agent decision making................................................................................................................... 65
Table 4-9. Summary of sources of model stochasticity associated with model development and
agent decision making................................................................................................................... 66
Table 4-10. Table of changes in variable costs and probability of adverse outcomes due to ACO
participation at each effort level. .................................................................................................. 69
Table 4-11. ACO parameters, values, and short description. ....................................................... 71
Table 4-12. Settings for market structure parameters in sensitivity analysis experiments. .......... 72
Table 4-13. ACO operational parameters varied in the sensitivity analysis. ................................ 73
Table 4-14. Provider and Insurer quality weight settings for sensitivity analysis. ....................... 74
Table 5-1. Theoretical and empirical relationships assessed in the base model grouped by agent
and outcome for comparison to base model results. ..................................................................... 82
Table 5-2. Difference in observable outcomes between ACO and baseline markets. .................. 87
Table 5-3. Comparison of ACO provider and non-ACO provider in ACO markets. ................... 91
Table 5-4. Percentage difference in observable outcomes between ACO providers and non-ACO
providers in ACO markets, analyzed by number of providers in market. .................................... 92
Table 5-5. Sensitivity of ACO market effects on the ACOEffortIncrease parameter. ................. 94
Table 5-6. Impact of care continuity multiplier on health outcomes. ........................................... 95
Table 6-1. Two configurations of simulated ACO markets defined by the number of providers to
illustrate the separable impact of care continuity and increased effort. ...................................... 106
10
1. Introduction
Initiatives put forward by the recent health law foster competition among insurers, coordination
among providers, and access for consumers, among many other changes. The health care system
is undergoing substantial restructuring as the full implementation of the Affordable Care Act is
nearing and efforts to repeal and replace it are increasing [2], [3]. Concurrently, health care market
conditions are constantly shifting with waves of horizontal and vertical, insurer and hospital
mergers [4], [5]. Additionally, health care spending per capita in the United States is the highest
amongst developed nations but quality and value of health care are not correlated with the growth
in costs [6]. The United States ranks near the bottom amongst 30 developed nation members of
the Organization for Economic Cooperation and Development in standard health measures [6].
Medical errors and weak preventative care are prevalent in the US health delivery system,
increasing the financial and economic burdens on payers and patients [7]. Rising health care costs
are partially driven by the fragmentation of health delivery outlets and an outdated fee-for-service
payment model [8]. Tracking and coordinating care are difficult, resulting in disintegrated systems
where accountability for health outcomes dissipates [9].
The recent health care reform in 2010 implements a series of legislations and initiatives to address
gaps in the US health care system and improves quality and accountability. Different aspects of
the health care system were restructured in the Patient Protection and Affordable Care Act (PP-
ACA, referred to as the ACA hereafter). The reform mandates health insurance coverage, increases
subsidies, stimulates insurance competition through establishing market places, and facilitates the
development of innovative payment and care delivery models [10].
The competitive stimulation enacted by the ACA is based on the concept of managed competition.
Alain Enthoven proposed managed competition that advocates using an “integrated framework
that combines rational principles of microeconomics with careful observation and analysis of what
works” to address failures in containing costs and promoting competition [11]. The idea leverages
on consolidating informed, paying consumers to improve the value of care and contain costs [11],
[12]. Along with mandatory insurance coverage for all individuals, the ACA established state-
level health insurance exchanges to allow for consolidated purchasing and aid consumers in
making informed choices based on costs, quality, and coverage. A minimum and standard
coverage are set to facilitate comparison between plans offerings by introducing metal tiers;
bronze, silver, gold, and platinum [10], [13]. Despite limitations in applications of managed
competition, theoretically, it would increase value of care and limit the growth of health care
expenditures in sufficiently large markets [11], [14]–[16].
The ACA restructures the supply side of care by endorsing the formation of Accountable Care
Organizations (ACOs). An ACO is a payment and care delivery model where a group of
independent providers voluntarily coordinate high quality care for their patients and agree to be
accountable for costs and quality of care and overall health of patients [17]. The ACO model aims
to realign payment arrangements to reward better health outcomes achieved as opposed to volume-
based traditional fee-for-service payments. The model rewards well-managed systems of care that
support patient-centeredness [18]. An ACO may involve primary care providers, multispecialty
groups, and hospitals [19]. The structure of an ACO varies greatly and coordination is loosely
defined. Since its emergence, the number of Medicare ACOs has been growing and has been
shown to contain spending and improve utilization and patient experience [20], [21].
11
When the formation of ACOs brings independent providers together, the resulting increase in
market power may be alarming. Providers’ cooperation in an ACO may increase their clout to
bargain for higher payments and lower risk [22]. There has been evidence relating increased
provider market power to increased prices in health care [23]. Questions have been raised as to
whether the growth of ACOs might be self-limiting because of the shifts in bargaining power that
might offset the effects of care coordination, accountability, and quality improvements on
expenditures [22], [24]–[26]. In fact, participation in ACOs as part of Medicare’s Shared Savings
Program (MSSP) is typically subject to an antitrust review enforced by the Federal Trade
Commission (FTC) and Department of Justice (DOJ) to minimize anti-competitive behavior [27].
Fragmentation and unaccountability in the US health care delivery system are being partially
remedied by the emerging ACO payment and delivery model, which consolidates providers.
Evidently, the competitive repercussions of this remedy raises many concerns. The potential
market changes from ACO formation might limit the likelihood of achieving the desired outcomes
of controlling spending and improving outcomes and patient experience. The way providers
behave when offered ACO incentives in each market is influential on the overall system costs,
quality, and competitive nature. ACO formations may generate value under some market
conditions but may be exploited by providers in other conditions. Understanding the market
conditions under which ACO value might be limited and is likely to incite anti-competitive
behaviors is critical to efforts for successfully ACO formations that achieve desirable goals.
Health care market outcomes are dependent on the dynamic interactions and decisions of players
or agents within this market. Agents in the health care market include patients, providers, and
payers. Health care markets can be thought of as a complex adaptive system (CAS) composed of
a population of autonomous agents that interact, adapt, and decide in endogenously and
exogenously determined market conditions. Therefore, the notion that global, static, and passive
evaluations are sufficient and effective at improving health care is challenged. Given the evolving
nature and heterogeneity of health care markets, there is a pressing need for continuous evaluation
and revaluation of proposed initiatives, competitive regulations, and antitrust policies. Due to the
complexity levels of agents and their interactions in the system, insightful examinations of the
system using reductionist mathematical and economic approaches alone are limited [28].
Therefore, it is proposed that Agent-based Modeling (ABM) offers a great platform for studying
competition in health care market as a CAS with autonomous, objective-driven agents [29]. This
tool has been used extensively in social sciences, economics, and other domains with extensive
behavioral components and decentralized decision-making, and has generated valuable insights.
To evaluate the potential anticompetitive implications of ACOs in different market conditions, an
ABM of health care markets with interactions between patients, providers, and commercial
insurers in the context of the insurance marketplaces is developed. This dissertation covers relevant
literature in the literature review section (chapter 2), outlines structural framework of the ABM
(chapter 3), describes methods for designing and building the ABM of competition in health care
markets and illustrative ACOs (chapter 4), provides model verification, validation, and scenario
results (chapter 5), and a discussion of the results, implications, limitations, and future work
(chapter 6 and 7).
12
2. Literature Review
To pave the way for how and why health care market competition should be modeled as a CAS,
literature is reviewed on the nature of health care markets, evolution and consequences of payment
models, and parallels to CAS. This literature review is divided into three sections: (1) the effects
of health care market structure and competition on outcomes and costs, (2) ACO formations and
potential market effects, and (3) the applications of CAS and ABM in various domains, particularly
in health care.
2.1. Health Care Market and Competition
Competition in health care markets does not abide by textbook definitions of competition [30]. In
fact, health care delivery, as a product, is drastically different from any other economic product.
Arrow (1963) elucidates on the contrast between the medical care market and other markets [31].
The locality of care provision, the assault on the individual’s physical integrity when acquiring
care, and information discrepancies between the provider and patient contribute to the complexity
of health delivery as an economic product [31]. Combined with ongoing health policy reforms,
waves of mergers among providers and insurers, and complex interrelationships, many aspects of
health care markets are elusive to conventional reductions economic approaches.
2.1.1. Health Care Market Composition
In economics, market structure is defined as the number of firms selling a product or service in a
market. The health care market is composed of buyers and sellers of health care services and
insurance. Insurers and providers usually compete for health coverage and delivery. To evaluate
competition, the focus shifts to privately insured individuals, typically because the effects of
competition on provider prices and expenditures are less substantial in public insurance programs
(i.e. Medicare, Medicaid, CHIP). Government insurance programs have predetermined
reimbursement mechanisms that are less likely to vary under different levels of competition among
providers. In the US health care system, multiple agents influence health care delivery for privately
insured individuals. Private insurers engage in price negotiations with provider groups in which
market conditions play a direct role in determining rates. Most privately insured individuals have
employee-sponsored insurance plans which means employers contract on their behalf with
insurance plans of various types [32]. The recent health care reform instituted health insurance
exchanges, federal subsidies, and individual insurance mandate. That, in turn, increased the
number of individuals with non-group coverage, making the health insurance exchanges an
interesting and promising arena for fruitful insurer competition. In these exchanges, plans sell
access to networks of providers to consumers. Many studies have focused on the performance of
health insurance exchanges as two-sided networks in different geographic, demographic, and
competitive conditions [33]–[36]. Therefore, for privately insured individuals, health outcomes
and costs are largely shaped by decisions made by insurers and providers. The environment in
which those agents exist, represented by the size of the market and presence of other insurers and
providers, also dictates how they behave and ultimately influences the system outcomes.
Industrial Organization is a strand of economic literature that offers a systematic framework to
study the structure of markets. It examines how a market is structured, how the structure of a
market affects the firms behavior and the performance, and how the behavior of firms influence
the structure and performance of the market [37]. The approach usually employed in this method
of analysis often relies on the Structure-Performance-Conduct (SPC) paradigm [38]. The SPC
paradigm is an approach that theorizes the feedback mechanism between market structure, market
13
performance, and firm conduct [38]. Analyses of the health market that utilize that approach often
identify measures of market structure and firm conduct, while omitting the behavioral link between
market conditions and system outcomes [28], [39]. Measures of market structure and competition
will be identified in the next section.
Due to the complexity of the health market, studies often focus on the relationship between the
structures of a limited number of sides of the market and performance and outcomes. Analyses
also regularly concentrate on the bilateral markets of the system, such as plan-provider markets,
patient-plan markets, and patient-provider markets, as opposed to a multilateral market that
encompasses all agent activity. While these studies made invaluable contributions to
understanding performance under different levels of competition, disciplinary bounds might limit
insights on health care market competition and its dynamics. Before an examination of literature
on the effect of market structure on outcomes and performance, the following subsection is a
discussion of the various ways in which competition is measured and operationalized in literature.
2.1.2. Measuring Competition
Increasing market competition promises increases in efficiency. However, competition itself is a
latent variable and is difficult to measure [40]. Different studies use different measures as a proxy
of the degree of competition in a market. These measures are employed in economics and antitrust
laws, which are concerned with curbing anti-competitive behavior. Competition relates to how the
behavior of one firm is influenced by the existence of another firm or of potential entrants to the
market [40]. The higher the level of competition, the lower the market power of firms, which is
the power to control market prices [41]. Therefore, approaches to assessing competition tend to
relate to quantifying market power and controlling prices. In the studies examined, several
measures are used to quantify and operationalize market competition, none of which appear to be
solely sufficient to describe the competitive conditions in markets.
2.1.2.1 Lerner index
The Lerner index, also known as the price-cost margin, is a measure of market power that relates
the price set by a firm to its cost [42]. The index is a ratio of the price margin to price and ranges
from 0 to 1. A value of 1 indicates high market power and low competition, and a value of 0
indicates no market power and a perfectly competitive market. This index is used in a prominent
study by Gowrisankaran et al (2013) which estimates a bargaining model to evaluate the effects of
hospital mergers [43]. Using this model requires knowing the marginal costs, which are not always
trivial [42]. While marginal costs may not always be known in real markets, it is possible to track
exact marginal costs in simulated markets. Therefore, this measure of competition and market
power can be useful in examining competition in simulated markets.
2.1.2.2 Herfindahl Hirschman Index
The Herfindahl Hirschman Index (HHI) is the sum of squares of each individual firms’ market
shares in the market. The HHI is inversely related to the degree of competition, so a lower HHI
indicates a more competitive market. Markets with high HHI values are concentrated markets;
meaning market shares are concentrated at fewer firms. This index is used by the DOJ and FTC
when evaluating mergers and acquisitions [40]. HHI’s measure of market concentration is
prevalent in literature relating market structure, competition and market power to performance and
market efficiency [28]. Some studies, however, emphasize the difficulty of geographically
defining markets and uniformly defining health products. For example, Baker (2001) expresses
14
concern with obtaining the necessary data and proper definition of markets and products when
using HHI after reviewing studies that utilize the HHI as a measure of the degree of competition
[44]. Since the HHI does not capture the number of firms, it is difficult to compare HHIs across
markets with different structures, but would be useful to compare HHI changes over time.
2.1.2.3 Market Structure
Market structure can be thought of as an indicator of the degree of competition [40]. It is defined
as the number of firms that sell the same product or service. It is often not sufficient to examine
the number of competitors in a market to evaluate the market power possessed and ability to raise
prices in a market, mainly because the market concentrations are not captured. Therefore, market
structure is often used along with the HHI to infer the competitive status of a market in antitrust
analyses [41].
2.1.2.4 Willingness to Pay
Willingness to pay (WTP) is a measure that was used to evaluate market power and competition
in several recent studies. WTP is a measure of social value of a product or a service, which is
represented by how much consumers are willing to pay. Estimates can be derived from consumer
choice patterns and behavior data. This approach is amenable to the heterogeneity in preferences
and indirect costs (i.e. travel or switching costs) when evaluating market competition. This
heterogeneity is not typically captured using traditional approaches that focus only on prices,
participants, or concentration [45]. A Study by Cory Capps et al (2003) estimates aggregate
consumer WTP as a measure of hospital value in an insurance network, and is therefore used as a
measure of market power [46]. In their model, a supplier, or a hospital, with a high WTP secures
higher prices from insurance plans, which is consistent with previous measures of competition and
market power.
2.1.3. Health Market Industrial Organization
This subsection provides a review of literature on agents involved in the health care market and
literature on competitive impact on their behavior and overall system performance and costs.
2.1.3.1 Providers
The hospital and physician sectors are among the largest industries in the US economy [4]. Along
with its growing importance, the hospital sector has been progressively more concentrated after an
upsurge in the number of mergers in the 1990s [47]. Average HHI for hospitals rose from 2,340 in
1987 to 3,161 in 2006 [28]. According to the recently updated DOJ-FTC Federal Merger
guidelines, markets with HHI>2,500 are classified as highly concentrated [48]. Another study also
finds resembling patterns in primary care, radiology, cardiology, and oncology [49]. Additionally,
more independent practitioners are becoming hospital employees. In 2000, less than 20% of
primary care physicians (PCPs) and 5% of specialists were working as hospital employees. In
2008, the proportion grew to over 30% and 15% respectively [50].
Provider competition and its effects on treatment prices, quality outcomes, and access have been
extensively studied in literature. The degree of integration of providers, horizontal or vertical, is
likely to shape bargaining leverage and the process of care [28]. It is observed that higher physician
and hospital prices are correlated with the degree of physician competition within this market [1],
[51]. White et al (2013) found that hospital prices for privately insured patients exceed Medicare
prices in markets with higher hospital concentration [52]. The same study also found that less
15
competitive markets also had more variable prices for privately insured plans [52]. In other words,
low provider competition is associated with increased variability and higher provider prices [51].
It has been hypothesized that the rise of managed care in the 1990s as a new delivery and payment
model has driven more provider consolidation and in turn increased spending, perhaps partly to
increase their ability to assume risks [53]. There has been some mixed evidence as to whether this
causal relationship exists [54], [55]. A similar question is currently being raised if ACOs
proliferation in private health care markets might lead to considerable changes in the competitive
landscape of private health care markets.
2.1.3.2 Health Care Insurance
Along with the growing concentrations of providers, insurers also appear to become growing more
concentrated [28]. Studying data from large employers depict an HHI that increased from 2,286 to
2,984 between 1998-2006, and the average number of plans dropped from 18.9 down to 9.6 over
the same time period [56]. The study also shows that more than three quarters of the health markets
studied had an increase of at least 100 points in the HHI between the three years from 2003 to
2006, and more than half of the markets had an HHI increase of more than 500 points [56]. Another
study on California finds increasingly concentrated health insurance markets, consistent with
findings from other studies [51]. Evidence on the relationship between the degree of insurer
competition and provider prices, premiums, and expenditures are mixed. When Schneider et al.
(2008) studied the correlation between health insurance concentrations and provider prices, there
was no significant link between increases in insurer concentration and decreases in provider prices
[51] or increases in premium growth [56]. Others, however, have found that higher insurance
market concentration is inversely related to hospital price health spending [1], [57], [58].
Currently, a dominant insurer’s inability to negotiate better prices might be partially explained by
a mandate enforcing network adequacy requirements established by the ACA which limits a plan’s
ability to exclude providers from their network [24]. Hence, increased insurer competition might
not equate to a reduction in health care spending, which may suggest that providers often have an
upper hand in controlling prices.
Furthermore, increasing health plan competition is likely to increase consumer surplus. With more
alterative plans to choose from, consumers can switch plans should their preferred providers get
excluded from a plan network. Therefore, increasing insurer competition potentially increases
provider negotiation power by “playing” insurers off one another [59]. This effect could offset the
aggregate potential reduction in premiums generated from facilitating health insurance
competition. The potentially countervailing effects exhibit the complexity of competition in
private health care markets under different conditions.
2.1.3.3 Employers and Consumers
The focus of this subsection is on the purchasers of coverage and patients. Employers typically act
as agents on behalf of their employees and purchase insurance plans. Large employers have
sufficient bargaining leverage to secure lower premiums for their employees, while small
employers and individual consumers are at a disadvantage [11]. In addition, employers might not
act as the best agents on behalf of their employees when choosing health care plans [46]. As part
of the ACA, state-based individuals and small businesses health insurance exchanges bridge this
disconnect by consolidating purchasers by facilitating individual choices and leveraging
purchasing power to control spending. Evidence from a panel dataset covering 10 million lives
16
indicated that marketplaces with employer contributions produce 13% welfare gains of premiums
compared to employer sponsored coverage, which often offers limited choices [32]. Theoretically,
implementing health insurance exchanges should increase value-based competition between health
plans by facilitating consumer choice. However, insurance exchanges led health care plans to
utilize different strategies and approaches to cost sharing and network breadth that might
compromise the value and level of competition [60]. Health insurance exchanges have been the
subject of heated debates on the future of the ACA primarily due to insufficient insurer
participation, premium hikes, and insurer exits [61], [62]. These exchanges vary widely across
states in terms of structure and effectiveness, prompting extensive data collection and analyses
since the future of the health law hinges on their success in controlling premiums and appealing to
insurers. Due to its dynamic nature, heterogeneity in composition and performance, and
embeddedness within the larger heath care delivery system, it is critical to study health insurance
exchanges.
Understanding previous theoretical and empirical efforts to investigate the impact and evolution
of competition in health care markets is critical to identifying research gaps and appealing areas
for research. The work discussed above highlights the trajectory of competition for different agents
within the health care market along with the relationships between various aspects of the market
and its performance. There remains plenty to be explored through the use of more comprehensive
approaches that capture the more sophisticated and adaptive nature of health care markets. Namely,
approaches that seize the heterogeneous, decentralized, and spatial aspects of health delivery and
how they relate to competition, novel delivery and payment model, and, most importantly, cost
and quality of care.
2.1.4. Health Care Markets
Researchers emphasize the convoluted nature of outcomes and processes within the health care
market. Economic analyses are often insightful in understanding the complex relationships
between certain aspects of the market and outcomes. Econometric studies typically employ
reductionist methods to study competition in health care markets wherein data is analyzed to
separate the effects of subcomponents [63]. However, reductionist approaches, such as SPC, often
treat the underlying system dynamics as a black box that generates observed outputs, given the
market structure and attributes. Studies mostly depend on macro measures of market structure, and
these studies overlook the rich differentiation and substitution patterns across consumers and plan
networks [46]. On the other hand, behavioral economic models often attempt to explain how
particular agents make decisions based on extracted data on agent behavior [64]. These static
approaches usually only encompass parts of the health care market and system, and rarely involve
multiple agents or sequential transactions consistently occurring in the health care market [28].
Potential lies in combining these approaches from various disciplines to recreate comprehensive
representations of health care markets aimed at investigating diverse competitive landscapes.
There appears to be a delicate balance of competition and market power in the various sides of the
market. Consolidating health care purchasing in the form of health insurance exchanges may be
central in facilitating access to care and improving the insurer risk pool. Also, studying market
power in health care markets is relative in terms of the balance between insurer and provider
market power, and shifts in the either levels of power appear to impact cost and quality of care. In
economic theory, markets that fall in competitive extremes, such as highly competitive markets
and monopolistic markets, tend to behave in a predictable manner. However, markets in-between
17
these competitive extremes, known as oligopolies in which there are a few firms competing, tend
to be less predictable and more representative of actual markets. Modeling health care markets as
complex adaptive system allows for the analysis of these “in-between” levels of competition while
accounting for the behavioral aspects of the agents, demand irregularities, and spatial nature of
health provision [65]. Integrating various disciplinary approaches highlights the prospective
insights that can be gained from a micro level, ground-up model of the health care market. The
model would leverage microeconomic concepts, and is explicitly defined using individual, agent-
level behavioral economic frameworks with an ability to learn and adapt in an ever-evolving health
care market. In recent developments relating to health policy in the US, supply and demand are
being structurally redefined and such simulation models can prove to be insightful in designing
legislation.
2.2. ACOs and Potential Market Effects
In 2011, Centers for Medicare and Medicaid Services (CMS) published regulation for its MSSP
that outlines the implementation of ACO’s with the shared saving payment model [66]. CMS’s
Medicare Shared Savings ACO program uses a shared savings model. The model specifies a
percentage of savings by an ACO shared between the provider and the payer. This incentivizes
collaboration among providers and reduces inefficiencies by aligning incentives between payers
and providers. Because Medicare covers 45 million beneficiaries, its regulations drive providers
to behave differently and have a sizable impact on the population and health care system trends.
Since CMS systematically determines its Medicare payment rates, competitive effects induced by
ACOs are less likely to impact prices for Medicare. However, since ACOs also function in private
health care markets and their presence is increasing, the impact of provider collaboration would
impact price negotiations for the privately insured.
The ACO model has been spreading and covering more individuals in recent years. Leavitt
Partners Center for Accountable Care Intelligence has been studying the growth and development
of ACOs since 2010 [67]. They maintain a database that is regularly updated from publicly
available information. As of January 2016, there were 838 active ACOs covering 28.3 million lives
covered, of which 17.2 million are from commercial payers [21]. Not only are ACOs growing in
numbers, but also in the size of their networks. More providers are participating in ACO networks.
For example, it is estimated that more than 18% of hospitals currently participate in some form of
ACO [68]. With changes in payment arrangements, increase in provider collaborations, and shifts
in risk bearing, there lies great potential for generating more value in health care delivery.
2.2.1. ACO Cost and Quality Improvements
ACOs leverage on claims that integrated delivery networks are capable of providing better
coordinated care at a lower cost [69]. The effectiveness of ACOs to slow down growth in health
care costs and improve outcomes for patients has always been a hotly debated issue. There has
been mixed evidence on the results of ACOs on quality, costs, and experiences. Although most
studies point to an improvement in patient outcomes, expenditures and utilization, the
sustainability of the improvement and the effects of integration are questionable. These doubts are
a result of the fact that shared savings in the MSSP ACO payment model are a function of the
reductions in yearly costs, and reductions in costs are theoretically expected to plateau with time.
ACOs improve delivery of care by creating financial incentives and laying a foundation for care
coordination. Integrated delivery networks, like ACOs, did not always yield efficiency gains in
18
studies that attempt to identify the advantages of care integration [70]. An analysis of 15 integrated
delivery networks (IDN) contributes to growing evidence that IDNs increase prices without
improving efficiency [69]. However, ACO contracts hold providers accountable to certain quality
standards by identifying key quality indicators, which should theoretically incentivize quality
improvement through care coordination and integration. In CMSs final rule on MSSP, there are 33
performance measures in 4 domains [71]. The measurement domains include patient/caregiver
experience, care coordination/patient safety, at risk population, and preventative care. Provider
reimbursements in that model are dependent on performance in these measures.
In the public health care market, a recent analysis by Nyweide et al. (2015) nudges the discussion
in favor of the effectiveness of Medicare ACOs [20]. The analysis is performed on one of the most
comprehensive datasets, which compares around 700,000 Pioneer ACO enrollees with more than
12,000,000 beneficiaries in the same market for 2 years. The growth in costs for the population
enrolled in Medicare’s Pioneer ACOs was smaller than the growth in the relative comparison
population, which equated to savings of nearly $35.62 per enrollee per month in 2012, and $11.18
in savings per enrollee per month [20]. These savings amount to nearly $280 million in 2012 and
$105 million in 2013. Inpatient services were impacted the most compared to other categories. In
this study, utilization and patient experience were significantly improved in ACO enrollees
compared with fee-for-service Medicare patients. The aforementioned study shows that there is an
evident reduction in expenditures and improvement in patient experiences for ACO enrollees.
However, it does not reject claims on the diminishing improvements for ACOs. Savings in the
second year were notably lower than those in the first year. Furthermore, these improvements are
established for pioneer ACOs, which are different compared to other Medicare ACOs. Given the
loose definition of an ACO, the generalizability of these improvements, particularly to the
structurally different private ACOs, is questionable.
2.2.2. Accountable Care Organizations and Anti-Competitive Concerns
The growth of numbers and sizes of ACOs is not limited to the Medicare setting, but extended to
private arrangements [26]. There are more than 287 commercial ACOs delivering care to
approximately 12.4 million lives [67]. California ranks first in commercial ACO arrangements as
of mid-2013. Market conditions, including provider and insurer market structure, influence the
pace of ACO development [72].
ACOs in the private sector often differ from public ACOs that are part of MSSP [73]. In fact,
several different payment models have been developed and are being tested in the private market
[74]. For example, private ACO plans have been giving patients financial incentives to seek care
within an ACO network, usually in the form of differential cost sharing [26]. Medicare does not
offer a similar incentive. In addition, many private ACO networks develop from narrow network
insurance plans. Narrow networks limit patient choice to a more confined network of providers in
exchange for lower premiums, and are increasing in popularity as a result of premium competition
in the market place. Other private ACO efforts are based on Preferred Provider Organization (PPO)
platforms [72]. This platform embraces patient choice and involves creating two tiers of providers:
(1) preferred provider networks, where obtaining care involves the low patient cost sharing, and
(2) out of networks care options that have higher cost sharing. ACO plans based on PPO platforms
introduce a third tier: ACO network with the lowest cost sharing, to incentivize patients to choose
such providers [68].
19
Private ACOs are usually based on the shared savings payment model, and sometimes include
shared risk and partial capitation [26]. There are no reported private ACOs that have fully capitated
payment models; all of which were based on a fee-for-service model [75]. In these private ACO
plans, there are usually various forms of shared risk payment models. Shared risk payments take
the forms of bonus payment for quality, partial population capitation, baseline revenue loss or
some combination of the various forms. For example, several ACOs offer providers 50% of the
savings generated by an ACO with the intention of moving towards a more capitated form of
payment model [26]. Because the ACO model is still developing, it is still early for a consensus
on an ideal payment arrangements that balances the risk, appeal, and growth of ACOs [26], [75].
Especially since some providers do not currently have the infrastructure to manage risk [75]. This
is particularly relevant since there are many quality improvement projects that are being
implemented, and it is important to make sure that the ACOs are perceived positively and are
associated with improvement [73].
2.2.3. ACO Concerns
As ACOs grow in size and numbers across the US health care system, concerns regarding their
consequences also grow. There are various questions that arise with the spread of ACOs, mainly
pertaining to the structure and competitive implications of ACOs.
2.2.3.1 Structure and Organization
It is important to note the various structural and organizational differences across ACOs.
Organizational experiments are commonplace with ACOs. Despite the fact that the ACO model is
relatively new, CMS has already tested 4 types of ACOs in the past: Pioneer ACO, Advanced
Payment ACO, Shared Savings Program, and Next Gen ACOs [17]. The models vary in risk
distribution between payer and provider, performance measures, and payment mechanisms. In
addition, ACOs in the private market employ various payment models which often share different
degrees of risk in different contracts and different reporting requirements [75]. Therefore, it is
difficult to define a uniform structure and organization that encompasses all ACOs.
ACOs are often sponsored by an active entity within the ACO that spearheads the initiative and
strategic planning. For example, physician groups sponsored more than 43% of all public and
private ACOs by the end of 2013. Hospitals sponsored nearly 40% of ACOs, while less than 10%
of all ACOs are sponsored by payers. Organizations, such as community organizations and practice
management companies sponsor the remaining ACOs [68]. Various sponsoring organizations
differ in structures and some question the influence of structures on ACO outcomes [75]. However,
evidence does not support this claim and suggests that ACO structure does not alter ACO outcomes
[24]. Instead, performance and results are dependent on other factors including strong
organizational leadership and market environment. Furthermore, there are several concerns with
ACO patient attribution, the process by which patients are counted as part of the ACO panel. From
a patient’s perspective, transitioning to an ACO is seamless. In fact, some patients might not be
aware that they are part of an ACO panel [24]. Patient attribution approaches have been evolving
in Medicare ACOs and have been approached differently in private insurance [26], [76]. It is
evident that organizational structures of ACOs vary widely across the different accountable care
arrangements, sponsoring entity, and designs.
20
2.2.3.2 Competitive Implications
Some researchers have expressed deep concerns in the resulting shifts in market share
concentration and recommend implementing price control policies to avoid the associated price
impact [22]. ACOs may create more market power, which may be exploited by providers to
increase prices without generating value or improving care [24], [77]. Key factors that determine
the extent to which ACOs can be used anticompetitively are market condition and competitive
landscape. It has been shown that the effects of integration on prices and bargaining is worse in
less competitive markets [70]. CMS has requested that antitrust agencies conduct a study to assess
the effects of MSSP ACOs on prices and quality in the private market [26]. The effects of market
power shifts vary depending on the market structure and level of consolidation. The questions that
arise are related to investigating how much market conditions determine the magnitude of possible
anticompetitive implications from ACO formations? Can ACO formations be used to increase
bargaining leverage for providers without adding value? Under what conditions should ACO
formations be endorsed or limited?
An antitrust statement released by the DOJ and FTC explains the review process required to join
Medicare Shared Savings ACOs [27]. This antitrust statement is a testament to the need for
protection against the risk of forming ACOs that lead to anti-competitive behavior. The statement
recognizes that not all ACOs may benefit consumers, instead, some may reduce competition, raise
prices, and lower the quality of care. An antitrust safety zone is designated in the DOJ and FTC
statement for providers that qualify for joining the ACO without review. To determine whether an
ACO falls in the safety zone, ACOs market share in 3 categories of services in its Primary Service
Area (PSA) is assessed. Independent ACO participants must have a market share of 30% or less
in each of the 3 service categories to qualify for the antitrust safety zone. There are exceptions for
rural areas along with several other rules and regulations to ensure the preservation of competition
in a market when ACOs are formed. It would be immensely valuable to evaluate these guidelines
and their effectiveness at preventing anticompetitive behavior or sustaining a competitive health
care market.
Competitive concerns are commonplace in ACO debates, and the antitrust statement is a concrete
attempt at minimizing anticompetitive behavior. The effectiveness of these measures and policies
has not been proven. In addition, the cutoffs chosen for the antitrust safety zone, rural exceptions,
and dominant participation limitations might appear to be empirically determined. It is possible
that some beneficial ACO formations will be rejected, while some other detrimental formations
are accepted. In heterogeneous markets, more context-specific, systematically determined cutoffs
would be more appropriate. To be able to understand the effectiveness of different anticompetitive
ACO policies, there needs to be a systematic way to evaluate market outcomes that account for
the complexity and evolution of health care markets.
2.3. Complex Adaptive Systems and Health Care Market Competition
Capturing social phenomena is very difficult using purely mathematical models. Studying large
numbers of agents in embedded, iterative systems quickly becomes intractable to analyze and
model mathematically. Traditional approaches have undeniable contributions to all fields of
science, but often fail to analyze complex emergent behavior and self-organization [78]. It is in
systems like these that complex systems thinking is most promising. Applying complex systems
thinking involves employing transdisciplinary sets of theories, tools, and methods devised in
computer science, systems engineering, and cognitive and social sciences to study a particular
21
problem [65]. Typical CAS involve autonomous agents that interact in complex, often non-linear
ways based on a set of rules and objectives in an environment. Agents learn and adapt according
to available information and past experiences. In CAS, control is not centralized and agents usually
self-organize and emergent patterns appear. Many social and economic phenomena can be
modeled in such a manner. This approach has been utilized in numerous domains, from
evolutionary cellular biology, ecology, and social sciences, to economic markets and behavior
modeling. The main strength of this modeling approach is the agents’ ability to follow adaptive
strategies and the ability to capture emergent behavior using ground-up, micro-level assumptions
to lead to global, macro-level phenomena [78]. Emergent behaviors are defined as macroscopic
coherent regularities, like identifiable distributions, patterns, equilibrium, and so forth, that emerge
over time [79]. Studies of CAS have been successful at exploring emergent behaviors and
properties that are often surprising and difficult to foresee [80]. In applying the CAS approaches,
the perspective shifts from breaking down system-level behaviors and patterns to constructing
them as a method of explanation [81], [82]. Insights will help to precisely identify rules and
conditions that give rise to emergent competitive dynamics in health care markets. The objective
is to “grow” health care markets in silico to demonstrate that a certain set of conditions (e.g.,
market structure, behavioral rules, regulations) is sufficient to generate macro phenomena and
outcomes of interest (i.e. anti-competitive pricing, narrow network plans, etc…).
2.3.1. Complexity & Economics
Studying economies and markets as CAS is not a new idea. Researchers have advocated this
approach with notable contributions like three volumes titled The Economy as an Evolving
Complex System [83]–[85]. Others have ascribed its suitability for health care operations,
conditions, organizations, and public health [86]–[94]. Here, complex systems thinking is used to
overcome disciplinary bounds in studying competition in health markets and the provision of care.
The combination of inherent complexity in individuals’ health conditions, behaviors, choices and
the continuous evolution of the health market conditions, regulations, and trends are especially
challenging to analyze. The interplay among these different facets remains largely unexplored.
Because of the importance of incentives, outcomes, and prices in health care markets, there is great
value in inspecting novel approaches to dissecting emergent trends, self-organization, and
synergistic relationships.
To qualify competition in health care markets as CAS, distinguishing features of CAS are
examined with parallels to characteristics in health care markets. Features of CAS that lend
themselves well to studying challenges in health care markets have been compiled from several
prominent sources on the subject of complexity [82], [85], [95]–[97].
Repeated Interactions
In CAS, heterogeneous agents in similar and different classes interact with one another in parallel.
The actions of agents may be a function of anticipated actions of other agents and the current state
of the system. These interactions may act as bidirectional feedback mechanisms that influence the
behavior of agents and the system. The agents co-create the aggregate state of the system. The
interactions among consumers, insurers, and providers in health care markets fit this characteristic
of CAS. Consumers choose from insurers and providers in scheduled interactions (e.g., open
enrollment periods, scheduled visits) and randomly occurring interactions (e.g., ambulatory care).
22
Decentralization
Aside from regulations and assigned roles that mediate agent interactions, there are no global
controllers in CAS. Instead, control is achieved through objective-driven agents that may operate
under mechanisms of competition and cooperation. In private health care markets, price, quality,
benefits, and provider networks are determined endogenously, yet they are still governed by
different provisions and incentive schemes (e.g., medical loss ratio, pay-for-performance
payments).
Cross-Cutting Hierarchal Organizations
CAS can account for multiple levels of organizations. Each sub-element serves as a component
that makes up a larger element higher up in the organizational scale. Cross-component interactions
are commonplace in these systems. In health care markets, provider organizations, which jointly
bargain, contract, and share infrastructure, and consist of individual providers that interact directly
with individual patients. Individual patients may exist in households, which collectively make
insurance and care decisions. These abstractions would represent the core of a CAS to study health
care markets.
Continual Adaptation
Recurring decisions carried out by agents are constantly enhanced with accrued experiences and
knowledge. These continuous reconsiderations mean agents are constantly adapting and the system
is evolving. Perpetually revised decisions in health care markets may mean that the structure,
organization, and, consequently, market power are continuously shifting. The dynamics that make
studying competition in health care markets especially difficult are central to CAS approaches.
Out-of-Equilibrium Dynamics
Due to persistent adaptations, these CAS typically operate sub-optimally. This is consistent with
the suboptimality of real economies that may operate far from a global equilibrium, as is the case
in health care markets. The shifts in bargaining power, information discrepancies, and innovation
mean that health care markets may have dynamic equilibriums. Even if health care markets, under
static conditions, are assumed to converge to some form of static equilibrium, CAS enables us to
investigate rates of convergence, variation, and sensitivity to initial conditions.
Emergence
In health care markets, insurers may continually develop new strategies (e.g., narrow network
plans) and providers may collaborate (e.g., accountable care organizations, or ACOs) to provide
new products and options for payers and patients. Studying the effects of these adaptations and
possible implications on competition in health care markets is challenging given the dynamic
nature of these adaptations. In employing the CAS approach, it is possible to study the initial rules
and conditions under which these collective agent behaviors emerge and sustain. Such insights
will be crucial to understanding possible legislation to limit or perpetuate such provider and plan
behaviors.
2.3.2. Agent-Based Modeling
CAS are often modeled and simulated using Agent-Based Modeling. ABM is a relatively new,
comprehensive method of modeling. It incorporates various types of modeling such as Discrete
Event and Systems Dynamics simulations. ABM is a more versatile approach that involves
designing autonomous, decision-making entities, called agents, with sets of behavioral rules and
23
objectives. The presence of other agents and environmental settings alter the decisions and
interactions within the system, which leads to system level changes and behaviors. Compared to
Discrete Event simulation, where entities are passive, simple, and limited in capabilities (i.e.
unable to learn or strategize), ABM allows for more intelligent, proactive, and adaptive agents.
This enables ABM to capture cascading effects that form from seemingly minute local interactions
[79].
To understand how simulation modeling fits in the scientific method, Axelrod (1997) explains that
there are two standard methods of doing science: induction and deduction [80]. Induction relates
to the discovery of patterns from empirical data and deduction involves designating rules and
axioms and proving resulting outcomes from those rules and axioms. ABM is a hybrid method that
builds on explicit assumptions, like deduction, and then generates data that can be analyzed
inductively. The goal of this model is to inspect trends and patterns in the generated data. ABM is
particularly useful when agents in the model use adaptive strategies, rather than optimization
strategies [80]. ABM may be viewed as “generative” science which bridges the micro-macro
divide [78], [95]. In applying such approaches, the perspective shifts from breaking down system-
level behaviors and patterns to constructing them as a method of explanation [81]. Insights will
help precisely identify rules and conditions that give rise to emergent competitive dynamics in
health care markets.
2.3.3. Uses of ABM
ABM applications span over a multitude of domains. In this literature review, the focus is on
applications in the field of economics and health care. Agent-Based Computational Economics
(ACE) is a strand of economics that focuses on applications of ABM to model economies as
evolving systems of complex adapting agents [29]. Tesfatsion (2002) describes decentralized
market economies as CAS made up a of large number of adaptive agents active in multiple
concurrent local interactions [29]. The local interactions then yield to the macroeconomic
symmetries such as behavioral norms, which then influence local interactions, resulting in a
complex dynamic system of convoluted causal chains linking distinct behaviors, interaction grids
and social welfare outcomes. Thus, the utilization of such approaches is gaining acceptance and
popularity in public policy and economics.
ABM has been used in validating existing theoretical economic equilibriums in a study of Austrian
market competition [98]. The authors claim that ABM developed for the study challenge general
equilibrium frameworks in policy models. A study by Bunn and Oliveira (2003) used ABM to
simulate the coordination game for evaluating potential abuses of market power in electricity
markets [99]. The study addresses a Competition Commission Inquiry and was a successful
contribution to research on market design, competitive policy issues, and market power. There are
follow-up studies on competition in the electricity market, all of which demonstrate the value in
using ABM to understand complex market dynamics and emergent properties [100], [101]. ABM
has also been used in various other fields within economics, such as auctions [102], supply chain
management [103], labor markets [104], financial markets [105], and product competition and
information regimes [106], [107].
The prominence of ABM is not limited to economic studies. Health care, one of the most complex
domains, has been a growing area for applications for various types of simulations [108]. The
Society for Simulation in Health Care highly emphasizes the current value and promising future
24
of simulation in health care [109]. ABM, with its versatility, is likely to overshadow the growth of
traditional simulation techniques in complex health care environments. In fact, ABM has already
been gaining momentum in health policy. Studies have employed the complex systems tools for
challenges like redesigning health care [86], [87], [110], and obesity epidemic [96] along with
other epidemiological challenges [89], [111].
This dissertation is an extension of a study that uses ABM to investigate the appeal of MSSP ACO
for providers and analyzes system behavior for different shared savings rates by Liu and Wu (2014)
[112]. In this model, providers can make decisions as to whether they choose to participate in an
ACO, given MSSP shared savings rates, demographic attributes of the population, and social
interaction. This model was insightful in understanding the effects of different shared savings
levels on different types of providers and relating it to overall system performance. Liu and Wu
(2014) modeled a single payer system, with passive agents, and no competitive markets. The
developed ABM identified key aspects of designing a shared savings ACO model that motivates
providers to maximize financial and quality outcomes. An extension ABM enabled the Medicare
patients in said model to choose providers to investigate the effect of patient choice on provider
ACO participation, expenditures, and quality [113]. Illustrative results show that patient choice is
associated with a reduction of yearly congestive heart failure patient costs while maintaining 30-
day readmission and mortality rates. The insights from ABM of abstractions of health care markets
exemplify flexibility of ABM and may be a blueprint for designing ABM with competitive or
cooperative provider agents in a market with private payers and actively choosing patients.
2.3.4. ABM for understanding the competitive implications of ACOs
Health care markets can be thought of as CAS composed of intelligent, autonomous agents whose
parallel, local interactions determine macroeconomic regularities and overall system performance,
which then feed back into determining local interactions. Due to the dynamic nature of health care
markets and the evolution driven by innovative agents and developing policy, there is a great
benefit in being able to deconstruct the health market’s black box to instigate more beneficial
competition and avert unintended consequences. Using classical economical approaches,
outcomes are difficult to predict in markets where prices and decisions are determined
endogenously [28]. Figure 2-1 below depicts a reductionist approach to study health policy and
market competition. Health care markets are studied as black boxes whose characteristics, such as
structure, concentration, and demographics, influence the effectiveness of policy implementations.
25
Figure 2-1. An illustration of health care markets studied using a reductionist approach.
ABM can be used to analyze the “in-between” levels and mechanisms of competition to understand
the effects of ACO formations given the market conditions. The competitive concern of ACOs in
the private health care market is a well suitable problem in which this modeling technique can be
useful. The complexity of the agent interactions and market characteristics in abstractions of health
market competition can be captured using ABM. Antitrust agencies need explanatory models and
tools to inform and evaluate policies on cooperating providers to prevent anticompetitive behavior
that account for evolving market conditions, demographic attributes, and adapting decision
makers. As shown in this literature review, there currently exists sufficient economic, behavioral,
and structural literature on the health care market for this research to be feasible. The complexity
of markets studied and the approaches it synthesizes places this problem on the current frontier of
computational difficulty. The increase in computational capacity and access to computational
power made the use of such modeling technique feasible [114]. Figure 2-2 illustrates a more
comprehensive view of health care markets as a CAS to study the effects of policy and accountable
care models. The influence of policies on all interrelated levels of the CAS is highlighted. Complex
systems thinking is amicable to using various disciplines and mechanisms to model the complex
influences of policy on outcomes.
26
Figure 2-2. An illustration of the natural components of an abstraction of health care markets when
utilizing a CAS approach.
3. Structural Framework
The aim of this research is to develop an ABM to study competitive dynamics in private health
care markets under different conditions and perform a case study on possible anti-competitive
implications of ACOs. The structural framework constitutes the skeletal layout around which the
ABM is built. It is intended to conceptualize the theories and mechanics abstracted to model private
health care markets and competition.
Simulation models are typically meaningful abstractions of actual systems that are meant to
capture phenomena of interests. Hence, there exists a tradeoff between a model’s level of detail
and its representativeness or generalizability. A model intended to generically study competition
in health care markets needs to be sufficiently complex to thoroughly capture interactions and rules
in the private health care market, yet sufficiently simple to compute, interpret, and code. This
chapter designates the nature of components, organization, and rules to establish a functional
model on the competitive dynamics in heterogeneous health care markets and possible
implications of ACO formations.
3.1. Health Care Market Model
Agent-based simulation models are composed of populations of autonomous agents that interact
locally in a dynamic environment. This model focuses on studying non-group insurance segment
of the private health care market. This segment has been growing after instituting the individual
mandate and establishing health insurance marketplaces in the ACA (also known as Health
Insurance Exchanges). Consumers purchasing non-group insurance are more sensitive to premium
changes and are therefore likely to be impacted directly by changes in prices and regulations.
Individuals, providers, and plans are to interact in the market characterized by managed
competition. In this market, insurance exchanges list insurances’ semi-standardized plans and
provider government subsidies for eligible individuals. Individuals purchase plans listed in their
exchange during a designated open enrollment period. Plans compete for enrollment on premiums,
benefits, and provider networks offered in the market place. All the plans’ differentiating aspects;
27
premiums, cost sharing levels, and provider networks, are determined after agreements have been
reached with providers. Individuals then seek care at their preferred provider within a network of
providers based on their choice of plan network and their preferences. This general overview of
agents, behaviors, and mechanisms is detailed in the subsections below.
3.1.1. Agent Structures & Components
Agents active in the described ABM of health care markets are individuals, providers, and plans.
Each type of agent engages in various types of personal and organizational interactions with one-
another in a manner inspired by the current private health care market. The degree of granularity
of agent abstractions is chosen such that key aspects are present to reconstruct the competitive
facets of the health care market while minimizing the irrelevant aspects of the organizational agent
structures. For key modeling decisions, theoretical, empirical, or practical justifications are
provided.
3.1.1.1 Individuals
Attention has been directed to ensure that the system’s main objective is to serve the individuals’
needs [18]. In this model, health care needs and decisions are modeled at the individual level.
Individuals are consumers when they choose a plan from the marketplace and a patient when they
choose and receive care from a provider. Modeling at this level is justified by the importance of
these individual level decisions and their implications [115]. These decisions are a key aspect of
the complexities in health care markets since most decisions are made prior to having knowledge
of actual needs and their implications carry over a longer time [116].
An individual agent is generated with health and demographic attributes that dictate their care
preferences, health care needs, and decisions. These attributes have been shown to be key
predictors of individuals’ behaviors [115], [117]–[121]. Consumers purchase insurance through
the marketplace, which lists all health care plans, prices, networks, quality of networks, and cost
sharing levels. Preferences of the individual, their geographic location, and perceived medical
needs determine their willingness-to-pay for single providers and provider networks. Health care
demands and needs are modeled at the consumer level to capture individual choice, substitution
effects, and information discrepancies [46], [116], [122]. Therefore, the two decisions individuals
repetitively make are (1) choosing a health insurance plan and (2) choosing a provider for every
episode of care. An individual’s decisions are a function of their location, providers’ locations,
insurance plan network, premiums, and out-of-pocket expenses. The focus of this model is on
individual choice of providers and insurers. There is no “do nothing” option in which an individual
may choose not to be covered or not to receive needed care. This omission is made to maintain
model simplicity and may be incorporated in future upgraded versions of the model if the decision
structure is known.
Health care needs and delivery are modeled at the episode-of-care level. The aim is to have generic
patients with diverse health needs at different levels of care. This level of analysis has been
examined by the RAND Health Insurance Experiment (HIE), a landmark study on health care
utilization under different insurance coverage [123]. While RAND HIE may be outdated, the
structure of their models to simulate health care utilization is ideal for this model’s purpose and
the study remains the largest experimental study on utilization patterns under different plan
designs. Utilization and demand for care can be generated from the RAND HIE in the form of
hospitalizations, acute, chronic, or wellness episodes [124]. Abstracting health care needs at the
28
episode level and having classifications allow for modeling wide-ranging needs and conditions
while retaining a simple structure. The episode unit of analysis is essential in the context of
continuity of care [115]. Therefore, for the intended modeling objectives, the individual and
episode levels of abstractions are ideal for capturing decisions and health needs.
3.1.1.2 Providers
Although there are various types of highly specialized care providers, this model focuses on a
generic representation of health care facilities to recreate the competitive nature of the market and
ACO formations. Given the classifications of episodes discussed in the individual subsections, it
is viable to have providers that are only be able to delivery certain types of episodes to mimic the
different types of facilities. However, this instance of the model will focus on hospitals that can
deliver all 4 types of episodes studied in RAND HIE.
Since the goal of this model is to examine the competitive implications of ACOs, and not mergers
or other forms of provider integration, hospitals in the simulated market are modeled as
independent providers. Hospitals are independent entities that bargain with insurers and have
decision-making abilities regarding quality investments. Effectively, these are the two decisions
providers repeatedly perform; (1) choosing investment levels and (2) determining service prices.
The two decisions performed by providers are a function of the competitive environment, which
influence prospective gains and losses associated with each decision. In a modeled market, the
distance, number, and quality of alternative providers determine the effect of increasing prices or
quality investments for each provider. Changes in prices and quality levels will affect insurer prices
and patient behavior. Additionally, changes in prices and quality would affect insurer contracting
and premiums. The tensions at play in these decisions are conceptualized in the next section and
the relationships are operationalized in the methods chapter.
3.1.1.3 Plans
Health plans are the products offered by insurers through the insurance marketplace for
individuals. Health plans connect patients to providers by offering a network of providers to their
enrollees to choose from when care is needed. This structure closely resembles that of a PPO.
Studying this type of plan network structure is informative because PPO networks serve as a good
foundation for ACO formations [72], [75], are dynamic compared to other network structures, and
have implications on care continuity [125]. Plans reimburse providers participating in their
network based on utilization and services rendered following a fee-for-service model in a PPO.
Plans choose premiums and cost sharing levels each year. This decision influences the number
subscribers, insurer negotiation leverage, and profitability.
The overview of structures and roles highlight the intricate nature of competition in health care
markets. Decisions with implications that feed back into future decisions, a key feature of CAS,
are apparent in these abstractions. Next, the environment in which agents and their interactions are
embedded will be outlined with the theoretical mechanisms at play.
3.1.2. Conditions and Environment of the Private Health Care Market
After delineating the organization of agents in the modeled health care market, this section
examines the nature of interactions between agents in the market that give rise to the economic
features. The agents interact in a market characterized by Managed Competition, a framework
partially implemented by the ACA [10], [126] and outlined by Alain Enthoven [11], [12]. Managed
29
Competition is founded on semi-standardized insurance products, informed and consolidated
demand, and increased accountability. The structure of the market is graphically represented below
in Figure 2-3Error! Reference source not found..
Figure 2-3. Structural framework of the ABM adapted from Managed Competition market framework.
Individuals buy coverage on the insurance marketplace, a virtual tool that informs individuals of
the different insurance products offered. Insurance marketplaces are assumed to provide perfect
information on premiums, provider networks, and cost sharing levels for each plan. In this
marketplace, plans compete on premiums, out-of-pocket costs, and networks. Premiums and
networks offered by plans are chosen based on whether agreements were reached with health care
providers in this year. Only one plan type is modeled; the silver plan which covers approximately
70% of the cost of care [13]. The focus is on silver tiered plans because they tend to vary across
markets and are a key in determining government subsidies in the ACA [10], [35], [127]. This
simplifies the insurer and consumer decisions in the model, but can be explored in future versions
of the model. If a provider and a plan disagree on prices, then the provider is excluded from the
plan’s network. This means plan enrollees can no longer receive care at that provider. While PPO
plans typically have tiered networks where out-of-network providers have higher out-of-pocket
expenses, tiered networks are omitted in this model. This omission might sacrifice some key
aspects of choice and competition within health care markets. However, such inclusions would
add substantial complexity to patient, insurer, and provider decisions, and would best be studied
in future iterations of the model.
30
Figure 2-4. Representation of an insurance rating area with a defined market structure for providers and
plans.
The simulated market represents one hypothetical Insurance Rating Area (IRA). IRAs have been
instituted as part of the ACA to limit the factors on which consumers can be charged differently
[128]. In each rating area, insurer premiums for each individual can only vary with age (3:1 ratio)
and tobacco use (1.5:1 ratio) [128]. A hypothetical IRA market structure is defined by the number
of providers, plans, and population, as shown in Figure 2-4. For modeling purposes, each market
is assumed to be a closed system, with no external influence. While actual markets competition,
prices, and performance are influenced by public insurance programs (e.g. Medicare, Medicaid)
and HMO presence (e.g. Kaiser Permanente) [59], this significantly enhances model trackability
and interpretability.
The simulated market is subject to overarching rules and practices in the ACA’s implementation
of Managed Competition. For example, consumers in the market choose their health plan from the
marketplace during a designated period in each year, usually called the open enrollment period.
Consumers are only allowed to switch plans during this period should they decide to [10].
Additionally, individual premiums for the same plan are only allowed to vary within the ratios
permitted by the ACA. This lays out the structural foundation on which the ABM of competition
in health care markets is laid upon.
3.2. Cooperation and Competition in the Health Care Market Model
The agents and structures interact with and influence one another in a variety of ways. This section
outlines the theoretical aspects of competition and cooperation present in the ABM. Several fronts
exist where agents would compete for profit and others that exhibit agent cooperation.
3.2.1. Competition
Classic microeconomic models help explain how competition relies on market structure. Cournot
and Bertrand competition models are examples of the few economic models that describe
oligopolies, markets where there are a few dominant sellers [129]. Cournot competition relates to
competition between sellers by varying the quantity produced of a homogeneous good or service;
whereas Bertrand Competition models price competition where sellers compete on the price of a
homogeneous product. Both classical models of competition are simplistic yet theoretically
insightful in understanding competition as an evolving concept [130].
The health care market is considered a multilateral market with multiple players sometimes acting
simultaneously as buyers and sellers. Mainly, individuals demand care, which drives the demand
31
for plans and providers. Meanwhile, providers are sellers to plans and patients, where plans are
buyers to providers and sellers to individuals.
3.2.1.1 Providers
Providers compete against each other on two fronts: patients and health plan contracts. There is
strong interdependence between patient demand and plan demand for providers. This
interdependence is driven by plans’ ability to steer patients towards providers and by patients’
preferences influencing the value a provider adds to the plan’s network. Therefore, providers
market themselves to individuals, competing on quality mostly since plans shield patients from
drastic price variation, and to plans, by competing on price and quality. Typically, patients are only
financially responsible for the deductible or copayment when receiving care, a fixed amount or
percentage of the cost of care. The percentage of copayment and deductibles depends on plan
benefits and whether a provider is within the PPO network of the plan.
Competition between providers is based on price and quality in a geographically defined area. This
competition can be described as stylized Bertrand competition [131]. It is modified due to the
difference in quality of their provided services, rendering the service partially heterogeneous.
Traveling costs associated with receiving care varies for each provider from a patient perspective.
Price and quality elasticity of demand differ with the number of alternatives available to plans and
individuals. In highly competitive markets, choices are available and patients and plans are
expected to be more responsive to changes in provider price and quality. In private health care
markets with market determined prices, there is no general theoretical framework to relate prices,
quality, and competition [23], [63].
Theoretically, in highly competitive markets, providers compete to deliver value for patients to
secure demand by creating strong patient preference and contracts with plans. Providers that
deliver less value often have weaker patient preferences and are more likely to be excluded from
networks [74]. However, given the locality of health care provision, it is unlikely that perfect
competition can be fully achieved. Often, choice of providers is limited and competition might not
always be a motivating driver for providers to deliver value. Provider quality attracts patients and
plans, however, quality investments increase costs and in turn reduce profits. This exhibits the
tradeoff for providers to invest in improving quality. This tradeoff is larger in less competitive
markets where the gains in demand are less likely to offset the investment costs because patient
preferences are less sensitive to quality when there are fewer provider choices.
Since competition relates to how the behavior of one firm is influenced by the existence of another
firm, the changes in behaviors for providers will be captured within a latent variable effort. The
effort construct captures the amount of resources invested by a provider to attain a particular level
of quality. Practically, higher provider effort could mean hiring additional staff and clinicians,
investing in IT systems, or other quality investments. The latent behavioral effort variable is
influenced by the market conditions, performance, and observable previous behaviors and
outcomes in the system. Behavioral changes that are induced by competition will be accounted for
by observing this variable. When assessing possible anti-competitive concerns, the relationship
between the effort variable and price will be key. If higher prices are associated with lower
provider levels in particular market configuration, this may affirm anti-competitive concern for
comparable market.
32
3.2.1.2 Plans
Option Demand Markets (ODM), or markets where access to a network of suppliers is sold by
intermediaries to consumers who are uncertain about their needs, have been studied by Capps et
al. (2003) [46]. When choosing health care plans, individuals usually choose an insurance plan
before knowing the type or intensity of care they will need with certainty. In their study, Capps et
al (2003), develop and validate a market power index for suppliers, or providers in this model, in
ODM [46]. The main market power index is based on aggregate WTP derived from the properties
of logit demand of providers demand data. More on using this method to estimate WTP will be
presented in the Methods chapter (Chapter 4). Essentially, this study examines the competitive
tension between providers and plans in the health care market, given the unique form of demand
for intermediary plans and providers.
In ODM, intermediary plans typically capitalize on their expertise and purchasing scale to identify
superior providers, obtain contract terms, and spread risks more efficiently than individuals
shopping for and obtaining care on their own. The objective of the plan is to deliver a broader
network of providers at a lower premium. There is an evident tradeoff between premiums and
network breadth. Broader networks offer more choice for the enrollees, but require additional
contracts increasing administrative costs. Broader networks increase risk variability and require
robust financial support, making this option viable mostly for larger insurers [46].
An example of how insurer competition resulted in different strategies in plans was evident in
narrow network plans proliferation. As insurance exchanges increase plan competition, narrow
network plans gained popularity. In 2014, 75% of health plans in the California exchange had
narrow networks defined as covering 25% or less of the providers in a given area [132]. Narrow
network plans aim to reduce premiums by narrowing down provider options for enrollees [133].
Issues regarding the reduction in access and cost implications of narrow network plans have arisen.
Narrow network plans can be thought of as an emergent trend driven by plan competition that may
be recreated using ABM.
3.2.2. Cooperation
In this model, there are various modes of cooperation between and across agent types. Namely,
the two key modes of cooperation are (1) price agreements between providers and plans and (2)
ACO formation between providers. Cooperation in adaptive systems has been extensively studied
in the context of game theory [80], [134]–[138]. In economic game theory terms, cooperation has
risks and rewards. In the case of an ACO, for instance, providers would be penalized for forming
ACOs with low quality, inefficient providers. On the other hand, providers would be rewarded for
collaborating to provide better care coordination, which improves quality and increases savings,
along with having higher bargaining leverage in an ACO. It is also possible to have providers
participate in an ACO and reap the benefits of securing demand and bonuses without increasing
effort. This exhibits what is known as the free-rider problem in economics. A game theoretic
approach to equilibriums would be beneficial in modeling providers’ choices related to
cooperation.
3.2.2.1 Provider-Insurer Price Agreements
Price agreements between providers and plans constitute cross-agent cooperation where both
agents would mutually benefit from enrollment and delivery of care. The price agreed upon by
both agents is the results of a bargaining process between each provider-plan pair. In the bargaining
33
process, a price is determined that maximizes each agents’ expected profits in the current market
conditions and balancing the tension between increasing prices and reductions in demand for care
[64]. In other words, increasing care prices would increase provider margins, but might lead some
insurers to exclude the provider, reducing the total volume of care provided. Figure 2-6 in the next
section depicts a causal diagram describing the pathways in which provider price and effort
decisions may alter health expenditures and outcomes. Factors that impact equilibrium prices
would be provider WTP, market share, alternatives, quality, and current market norms [46], [139].
The bargaining process results in a price offer from the provider to the insurer and then the insurer
decides whether to accept the price offer and include the provider in its network. Each provider-
insurer pair engage in bargaining and the result is a configuration of provider network for each
insurer and a set of agreed upon prices for care. Figure 2-5 depicts a hypothetical configuration of
provider network for each insurance plan. In this illustrative market, there are 3 plans and 4
hospitals, 2 of which participate in an ACO. In the figure, the blue lines represent agreements
between providers and insurers, while the orange line represents an ACO contract between the 2
participating hospitals and insurers.
Figure 2-5. Illustration of a potential network configuration of agreements in a market a market with 3
insurance plans and 4 hospitals, 2 of which form an ACO.
As mentioned earlier, bargaining and price agreements between providers and insurers are
influenced by numerous factors, many of which are complex and highly embedded. For instance,
consider the impact of the WTP component, which reflects the perceived social value of a provider.
The WTP would be a function of the distance between a patient and the provider, the density of
patients around a provider, or the number and distance of alternative providers. The market share,
on the other hand, is the actual proportion of care delivered at a provider, which is influenced by
the WTP, provider prices, and alternative providers. Profitability is directly computed from the
34
market share and agreed price, while prices are impacted by WTP and prospective market shares.
The embeddedness is evident above and the perpetual feedbacks in market evolution is to be
accounted for when modeling price agreements.
The insurer-provider bargaining has been studied extensively in economic literature [46], [139]–
[141]. A stylized theoretical bargaining model developed by Ho and Lee (2013) best suits this
model [59]. In Ho and Lee’s model of bilateral negotiations between insurers and hospitals,
hospital price is determined then insurer premiums are set. The structure of this bargaining model,
which was used to examined Managed Care Organizations, fits the purpose of the proposed ABM.
This bargaining model assumes a sequence of events that is amicable to ABM and uses inputs that
can be easily collected from other agents (i.e. aggregate WTP, utilization). Therefore, the
bargaining model utilized in this ABM is based on the theoretical bargaining model developed by
Ho & Lee (2013). The Nash product solution is used to obtain the equilibrium price at which
hospitals and insurers maximize their corresponding disagreement tradeoffs. The structure of this
bargaining model, as outlined by Ho and Lee (2013) and adapted to his model, satisfy the axioms
needed to allow for the use of the Nash product solution [142]. After determining the hospital
prices, insures proceed to form networks based on a utility and profit maximizing two-stage
optimization problem. Mathematical details relating to operationalizing the bargaining and
network formation modules will be provided in the methods chapter.
3.2.2.2 Cooperation in ACOs
This model is to be used to examine possible effects of ACO formation to illustrate potential uses
hypothetical markets in silico. Modeling the effect illustrative ACO formations would demonstrate
the utility of this model from the perspective of the anti-trust agencies when reviewing ACO
proposals. Thus, ACO formations are induced in simulated markets, then compared to an identical
market with no ACO formation. Comparing the ACO markets to non-ACO markets would separate
the impact an ACO may have on competition and outcomes of health care markets. This subsection
outlines the incorporation of ACOs in the ABM.
As shown in the literature review, the term ACO is very flexible and may incorporate many
different provider integrative arrangements. In this model, an ACO is defined as two or more
providers that collaborate and are held accountable financially for health outcomes and costs of an
enrolled patient population. These providers receive fee-for-service payments for services
provided, shared risk payments depending on total patient costs [66], [75]. ACO providers and
plans share ACO savings or additional costs at the end of the year. Participating providers are
handled as a single entity in the bargaining and network formation process. ACO providers
achieve better health outcomes by providing continuous care and collaborating with one another.
Therefore, there is an increase in service costs from a provider perspective which is associated
with the additional resources for care continuity and collaboration, modeled here as a fractional
increase in the effort variable [25], [143]. The additional resources may be needed for investing in
quality training or health information technology (IT) systems [143]. ACO formations, as defined
here, would impact the aggregate health expenditures via several different pathways. First, ACO
formations enable providers to bargain collectively with insurers. This means that negotiated
service prices are likely to be higher for providers in an ACO arrangement. Consequently,
expenditures would increase because of this effect. However, the quality improvements in care
delivery and coordination might reduce total utilization through reduction in hospitalizations and
readmissions [19]. Improvements in outcomes can be achieved through increased training,
35
investment in disease registries and health IT. Improved care continuity among providers has been
shown to reduce utilization and expenditures when sharing patients for some health conditions
[144]. Continuity of care can be measured using a variety of different indices that study duration,
density, dispersion, and sequence of care at different providers [145]. A simple measure of
continuity of care that can be used to gauge the continuity of care provided at an ACO would be
density of care. Effectively, seeing different providers within the ACO would count as seeing the
same health care provider due to coordination and sharing of patient data among providers.
Fundamentally, changes in prices and utilization are the two pathways in which ACO formations
can affect aggregate health expenditures. Figure 2-6 is a causal diagram that visually represents
the assumed sequence and direction of changes induced by ACO formations in the model.
Figure 2-6. ACO effects and pathways for altering health expenditures in the ABM that studies
competition in health care markets.
Figure note: A “+” sign represents a direct relationship where increasing one factor increases
following outcome, and a “-” represents an inverse relationship such that increasing one factor
decreases the subsequent outcome. ACO formations enable a pathway to reduce health
expenditures by increasing care coordination.
36
Competitive conditions around which an ACO is formed dictate the magnitude of the relationships
illustrated in Figure 2-6Error! Reference source not found.. Participating providers’ combined
market share and WTP relative to market would determine the magnitude of price increases due
to the increased bargaining leverage. Using ABM, the objective is to separate provider price
increases due to ACO investments and increased efforts from increases due to the additional
bargaining leverage. It would be ideal to see the ratio of improvements in quality outcomes to
prices increases achievable by ACO formations under different market conditions. The ratio would
be different because ACOs will have different realized increases in bargaining leverage and quality
investment under different market conditions. Such insights could be valuable to understand under
which conditions are ACO formations more beneficial.
3.3. Delimitations of the Model Design
The model represents a contribution from systems engineering and complexity science to study
competition in health care markets. Naturally, this tool has deliberate omissions of aspects of the
system of interest for simplicity, interpretability, and computability. In this section, the structural
delimitations of the model are listed and described.
3.3.1. Nature of Consumer Choices
This model focuses on choices of consumers among alternative plans and providers, not on
whether consumers choose to acquire care or coverage. It is assumed that individuals will purchase
coverage every year and will receive care for every episode generated. While the “choose nothing”
option is a real alternative for actual individuals, incorporating such decision would require
substantially more data and model calibration. This is especially challenging given the limited
amount of experimental work in this area of consumer choice [4], [116]. Also, some economic
researchers have focused on the similar aspects of consumer choice [116]. Effectively, consumers
in this ABM model of competition only choose among alternative providers and plans.
This delimitation means that the model cannot be used to study uninsured rates. This may also
overstate provider and plan market power because consumers are forced to choose from them.
Theoretically and practically, consumers may forgo care or coverage if the cost of all alternatives
are higher than their willingness to pay. However, consumers are forced to choose among
providers and plans, raising the latter’s abilities to raise prices and premiums. While the individual
mandate instituted by the ACA may increase the appeal of coverage and care, consumers are still
free not to buy coverage.
3.3.2. External Competitive Influences
The simulated health care market represents a standalone, closed system with no external
influence. Modeled provider prices and insurer premiums are determined endogenously based on
each agent’s market leverage and costs. Other choices, such as plan network formations and
provider efforts, are a function of other agents’ decisions and the profitability of the simulated
consumer population. In real systems, however, there are several external factors that influence
the private health care market. Public insurance programs, like Medicare, Medicaid, and CHIP,
play a substantial role in influencing pricing and behavioral norms due to the volume of care
delivered through these programs [63]. Additionally, the presence of prominent delivery and
payment models, like Kaiser Permanente’s HMOs in California, is also large enough to alter
private market norms and prices [59], [72], [146]. The influence of the abovementioned factors on
37
private health care markets has been extensively studied. In addition, Medicare ACOs have
spillovers on quality and prices in private markets they serve [147]. There is value in studying the
impact of ACOs formations when accounting for these factors.
External factors that impact competition in health care markets can be accounted for in future
iterations of the ABM. This model represents the first step towards recreating representative
competitive dynamics in health care markets. A traceable, simple ABM and theoretical foundation
can be conveniently verified and validated. These factors have been omitted for modeling
convenience and because an evaluation of ACO formations can still be carried out in markets with
such exclusions.
3.3.3. Differentiated Products and Market Exits
Health care providers and health plans vary on different dimensions. For instance, providers vary
in bed capacity, waiting times, and specialization. Health plans offer different benefits and tiers of
providers. In this model, health care facilities are standardized and health plans do not have as
much dimensions to differentiate their services or products. Health care providers deliver all levels
of care for all patients, and only vary in out-of-pocket amounts and quality. ACO formations would
add another layer of provider and plan differentiation. The avenues for product and service
differentiation in the model are fewer than actual health care markets. However, models are
abstractions and investigating the complex effects of the modeled avenues on market trajectories
would be informative. Exploring other ways of differentiations would be useful in the future, but
would require substantial agent complexity and theoretical rigor, beyond the scope of this model.
The market structure is intended to be exogenous in the ABM to explore competition in private
health care markets. Thus, a simulated market is defined by the number of insurers and providers
in the market. In each simulated market, providers and insurers do not have a choice to exit the
market. This delimitation is in place to block exits from distorting the parameterization of markets
by its agent structure. Modeling a market for shorter periods of time and monitoring agent revenue
and margins would signal the likelihood of insurer or provider exit. In addition, market entry and
exit does not a play a direct role in understanding the short-term impact of ACO formations, which
is the aim of the case study in research.
4. Methods
The health care market ABM simulation components, interactions, sequences, and scenarios are
operationalized in this section. The proposed structural framework is implemented by formalizing
the underlying relationships mathematically and iteratively to draw insights on the competitive
dynamics in health care markets and potential effects of ACOs. The model is programmed on
AnyLogic Multi-Method Simulation Software v.7.1.2 [148]. For the agents to perform the roles
outlined in the structural framework, each agent is equipped with modules that enable decisions
and progression in the system. Figure 4-1 below depicts the modules that make up each agent to
carry out their role where every box inside an agent corresponds to a module. The model is
designed to be modular in nature such that each module functions independently given a certain
structure of module inputs from other modules, agents, and system states. This flexible approach
means the model is designed to permit module modification, enhancement, or replacement.
38
Figure 4-1. The organization of modules outlined in this model that allows each agent to play the
designated role in the agent-based model. Diamond shaped modules correspond to decision
modules.
4.1. Components of the Health Care Market Competition ABM Simulation
The health care market ABM includes 3 types of active agents. These agents interact frequently
and contribute to other agents’ decisions and system outcomes. This subsection initializes the
agents and environment settings by elaborating on the individual agent attributes, related data
sources, and general system rules.
4.1.1. Initial Agents and Key Attributes
The active agents in the health care market model are the consumers, providers, and health care
plans. When the model is first initialized, each agent is populated with the operationalized
attributes and roles presented in Table 4-1.
Table 4-1. A summary of the active agents, attributes and behaviors in the health care market ABM.
Agent Population Key Attributes Key Behaviors
Consumer I Consumers
1,2, … , I
X i: Full set attributes for individual i
D i ⊆ X i: Demographic attributes
Drive health care
demand
39
H i ⊆ X i: Health attributes
N i ⊆ X i: Health needs
P i ⊆ X i: Preferences for providers
G i : Geographic location for individual i
Provider i: Preferred provider
Insurer i: Chosen Insurer
Choose providers and
plans
Utilize health care
system.
Provider
J Providers
1,2, … , J
Z j: Full set of provider attributes for
provider j
A j⊆ Z j: Provider decision attributes for
TPB
O j⊆ Z j: Quality/profit orientation
Q j⊆ Z j: Current quality measures
Price j,k: Price provider j charges insurer k
Effort j: Effort level for provider j
C j(effort): Cost function for an effort level
G j: Geographic location
Health care for
patients
Compete for patients
and health plans
Choose effort level
Plan
K Plans
1,2, … , K
O k: Quality/profit orientation parameter
n l: Network of providers
OP l: Out-of-pocket cost
prem l: Premium
Selects providers
included in networks
Set premiums and
cost sharing levels
Offer coverage to
individuals
Market Area: Area of Insurance Rating Area
Mortality Rates: Market-wide mortality
rates per capita
Comp: Market-wide set of competitive
indices
HHI i: Insurer HHI
MS i ⊆ Comp: Insurer market shares
HHI p: Provider HHI
MS p ⊆ Comp: Provider market shares
Unit for defining the
health care market
Unit of analysis for
market performance
Basis for defining
spatial distances
The attributes influence how agents perform their key behaviors and decisions which impact other
agents’ decisions and system outcomes. The above attributes and key behaviors are organized in
the flowchart shown in Figure 4-2 below.
40
Figure 4-2. A flowchart representing the relationships between agents’ modules in the health care market
ABM simulation.
As shown in Figure 4-2, consumer attributes predict the demand for health care. Demand is also
forecasted by providers and plans to lead to a sequence of decisions that end with the observable
outcomes that are shown with a grey glow in Figure 4-2. Provider attributes influence attitudes
and norms that predict effort investments a provider makes given their perceived demand. Based
on a plan’s risk and quality attitudes, a network of providers is formed and offered to consumers.
The observable outcomes partially determine next year’s decisions and events.
4.1.2. Initial Distributions and Agent Parameters
Key attributes of the generated agents are obtained from pertinent datasets and sources to produce
comprehensive health care markets with suitable consumer, provider, and plan attributes. These
attributes are used to generate agents during model initialization and to parameterize key decision
modules. Building representative models of complex systems is typically bound by the availability
of data and theoretical work at the different levels of the system. Here, data from various sources
41
are employed to enable a suitable abstraction of health care markets, and heterogeneity in needs
and decisions. Table 4-2 lists key model inputs, related modules, external sources, and the
associated model variables.
Table 4-2. Table of distributions and parameters involved in the model and sources.
Model Inputs Attributes &
Module
Source Variables/Distributions
Consumer Demographics Demographic
[D i]
National Health and
Nutrition Examination
Survey (NHANES)
2013-2014
Demographics Data
[149]
Age
Gender
Race
Education
Income
Consumer Health
Conditions
Health Conditions
[H i]
NHANES 2013-2014
Questionnaire Data [149]
Disease Count
Disability
Health Score
Mental Health Score
Previous Year Visits
Previous Year
Hospitalizations
Consumer Health Needs Health care Need
and Costs
[N i]
RAND Health Insurance
Experiment [150]
Frequency of Health
Episodes
Costs of Health Episodes
Providers & Plans Preferences
[O j]
Assumed Risk Attitudes
Quality Orientation
Decisions Weights
Provider Operations Cost of Effort
[C j(effort)]
[151], [152] Operational Costs
Effect of Effort on Quality
Provider Outcomes Health care
Provider Operations
CDC Statistics [153],
[154]
Mortality
Readmissions
Market Premiums & WTP Premiums HIX [155]
4.1.3. Initial Environment and System Parameters
Agents defined and populated in the previous subsection interact in a health care market of a
defined uniform rectangular area of size m by m. I consumer, J hospitals, and K plans will be
generated in market M composed of all agents I, J, and K ⊆ M. Provider and consumer agents are
scattered randomly in the simulated IRA of size area. Time t in the model is measured in yearly
increments, and the model is initialized such that t=0. All health plans K are listed in a marketplace
for the designated market. Each insurance plan k∈ K is listed with its premium premk, provider
network nk, out-of-pocket costs OPk and observable quality outcomes and attributes Yj for each
provider j ∈nk. Year 0 of the model is assumed to correspond to the year 2013, mainly because that
is the year ACA health exchanges came into operation and their data is available.
Consumers only obtain care from providers in the network of their chosen health plan. Based on
the plans premiums and networks, consumers make a choice of health plans once a year during the
open enrollment period, and make a choice of providers from the plan’s networked providers for
every episode of care.
42
4.2. Agents and Modules
With the key attributes, environment, and initializations laid out, agent capabilities and decision
mechanisms are devised in this subsection. This model is designed to be modular, as illustrated in
Figure 4-1, so that each module enables a certain capability and can be constructed independently
so long as a module takes in compatible inputs and generates outputs of appropriate structure. A
module relates to a behavioral capability or operational mechanism for an agent. Therefore, the
modular model allows for refitting, updating, and improving modules as empirical and theoretical
studies on the behavioral and progression mechanisms become available. The proposed ABM
synthesizes well-supported theoretical and empirical models on behaviors and progression with
assumed formulations of decision-making mechanisms to recreate agent interactions and
competitive forces in the private health care market.
Consumers, providers, and plans, each have a unique set of decisions, operational modules, and
implications on the progression of the system and observed outcomes. The modules relating to
each agent are presented in the following subsections.
4.2.1. Consumers
Consumers drive the demand of the health care market. There is extensive economic literature on
modeling demand for health care and consumer choice of providers and health plans. Existing
models of health care need and demand are typically complex, condition specific, and multi-
staged. Some models examine the decision-making process of obtaining health care [156], while
others model the various classes of health care need in a population [157], [158]. The need for
health care is operationalized in this subsection along with preferences that are a function of the
perceived demand.
Consumers in this model are sampled from the National Health And Nutrition Examination Survey
(NAHNES) data from the years 2013-2014 [149]. The set of attributes of each generated agent is
sampled from the attributes of an individual aged between 18 and 64 in the NHANES dataset. The
sampled attributes and variable transformations are listed in Table 4-3:
Table 4-3. List of individual agent's attributes along with source variables sampled from NHANES 2013-
2014.
Type Model Variable NAHNES Variable NHANES Variable
Identifier
Comments
Demographic
Attributes
Age Age RIDAGEYR Range between 18 to 64
years of age
Gender Gender RIAGENDR Binary variable
Ethnicity Ethnicity RIDRETH1 5 groups: Mexican
American, Other Hispanic,
Non-Hispanic white, Non-
Hispanic Black, Other
Education Education DMDEDUC2 Transformed highest
degree into years of
education
Income Annual Income INDHHIN2 12 categories of annual
income
Health Attributes
General Health
Score
General Health
condition
HUQ010 Transformed from a 5-
point scale to 100-point
scale
43
Mental Health
Score
Depression Screening
results
DPQ_H Transformed from a PHQ
score to 100-point scale
Disease count Health Conditions MCQ_H, DIQ,
BPQ, HEQ, OSQ,
SLQ, SMQ, CDQ
Count of all reported
conditions for an
individual
Last year visits Times of care in past
year
HUQ051 Reported times of care are
interpreted as provider
visits
Disability Work limitation PFQ051 Reported work limitations
are interpreted as disability
The sampled variables have been transformed to match the health needs generation and provider
preferences modules, which are outlined in the following subsections. For example, the health
score is used to predict health care needs is a 100-point score, while NHANES subjects report their
perceived health on a scale from 1 to 5. The variables extracted from NHANES dataset have been
chosen to best match the independent variables that are used as inputs in the health needs
generation and patient preferences modules.
4.2.1.1 Consumers Health Care Need and Utilization Module
This module computes the need for health care services for consumer i. Health needs are defined
by Fries (1997) as the illness burden of a population [159]. At an individual level, it can be thought
of as the sum of all conditions, such as stokes, heart attacks, and arthritis, that requires an individual
to obtain care. The actual need for health services is influenced by consumer demographic
attributes, health conditions, and previous need [159]. The level of analysis for modeling health
needs is at the episode of care. For each consumer, rate of episodes of care per year and costs per
episode are generated. Keeler et al (1988) estimated demand for health care using data from the
landmark RAND Health Insurance Experiment data for 4 types of episodes: wellness episodes,
acute episodes, chronic care episodes, and hospitalizations [124]. The first 3 types are considered
outpatient episodes, and hospitalizations are inpatient.
According to Keeler et al (1988) [124], an individual i's episode frequency of type h in a given
year follows a negative binomial distribution with a mean 𝜇 , . For the 4 types of episodes, Table
4-4 below lists the relevant coefficients from [124] that were used in estimating ln (𝜇 , ).
Table 4-4. Coefficients from Keeler et al (1988) [110] used in negative binomial regressions equations for
predicting episode rates.
Hospital Acute Chronic Wellness
Intercept -1.1 0.097 -0.57 -3.4
Health score -0.012 -0.0052 -0.0075 0.0027
Disability 0.34 0.11 0.17 0.052
Disease Count 0.0025 0.011 0.015 0.0062
MD Visits Missing 0.18 0.14 -0.044 0.36
Woman 0.31* 0.79*
Woman age 18-40 0.53* 0.44*
Woman age 41-65 0.3 0.99
Man age 46-55 0.36 0.058 0.71 0.51
Man age 56-65 0.55 0.094 0.89 0.53
Coinsurance 25% -0.21 -0.28 -0.37 -0.24
Log(MD Visits) -0.25* 0.28* 0.31* 0.12*
44
Log(income) 0.04 0.4 0.13 0.08
Education 0.04 0.4 0.13 0.08
Black -0.20 -0.54 -0.40 -0.29
Where 𝑆𝐸𝑆 = log(𝑖𝑛𝑐𝑜𝑚𝑒 ) + 0.2 ∗ (𝑦𝑒𝑎𝑟𝑠 𝑜𝑓 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 ) − 11.1 and log(MDVisits = 0) = 0.
* p-value<0.05
** p-value<0.01
For each generated episode of each type, the accompanying log(episode cost) is obtained using the
coefficients shown in Table 4-5. These generate positive cost estimates for each episode based on
the episode type and patient’s health and demographic attributes.
Table 4-5. Coefficients from Keeler et al (1988) [110] used to generate log(episode costs).
Hospital Acute Chronic Wellness
Intercept 7.953 4.503 4.436 4.741
Health score -0.023* -0.018* -0.032* -0.017*
Socioeconomic
Status (SES)
0.03 -0.04* 0.05 0.00
Log(MD Visits) -0.03 0.00 0.09* -0.03
Black 0.14* 0.01 -0.17 0.00
Woman 0.11 -0.15* -0.23* -0.07
Man aged 46-65 0.26* -0.08 -0.00 0.21**
Where 𝑆𝐸𝑆 = log(𝑖𝑛𝑐𝑜𝑚𝑒 ) + 0.2 ∗ (𝑦𝑒𝑎𝑟𝑠 𝑜𝑓 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 ) − 11.1 and log(MDVisits = 0) = 0.
* p-value<0.05
** p-value<0.01
The coefficients of independent variables that were statistically significant in the HIE regressions
were used in the model. The simulated health care episode costs are in 1986 US dollars, and are
inflated to 2013 dollars, assuming 3% yearly inflation rate. Therefore, all generated costs are
multiplied by 2.2213 or (1.03
).
The number of episodes generated by these regressions are used as next year’s MD visits for a
patient, which are assumed independent variables in Keeler et al (1988) [124]. While this may be
problematic for error propagation, possible issues may be mitigated when modeling only a few
years; less than 5 years. Also, each acute or hospitalization episode can probabilistically increase
the disease count for a patient. A random probabilistic process determines if each hospitalization
or acute episode increases the disease count for a patient. The probability of such event has been
chosen to maintain the same approximate distribution of disease counts in the population. The
calibrated probability is at about 8.5% probability that a hospitalization or acute episode adds
one disease to a patient’s disease count.
4.2.1.2 Consumer Utility and Willingness-to-pay Module
Consumers’ utility in the health care market and the corresponding WTP represent the social value
a provider adds by delivering an episode of care. The WTP is used to predict how consumers select
plans and providers when care is needed. This approach has been utilized in Option Demand
Markets studied by Capps et al [46]. The approach used in this model is inspired by the same
conditional choice structure through obtaining a consumer’s WTP for providers and plans.
1. Utility of individual i when utilizing provider j for episode of intensity EI:
45
𝑈 , , = 𝑈 𝑋 , 𝑌 , 𝐸𝐼 , 𝑂𝑃 = 𝛼𝑦 + 𝛽𝑥 𝑦 + 𝛿𝐸𝐼 − 𝛾 ∗ 𝑂𝑃 , , + 𝜀 ,
Where
α is the unconditional marginal values of provider j attributes that are well known and
equal for all patients. There parameters in α include the effect of provider j’s attributes,
such as provider quality and out of pocket costs, on utility.
β are the marginal values of provider j’s attributes conditional on consumer i's
attributes. These attributes include the effect of the distance between a provider and the
patient, conditional on the patient’s attributes, like age, gender, race, income, and
education.
𝛿 are the marginal effects of episode intensity on consumer preferences.
γi is a function that converts money to utils for individual i. The value individual i
places on $1.
OPi,j,EI is the out of pocket cost a patient pays at provider j for an episode of intensity
EI.
ε is a patient and provider specific error term distributed with the standard double
exponential distribution
2. Interim expected utility (up to an arbitrary constant) for access to a plans network nl for
individual i with attributes Xi is:
𝑉 , , (𝑛 , 𝑋 , 𝐸𝐼 ) = 𝐸 𝑚𝑎𝑥 ∈ [𝑈 (𝑋 , 𝑌 , 𝐸𝐼 , 𝑂𝑃 ) ]
= ln exp 𝑈 (𝑋 , 𝑌 , 𝐸𝐼 , 𝑂𝑃 ) ∈
The second equation is derived by relying on the mathematical result that ln[Σexp(um)] is
the expectation of um if it they are systematic components of utility [46].
3. The probability that individual i chooses provider j for episode intensity of EI, among
providers in the same plan network nl is represented using a demand logit function:
𝑠 , , , 𝑛 , 𝑋 , 𝐸𝐼 , 𝑂𝑃 , =
exp [𝑈 𝑋 , 𝑌 , 𝐸𝐼 , 𝑂𝑃 , ]
∑ exp [𝑈 (𝑋 , 𝑌 , 𝐸𝐼 , 𝑂𝑃 )]
∈
4. Provider j’s contribution to interim utility is obtained through the difference of
ΔVj
IU
= V
IU
(M,Xi,Gi)- Vj
IU
(M/j,Xi,Gi) for consumer i with episode intensity EI. Where nl/j is
network nl without provider j:
Δ𝑉 𝑛 , 𝑋 , 𝐸𝐼 , 𝑂𝑃 , = 1
1 − 𝑠 𝑛 , 𝑋 , 𝐸𝐼 , 𝑂𝑃 ,
5. Converting equation (5) to monetary terms yields consumer i's willingness-to-pay to
retain provider j in network l for an episode of intensity EI:
ΔW
, , 𝑛 , 𝑋 , 𝐸𝐼 , 𝑂𝑃 , =
Δ𝑉 𝑛 , 𝑋 , 𝐸𝐼 , 𝑂𝑃 , 𝛾
46
This module computes a consumer’s ex post utility and ex ante WTP for a provider to be included
in a network. Given the consumer’s attributes and intensity of the episode, the output of this model
is the consumers WTP and utility. This utility, combined with the perception of need that
consumers have, drive their preferences and ex-ante choice of plans.
Patient choice parameters have been estimated in several different studies [121], [160]–[164].
However, most of the studies estimate the parameters from populations that do not compare to the
population studied in this model. For instance, some studies observe the Medicare population,
others estimate preferences in European countries, and some estimate the effect of distance and
severity of episodes on choice. Therefore, this model uses parameters that are synthesized from
some of literature, but acknowledges the limitation in not having clear estimates, given the
importance of these parameters in estimating utility, WTP, and consequently impact bargaining.
Below is a list of the patient choice parameters.
Table 4-6. Parameters used to estimate patient utility of received care at a given provider.
Parameter Coefficients Source Comments
Distance -0.000869 [121] – Medicare population Per mile to provider
Number of
Alternative
Hospitals
0.00744 [121] – Medicare population Per alternative provider
Medicaid
Eligibility
-0.3206 [121] – Medicare population If consumer yearly income is less than $12,000,
corresponding to the Federal Poverty Line.
Age -0.0004 [164] – Females population Binary variable corresponding to the gender of
the agent
White 0.0614 [164] – Females population Binary variable corresponding to whether the
agent is ethnically white
Severity -0.0296 [164] – Females population At source, this value corresponds to having
“poor health”, in the model it used per 1000
units of episode intensity
Mortality Rate -0.2 Assumed Average market mortality rates used as
reference. For the difference between market
average and each provider’s average
Readmissions
Rate
-0.2 Assumed Average market readmissions rates used as
reference. For the difference between market
average and each provider’s average
While the lack of suitable estimates to these parameters might be concerning, the model results
will be assessed for sensitivity to these parameters. In the Table 4-6, negative coefficients indicate
that adding to the attribute makes an alternative less desirable. For instance, providers that are
further away or have higher mortality rates are less favorable. Also, in markets with more choice,
consumers are more likely to switch. In the case of a consumer attributes, such as Medicaid
eligibility, a negative coefficient indicates a reduction in sensitivity to other attributes. This
provider choice model is used as shown in next subsection to compute provider WTP, assess
insurer offerings, and choose plans.
4.2.1.3 Consumers Choice of Plans
Consumers choose a plan during the open enrollment period every year. The process of evaluating
each insurer involves evaluating the insurer’s network, premiums, and associated out-of-pocket
costs. Therefore, the consumers weigh their WTP for each insurer’s network of providers against
47
the insurer’s premiums and out-of-pocket costs. Uniform cost sharing rates across all insurers, as
described in the structural framework, means the comparison between the ratio of an insurer’s
network WTP and premium for each consumer suffices to evaluate the perceived value of each
plan. A consumer computes a ratio of their WTP given their expected utilization to expected costs
for each insurer, both in monetary units. This ratio for each insurer is used as the ex-ante
𝑈𝑡𝑖𝑙𝑖𝑡𝑦 , , , , and a conditional choice structure is used to determine the probability of
choosing each insurer.
At the beginning of year t, consumer i picks plan l with the following probability:
𝑃 (𝑖 𝑒𝑛𝑟𝑜𝑙𝑙𝑠 𝑖𝑛 𝑙 ) =
𝑒 ∗ , , ∑ 𝑒 ∗ , , ∈
where 𝜑 is a smoother parameter that is calibrated to approximate insurer switching rates on the
ACA health exchanges. It is estimated that approximately 43% of consumers switched insurers in
2016 on Health care.gov, the federally facilitated health insurance exchanges [165]. The
corresponding calibrated smoother value is approximated to 0.1 to hit the average provider
switching rates. The probability function here is derived from the conditional logit probability
model, where the higher the utility from plan l, the higher the likelihood that individual i enrolls.
4.2.2. Health Care Providers
Health care providers compete for demand from health plans and patients. To attract more patients,
providers aim to contract with more health plans by keeping their prices competitive and quality
levels high. This module outlines how providers compete on multiple fronts for consumers and
health plans.
4.2.2.1 Health care Provider Financial Module
For health care provider j, effort is defined as effortj, a discrete variable with 3 levels. Increased
effort leads to quality and efficiency improvements. Effort is modeled to proxy the relationship
between investments, costs, and outcomes. In a study by Lee and Zenios (2012), provider efforts
are studied in a principal-agent framework [151]. A similar approach is adopted for this model. It
is assumed that the effort decision for health care providers determines process and health
outcomes. Effort is a latent behavioral variable which will be studied in the system to investigate
the effects of competition and ACO formations.
Quality vector Q refers to all observable health and process outcomes for provider j, Process
outcomes would be enrollment smoking programs, telephone follow-ups, and disease management
programs, which are modeled in this ABM as wellness episodes outline later in this subsection.
The effects of efforts on health outcomes (HO) is presented below.
𝑄 ~ 𝑒𝑓𝑓𝑜𝑟𝑡 + 𝜀
For health care provider j, the cost per unit of intensity function is assumed to be quadratic where
𝑐𝑜𝑠𝑡 (𝑒𝑓𝑓𝑜𝑟𝑡 ) = 𝐶 ∗ 𝑒𝑓𝑓𝑜𝑟𝑡 + 𝐶
The cost function has been calibrated to maintain the same form and fitted to the range of margins
obtained from empirical work. Constants Cj and C0 have been approximated at
0.5/((EffortSize+1)
2
) and 0.5 as calibrated to a range of hospital margins [152]. In a FFS setting of
PPO, provider profits are a function of episode volume, episode intensity, provider costs and price
48
paid by health plans. At the end of the year, health provider j delivers El,j episodes of care for
enrollees in a plan l, each episode of intensity EI for each plan l∈Lj which provider j contracts with.
The cost of one episode of intensity EI is CPEj,EI. The price charged by health provider j to health
plan l for an episode of intensity EI is PPEl,j,EI. The following formula models the profit function
for health care provider j in PPO contracts:
𝜋 = 𝑟𝑒𝑣𝑒𝑛𝑢𝑒 − 𝑐𝑜𝑠𝑡
𝜋 = 𝑃𝑃𝐸 , , − 𝐶𝑃𝐸 , ∈ , ∈
Where CPEj,El is a function of the cost of effort per unit of intensity costj(effortj) such that:
𝐶𝑃𝐸 , = 𝐸𝐼 ∗ 𝑐𝑜𝑠𝑡 𝑒𝑓𝑓𝑜𝑟𝑡 + 𝜀 ,
It is assumed that episode intensity has a linear effect on the cost of the episode along with a
provider and intensity specific noise variable. PPEl,j,EI is also assumed a linear function of the base
price per unit of intensity set in the agreement with a health plan and the intensity of the episode.
𝑃𝑃𝐸 , , = 𝐸𝐼 ∗ 𝑝𝑟𝑖𝑐𝑒 ,
The error term 𝜀 , exists in the provider cost function but not in the price charged per episode
because it is assumed that providers and plans agree on service prices, regardless of the variations
within the service provided.
4.2.2.2 Provider Effort Selection
Provider effort is a behavioral variable in is a key variable outcome of this simulation model. Effort
levels reflect the amount of investment providers make to improve processes and operations, and
consequently health outcomes. As shown in the financial module for providers, increasing effort
relates to a quadratic increase in costs.
Unlike consumer choice, provider efforts are hard to measure and observe empirically. Therefore,
there is limited behavioral economics literature to build a model for providers’ choice of levels of
effort. In this case, an established behavioral framework called the Theory of Planned Behavior
(TPB) is used to model providers choice of effort levels. The decision is performed once a year
before the bargaining period.
The TPB is a well-supported behavioral framework to understand how utility drives intentions
[166]. In this framework, intentions are derived from attitude, perceived control, and subjective
norm. The components of the intention-driven decision module are illustrated in Figure 4-3.
Intentions for behavior change are driven by the perceived added benefit, social normative
pressures, and perceived behavior control over benefit when behavior is changed. This framework
will be utilized in modeling provider effort choice. Medicare ACO ABM simulation model by Liu
and Wu (2014) utilizes the same framework for provider ACO formation [112].
49
Figure 4-3. Components that make up the intention for the provider agent for each effort level
according to the Theory of Planned Behavior.
In the theory of planned behavior, intentions are a weighted sum of attitude (A), subjective norm
(SN), and perceived control (PC). Utility, subjective norm, and perceived control accentuate the
ability of agents in the ABM to autonomously decide, influence one another, and build on past
intentions. An agent’s decision to pursue a behavior option will have an uncertain effect on the
agent’s utility, perhaps through additional profit or better care for patients. Agents learn from past
behavior changes and make better decisions building on prior intentions.
Intention for behavior b option by agent a in TPB is modeled as follows:
𝐼 , = 𝛼 , 𝐴 , + 𝛼 , 𝑆𝑁 , + 𝛼 , 𝑃𝐶 ,
𝐴 , 𝑆𝑁 , 𝑃𝐶 , 𝐼 ∈ [0,1], 𝛼 , = 1
Attitudes A are usually driven by perceived benefit. Subjective norm SN is an agent’s perception
whether other related agents believe they should perform the behavior change. Subjective Norm
captures the peer pressure and social norms between agents. Perceived behavior control PC
captures the difficulty in performing the behavior change and reaching the desired outcomes. The
final intention for a given year is a time-weighted function of the most recently computed I
t
intention and previous final intentions I
t-1
. This updating mechanism is used to account for an
agent’s past intentions. The composite intention I
*t
is then used to compute the probability an
action that matches this intention is performed. The probability that action w is performed is
modeled below.
𝐼 , ∗ = 𝜔 𝐼 , − (1 − 𝜔 )𝐼 ,
𝑃 (𝑏𝑒 ℎ𝑎𝑣𝑖𝑜𝑟 = 𝑥 ) =
exp (𝐼 ∗ ∗ 𝜁 )
∑ exp (𝐼 ∗ ∗ 𝜁 )
Where ζ is a smoothing coefficient that determines the sensitivity of the probability to changes in
intentions I.
50
Effort has discrete, ordinal levels: effort ∈ EFFORT. Each provider is generated with quality
preference Oj, an initial effort level effortj
0
= 2, with an effort space EFFORT= {1, 2, 3} The TPB
models how providers choose the effort level they invest in a year t+1. Intentions to change effort
levels for providers are driven by provider attitudes, subjective norm, and perceived control.
Provider attitudes on effort change are a function of utility at different levels of effort. Changes in
effort levels induce demand changes, quality changes, and cost changes. Therefore, providers
forecast changes in profits and quality across all potential effort levels in EFFORT. Attitude for a
change in effort level is modeled as:
𝐴 , = 𝑂 ∗ 𝑈 (𝐹𝑄 , ) + 1 − 𝑂 ∗ 𝑈 (𝐹𝜋 , )
In the equation above, FQj,effort and Fπj,effort are forecasted quality levels and profits from changing
effort level. Uj(X) is a function that converts monetary and quality values to utils for provider j.
Changes in effort would change profits through changes in service costs, service prices, and
changes in demand. Changes in prices due to effort will be described in more detail in the section
relating to provider-insurer bargaining. For now, it suffices to know that increasing effort can, in
some cases, increase the agreed service price, therefore offsetting the increase in incurred costs
due to increasing effort. Assuming changes in effort do not have immediate effects on changes in
demand, providers do not account for changes in demand due to changes in effort. This is because
effort is not observed by consumers, only outcomes are. The effect of effort on outcomes is
captured by the forecasted quality change element, FQj,effort, of the attitude component, 𝐴 , .
Therefore, the forecasted profit change for each effort level corresponds to the percentage change
in profit per episode with the current year’s profit per episode used as baseline. The change in
profit can be positive or negative, depending on the current effort level and the level evaluated.
Expected changes in health outcomes QHO are a function of process effort. The utility from changes
in quality due to changes in effort are computed from the percentage change from current quality
for each effort level. This a reflection the relationship between effort and quality conceptualized
above, and will be operationalized in the module relating to provider operations. See
EffortModifier in 4.2.2.4.
The subjective norm component is a representation of where provider j stands with respect to the
performance of other providers. For provider j, the subjective norm is a function of the effort levels
for all other providers in the market. The following function is derived by intuition based on the
nature of peer pressure and market norms. It is assumed that providers’ pressure diminishes with
the distance. For example, a nearby provider with high quality imposes more peer pressure to raise
quality than a more distant one. If other providers in market M (k in M, k ≠ j) deliver high quality
care, the subjective norm component makes increasing effort more favorable for provider j.
Conversely, if health care quality levels are poor in the market, there is less pressure for provider
j to increase effort. Therefore, subjective norm is modeled as follows:
𝑆𝑁 , = 𝑒𝑓𝑓𝑜𝑟𝑡 − 𝑒𝑓𝑓𝑜𝑟𝑡 𝑠𝑖𝑧𝑒 (𝐸𝐹𝐹𝑂𝑅𝑇 )
∗ 𝑈 𝐸𝑓𝑓𝑜𝑟𝑡 − 𝐸𝑓𝑓𝑜𝑟𝑡 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 , ,
The effect of the difference between the quality of the rest of the providers and provider j’s quality
diminishes with distance, Tkj(Gk,Gj). Based on this formula for subjective norm, SN is positive if j
is performing weaker than neighboring providers for an effort level that is higher than its previous
effort level effortj
t-1
. Also, SN is positive if j is performing better than neighboring providers for
an effort level that is lower than its previous effort level effortj
t-1
.
51
Perceived control for provider j over the effects of changing effort levels are driven by the noise
levels between efforts and outcomes. If the relationship between effort and outcomes are less noisy
(smaller ε of health outcomes), then perceived control is higher. Therefore, perceived control is
modeled as 1-logit(ε), which approaches 0 as ε increases, and approaches 1 as ε decreases.
Therefore, provider j’s perceived control is modeled as follows:
𝑃𝐶 , = 1 − 𝑙𝑜𝑔𝑖𝑡 (|𝜀 |)
The intention for each effort level is computed by summing over each component: A, SN, and PC.
For each level the composite intention Ieffort
*t
is updated based on a time-weighted average of past
intentions. The probability of choosing effort level e for the next year is as such:
𝑃 (𝑒𝑓𝑓𝑜𝑟𝑡 = 𝑒 ) =
exp (𝐼 ∗ /𝜁 )
∑ exp (𝐼 ∗ /𝜁 )
4.2.2.4 Health care Provider Operations Module
This module relates to the types of episodes providers deliver, the effect of provider effort, and
possible outcomes. This model is intended to account for the various types of episodes and capture
the potential effects of coordination and increased effort. Using the utilization estimates from
RAND HIE [150], the likelihood of adverse outcomes per episode are computed from adjusted
mortality rates and readmissions rates. Figure 4-4 illustrates the categorization of episodes,
potential outcomes and the relationship between episodes and observable health and process
outcomes.
Figure 4-4. The relationship between types of episodes, health care need, and health outcomes.
Providers delivery 4 types of episodes, two of which have 3 possible health outcomes. Acute
episodes and hospitalizations can be successful, results in death, or readmissions. The probability
of each event is estimated from the Centers for Disease Control and Prevention (CDC) reports on
mortality rates, cause of death, and readmissions rates reports. The yearly probability of death in
an episode is obtained as follows:
𝑃 𝑑𝑒𝑎𝑡 ℎ
@𝑓𝑎𝑐𝑖𝑙𝑖𝑡𝑦 | 𝑥 = 𝑃 (𝑑𝑒𝑎𝑡 ℎ@𝑓𝑎𝑐𝑖𝑙𝑖𝑡𝑦 |𝑑𝑒𝑎𝑡 ℎ)
( )
∗ 𝑃 𝑑𝑒𝑎𝑡 ℎ
𝑥 ( )
52
Effectively, the proportion of deaths that occur at a medical facility, (i) in the equation above, and
adjusted mortality rates, (ii) in the equation above, as described by the CDC cause of death dataset
[154] and CDC mortality statistics dataset [153], respectively, are multiplied to obtain the
probability of death at a provider in any given year. The probability of death in a given episode of
intensity EI is then obtained as follows:
𝑃 𝑑𝑒𝑎𝑡 ℎ
@𝑓𝑎𝑐𝑖𝑙𝑖𝑡𝑦 | 𝑥 , 𝐸𝐼 = 𝐸𝐼 ∗
𝑃 𝑑𝑒𝑎𝑡 ℎ
@𝑓𝑎𝑐𝑖𝑙𝑖𝑡𝑦 | 𝑥 𝐸 (𝑦𝑟𝑙𝑦𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 )
This formulation assumes that the probability of death in an episode is proportional to the intensity
of an episode.
Readmissions refer to unplanned hospitalizations in 30 days after discharge from acute care or a
hospitalization. CMS public reports on readmissions are not adjusted to patient attributes and is on
Medicare beneficiaries. These estimates may be too high for the population studied in this model,
but this is the only extensive dataset that is publicly available. Therefore, data from CMS Compare
on hospital-wide 30-day readmissions (Variables name: READM_30_HOSP_WIDE) is obtained and
fitted to a probability distribution [167]. Readmissions scores refer to percentage of episodes that
end with readmissions, and the values range from 10.8 to 19.9% with a median of 15.5%. The
readmissions probability per episode is sampled from a normal distribution with mean 15.5% and
standard deviation 0.825%.
Effort impacts the probability of adverse outcomes as shown in the choice of effort subsection.
Effectively, 𝑄 ~ 𝑒𝑓𝑓𝑜𝑟𝑡 + 𝜀 . Therefore, an effort multiplier is introduced such that the
probability of any adverse outcome, as shown above, is multiplied by the following effort modifier:
𝐸𝑝𝑖𝑠𝑜𝑑𝑒𝐸𝑓𝑓𝑜𝑟𝑡𝑀𝑜𝑑𝑖𝑓𝑖𝑒𝑟 = 𝑒𝑓𝑓𝑜𝑟𝑡𝑆𝑖𝑧𝑒 −
𝑒𝑓𝑓𝑜𝑟𝑡 𝑒𝑓𝑓𝑜𝑟𝑡𝑆𝑖𝑧𝑒
For effort space EFFORT= {1, 2, or 3}, this function ranges between 1.15 at effort = 1 to 0.7321
at effort = 3. Figure 4-5 below shows a plot of the relationship between effort and the multiplier to
probability of adverse outcome.
Figure 4-5. Plot of Effort Multiplier to probability of adverse outcome, namely probability of
mortality or readmissions.
53
This plot shows that highest effort reduces the probability of adverse outcomes in a given episode
by approximately 25% (Multiplier<1), while lower effort can increase the probability of such
events by 15% (Multiplier>1).
4.2.3. Health Plans
Health care plans contract with individual providers to form a network that is then offered to
consumers. The premium, breadth of network, and the level of cost sharing are decided by the plan
itself. A plan’s financial module is developed to model the mechanisms by which health plans
operate and execute decisions.
4.2.3.1 Health Plan Financial Module
The revenue for the health care plan is the premiums received from enrollment. Therefore, the
more consumers enroll, the higher the revenue. Costs for a health care provider are variable per
patient payments to providers as per agreed prices. Therefore, every time a patient utilizes a
provider, the health care plan incurs a cost totaling in the expense paid to the provider.
At the end of the year, health plan l covers Ej,l episodes of care at each contracted provider j∈nl, in
an amount equal to the total intensity of care delivered multiplied by the price per unit of intensity.
The cost of one episode of intensity EI is the price charged by health provider j to health plan l for
an episode of intensity EI is PPEl,j,EI = pricel,j * EI. The following equation models the profit
function for health plan l in a FFS contract as the sum of premiums received less the amounts paid
for every episode in provider expenses and administrative costs. The profit function is as follows:
𝜋 = 𝑝𝑟𝑒𝑚 ∈ ( )
− ⎣
⎢
⎢
⎢
⎡
((𝑝𝑟𝑖𝑐𝑒 , − 𝑂𝑃 , ) ∗ 𝐸𝐼 , )
( ) ⎦
⎥
⎥
⎥
⎤
∈
Revenues for health plans are premiums received for enrollees of demographic attributes Di
modeled in component (i). Costs for each episode is PPEl,j,EI = pricej,l*EI less the out-of-pocket
proportion OPl,j which is paid by the patient modeled in component (ii). The formulation above
highlights that profits increase with increasing enrollment and premiums or reducing costs and
utilization. Hence, it is important for a health plan to offer an attractive network of providers that
offer competitive prices. Premium pricing and out-of-pocket allocations are also critical in
attracting consumers to plans and in steering patients away from some providers. The modules for
plan pricing will be discussed in more details in the following subsection.
4.2.4. Health care Provider and Plan Agreements
This module describes the sequence of events and processes by which health plans end up
providing a network of providers for an agreed price. Contracts are defined as agreements between
providers and plans on a price per unit of intensity and inclusion of provider in insurer network. It
is essential for this module to account for the role of market power on agreed prices and the
resulting effects on health plan premiums.
4.2.4.1 Plan-Provider Prices
This bargaining module determines the prices offered to insurer l by provider j by computing the
disagreement tradeoff, which is the change in profits for each bargaining party with and without
the other. Let ∆
𝑃𝑟𝑜𝑓𝑖𝑡 be the disagreement tradeoff for insurer l for not agreeing with provider
54
j at a given price p, and ∆
𝑃𝑟𝑜𝑓𝑖𝑡 is the disagreement tradeoff for provider j for not being
contracted by insurer l at the same price.
Insurer Disagreement Tradeoff
In this model, the insurer’s disagreement tradeoffs are modeled as follows:
∆
𝑃𝑟𝑜𝑓𝑖𝑡 = 𝑀𝑆 , , (𝑃𝑟𝑒𝑚 − 𝐴𝑉𝐶 ) − 𝑀𝑆 , , (𝑃𝑟𝑒𝑚 − 𝐴𝑉𝐶 )
Where 𝑀𝑆 , and 𝑀𝑆 , are the market shares for insurer l with and without provider
j, respectively. 𝑃𝑟𝑒𝑚 and 𝐴𝑉𝐶 are the premiums and average variable cost per enrollee with and
without provider j. This formulation captures the contribution of provider j to insurer i's revenue
and costs. Using the formula for the optimal premium for insurer l for a given network, the change
in profits for insurer l for including provider j is modeled as a function of this provider’s
contribution to the average variable cost and to the aggregate network’s WTP. In the formulation
above, premiums and average variable costs are obtained in the following manner:
𝑃𝑟𝑒𝑚 =
𝑘 ∗ 𝑁𝐸𝑇𝑊𝑂𝑅𝐾 % + 𝐴𝑉𝐶 2
𝑃𝑟𝑒𝑚 =
𝑘 ∗ (𝑁𝐸𝑇𝑊𝑂𝑅𝐾 − 𝐽 % ) + 𝐴𝑉𝐶 2
𝐴𝑉𝐶 = (1 − 𝑂𝑂𝑃 ) ∗ 𝐴𝑣𝑔𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 ∗ 𝑀𝑆 ∗ 𝑝 + 1 − 𝑀𝑆 ∗ 𝑝
𝐴𝑉𝐶 = (1 − 𝑂𝑂𝑃 ) ∗ 𝐴𝑣𝑔𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 ∗ 𝑝
The above formulas can be used to estimate the average profit per enrollee for insurer l with and
without provider j where k is a scalar constant estimated in the insurer premium setting module,
subsection 4.2.4.3. It is assumed that patients flow to providers is proportional to the provider’s
market share. Consider the graph in Figure 4-6 that depicts how premiums with and without
provider j are determined. In this example market, the premiums are determined for two
configurations, one where all 5 providers in the market are in the network (blue line), and one
where provider j not contracted (red line). Assuming provider j has 15% market share, 10% of the
market’s WTP, and charges $1.2 per unit of intensity. The rest of the providers charge $1 per unit
of intensity and the average intensity per enrollee is 450 units of intensity.
55
Figure 4-6. A plot of an example insurer’s per enrollee profits with and without a provider j.
To estimate the total change in the insurer’s market share with and without provider j assuming
uniform cost sharing levels among all plans, the following function is used:
𝑀𝑆 , = 𝑀𝑆 , ∗ 1 − 𝑊𝑇𝑃 𝑃𝑟𝑒𝑚 1
𝑃𝑟𝑒𝑚
This function specifies that the change in insurer market share is a function of the remaining WTP
captured in the insurer network, the change in premium, and the current insurer market share with
the provider.
Provider Disagreement Tradeoff
For provider j, the change in profits due to a disagreement with insurer l is modeled as follows:
∆
𝑃𝑟𝑜𝑓𝑖𝑡 = 𝐴𝑣𝑔𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦 ∗ 𝑀𝑆 𝑀𝑆 𝑝 − 𝑐𝑜𝑠𝑡 , − (𝑀𝑆 − 𝑀𝑆 ) 𝑝 − 𝑐𝑜𝑠𝑡 ,
where 𝑐𝑜𝑠𝑡 , is the cost per unit of intensity at the provider’s current effort level and 𝑝 is
the current average market price. This formulation represents the change in profits as the profits
made from reaching an agreement at price p less the profits made from those enrollees switching
away from insurer l to receive care at provider j for the current average price.
Finding the equilibrium price to guide provider price direction
Using the above formulations provider and insurer disagreement tradeoffs, the equilibrium price
offered by each provider j to insurer l can be obtained under Nash bargaining as the bilateral price
56
that maximizes the Nash product of computed as the argument that maximizes the agreement
surplus. Therefore, the price 𝑝 ∗
is the price that maximizes the product of the disagreement payoff
such that
𝑝 ∗
= arg max
, ∆
𝑃𝑟𝑜𝑓𝑖𝑡 ∗ ∆
𝑃𝑟𝑜𝑓𝑖𝑡
Where 𝛽 ∈ [0,1] is bargaining power parameter for the provider’s relative bargaining leverage
over insurers. For simplicity, providers market share, 𝑀𝑆 , is used as the parameter 𝛽 . Figure 4-7
below plots the function ∆
𝑃𝑟𝑜𝑓𝑖𝑡 ∗ ∆
𝑃𝑟𝑜𝑓𝑖𝑡 over price for a same provider that
has 15% of the markets WTP but different 3 different market shares [0.1, 0.3, and 0.5] in market
with 3 providers, average insurer enrollee intensity of 350, provider operations costs as 0.8 per
unit of intensity, and an average price per unit of intensity at $1 offered by other providers.
Figure 4-7. Plot of optimal prices in reference to the $1 per unit of intensity baseline to illustrate the effect
of provider market share on equilibrium prices.
As shown in the example above, the optimal price is higher than the current market average price
when the provider has 𝑀𝑆 ≥ 0.3, but an optimal price that is lower than the average market price
when 𝑀𝑆 = 0.1. In this plot, the effect of the bargaining parameter is evident, as the provider
market share increases, the parameter 𝛽 increases and ∆
𝑃𝑟𝑜𝑓𝑖𝑡 decreases approaching 1,
which in turn reduces the product of the disagreement tradeoffs. The graph above is used to obtain
the cut-off provider MS above which the provider can increase price. If the provider does not meet
this MS threshold, then the price offer decrease. The new price is obtained using the following
sequence:
57
1. For each provider-insurer, the cutoff provider MS, MScutoff, is obtained by finding the MS
at which the optimal price of the Nash product exceeds the current price.
The MScutoff, is obtained for the provider’s WTP, average market price, average
intensity per enrollee for insurer j, and provider cost
2. If the current provider MS > MScutoff, PriceDirection = +1, else PriceDirection = -1
3. Set 𝑃 , = 𝑃 , + 𝑃𝑟𝑖𝑐𝑒𝐷𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 1 − 𝑂 ∗ 0.1
These steps move the price offer an amount proportional to the provider’s profit orientation in the
direction of the equilibrium price obtained from the Nash product solution. This formulation can
be interpreted that more profit oriented providers are more aggressive at adjusting their prices to
market conditions compared to more qualitied providers. Provider quality orientation, Oj, range
from 0.3 to 0.7, effectively limiting the yearly price offer change to ±7% of the baseline price.
4.2.4.2 Plan Network Selection of Health Providers
Insurers choose to accept or reject the price offers that are the guided by the Nash product to the
bargaining problem. The network offered by insurer l is a group of providers formed based on the
value the insurer determines in contracting with this provider. Therefore, each insurer evaluates
the utility of contracting with a provider based on the provider’s attributes, prices, and insurer’s
risk orientation. Effectively, insurers compute a minimum utility threshold, then contracts with
providers whose utility is higher than this threshold. This formulation is acceptable since there are
no contracting costs for each provider. For an insurer, the utility of contracting with a provider j is
a function of this provider’s quality attributes, WTP contribution, and price. An insurer’s utility
function for contracting with provider j is as follows:
𝑈 = 𝑂 𝑈 𝐹𝑄 + (1 − 𝑂 )𝑈 𝐹𝜋
Where
𝐹𝜋 = 𝑀𝑆 %,
𝐹𝑄 = 𝑙𝑜𝑔𝑖𝑡 , see Figure 4-8
For a positive constant c, 𝑚𝑜𝑟𝑡𝑅𝑎𝑡𝑒 + 𝑟𝑒𝑎𝑑𝑅𝑎𝑡𝑒 > 0, and 𝑀𝑆 %, ∈ [0,1]. Note that the quality
utility function decreases with higher mortality rates and readmissions rates. Higher mortality rates
translate to lower insurer utility. At higher mortality rates, a small increase in mortality results in
faster decrease in utility. The utility for each provider is a function of this provider’s current market
share and quality measures.
58
Figure 4-8. Plot of insurer utility for a provider's mortality rate depicting the shape of the insurer utility
function.
After, each utility is divided by the price, 𝑝 , , to obtain the utility per dollar. This utility per dollar
for each provider, unique to insurer l, is then used as such:
1. Start with an empty network 𝑛 = ∅
2. Compute
, for all providers 𝑗 = 1, … , 𝐽
3. Find the average value for the above measure and standard deviation
𝑈 = 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 (
, , … ,
, )
𝜎 , = 𝑠𝑡𝑑𝐷𝑒𝑣 (
, , … ,
, )
4. Set insurer i’s threshold 𝑇𝑙 = 𝑈 + 𝑂 ∗ 𝜎 ,
5. If
, ≤ 𝑇𝐻 , add provider j to insurer l’s network 𝑛
This approach guarantees at least one provider in each insurer’s network, which is assumed as the
minimum network requirements for the simplicity of this model. This formulation is also
consistent with existing laws such as “Any Willing Provider” laws [168]. These laws require that
all willing providers can be members of carrier’s networks if certain conditions are met.
4.2.4.3 Plan Premium Setting
The ACA limits individual premium rating among enrollees and prohibits adjusting for pre-
existing conditions. Therefore, in the context of this model, since all consumers are assumed within
the same IRA, insurers charge consumers based on their age only.
For the network of health care providers chosen in the plan network selection module, plans
maximize their profits by selecting the optimal premium and cost sharing levels to increase
enrollment. In this model, only model Silver tier PPO plans are modeled, which cover
approximately 70% of health care costs. Higher premiums increase per enrollee profit but decrease
59
overall enrollment in plans. Lower premiums increase enrollment and market share, but reduce per
enrollee profit. Therefore, plans select the optimal preml to maximize profits such that:
𝑝𝑟𝑒𝑚 = 𝑎𝑟𝑔 max
⎝
⎜
⎛
𝑑𝑒𝑚𝑎𝑛𝑑 , ∗ 𝑃𝑟𝑒𝑚 − 𝑀𝑆 ∗
∈ 𝑁𝐸 , ∗ 𝐸𝐼 , ∗ 𝑝𝑟𝑖𝑐𝑒 , − 𝑂𝑃 , ⎠
⎟
⎞
In this previous formulation, premiums change the demand dmnd for a plan, the profit margin
[prem-AVC]. Maximizing profits can be achieved through maximizing demand and maximizing
the margin. However, there are correlation between increasing profit margins and decreasing
enrollment. Therefore, the optimal levels of premiums are determined by maximizing for the above
function. If cost-sharing levels are uniform across all plans, then demand is assumed to be directly
proportional to the aggregate WTP for the network of insurer i minus the premium charged.
Therefore, profits are proportional to:
max 𝑃𝑟𝑜𝑓𝑖𝑡 , = (𝑘 ∗ 𝑊𝑇𝑃 %
(𝑛 ) − 𝑝𝑟𝑒𝑚 )
( )
∗ 𝑝𝑟𝑒𝑚 − 𝐴𝑉𝐶 ( )
for a given network 𝑛 where k is a positive constant. Because (i) is proportional to 𝑑𝑒𝑚𝑎𝑛𝑑 and
(ii) is the average profit: revenue minus average costs. Demand is correlated with the percentage
of WTP for providers captured in the insurer’s network with a coefficient adjusting for the number
of providers in a market, 𝑘 ~log (#𝑜𝑓𝑃𝑟𝑜𝑣𝑖𝑑𝑒𝑟𝑠 ). The above equation is a convex function on
𝑝𝑟𝑒𝑚 , see Figure 4-6. Therefore, using the first-order condition, the value 𝑝𝑟𝑒𝑚 ∗
that maximizes
the above function is:
𝑝𝑟𝑒𝑚 ∗
=
𝑘 ∗ 𝑊𝑇𝑃 %
(𝑛 ) + 𝐴𝑉𝐶 2
Where 𝑊𝑇𝑃 %
(𝑛 ) is the aggregate WTP captured in insurer i’s network, and 𝐴𝑉𝐶 is the average
variable cost, or the expected intensity for each enrollee times the expected price per unit of
intensity offered by providers. Assuming patients flow to each provider is proportional to the
provider’s market share, the 𝐴𝑉𝐶 is obtained by multiplying each provider’s price with the
provider’s proportional market share in the network.
Estimating Module Parameters and Ranges
To estimate the constant k, it has been shown that median premiums for bronze, silver, and gold
plans were 15% higher than narrow plans [169]. Narrow plans have 31% to 70% hospital
particiation [169]. Assuming narrow networks capture 31% to 70% of aggregate WTP for hospitals
in a rating area and AVC is the same for narrow and broad networks, 𝑘 ∈ [0.2432 ∗ 𝐴𝑉𝐶 ,
0.7692 ∗ 𝐴𝑉𝐶 ].
The range for premiums for bronze PPO plans in health insurance exchanges (for individuals aged
27) is between $150 and $750 between the years 2014 to 2017 obtained from the HIX dataset
[155]. This range is used to standarized the AVC for individuals to the percentage of aggregate
60
WTP captured in the plan’s network, AVC ranges obtained are between 200 and 1400 dollars per
person. At baseline prices where price per unit of intensity is $1, this will be equivalent to the
2013-adjusted average cost per individual generated from RAND HIE estimators.
Intial Insurer Settings
Consumers are randomly assigned to an insurer at model initialization. After year 0, consumers
compute their WTP for networks and assess insurer offerings, and switch according to their
preferences. At year 0, consumers do not any insurer switching costs. All premiums start with a
premium of 400 (for individuals 27 years of age), and then adjust their premiums the following
years according to above descirbed optimization problem.
Insurer Optimal Premium Adjustment
Since the optimal premium for an insurer does not account for other insurers’ offerings in the
simulated health care market, an ad-hoc adjusting mechanism is introduced. The goal of this
adjustment is to increase the value offered by an insurer if an insurer has a very small market share
and charges more than other insurers per WTP captured in their network. Effectively, a condition
is introduced to reduce premiums for the chosen network if the insurer has a market share that is
less than 1/(# of Insurers +1) in the market and their %WTP/premium captured is less than the
market average.
This concludes the provider-plan interaction in the context the simulated health care market.
Insurers form networks and set premiums based on provider prices and effort decisions.
Consumers then choose a plan and obtain care as needed, choosing a provider at each episode
depending on episode severity and their preferences. ACO formations and care coordination layers
on top of the above discusses modules and mechanisms and will be described in the following
section.
4.3. Yearly Interaction and Event Sequence
This section defines the sequence of interactions that occur in the health care market in a given
year. The interactions are sequential to reflect the nature of decisions and agreements in actual
health care markets. A similar, but less comprehensive, sequential economic model was studied
by Ho (2009) to understand plan network formations [64]. The sequence proposed here is an
extended and stylized adaptation of the sequence studied by Ho (2009). In each step, the agents
involved utilize decision modules that were described above. Each year t, the agents in the system
engage with each other in the sequence described below.
1. Patients calculate utility and WTP for all providers
Consumers estimate their perceived WTP all providers
2. Providers set effort levels
Providers develop their intentions changing their effort, or investments, before the end the
year. The decision involves current market norms, profit changes, and quality changes a
change in effort.
3. Providers make price offers to plans
For each provider-plan pair j,l, pricej,l is computed guided by the Nash solution to the
bargaining problem. The price offer is a function of provider market share, WTP, current
market prices, and patient utilization.
4. Plans choose their network of health care providers
61
Plans review price offers from providers and ACOs for inclusion in their network. Plans
choose providers according to maximize their utility on provider prices and their health
outcomes. For each plan l ∈L, a network of provider nl ⊆J is formed.
5. Plans set premiums
Once the network of providers by plan l, then l optimizes preml to maximize profits as
described in the premium and out-of-pocket cost setting module.
6. Consumers choose a health plan
Consumers use provider WTP and preferences to compute the most suitable plan during
the open enrollment period as per the consumer choice module.
7. Sick patients obtain care
For every episode generated using the health needs module, patients obtain care from their
choice of provider j in nl. The probability that a patient chooses provider j for episode EI
is outlined in the patient utility and WTP module.
8. Payment reconciliation and outcomes reporting
At the end of the year t, providers and the ACO receive the agreed upon payments for each
episode delivered during the year. Savings from ACOs are computed and shared according
to agreed structure. Providers and plans learn more on the perceptions of consumers and
other agents.
9. Population demographics and health conditions are updated.
Demographic and health conditions are updated and new consumers are born and existing
consumers age, progress through conditions, and die.
4.4. Observable Outcomes
The simulation model proposed in this research produces extensive and rich datasets on outcomes
of the model’s run. The simulated markets would be observed for relationships between agent
decisions, outcomes, prices, and market characteristics. For instance, anti-competitive provider
behavior is captured in this model by relating changes in provider prices and effort levels.
Outcomes that can further deepen the understanding of ACOs, market conditions, and competition
will be assessed. In this section, observable outcomes that would be examined at the end of each
simulation run are outlined to understand the potential relationship between market conditions and
anti-competitive concerns from ACO formation.
1. Health Outcomes
Observable health outcomes refer to health outcomes and utilization across the different
simulated health care markets.
2. Effort Levels
Effort levels and how efforts change over time are of particular interest in this ABM of
health care market. As mentioned above, if low effort levels are associated with increased
reimbursements in ACO formations in particular market condition compared to an identical
baseline market where no ACO formations are allowed, then this indicates potentially
observed anti-competitive behavior by forming ACOs.
3. Health care Payments
Payments that make up the cost of health care are observed. Prices set for provider services,
premiums set by plans, and consumer expenses are of particular interest. It is necessary to
observe how prices, premiums, and total expenditures change in different market
conditions, and how they respond to ACO formations. ACO arrangements typically
involve shared savings agreements, where savings are divided by plan and provider. These
62
savings will be observed under different market structures to understand the impact of
market settings on ACO effectiveness.
4. Competition statistics
The competitive landscape can be described via several different indices. In this model,
provider market share, HHIs, and prices are observed for indication of market power.
Increased prices indicate that provider gain higher margins, and this is expected to correlate
with higher HHIs and market share. Insurer market shares are also of interest to observe if
higher insurer market power translates into pressure to reduce provider prices or into higher
premiums.
4.5. Base Model Verification and Validation Methods
To ensure the model is accurately modeling the system of interest and is behaving in the manner
intended, the ABM of the health care market undergoes a series of validation and verification steps.
This section outlines the methods used to test how closely the modeled health care market
functions compared to actual health markets and whether components and modules are performing
the way they are expected to perform.
4.5.1. Verification Methods
The verification process refers to ensuring that components and modules in the ABM behave the
way they are intended to behave. To help ensure that, aspects of the model are compared to data
and parameters used in initializing the model. The verification procedures are intended to compare
the envisioned module behavior to the actual module behavior. The verification component of this
model is divided into the verification procedures for each agent and their corresponding modules.
Consumer Agent
Verification for the patient agent involves comparing the sampled agents to the source dataset
(NHANES 2013-2014) and the generated utilization to the expected utilization from the RAND
HIE. A visual inspection of the distribution of the attributes of the modeled computation and those
of the NHANES sample should be sufficient to verify that the sampling mechanism is working as
intended.
Also, the generated utilization rates and costs will be compared to the expected utilization
estimated by the RAND HIE estimates. While the modeled population differs from the population
studied in RAND HIE, it is possible to perform calculations to obtain estimates relating to the
population in the model.
The model is equipped with a visual graphical interface where it is possible to inspect every
consumer agent to examine the utility and WTP modules. Therefore, the utility and WTP modules
will be verified to ensure that utility and WTP respond to provider distance, for instance. The
verification is performed on the all aspects of the utility and WTP module in the consumer agent.
The verification also extends to the conditional logit choice model that is based on the WTP.
Effectively, the conditional logit choice model will be examined to ensure that the probabilities
generated corresponding to the right WTP values and the current provider network available to the
consumer agent. In addition, other verification checks include comparing the average distance per
episode patients travel across markets of different size and number of providers. More providers
63
in a market with a constant area should translate into less distance traveled per episode, while
larger areas should be correlated with more distance traveled per episode.
Provider Agent
The provider agent modules are verified by examining the expected relationships between provider
related outcomes. The relationships that will be examined relate the verifying the relationship
between the following components:
Provider effort and provider cost
Provider effort and provider outcomes
Consumer utilization and provider costs
The relationships between the outcomes listed above are governed by the mathematical equations
listed in various subsections above. Therefore, these relationships will be compared against these
equations.
Aspects of the provider pricing and effort decision modules will be inspected visually through
taking instances of the model to ensure that their components work as represented mathematically.
This inspected involves freezing the model run and examining the model generated expectations
and forecasts in the decision models and comparing them to computed outcomes.
Insurer Agent
The verification of insurer modules involves evaluating the relationships between components of
the premium setting and network selection modules. For instance, the optimization problem
involving the premium setting reveals a relationship between premiums and WTP captured in the
provider network. To verify these expected associations, the following relationships will be
examined:
Insurer premiums and number of provider in insurer network
Insurer premiums and patient utilization
To investigate the impact of utilization on insurer premiums, patient utilization is artificially
inflated using a UtilizationMultiplier where the rates of each episode type is multiplied by this
factor. A UtilizationMultiplier = 2 means the rates of hospitalizations, acute, and chronic episodes
are doubled for every patient in every year, raising the average variable costs from the insurer’s
perspective. It is expected that increasing utilization would increase premiums, as shown in the
premium setting module in 4.2.4.3. Assessments of the network formation module will be
inspected visually to identify if the utility modules and the selection mechanisms work as intended.
These inspections would compare the model utilities to computed utilities, given the runs current
conditions.
4.5.2. Validation Methods
The validation process refers to ensuring that the model is representative of the intended system
specified in the objective of this study. To validate the competitive relationships observable in the
model, a comprehensive review of relevant literature is performed. The associations observed in
simulated health care markets are compared to empirical associations in studies of actual private
health care markets. This comparison is aimed at establishing the face-validity of the model.
Because of the complex and generic nature of the model, pursuing more elaborate methods of
64
validation, such as cross validation, requires more modeling precise populations, conditions, and
market characteristics.
Empirical studies have been identified systematically to gauge the strength of evidence and
consensus on relationships between any two outcomes in health care markets. For example,
consider the relationship between provider concentration and service prices. Several studies have
found a positive correlation between provider concentration and provider prices in private markets.
Studies that investigate such relationships have been systematically identified and compiled to
evaluate the strength of evidence on this association.
The categorical search is performed on Google Scholar to obtain peer-reviewed publications on
private health care markets that study the relationship between primary (market structure and
concentration measures) and observable outcomes. Publications on national level data and from
the year 2000 onwards are considered for assessing the face-validity of the model. Table 4-7
summarizes the results of the search
Table 4-7. Systematic search on primary variable and observable model outcomes to establish model
face-validity.
Agents Outcome 1 Outcome 2 Relationship Source
Provider-
Individual
Providers
Concentration
Patient
Expenditures
Physician Organization concentration is associated
with higher physician prices
Schneider et al
(2008) [51]
Physicians in more concentrated markets charge
higher service prices
Dunn & Shapiro
(2012) [1]
Insurer-
Individual
Insurer
Concentration
Patient
Expenditures
Insurance market concentration is inversely related
to spending
McKellar et al (2013)
[57]
Health plan concentrations not significantly
associated not payer price difference
Schneider et al
(2008) [51]
Insurer-
Provider
Insurer
Concentration
Provider
Prices
Increases in insurance market concentration are
significantly associated with decreases in hospital
prices
Moriya, Vogt,
Gaynor (2010) [170]
Insurance market concentration is inversely related
to hospital prices
McKellar et al (2013)
[57]
Higher health plan concentration is associated with
lower hospital prices
Glenn et al. (2011)
[58]
Insurance carriers in more concentrated health
insurance markets pay lower fees to physicians
Dunn & Shapiro
(2012) [1]
Health plan concentration does not appear to be
significantly associated with higher outpatient
commercial payer prices
Schneider et al
(2008) [51]
No evidence that Insurance premiums grow faster in
more concentrated insurance markets
Dafny et al. (2012)
[56]
Insurer-
Provider
Provider
Concentration
Insurer
Premiums
Insurance premiums are higher where insurance
markets are more concentrated
Trish and Herring
(2015) [171]
Insurer-
Insurer
Insurer
Concentration
Insurer
Premiums
More insurers mean lower premiums
Dafny et al. (2015)
[172]
There is no statistically significant correlation
between Insurer HHI and variation in health plan
network breadth
Dafny & Ody (2016)
[173]
Additional insurer carriers reduce premiums in a
rating area
Samuel et al. (2015)
[174]
65
As the number of insurers in a rating area fall,
average and median premiums are higher
Holahan et al. (2017)
[175]
Increases in provider market concentration are not
significantly associated with changes in hospital
prices
Moriya, Vogt,
Gaynor (2010) [170]
Insurer
Concentration
Insurer
Network
breadth
Variation in hospital prices increase with increased
hospital market power
White et al (2013)
[52]
4.6. Base Model Run Settings
After the model is verified, scenarios are run to establish model face validity, and then to test the
impact of the illustrative ACO formation on the modelled competitive landscape. This subsection
elaborates the settings for the base model. The main premise for the scenarios is to vary the market
structure and IRA size and observe the changes in model outcomes. Essentially, the number of
insurers, providers, and consumers in the rating area are parameterized along with the physical
size of the rating area. The outcomes, outlined in subsection 4.4, are to be observed in relation to
these parameters.
The initial settings and parameters varied in the base model experiment are aimed to create a
heterogeneous set of hypothetical health care markets to examine the range of possible model
outcomes and trajectories. Hence, some parameters are considered model parameters to investigate
their effects, while others might be varied as a source of inherent market heterogeneity.
Market Structure Scenarios
This presents the main lever of change in the ABM health care market simulation model. In this
group of scenarios, different market structures are imposed on the health care market. It is assumed
that a fixed population with predetermined demographics populates an IRA of size m x m. For
each scenario, a set number of consumers I, plans J, hospitals K, interact in the health care market
of size M. The previously mentioned parameters are varied according to the following ranges and
steps:
[3, 7] Providers, varied in steps of 1
[3, 7] Insurers, varied in steps of 2
{30,000, 45,000} consumers
{200, 400} area of health care market
This base model is initialized with several initial conditions and heterogeneity among agents. For
instance, consumer and provider agents are scattered randomly in the hypothetical IRA, making
their location a form of heterogeneity. Table 4-8 below summarizes the source of heterogeneity in
the experiment used for model verification.
Table 4-8. Summary of sources of model heterogeneity associated with model development and agent
decision making.
Agent Aspect of
Heterogeneity
Details Source
Consumers
Location Uniform randomly scattered in
area
Assumed
66
Attributes Sampled from NHANES [149]
Provider and
Insurer Switching
Cost
Calibrated to match switching rates
– Equivalent to default bias
Provider Switching: [117]
Insurer Switching [165]
Initial insurer Uniform random across all insurers Assumed
Insurer
Quality Weight uniform (0.3,0.7) - Higher weight
means more quality oriented
Assumed
Provider
Location Uniform randomly scattered in
area
Assumed
Quality Weight uniform (0.3,0.7) - Higher weight
means more quality oriented
Assumed
Model Initialization
At initialization, the model assigns consumers to arbitrate insures and omits switching cost from
the first open enrollment period to allow free switching in the first year. Model initializes with
insurer premiums set to $400 for a 27-year-old individual, as approximated from HIX dataset on
ACA health insurance exchanges in the year 2013 for Silver PPO plans [155]. Provider price factor
is initialized at 1 per unit of intensity, and initial effort level is set to 2, the medium level. Other
model initialization settings can be found in each agent’s corresponding subsection or in the
comprehensive parameter settings in appendix A.
As the model evolves, starting with the initial parameter and conditions outlined above, there are
several sources of stochasticity that may alter the trajectory of health care market. These sources
of stochasticity are described in Table 4-9 below.
Table 4-9. Summary of sources of model stochasticity associated with model development and agent
decision making.
Agent Aspect of
Stochasticity
Details
Consumers
Choice of Provider Conditional Logit Choice Model – Estimated & assumed
parameters from relevant literature
Choice of Insurer Conditional Logit Choice Model – Assumed & calibrated
structure of choice, based on provider WTP
67
Episode Rates &
Intensity
From RAND Health Insurance Experiment Data
Episode Frequency: Negative-binomial distribution generator
Episode Cost: Exponential distribution generator
Episode Outcomes CDC hospital mortality and readmissions data (1999-2015)
Insurer
Network of
Providers
Utility maximizing network formation based on provider
prices, quality, and WTP
Premium
Adjustment
Premium Reduction for below average value of network and
low market share
Provider
Effort Level Based on theory of planned behavior
Price Offer Equilibrium guided price direction + random multiplier
This describes the model initialization, heterogeneity, stochasticity and scenario parameters that
are modeled. The aim of this experiment to generate comprehensive market conditions to assess
the model’s face validity by comparing the modeled relationships to empirical and theoretical
relationships. The model will be run for 5 years, and observed during the last 3 years in each run.
Since the model does not have an elaborated disease progression and demographic evolution
module, the modeling time is set to be less than 5 years to mitigate the impact of demographic,
etiological, and epidemiological changes. The model is observed for 3 years to allow agents to
depart from initial assignments, learn the market, and perform key decisions. The model is
assumed to ‘warm-up’ in first two modeling year and the trajectory of key model outcomes
stabilize over the last 3 years constituting a ‘steady-state’. The trajectory of significant model
outcomes over modeling years can be reviewed in appendix B. For each configuration of market
structure {providers, insurers, consumers, area}, 5 replicates are run.
A flow chart of the model sequence has been developed to outline the sequence of events of the
model in more details. The sequence outlined in 4.3 is illustrated in the Figure 4-9 below.
68
Figure 4-9. Simulation sequence illustrated in a flow chart. Each loop constitutes a year, and the model is
observed for 5 years in total, observing only the last 3 years.
69
4.7. Case Study: Accountable Care Organizations
In the base model, providers compete and offer direct alternatives for each other’s services. This
competition is hypothesized to reduce prices, spread WTP, and improve efficiency [176].
Continuous care provision by integrated provider networks is also hypothesized to improve care
delivery and costs. This improvement it achieved through reducing redundant procedures,
providing better preventive, and discharge planning. In the model, ACO formation is a cooperative
model that can be used to achieve the hypothesized improvements. This model examines a case-
study on an illustrative ACO definition and illustrative effects. This subsection operationalizes the
ACO arrangements in the simulated market outlined in 3.2.2.2.
4.7.1. ACO Structure and Arrangements
The model is aimed at investigating the impact of ACO formations on the competitive landscape
of private health care markets. Hence, in the above described market, an ACO formation is induced
such a group of providers participate in an accountable care arrangement that is offered to insurers.
The number of providers that participate in the ACO is referred to as 𝑆𝑖𝑧𝑒 ∈
[2, # 𝑜𝑓 𝑃𝑟𝑜𝑣𝑖𝑑𝑒𝑟𝑠 − 1]. One of the participating providers sponsors the ACO, and the sponsor
enters the price setting process, with an aggregated WTP, MS, and costs. Consequently, the
resulting price offer will be higher than from the ACO compared to separate provider entities. The
only role of the sponsoring provider is to enforce its profit/quality orientation Oj on the magnitude
of price change in the resulting price direction described in 4.2.3.1 for all providers in the ACO.
Additionally, insurers evaluate the ACO as a single entity such that an agreement will be made
with the whole ACO, if an agreement were to be made.
4.7.1.1 ACO Investments
Being part of an ACO introduces a fractional increase in each participating provider’s effort equal
to ACOeffortIncrease. The ACOeffortIncrease is a multiplier set to 1.05, equating to a 5% increase
in effort. This is an illustrative multiplier on a conceptual effort variable, but the actual change on
probability of adverse outcomes, obtained from subsection 4.2.2.2, are computed in Table 4-10
below:
Table 4-10. Table of changes in variable costs and probability of adverse outcomes due to ACO
participation at each effort level.
ACO
Effort
Level
ACO
Cost(effort)
ACO
EffortModifier
Non-
ACO
Effort
Level
Non-ACO
Cost(effort)
Non-ACO
EffortModifier
∆
%
Cost(effort)
∆
%
EffortModifier
1.05 0.5689 1.1404 1 0.5625 1.1547 1.14%
Increase
1.23%
Reduction
2.1 0.7756 0.8954 2 0.7500 0.9156 3.42%
Increase
2.20%
Reduction
3.15 1.1202 0.7074 3 1.0625 0.7321 5.43%
Increase
3.37%
Reduction
The diminishing marginal returns (quality improvements) on costs for ACO participation at each
level are evident in Table 4-10. The percentage change ∆
%
have been obtained by looking at the
percentage difference in the EffortModifier and Cost(effort) variables between 1 and 1.05, 2 and
70
2.1, and 3 and 3.15 effort levels, see Figure 4-5 and cost equation in 4.2.2.1. The illustrative ACO
effects on effort are set to 1.05 because a smaller effect means the change in quality might be
statistically undetectable, while a higher effect might inflate the difference. The values in Table
4-10 represent reasonable illustrative effects to explore possible changes in balances and outcomes
in the simulated health care markets.
4.7.1.2 ACO Care Continuity
ACO formations deliver the full continuum care in this model. The illustrative ACO formation can
reduce the likelihood of adverse outcomes through the increase in effort and through continuity of
care. This improvement is realized in an episode delivered at an ACO provider if the patient has
received care at any of the ACO providers in the past year. The CareContinuity multiplier with a
value of 0.95 reduces the probability of adverse outcomes when the previous condition is met. This
is equivalent to a 5% reduction in the probability of adverse outcomes for patients that are part of
the ACO panel. The multiplier also reduces the rate of hospitalization by 5% if a patient’s preferred
provider is an ACO provider. The rates generated using the regressions from the RAND HIE for
hospitalizations and acute episodes are also multiplied by this multiplier. This reduces the rate of
hospitalizations by ACO patients by 5% if the patient consistently sees the ACO provider.
The continuity of care multiplier means that the actual difference in mortality and readmissions
rates between ACO and non-ACO providers can be larger than the change explained by the
difference in effort. Effectively, these improvements would be higher if more patients consistently
receive care at ACO providers. Using a density index for the continuity of care, the more care is
concentrated in the ACO, the better the outcomes. Benefits of continuity of care are realized,
regardless of ACO effort. This is for two reasons, (1) impact of effort on quality is captured in the
provider operational module, and (2) care continuity may theoretically be useful without additional
investments. Having more comprehensive patient records reduces duplicate testing, medical
errors, and facilitates discharge planning. Hence, impact of care continuity is realized when more
care is concentrated at ACO providers, regardless of the ACO effort. The benefits of care
continuity combine with incremental increase in ACO provider effort (ACOeffortIncrease) to form
the net ACO benefit.
4.7.1.3 ACO Shared Risk & Attribution
ACO arrangements typically involve a shared risk agreement where the risk from reduced
utilization are shared between insurers and plans [25], [74]. These agreements involve patient
panel, tracked expenditures, and a shared rist rate. This subsection outlines the conditions that
make patients part of the ACO panel and how expenditure targets. Patient attribution to an ACO
can take multiple forms: prospective and retrospective attribution. Concerns with the different
attribution methods have been discussed in 2.2.3.1. This model utilizes retrospective attribution
with AttributionCriteria = 50%; such that a patient that receives more than 50% of their care at an
ACO provider is considered part of the ACO panel. At the end of the year, patients in the ACO
panel have their expenditures aggregated and compared to the expenditures of the remaining
patients in the market being served by non-ACO providers.
The SharedSavings rate is set to 50%. This means that if per patient expenditures of the ACO panel
is higher than their non-ACO counterparts, then providers pay 50% the difference to insurers. If
ACOs reduce expenditures for its patient panel compared to non-ACO patients, then the insurers
pay 50% of the saved amount to the provider. This reconciliation occurs are the end of the year.
71
In summary, ACO formations are parameterized by the number of participating providers
(𝑆𝑖𝑧𝑒 ) , increase in effort (ACOeffortIncrease), retrospective attribution criteria
(AttributionCriteria), ACO savings rate (SharedSavings), and improvements due to care continuity
(CareContinuity). The illustrative ACO formation parameters are highlighted in the Table 4-11
below:
Table 4-11. ACO parameters, values, and short description.
Parameter Value Remarks
Size ACO 2 At least two providers can make an ACO
ACOeffortIncrease Multiplier 1.05 Translates into increase in costs and improvement outcomes.
AttributionCriteria 50% The percentage of utilization delivered at an ACO provider to
include a patient in the ACO panel
SharedSavings Rate 50% Percentage of risk shared by providers on the average costs
per ACO panel patient
CareContinuity Multiplier 0.95 Improvement in episode outcomes due to patients
consistently receiving care at ACO
4.7.2. ACO Formation Verification Methods
The verification procedures would insure that the ACO pathways function as intended. There are
several different facets to verify: (1) the bargaining effects, (2) the increased ACO costs, and (3)
improved quality.
ACO formations allow the participating providers bargain as a single entity, therefore, the
bargaining process is observed to ensure that the prices are determined using the combined ACO
market share and WTP. In addition, the costs per episode should be higher for ACO providers
because of the fractional increase in effort. Tests will be performed to compare the expected
provider costs difference to the modeled provider costs. Also, the outcomes are modeled to be
better for ACO providers due to the increased effort and continuity of care pathways. Therefore, a
test will be run to verify that higher ACO market share translates into better outcomes through the
continuity of care pathway. Because the continuity of care is modeled and measured using the
density index, ACO provider outcomes should be better at higher ACO market shares.
4.7.3. ACO Experiment Settings
Once the model has been rigidly tested for consistent behavior and outcomes, scenarios are
designed to test the hypotheses and attain the model objectives. This study aims to understand the
market context under which ACO formations might be concerning. Each scenario will be run for
two identical markets, one where ACO formations are induced and another baseline market where
there are no ACO formations. The following general multi-level framework of scenarios is
outlined to achieve a better understanding of market conditions and potential anti-competitive
ACO formations.
The ACO experiments extend the base model experiment discussed in the prior subsection.
Effectively, the scenario parameters, heterogeneity, stochasticity, and sequence of the model
remain the same. However, two providers now coordinate in the form of the ACO with effects and
behaviors explained in 4.2.5. In each modeled market, two arbitrary providers form the ACO, with
72
the retrospective attribution criterion set at 50%, shared risk rate of 50%, 5% increase in effort
levels, and improved outcomes through care coordination.
Levels of Analysis for ACO Effects
The effects of ACO formations will be inspected from two perspectives; (1) overall market effects
and (2) within-ACO-market effects. For each market setting, the impact of ACO formations is
separated by comparing markets with ACOs to baseline markets. This comparison explains the
market-level effects of ACO formations with respect to the observable outcomes listed in 4.4. In
addition, a comparison within ACO markets will be performed to compare ACO providers with
non-ACO providers in ACO markets. The comparison would also extend to compare ACO
performance in different market settings. This comprehensive analysis would deconstruct the
inherently complex nature of heterogeneous health care markets and ACO formations from
through all the avenues that this model enables.
4.7.4. ACO Sensitivity Scenarios
Effects of the illustrative ACO experiment parameters can be further understood through
investigating the sensitivity of the observable outcomes to key components in the model. In this
subsection, a procedure to investigate the impact of various aspects of the model on the impact of
ACO formations. The parameters that are varied in this analysis are grouped by agent. First,
parameters relating to ACO formations are varied, then parameters relating to consumer
preferences, and lastly, provider quality orientation. Due to the large number of parameters and
outcomes, only a subset of parameters are varied. These parameters are chosen because of their
practical or theoretical importance. This sensitivity analysis is classified as a one-way sensitivity
analysis, meaning that only parameter is varied at a time. Interactions between studied parameters
are elusive in one-way sensitivity analyses, but the quantity of parameters and outcomes make this
task beyond the scope this analysis. However, the interactions between the parameters studied in
the sensitivity scenarios and key model inputs (market structure parameters) can be investigated
since each sensitivity scenarios experiment involves varying some of the market structure as well.
Each sensitivity analysis run involves running an ACO market and a baseline market to enable an
assessment of the robustness of the findings to key model parameters.
A sensitivity analysis experiment is run for each parameter discussed in the following subsections.
In each experiment, the number of providers, number of insurers, area, and population sizes are
varied, as with the base model setting discussed in section 4.6. However, for brevity, all market
structure parameters will be varied across two settings only; high and low, as shown in Table 4-12.
Table 4-12. Settings for market structure parameters in sensitivity analysis experiments.
Market Structure
Parameter
Varied? Low Setting High Setting
Providers Yes 3 7
Insurers Yes 3 7
Consumers No- 35,000
Area Yes 200 400
For each configuration of value of the parameters shown in Table 4-12 along with a value for the
parameter of interest in the sensitivity analysis, 5 replicates are run. The parameters studied in this
73
sensitivity analysis can be categorized into the following three categories; ACO parameters,
consumer perception parameters, and provider and insurer quality orientation parameters.
4.7.4.1 ACO Parameters
The ACO is parameterized by 4 variables; 2 structural parameters (Number of participating
providers and shared savings rate) and 2 operational parameters (ACO effort increase and care
continuity improvements). This sensitivity analysis focuses on the operational parameters of the
ACO, and leaves the remaining for a possible structural sensitivity analysis in the future. A subset
of the ACO parameters, are listed in Table 4-11, is varied to understand the robustness of the
results relative to these parameters. Namely, investments associated with ACO and continuity of
care effects. This translates into varying the ACOeffortIncrease and the CareContinuity multipliers
while maintaining the remaining components of the model as is. Each parameter is set to a high
and low value compared to the baseline experiment settings, as shown in Table 4-13.
Table 4-13. ACO operational parameters varied in the sensitivity analysis.
Parameter Baseline in
ACO
Experiment
Low High
ACOeffortIncrease Multiplier 1.05 1.025 1.075
CareContinuity Multiplier 0.95 0.975 0.9
The observable outcomes will be studied for each setting of these parameters to evaluate the impact
of these value. ACOeffortIncrease represents the cost of ACO formations and the potential for
ACOs to improve quality. Effectively, at higher ACOeffortIncrease, ACO formations are costlier
to providers but can achieve better outcomes. In other words, this parameter captures the potential
for ACOs to improve quality, at a higher cost.
Sensitivity of ACO results to the expected utilization of the patient population is also investigated.
In this experiment, consumers in the market have higher utilization rates on multiplicative factor;
UtilizationMultiplier>1. Effectively, the yearly rates of each episode type are multiplied by this
factor for every patient. An UtilizationMultiplier = 1.5, for instance, means that the yearly rates of
hospitalizations, acute, and chronic episodes are now 50% higher. This sensitivity analysis
experiment aims to explore how patient needs can impact the realization of ACO benefits. Hence,
the following values of UtilizationMultiplier are tested: {1.5, 2, 2.5, 3}. For each configuration
shown in Table 4-12 and UtilizationMultiplier, 5 ACO markets and 5 baseline markets are
simulated and compared.
4.7.4.2 Consumer Parameters
Parameters relating to the consumers’ perception of quality of a provider are tested in this test of
sensitivity. Effectively, consumers in these runs are not aware of any provider quality attributes or
strongly favor high quality providers. Therefore, only the distance and attributes of the consumers
themselves dictate their choice of providers when consumers do not have quality preferences.
Aspects relating to consumer perceptions of providers and quality reporting have been a subject of
theoretical and practical literature with conflicting empirical results. Parameters used to model
consumer sensitivity to provider quality are not empirically estimated in this model, therefore, it
is important to evaluate the sensitivity of the model results to these assumed values.
74
4.7.4.3 Provider & Insurer Quality Orientation Parameters
Important decisions performed by providers and insurers are impacted by their quality orientation.
Therefore, it is important to evaluate impact of these parameters on the outcomes. The values of
these parameters are sampled from the uniform distribution for each agent in each run (see Table
4-8). The mean of the uniform distribution from these values are sampled from will be tested at a
higher and lower setting. The value for the high and low settings are shown in Table 4-14.
Table 4-14. Provider and Insurer quality weight settings for sensitivity analysis.
Parameter Baseline Low High
Provider Quality Weight Uniform(0.3,0.7) Uniform(0.1,0.5) Uniform(0.5,0.9)
Insurer Quality Weight Uniform(0.3,0.7) Uniform(0.1,0.5) Uniform(0.5,0.9)
Because these parameters are assumed and influence a multitude of decisions, the sensitivity of
model outcomes to these parameters will be evaluated.
This sensitivity analysis procedure would assess the sensitivity of the results several key
components of the model. Primarily, these parameters are conceptual in nature and their impact on
direct aspects of the model can be predictable. However, given the complexity of the modeled
market, the impact of these variable on observable outcomes and ACO value would reveal
theoretical and practical insights.
5. Results
The simulation model has been built and the experiments detailed in the previous section were run
to verify, validate, and analyze model output. In the results section, an assessment of verification
procedures is discussed first. Model’s face validity was then examined to compare the base model
results to empirical and theoretical relationships in health care markets as outlined in subsection
4.5.1. Lastly, the impact of ACO formations was analyzed on a market-level and provider-level
of. The impact was then tested for sensitivity to ACO parameters and agent-specific attributes.
5.1. Model Implementation Overview
The simulation was developed with visual dashboards to enable swift verification of key aspects
of the model and infer the simulated market’s status at a glance. Snapshots of the outcomes
component of the dashboard can be seen in Figure 5-1, where charts depicted the population size,
access, utilization, mortality rates, and expenditures for an illustrative simulation run. In each
chart, the x-axis corresponds to the simulation year. The size of the population can be seen in the
top-left of Figure 5-1. The top-center figure depicted the average distance patients travel per
episode of care, while a cumulative episode counter was shown in the top-right figure much that
each color represents an episode type. The aggregate costs for each episode type were represented
in figure in the bottom-right. The corresponding expenditures, in the form of premiums and out-
of-pocket expenses, were represented in the bottom-center chart. Lastly, mortality rates were
shown in the bottom-left.
75
Figure 5-1. Snapshot of the outcomes dashboard of the simulation model built on Anylogic v.7.1.2.
A virtual health IRA was also visually represented to verify that agents were uniformly scattered
across the area and the number of modeled providers and insurers match the model inputs. An IRA
generated by the model can be seen in Figure 5-2. In an example market, there were 5 providers
visible in the square area, 2 of which were blue; meaning these providers participated in an ACO.
The dots uniformly scattered throughout the area correspond to consumer agents in the health IRA.
On the left of the area, 4 insurers were visible, and the size of the insurer corresponds to its market
share. Above each insurer and provider, certain visible values reflected current statuses of these
agents, which can be inspected in more details be accessing agent-specific interfaces.
76
Figure 5-2. Snapshot of the virtual IRA as generated by the model using Anylogic v. 7.1.2.
Average market performance and competitive indices were computed yearly and depicted
graphically in the market status component of the visual model dashboard for a simulation run. A
snapshot of market status dashboard is shown in Figure 5-3. In this dashboard, provider and insurer
concentration trends, in HHI, were represented. Average provider prices, effort, and inclusion in
insurer networks were also depicted.
Figure 5-3. Snapshot of the market status component of the model's visual dashboard depicting market-
related measures over the simulation period as generated by Anylogic 7.1.2.
77
Outcomes relating to the health insurance exchanges were summarized at the end of the run to
show insurer premiums and corresponding provider network size. For the illustrative simulation
run, Figure 5-4 shows a summary of the range and average yearly insurer premiums and network
size. Lastly, the run’s observable outcomes were summarized in the end-of-run summary shown
in Figure 5-5, which were computed from the last 3 years of the simulation run.
Figure 5-4. Snapshot of the summary of health exchange related outcomes on insurer premiums and
network size.
Figure 5-5. Snapshot of the end-of-run observable outcomes averaged over the last 3 years of the model
as generated by an illustrative model run in Anylogic 7.1.2.
This overview covers the graphical interface of the model built on Anylogic 7.1.2 to implement
the framework detailed in this dissertation (Chapter 3). The model was built to provide extensive
details on a simulation run. This enables the user to visualize the system being modeled and verify
the functionality and relationships between system components. With the software implementation
of the modeled covered in this subsection, the base model experiment was run and the verification
and validations procedures were performed.
5.2. Base Model Results
The base model experiment involved simulating a series of health care markets with different
market configurations parameterized by {providers, insurers, consumers, area}. Each market
configuration was simulated 5 times to obtain statistical bounds on averages and perform
hypotheses testing. A total of 400 markets were simulated and the data was analyzed. These
78
simulated markets were used to perform the verification and validation procedures, the results of
which are covered in the following and subsequent subsections, respectively.
5.2.1. Base Model Verification
The verification procedures on the base model were grouped by agent. For each agent, the pertinent
modules and decisions were verified against the expected relationships. The key results of the
verification procedures are presented in this subsection, while more detailed verification results
can be reviewed in appendix C.
5.2.1.1 Consumer Agent Verification
The distribution of consumer demographic and health attributes were sampled from individuals in
the NHANES sample. Hence, the distribution of demographic and health attributes in generated
consumer population in the base model were visually compared to NHANES distributions of the
attributes. Indeed, the sampling mechanisms worked as intended when a visual comparison
between the distributions revealed identical shapes and spreads (appendix C).
In verifying the modeled episodes to the expected episode frequency and costs, the model
generated consistent proportions of episodes with those estimated in the RAND HIE [124]. Using
the regressions obtained from Keeler et al. (1988), the estimated costs and frequency and cost of
each type of episode were consistent with those generated in the model. Hence, the health care
needs module functions as intended and accurately captures the regressions from Keeler et al. 1988
[124].
A visual inspection of the agent-specific interface revealed that the WTP and conditional choice
model also indicate proper function. Upon pausing the model, verifications of the real-time
estimates and probabilities were consistent with externally calculated estimates given the current
system settings. Estimates of distance traveled per episode present the expected relationships with
the number of provider options in the market and the size of the area, see Figure 5-6. The distance
per episode decreased as the number of providers increase in a market of a constant area. Larger
areas, which were less dense, had higher average distance per episode.
79
Figure 5-6. Verification of the expected relationship between average patient distance per episode.
The consumer agent modules and expected relationships confirmed that the utilization and choice
modules functioned as outlined in 4.1.2. The various aspects, including sampling attributes,
generating health needs, and decisions, have been verified by inspecting consumer-specific
variables during illustrative runs and analyzing resulting base model relationships.
5.2.1.2 Provider Agent Verification
Provider verification procedures included visually examining the bargaining and pricing behaviors
and observing average relationships between effort, outcomes, and utilization. By closely
examining the provider agent in illustrative runs, modules relating to pricing and effort selection
functioned as intended. External calculations were consistent with effort module computations of
yearly intentions. The pricing module’s cutoff market shares, as described in 4.2.4.1, were verified
to be correct based on the outlined structure, and the price offers worked as intended.
Verifying the relationship between average provider effort and average provider costs (Figure 5-7)
disclosed a relationship consistent with the quadratic cost function over effort in subsection 4.2.2.1.
50 100 150 200 250
3 4 5 6 7 3 4 5 6 7
Area: 200x200 Area: 400x400
Average Distance Per Episode
Average Patient Distance Per Epsiode Over Number of Providers, by Area
Number of Providers In Market
80
Figure 5-7. Verification of the quadratic relationship between provider effort and costs.
The relationship between average provider effort were also verified to be consistent with the
EffortMultiplier plotted in Figure 4-5. Improvements in quality were proportional to 𝑒𝑓𝑓𝑜𝑟𝑡 , and
a scatter plot of mortality rates and average effort revealed a consistent relationship in Figure 5-8.
Figure 5-8. Verification of the diminishing relationship between provider effort and mortality rates.
These verification procedures indicated that the provider agent was indeed functioning in
accordance with the intended mechanisms and modules outlined in Chapter 4.
5.2.1.3 Insurer Agent Verification
Insurer agent verification involved examining the various modules and decisions are described in
section 4.5.2. The key decisions executed by an insurer agent were choosing providers to include
in network and premium setting. The network selection decision was verified by tracing the
200 300 400 500 600
Provider Cost Per Patient
1 1.5 2 2.5 3
Average Provider Effort
Provider Cost & Effort
.4 .6 .8 1 1.2
Mortality Rate (per 100)
1 1.5 2 2.5 3
Average Provider Effort
Provider Effort & Mortality Rate
81
decisions and computations and comparing them to externally calculated insurer utilities for each
provider. The sampled model calculations consistent with external calculations. Once the network
was selected, the subsequent decision by the insurer was premium setting. The modeled
relationship required that insurer premiums be directly related to the WTP captured by the insurer.
By using the number of providers included in insurer networks as a proxy for the WTP captured
in insurer network, the linear relationship between provider inclusion and insurer premiums (from
4.2.4.2) was verified in Figure 5-9.
Figure 5-9. Scatter plot depicting the direct relationship between average proportion of providers included
in insurance plan networks and plan premiums.
Insurer premiums were also modeled to be a function of the average variable costs of their
enrollees. By artificially inflating each consumer’s utilization rates with a multiplier, insurer
premiums should increase in response to verify the modeled relationship between average enrollee
costs and insurer premiums. Figure 5-10 verified the expected linear relationship between insurer
premiums and average variable enrollee costs as described in 4.2.4.3. Higher factor meant higher
utilization, higher per enrollee insurer cost, and consequently higher insurer premiums, verifying
the modeled relationship between utilization and premiums.
.4 .6 .8 1
Proportion of Market Providers in Insurer Network
500 600 700 800 900
Insurer Premiums ($)
Insurer Premiums and Proportion of Provider Included in Networks
82
Figure 5-10. Box plot of insurer premiums for artificially inflated consumer utilization rates
(UtilizationMultiplier factor).
The verification procedures verified that various components of the model performed the intended
tasks demonstrating a program consistent with the desired functionalities detailed in the methods
section. The following step involved comparing the relationship between various outcomes in the
model with theoretical and empirical relationships obtained through a categorical review of
relevant literature.
5.2.2. Base Model Face-Validity Assessment
With the mechanisms and mathematical relationships verified by running the verification
procedures, this subsection compares the base model relationships between competitive indices
and observable model outcomes to those found in empirical literature. To perform the comparison,
relationships and data sources listed in Table 4-7 are grouped into agent-specific categories for an
assessment of the consensus regarding the empirical relations between outcomes. The grouped
relationships are shown in Table 5-1 along with the base model results compared to the theorized
relationships.
Table 5-1. Theoretical and empirical relationships assessed in the base model grouped by agent and
outcome for comparison to base model results.
Outcomes Theorized
Relationship
Sources of
Empirical
Relationships
Strength of
Empirical
Evidence
Supporting
Agreement
of Model
Results
Compared
Comments on
Model Results
500 1,000 1,500
Insurer Premiums ($) for 27 yo
1.5 2 2.5 3 3.5
Insurer Premiums For Artifically Inflated Utilization
Utilization Multiplier
83
Theorized
Relationship
to Theorized
Relationship
Provider
concentration
and service
prices
Higher provider
concentration is
associated with
higher service price
& expenditures
[1], [51], [52],
[177], [178]
Very Strong Very Strong Consistent,
statistically
significant
association across
all simulated
markets
Insurer
concentration
and insurer
premiums
Higher insurer
concentration is
associated with
higher premiums
[170], [172]–
[175]
Strong Mixed Not statistically
significant
association. More
visible relationship
in certain markets
Provider
concentration
and provider
effort
Higher provider
concentration is
associated with
lower quality
[63] Mixed Mixed Statistically
significant
theoretical
relationship
observed in markets
of larger areas
Insurer
concentration
and health
care spending
Higher insurer
concentration is
associated with
lower health
expenditures
[51], [57] Mixed Mixed Statistically
significant
theoretical
relationship
observed in markets
of smaller, densely
populated areas
Insurer
concentration
and service
prices
Higher insurer
concentration is
associated with
lower service price
[1], [51],
[56]–[58],
[170]
Mixed Mixed Not statistically
significant
association. More
visible relationship
in certain markets
Table 5-1 illustrates that the base model produced relationships that were consistent with
relationships observed in comparable empirical studies for most theorized relationships. The
relationships were tested for statistical significance. A “strong” agreement between model results
and theorized relationship in Table 5-1 meant there was a statistically significant relationship
between outcomes assessed. A “strong” agreement between empirical relationships and theorized
relationships in Table 5-1 indicated that most of the categorically-selected studies evaluated found
statistically significant relationship between the two outcomes, consistent with the theorized
relationship. A “mixed” result indicated that empirical relationships were not entirely consistent
with the theorized relationships, or that the base model exhibited statistically significant theorized
relationships in some market conditions.
To elaborate on the results of the base model, the remainder of this subsection represents key
observations that would corroborate theoretical and empirical consistency of this model. The
84
strongest empirical and theoretical relationship observed in private health care markets related to
the provider market power and provider service prices. Most empirical studies use provider
concentration as an index of provider market power. There was a strong consensus on the direct
relationship between provider concentration and provider market prices. Figure 5-11 fitted a linear
relationship between provider HHI and provider price factors in the base model experiment on
which the positive correlation was visible (p-value<0.001).
Figure 5-11. The base model experiment's direct relationship between provider concentration (HHI) and
provider service prices (p-value<0.001).
With regards to the first element in the face-validity Table 5-1, the results of the base model
presented strong agreement with theoretical relationship between the two outcomes because of the
statistical significance of the association. Also, in this element, the base model was consistent with
empirical research in which also agreed with the theorized relationship.
On the other hand, the relationship between provider concentration and provider effort, the 3
rd
row
in Table 5-1, was not consistently observed in the base model experiment results. In the base model
experiment, there was no observed association between provider concentration and provider effort,
contrary to the theorized association. Figure 5-12 depicts the insignificant association between
provider concentration and provider effort. However, upon further inspection of the association by
area of the simulated market, the theorized relationship was observed in markets of larger areas.
The difference in association between provider concentration and provider effort under different
market areas can be seen in Figure 5-13. There was a statistically significant inverse relationship
between provider concentration and average provider effort in markets with area = 400x400 (p-
value<0.05).
.8 .9 1 1.1
Average Provider Price
2000 4000 6000 8000
Provider HHI
Provider Concentration & Provider Price
85
Figure 5-12. Base model experiment results depicting the “mixed” association between provider
concentration (HHI) and average provider effort.
Figure 5-13. Base model experiment results on the association between provider concentration (HHI) and
provider effort separated by market area representing different directions of associations.
1 1.5 2 2.5
Average Provider Effort
2000 4000 6000 8000
Provider HHI
Provider Concentration and Provider Effort
1 1.5 2 2.5
2000 4000 6000 8000 2000 4000 6000 8000
Area: 200x200 Area: 400x400*
Average Provider Effort
Provider HHI
* p-value<0.05
Provider Concentration and Provider Effort, by Market Area
86
A similar procedure was performed for the remaining rows of Table 5-1 to assess the performance
of the base model experiment results compared to the theorized and empirical associations from
pertinent research. In most cases, relationships and associations produced by the base model were
comparable to those observed in empirical literature, at least in certain market configurations. For
a full assessment of the results used in the comparison between the base model results and
theoretical associations, please refer to appendix D. The results of this procedure are summarized
above in Table 5-1.
The base model produced relationships that were comparable to published empirical relationships
obtained from categorically selected literature. The relationships and dynamics were sensible and
do not indicate unexpected or “out-of-the-ordinary” trends or behaviors. These results were
attained through modules and mechanisms in the base model that have been synthesized from well-
grounded empirical and theoretical work that have been verified in the base model. Hence, the
results of the assessment indicate that the base model was a reasonable abstraction of health
markets on which market dynamics can be studied and interventions can be explored.
5.3. ACO Case Study Results
The ACO case study experiment involved testing the impact of illustrative ACO formations on the
competitive dynamics of the simulated health care markets studied in the base model. ACO
formations impacted market level outcomes and within-ACO-markets provider outcomes. As
shown in section 4.7.3, several different market configurations have been simulated and replicated
for statistical estimates and testing. In this subsection, ACO formations were verified, investigated,
and tested for sensitivity to key model parameters. Data from 800 simulated health care markets,
half of which had ACO formations, was analyzed in this subsection. The simulation experiment
took approximately 30 hours to run on a personal computer (Intel i7 3.6 Ghz Processor and 16 GB
RAM).
5.3.1. ACO Formation Verification
Various facets of the illustrative ACO formation were verified by examining the changes in ACO
costs and quality. The ACO formation, as described in Section 4.7, impacted provider effort
through a fractional increase that was associated with an increase in provider costs. Hence, a test
of ACO provider costs and non-ACO provider costs should reflect this increase. The average cost
per episode for an ACO provider was about 4.5% higher than a non-ACO provider. This increase
was within the expected changes in ∆
%
Cost(effort) shown in Table 4-10.
Verifying the impact of ACO formations on quality was not straightforward, as there were multiple
avenues to improving quality through ACOs: (1) fractional increase in effort and (2) improved
care continuity. The care continuity component of quality improvements was observed by relating
the quality outcomes to the proportion of care delivered in an ACO. The reduction in adverse
outcomes was attained using the density index of continuity of care; meaning that higher
concentration of care delivered in ACO should be associated with less adverse outcomes. Figure
5-14 verifies the expected relationship between mortality rates and the percentage of care delivered
at ACO providers. Inspecting other quality measures, such as readmissions rates and
hospitalization rates, reveal an identical relationship.
87
Figure 5-14. Verification of continuity of care benefits which were realized when a higher proportion of
the care was delivered in an ACO.
The impact of care continuity on the remaining quality measures can be reviewed in appendix E.
The effect of ACO formations on the bargaining process was inspected by observing the agent-
specific interface and tracing the progress of the process. The review of this process confirms it
functions as described in 4.7.1, and correctly combines participating provider market shares and
WTP. The aim of the verification process was to verify the various balancing aspects of ACO
formations so ACOs can have tangible benefits, costs, and impact the bargaining process.
5.3.2. ACO Market Effects
ACO formations in the simulated health care markets were analyzed to infer their impact on market
level outcomes. This analysis compared markets with ACO formations to baseline markets that
had no ACO formations. Hypothesis testing was used to establish statistically significant
differences between the two simulated market types under different market configurations. Table
5-2 summarizes the differences in key observable outcomes across health care markets with ACO
formations and baseline markets.
Table 5-2. Difference in observable outcomes between ACO and baseline markets.
Average Provider
Price
Average Provider
Effort
Mortality Rates
(per 10,000)
Baseline Markets 0.93 2.00 65.40
ACO Markets 1.01 2.04 62.31
Difference +0.078* +0.044* -3.09*
% Difference +8.37% +2.22% -4.73%
.5 .55 .6 .65 .7
Mortality Rates (Per 100 Individuals)
.2 .4 .6 .8 1
ACO Market Share
ACO Care Density & Market Mortality Rates
88
* p-value<0.05
Markets with ACO formations had higher average provider prices (8.37% higher) than simulated
baseline markets. Markets with ACOs also had higher average provider effort and lower per capita
mortality rates (4.73% lower). Effectively, ACO formations produced improvements in quality for
higher prices. To investigate how ACO formations altered value-based pricing compared to
baseline markets, the relationship between price and effort was examined under different market
attributes.
Figure 5-15. Correlation between average provider price and effort in ACO and baseline markets
separated by market attributes (area & population size).
Figure note: This scatter plot infers the status of value-based competition in the simulated health
care markets of different areas and population sizes, with and without ACO formations. Flat
associations between price and effort indicate that variation in prices were not explained by effort,
thus the were no indications of value-based competition. Correlations between provider effort and
price were shown for each market with the corresponding color.
A scatter plot with a higher positive correlation between average provider price and provider effort
signals value-based pricing. This indicated that provider quality increased with provider price.
However, if price changes were not associated with effort, then there was indication of weak value-
based competition. Even though provider price and effort ranges were similar across markets, the
correlation between provider price and effort varied. A closer look at the impact of ACO
Correlation: 0.2588
Correlation: 0.1701
Correlation: 0.2578
Correlation: 0.1877
Correlation: 0.0853
Correlation: 0.3662
Correlation: 0.0888
Correlation: 0.2985
.8 1 1.2 .8 1 1.2
1 1.5 2 2.5 3 1 1.5 2 2.5 3
Population: 30,000 - Area: 200x200 Population: 30,000 - Area: 400x400
Population: 45,000 - Area: 200x200 Population: 45,000 - Area: 400x400
Baseline Markets ACO Markets
Provider Price
Average Provider Effort
Provider Value-Based Pricing: By Area & Population Size
89
formations on value-based competition under markets with different areas and population sizes
revealed different degrees of value-based pricing in simulated markets with different attributes.
Figure 5-15 shows that markets with a smaller population size had stronger indications of value-
based competition without ACO formations (top two plots by looking at blue dots with green fitted
lines). This was inferred from the statistically significant positive relationship between provider
price and effort (p-value<0.05). The degree of this association in the same markets were altered
with ACO formations, but the difference was not statistically significant. However, in markets
with population size 45,000, the base-model showed weaker indications of value-based pricing
(bottom two plots by looking at the blue dots with green fitted line). The association between
average provider price and effort in baseline markets with larger populations was not statistically
significant. But the same markets with ACO formations had stronger indications of value-based
competition (by looking at red dots with orange line) because of a statistically significant
association (p-value<0.05).
Effectively, ACO formations were shown to be associated quality improvements in quality at the
aggregate market level along with increases provider prices. A closer inspection of the correlation
between average provider price and effort revealed that ACO formations were strengthened value-
based competition in markets with larger population size.
ACO Formations & Market Power
ACO formations raised concerns relating to the increase in provider market power. This subsection
investigates features of the increase in provider market power due to ACO formations. Indications
of provider market power in the simulated baseline market were seen in the positive relationship
between provide concentration and provider prices. In markets with higher HHI, the average
provider price was higher in ACO markets (see Figure 5-11). In ACO markets, this association
was amplified compared to baseline markets, as seen in Figure 5-16.
Figure 5-16. The association between provider concentration and provider prices in baseline and ACO
markets.
.8 .9 1 1.1 1.2
Provider Price
2000 4000 6000 8000 10000
Provider HHI
Baseline Markets
ACO Markets
Average Provider Concentration & Price
90
The difference in the coefficients of a regression between provider HHI and provider prices was
statistically different in ACO and baseline markets (p-value<0.001). Therefore, this finding
supports concerns relating to ACOs intensifying provider clout to negotiate higher prices. Another
facet of providers’ market power was the proportion of providers were excluded from insurer
networks. When providers had high market power, less providers were excluded because of the
market share and WTP these providers capture. Therefore, comparing provider inclusion in
baseline markets and ACO markets indicated shifts in market power associated with ACO
formations. Figure 5-17 shows a box plot of the proportion of market providers included in insurer
networks in baseline and ACO markets. Excluding providers from their networks was one way
insurers might exercise market power and reduce provider prices. In ACO markets, insurers
include approxmately 12.45% more providers in their network, signalling an increase in provider
market power.
Figure 5-17. Average proportion of market providers included in insurer networks in baseline and ACO
markets.
There was an evident increase in provider market power manifested in the form of higher provider
inclusion in insurer networks. Combined with increased provider market power in the form of
increased provider prices, ACO formations had measurable increase provider market power.
However, ACO formations also improved market level provider effort and quality compared to
baseline markets. ACO markets had stronger indications of value-based pricing degrees for
changing population sizes and market area. Meanwhile, baseline markets exhibited a correlation
between provider price and effort in certain market configurations only. Exploring the impact of
ACO formations on provider behavior and within-market outcomes would help gain insights on
.4 .6 .8 1
Baseline Markets ACO Markets
Proportion of Market Providers in Insurer Networks
Provider Inclusion in Insurer Networks
91
how ACO markets differ from baseline markets, and when ACO formations were more valuable
versus potentially anti-competitive.
5.3.3. ACO Provider Results - ACO vs. Non-ACO providers
Exploring within-market provider differences involved comparing ACO providers to non-ACO
providers in ACO markets. Non-ACO providers operated in a conventional manner, identical to
providers in baseline model. Meanwhile, two providers participating in the ACO had combined
WTP, MS, incrementally higher effort per level, and benefitted from continuity of care. The results
would investigate the differences between ACO and non-ACO providers under various market
conditions. A comparison of observable outcomes across ACO and non-ACO providers is
presented in Table 5-3.
Table 5-3. Comparison of ACO provider and non-ACO provider in ACO markets.
Providers Provider
Price
Provider
Effort
In-Hospital
Mortality Rates
(per 100 episodes)
Readmissions
Rates
(per 100 episodes)
Non-ACO
Providers
0.935 2.009 0.384 0.0504
ACO
Providers
1.118 2.097 0.358 0.0465
Difference 0.1835* 0.0879* 0. 0260* 0.0039*
% Difference +19.6% +4.4% -6.8% -7.7%
* p-value<0.001
ACO provider prices were substantially higher than non-ACO providers, approximately 20%
higher. In return, ACO providers invested higher effort (5% higher) and achieve measurable
improvements in in-hospital mortality rates and readmissions rates. To separate the difference
between ACO providers and non-ACO providers, provider prices were examined in different
market conditions. In Figure 5-18, the difference in provider prices for ACO and non-ACO
providers evidently grew with the number of providers in the simulated market.
92
Figure 5-18. A comparison of average provider prices in baseline markets with ACO and non-ACO prices
in ACO markets.
In Figure 5-18, it is apparent that the baseline provider prices decreased as the number of provider
increase. ACO prices, however, appeared to increase at first as the number of providers increased,
then decreased as the number of providers increased beyond 5 providers (shown in the green boxes
in Figure 5-18). This suggested that ACO prices were the highest in markets with 5 or 6 market
providers. A closer look at markets by the number of providers supported this notion. The analysis
focuses on the percentage difference (∆
%
) between ACO and non-ACO provider outcomes. When
assessing the percentage difference across ACO and non-ACO providers, it is important to note
that with lower non-ACO outcomes (e.g. provider price different in markets with 7 providers), the
percentage increase in outcomes will be larger because of smaller denominator. It is important to
keep this point in mind when assessing Table 5-4. This table is concerned with the percentage
difference between the outcomes non-ACO and ACO providers (e.g. for ∆
%
provider price, the
difference between red and green boxes in Figure 5-18).
Table 5-4. Percentage difference in observable outcomes between ACO providers and non-ACO
providers in ACO markets, analyzed by number of providers in market.
# of
Providers in
Market
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
Provider Price
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
Provider Effort
Average ACO
Market Share
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
In-Hospital
Mortality Rates
3 +6.4%* +3.0% * 73.6% -5.3%*
4 +18.1%* +2.8% 51.3% -6.2%*
5 +25.7%* +5.5% * 41.0% -7.6%*
6 +24.4%* +7.9% * 35.1% -8.9%*
.6 .8 1 1.2 1.4
Provider Price
3 4 5 6 7
Average Provider Prices
Baseline Market Price
Non-ACO Provider Price
ACO Provider Price
Number of Providers in Market
93
7 +25.5%* +2.9% 31.1% -4.2%*
* p-value<0.05
ACO providers had different prices, effort, and quality when compared to non-ACO providers in
the ACO market. ACO prices were substantially higher than their non-ACO counterparts, but the
percentage difference in these prices varies with the number of providers in the market (column 2,
Table 5-4). Meanwhile, observable quality measures, as influenced by provider effort and
continuity of care, were also higher for ACO providers. Even with minimal effort difference in
markets with few providers (top two rows in Table 5-4), ACOs improved quality through
continuity of care. The improvement in quality increased as the difference between ACO and non-
ACO providers increased with more providers in the market.
When comparing ACO providers across different markets, the biggest percentage difference was
the ACO price. The price varied largely for different number of providers in the market. ACO
effort and outcomes were not significantly different from each other in different simulated markets.
Different ACO prices in different markets has implications on the ACO’s ability to generate
savings through reduced patient utilization.
ACO Savings
In all the simulated ACO markets, ACO providers did not generate any saving through the
reduction of patient utilization pathway. It appears that the savings produced through the reduction
of patient utilization does not offset the increase in provider prices due to the increase of market
power. The explanation for the lack of ACO savings due to high provider cost is best illustrated in
Figure 5-19. There was a strong association between higher ACO prices and lower ACO savings
(more negative savings).
Figure 5-19. Relationship between ACO prices and ACO savings per ACO enrollee.
Figure 5-19 supports the notion that the modeled ACO formations were associated with too-high
of an increase in prices that reduction in utilization do not offset this increase. Therefore, the
illustrative ACOs failed to generate any savings. The sensitivity of ACO savings to patient
utilization was investigated closely in the ACO sensitivity scenarios (subsection 5.3.4.1).
-600 -500 -400 -300 -200 -100
ACO Savings ($ per ACO enrollee)
.9 1 1.1 1.2 1.3
ACO Provider Price
ACO Price and ACO Savings
94
5.3.4. ACO Sensitivity Scenarios
In the sensitivity assessment, additional markets were simulated while varying some of the key
model parameters. For each agent, key parameters were chosen to study and investigate the model
outcomes’ sensitivity. ACO parameters, patient perception of quality and utilization, provider and
insurer quality orientation were studied. Market-level changes induced by ACO formations were
assessed at each level of these sensitivity analysis parameters. Most parameters were set at high
and low values around the value tested in the ACO experiment (except patient utilization).
Differences in the model results at high and low settings of a parameter were tested for statistical
significance. This subsection summarizes the impact of the parameters of interests on key
observable outcomes and agent behaviors. Full results of the sensitivity analysis can be found in
appendix F.
5.3.4.1 ACO Parameters
The sensitivity analysis assesses the impact of the operational ACO parameters on the effect of
ACO formation on the market-level outcomes. The two parameters of interest relate to the change
in effort when participating in ACO, ACOEffortIncrease multiplier, and care continuity
improvement when receiving care at the ACO consistently, CareContinuity multiplier.
ACO Effort Increase
ACO Provider effort increase impacted several different components of the provider’s functions.
Additional effort impacted provider price bargaining, episode costs, and health outcomes.
Therefore, the impact of this parameter on the difference between ACO and baseline markets was
assessed and summarized in the Table 5-5. Low ACOEffortIncrease translated into a statistically
undetectable difference between ACO and baseline markets. Even at higher ACOEffortIncrease,
there was no statistically significant different between ACO market effort and baseline market
effort. This suggested that ACO providers tended to adjust to this high incremental increase in
effort by reducing their effort level. When ACOs had a higher potential to improve quality through
higher costs, ACO markets were less likely to achieve this potential compared to when costs and
quality potential were at baseline levels.
Table 5-5. Sensitivity of ACO market effects on the ACOEffortIncrease parameter.
ACO Effort
Increase
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
Low – 2.5% +8.77%* +0.61% -3.34%*
Baseline – 5% +8.37%* +2.23%* -4.73%*
High – 7.5% +7.42%* +1.57% -3.84%*
* p-value<0.05
The changes in provider effort across the ACO and baseline market is further illustrated in the box
plot that shows how at higher ACOEffortIncrease, providers might compensate the added variable
cost by decreasing overall effort (right-most red box extends below the blue box in Figure 5-20).
95
Figure 5-20. Sensitivity of market effort to ACO effort increase multiplier.
These results suggest that ACO providers might lower overall effort if the variable costs
association with ACO participations were too high, as with the 7.5% increase.
Care Continuity Multiplier
Continuity of care multiplier was the pathway for ACOs to reduce future patient utilization and
generate savings through continuity of care. Because the continuity of care was modeled to be
realized as proportion of care was delivered at an ACO (density continuity of care index), the
effects of continuity of care were independent of provider effort. The impact of this multiplier on
the health outcomes in ACO markets was as expected, shown in Table 5-6. Higher care continuity
multiplier increased the difference between ACO markets and baseline markets.
Table 5-6. Impact of care continuity multiplier on health outcomes.
Care Continuity
Reduction in
Adverse Outcomes
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Hospitalization Rates
Low – 2.5% -1.10% -1.37%*
Baseline – 5% -4.73%* -2.61%*
High – 7.5% -6.04%* -4.12%*
* p-value<0.05
As expected, changes in hospitalization rates and mortality rates impacted ACO savings. Reducing
future hospitalizations and adverse outcomes translated into savings for ACO payers. However,
even at the highest tested care continuity effects, all ACOs failed to generate savings. Figure 5-21
1 1.5 2 2.5 3
Average Effort
.025 .05 .075
Sensitivity of Market Effort to ACO Effort Increase Parameter
Baseline Markets ACO Markets
ACO Effort Increase Multiplier
96
shows ACO savings for different CareContinuity multiplier values. Higher multiplier values
improved ACO savings, but failed to generate positive ACO savings.
Figure 5-21. ACO savings for higher CareContinuity multiplier levels were higher but not positive.
ACO operational parameters altered the impact of the ACO on a market level, as expected. The
results were sensitive to changes in the illustrative settings. These operational settings were
illustrative in nature and warrant additional empirical work to obtain grounded estimates and
structural links between parameters.
5.3.4.2 Consumer Parameters
The sensitivity analysis on the consumer modules were on consumer awareness of quality in the
market and care utilization. The impact of customer preferences of provider quality is described
first, then the impact of increases in patient utilization rates.
Consumer Quality Preferences
Switching off consumer preferences for quality or doubling it did not have a statistically significant
impact on market or ACO outcomes. Even though the average distance per episode increased when
patients were simulated to be more sensitive towards provider quality, the impact was not sizable
effect to alter provider or insurer behaviors. Average prices, effort, and outcomes did not vary in
ACO or non-ACO markets when consumer quality preferences were changes. However, the
correlation between provider prices and effort increase when consumers were more sensitive
towards provider quality. Figure 5-22 depicts how the correlation between provider quality and
price steepened with stronger consumers preference for quality increases.
-600 -500 -400 -300 -200 -100
ACOsavings
.025 .05 .075
ACO Savings and Impact of Continuity of Care
Reduction in Adverse Outcomes for Continous Care
97
Figure 5-22. The correlation between provider quality and price for different degrees of consumer
preferences of provider quality.
Figure note: As the preferences of consumers increase, the correlation between provider prices and
quality increased, suggesting that providers with higher quality charged more than lower quality
providers when consumers were highly sensitive to quality. There was a statistically significant
difference in the slope of the fitted line between ‘No Quality Preference’ and ‘High Quality
Preference’. There was also an apparent increase in correlation between provider prices and effort
as consumer quality preferences increase.
While different consumer preference might not change the average provider prices and health
outcomes, the correlation between provider price and effort responded to consumer preferences
and awareness. This finding might have efficiency implications in health care markets.
Patient Utilization
In this sensitivity analysis scenarios, patient utilization rates were increased by a multiplicative
factor; UtilizationMultiplier. This has implications for understanding the impact of patient needs
on ACO costs and benefits. Increased utilization did not have an impact on provider pricing or
quality of care. The impact of increased utilization impacted insurer premiums (see Figure 5-10),
which was a function of average enrollee utilization (section 4.2.4.3).
The simulated ACOs did not generate any savings in the ACO experiment. This sensitivity analysis
explored whether ACOs were more beneficial for higher patient needs. Indeed, ACO savings
appeared to increase linearly with increased utilization, as shown in Figure 5-23.
Correlation = 0.1193 Correlation = 0.259 Correlation = 0.379
.8 .9 1 1.1 1.2
1.5 2 2.5 3 1.5 2 2.5 3 1.5 2 2.5 3
No Quality Preference Baseline Quality Preference High Quality Preference
ACO Markets
Provider Price
Average Provider Effort
Consumer Quality Preferences & Value-Based Competition
98
Figure 5-23. ACO savings increase for higher expected patient utilization rates.
At approximately twice the baseline utilization, the majority of ACO arrangements were cost
saving. All ACOs became cost savings for utilization that was 2.5x baseline utilization. In
summary, consumer preferences did not have a measurable impact on average market level
outcomes, but shifted the correlation between provider price and effort. Also, increased patient
utilization increased the likelihood of ACO savings.
5.3.4.3 Provider Quality Orientation
Provider quality orientation parameter played a key role in choosing effort levels and the
bargaining processes. This parameter, which ranges between [0,1] related to how much weight
providers place on quality versus profits. A low setting for this parameter represented a profit
oriented provider, while the high setting represented a quality oriented provider. At its different
levels, the parameter changed the impact of ACO formations in the simulated health care market.
Compared to the baseline market, ACO markets with profit oriented providers had the highest
increase in effort (p-value<0.05). This finding suggests that providers that were more profit
oriented had lower baseline effort, and increased it due to risk-sharing in ACO formations. Better
alignment of financial incentives and quality compelled profit oriented providers to increase effort
investments.
-200 0 200 400 600 800
ACO Savings ($/Person)
1.5 2 2.5 3
Utilization Multiplier & ACO Savings
Patient Utilization Rates Multiplier
99
Figure 5-24. Box plot of average provider effort in baseline and ACO markets for different provider
orientations.
Figure notes: The left most boxes were statistically different at p-value<0.05, while the balanced
baseline effort was statistically different from balanced ACO markets (p-value<0.05). Quality
oriented ACO markets were not statistically different from quality oriented baseline markets.
For quality oriented providers (right-most pair in Figure 5-24), ACO formations did not produce
significant changes in effort compared to baseline market. Quality oriented providers already
invested in higher effort that better alignment of incentives did not produce significant changes in
their investments. There was also an apparent reduction in provider effort variability with more
quality providers. This may be translated into more provider quality standardization.
Other observable outcomes and the effect of ACO formations remained unchanged for different
provider orientations. Despite the difference in effort across different markets and provider
orientations, the observable quality outcomes were consistently better in ACO markets compared
to baseline market. Essentially, this sensitivity analysis suggested that provider orientation played
a key role in their effort choice and the effectiveness of ACO formations in pushing providers to
invest more effort. Profit oriented providers invested more effort in ACO markets. While quality
oriented providers already invested in high effort that ACO formations did not induce significant
additional investments in effort.
5.3.4.4 Insurer Quality Orientation
The insurer quality orientation parameter impacted insurer network selection, and consequently
changed provider market shares and prices. Hence, it was possible that several aspects of the ACO
market differ from baseline under different insurer quality orientation. At higher quality orientation
1 1.5 2 2.5 3
Profit Oriented* Baseline* Quality Oriented
Provider Effort
Baseline Markets ACO Markets
Provider Effort & Provider Orientation
* p-value<0.05
100
insurers were more stringent about including lower quality providers in their networks. As a result,
it was expected that more provider exclusions will occur for a higher insurer quality parameter.
Indeed, the results show that more provider exclusions occured at a higher insurer quality
orientation visible in Figure 5-25. In Figure 5-25, moving from left to right, more quality oriented
insurers were more likely to exclude lower quality providers from their networks.
Figure 5-25. Inclusion of market provider in insurer networks for different settings of insurer quality
orientation.
More quality oriented providers were more likely to exclude lower quality providers from their
network (right most boxes, p-value<0.05). In ACO markets (red boxes), the reduction in provider
inclusion was less apparent for more quality oriented insurers, yet statistically significant
compared to ACO exclusions in markets with insurers with lower quality orientation (p-
value<0.05).
The difference between ACO and baseline markets was not significantly different for varying
levels of insurer quality orientation. Provider exclusion from insurer networks was a pathway for
insurers to exercise market power. This pathway reduced provider prices, since insurers steered
patients away from providers. Therefore, a closer look at provider prices showed that more quality
oriented insurers put downward pressure on provider prices. In fact, insurer pressure translated to
a 2.2% decrease in provider prices when comparing low quality oriented insurers to highly quality
oriented insurers. The reduction in provider prices was noticeable (both in the ACO and baseline
markets) as insurer quality orientation increased in Figure 5-26. The impact, however, did not
change the difference between ACO and baselines market prices.
.4 .6 .8 1
Low Quality Orientation* Baseline Quality Orientation* High Quality Orientation*
Baseline Markets ACO Markets
Proportion of Providers In Insurer Network
Insurer Quality Orientation & Inclusion of Providers In Network
* p-value<0.05
101
Figure 5-26. Insurer quality orientation and effects on provider prices in ACO and baseline markets.
Figure Note: Higher insurer quality orientation reduces provider prices in baseline and ACO
markets, moving from left to right in the figure. The magnitude of the difference between ACO
and baseline market prices, however, remain unchanged for different insurer quality orientation.
Variation of provider prices increase for higher insurer quality orientation (Chi-squared test on
variance p-value<0.05).
It was also observed that insurer quality orientation impacted the variability of provider prices in
baseline and ACO markets. Figure 5-26 shows that the boxes corresponding to markets with
insurers having high quality orientation (right) were longer than boxes corresponding to markets
with lower insurer quality orientation (left). A Chi-squared test on the variance between markets
with high and low insurer quality confirmed a statistically significant increase in variance
associated high quality oriented insurers (p-value<0.05).
The sensitivity analysis emphasized the importance of parameters that change ACO experiment
results. The impact of the parameters studied had measurable effects on direct agent choices and
health outcomes. The magnitude of the effect was of interest because it would inform on the
robustness of model insights and guide future empirical work for less-known, but influential
parameters, such as provider quality and profit orientations. The sensitivity analysis also exposed
complex relationships between model parameters and outcomes, such as the relationship between
insurer quality orientation and provider price variability, along with several other interesting
observations discussed in the following chapter.
6. Discussion
This chapter discusses the model development, contributions, findings, implications, limitations,
and future work. The research objective was to draw insights on the competitive impact of ACO
.8 .9 1 1.1 1.2
Low Quality Orientation Baseline Quality Orientation High Quality Orientation
Baseline Markets ACO Markets
Provider Price
Insurer Quality Orientation and Provider Prices
102
formations on private health care markets. Complex systems thinking was employed and ABM
was constructed to recreate abstractions of private health care markets. In doing so, this research
demonstrated the suitability of complex systems approach, the utility of ABM, and generated key
insights into the impact and value of ACO formations in private health care markets.
ACO markets were compared to baseline markets, where there were no ACO formations, to
understand their market-level impact. This required developing the base model, verifying it, and
assessing its face-validity. The base model was developed using the well-grounded conceptual
framework described in Chapter 3. Additionally, the model is equipped with decision, progression,
and interaction modules that were theoretically described in chapter 4. The programming of model
components was verified, and confirmed that the tested components were consistent with their
corresponding functions. A comparison of base model results reveals relationships that were
analogous to those observed in empirical studies. The comparison involved published literature
that was categorically selected to establish model face-validity. The base model’s intricate
relationship between several outcomes in private health care markets corroborates model’s
sufficiently complex and meaningful abstraction. In other words, the base model is developed
based on an informed conceptual framework, is verified, and passes the face-validity assessment.
Additionally, the base model results did not have results that were out-of-the-ordinary or
unexpected. Therefore, the base model is considered sufficiently representative of key features of
private health care markets and suitable to study and test the impact cooperative payment models.
6.1. Studying Health Care Markets as Complex Adaptive Systems
The approach and methodology employed in this research is devised by complexity science and
systems engineering. Health care markets were qualified as complex and adaptive systems by
drawing parallels between distinguishing aspects of CAS and health care markets. Hence, agent-
based modeling was chosen as the method of choice to model private health care markets, which
is amicable to previously mentioned characteristics of CAS. The resulting models represented
hypothetical markets that were decentralized, adaptive, have repeated interactions, highly
embedded feedback loops, hierarchal structures, and out-of-equilibrium dynamics. The result
markets were ground-up, generative model with explicitly defined rules, interactions, roles, and
sequences to create an artificial health care market.
This unique model presented a different perspective to study health care markets. Hypothetical
markets were “grown” in silico and all the explicitly defined rules and interactions represent the
set of sufficient conditions that give rise to system behaviors and trends in observed outcomes.
Effectively, the model and the analysis performed in this research was an implementation of the
complexity theory approach to explain system trends and emergent behaviors. This approach
entailed developing explanations by reconstructing the system-level trends and behaviors instead
of breaking-down the system to explain behaviors (reductionist approach). This research furthered
this method of explanation devised in complexity theory and systems sciences through an
implementation on competition and private health care markets. Additionally, it demonstrated the
richness and rigor of this approach. The results provided insights that were elusive to conventional
empirical studies.
6.2. ABM of Private Health Care Markets
At the time of this dissertation, this model is the only ABM concerned with studying competition
in private health care markets. ABM was used as a tool to study health care markets as a CAS, and
103
was programmed using JAVA on Anylogic v.7.1.2 [148]. From a technical standpoint, this model
embodied many of the modeling technique’s advantages that make it suitable to study a problem
with such complexity. On the other hand, the ABM technique had several disadvantages and
limitations that were also present in the model developed for this research. This subsection focuses
on discussing the advantages and disadvantages of using the ABM technique to achieve the
research’s objective.
The key technical advantages of the ABM technique utilized in this model were its modularity,
agency, and richness. Agents and their corresponding decisions, interactions, and progressions
were programmed to be modular in nature. This means that any module corresponding to agent
operations, interactions, or decisions, can be upgraded, added, or completely switched off. If the
replacement module satisfies the input and output structures, the model would still function. This
paves the way for upgrades and adaptations of the model that would repurpose the simulated
market to study to accommodate diverse forms of interventions or market policies. From a
technical standpoint, this model demonstrated the modeling technique’s flexibility compared to
other techniques. Mathematical and statistical models, for instance, do not have such flexibility.
Also, ABM, as the name suggests, involved agency in the modeled entities. This means that agents
can be programmed to execute decisions and act autonomously. Consumers, providers, and
insurers in the model were represented as agents with a distinct set of decisions and interactions.
The agents were semi-intelligent, goal-oriented, and equipped with choice and behavioral models
to perform specified roles. Other modeling techniques, such as discrete event and systems
dynamics modeling, do not enable such agency and autonomy. Lastly, ABM enabled the
development of rich artificial markets. These markets have spatial features, agent-level, and
system-level outcomes at any desired level of granularity. This richness enabled the creation of
organic, identifiable health care markets that can be engaging for decision makers and researchers.
The modeling technique also generated extensive data that can unveil deep insights, as seen in the
ACO illustrative case study. Studying emergent behaviors is only possible through this level of
richness in model structure and data generation. Other modeling techniques are not amicable to
spatial features in modeling and are limited in the level of abstraction and granularity.
While ABM offers appealing advantages that make it suitable for this research’s purpose, various
disadvantages were evident throughout this research. Some of the disadvantages within the model
designed for this research were related to its computational cost, data and theoretical requirements,
and “combinatorial explosions”. First, ABM was computationally costly compared to other
modeling techniques. For instance, simulating 800 markets with nearly 40,000 consumers took
approximately 30 hours to run on a personal computer (Intel i7 3.6 Ghz Processor and 16 GB
RAM). Each consumer agent was associated with their corresponding episodes, providers, and
insurers; all of which added to the computational burden of the model. Second, the model is data
and theoretically intensive. Because the model follows a ground-up approach, elaborate definitions
and rules require substantial empirical and theoretical foundations. Data and theory related to more
model parameters and decisions would have been more desirable. Compared to other modeling
techniques that are less granular, ABM require more data and theory. Lastly, complex ABM can
encounter combinatorial explosions, given the number of parameters and scenarios that can be
varied. This means the number of combinations of model parameters and scenarios follows
polynomial growth. The richness of the ABM has this tradeoff; the number of scenarios and
analyses can be difficult to comprehensively cover, and increases substantially for each additional
104
parameter. The model constructed for this research encountered this issue due to its scale, and was
mitigated by focusing on market structure parameters and observing a specified set of outcomes.
6.3. ACO Model Findings
After the suitability of the base model for testing the impact of cooperative payment models was
assessed, an illustrative ACO case study was modeled. The shifts in market power involved with
ACO formations were quantified using appropriate competitive indicators, market power indices,
and observable outcomes. The magnitude of the relationships and associations between market
parameters, competitive indicators, and outcomes varied largely in different simulated market
configurations. This section focuses on discussing the illustrative ACO model findings from 3 key
perspectives: market power, quality, and value.
6.3.1. ACOs & Market power
On average, ACO prices were consistently higher than non-ACO prices in all the simulated
markets. Because the illustrative ACO combines each provider’s market share and WTP, it is
expected that the resulting ACO prices will always be higher. This result is explained by the shape
of the optimal price function on provider market share and WTP obtained from equilibrium price
through the Nash product.
Provider Market Power: Lerner Index
The extent of the price increase, however, varied depending on the market configuration. The
Lerner index, as described in 2.1.2.1, measures the price-cost margin for providers as a proxy for
market power [42]. Using this index, providers in markets with fewer providers (4 or less) gained
less additional market power from an ACO formation than providers in markets with more
providers (6 or more), as seen in Figure 5-18. Because provider costs were the same for all ACO
providers and ACO formations raised ACO prices more in markets with more providers, it means
that ACO providers in markets with a high number of providers achieve a larger increase in market
power using the Lerner index. This result may appear counterintuitive, but Figure 6-1 shows that
providers in markets with more providers gained more market power than providers in markets
with fewer providers. The gain in market power can be seen in the difference between the red and
blue boxes for each market type.
105
Figure 6-1. ACO provider prices compared to non-ACO provider prices grouped by number of providers
in market.
Figure Note: This analysis is grouped by number of providers in market. Providers in markets with
fewer than 5 providers gain less additional market power from ACO formations than providers in
markets with more than 5 providers. The increase in provider market power is shown in this graph
by the difference between the red and blue boxes in each market type.
Providers in markets with < 5 providers had higher baseline prices (right blue box in Figure 6-1).
Therefore, the percentage difference in price was lower for ACO formations in these markets.
Effectively, ACO-induced increases in market power produced smaller price differences in
markets with higher baseline market power. Providers in markets with > 5 providers had diluted
market power and combining market shares and WTP translated into higher increases in prices.
This finding indicates that the gain in market power from ACO formations relates closely to the
baseline market price without the ACO formations. If the baseline prices were already high, then
ACO formations translate into a smaller marginal increase in market power. While in competitive
markets with lower baseline prices, ACO formations allow participating providers to increase their
prices more.
Insurer Market Power: Excluding Providers from Insurer Networks
Another indication of shifts in market power due to ACO formations was observed in provider
exclusion from insurer networks [64]. ACO formations significantly increased provider inclusion
in insurer networks (Figure 5-17), indicating that insurers lost some of their market power to the
illustrative ACOs in the market. As a result, insurers did not put downward pressure on provider
prices by excluding the providers from their network, steering away some of the providers’ market
share, and, consequently, reducing the providers’ price. In the baseline market, the proportion of
providers excluded was higher for markets with a higher number of providers. ACO formations
dampened this pressure, increasing the number of providers included in insurer networks. This can
Provider Price
.6 .8 1 1.2 1.4
Providers in Market > 5 Providers in Market < 5
non-ACO Providers ACO Providers
ACO Price and Non-ACO Price by Number of Providers in Market
106
be explained by the mechanism in which insurers were modeled to form their networks. This
mechanism involves comparing providers to average price and quality. When average market
prices increased due to ACO formations, insurers included more non-ACO providers in their
networks because the average market price is higher. Since network formations were an
implementation of any-willing-provider laws, insurers would include more non-ACO providers
when accounting for the higher ACO providers’ prices.
Insurer market power increased when insurers were more quality oriented. Higher insurer quality
orientation translated into more exclusions of lower quality providers from insurer networks. This
is evident in the insurer quality orientation sensitivity analysis (Figure 5-25), where provider
inclusion was lower (exclusion is higher). The relationship between higher insurer quality
orientation and lower provider prices was confirmed in Figure 5-26. From the sensitivity analysis
performed on insurer quality orientation, it was possible to infer that insurers that were stringent
in choosing providers in their network play a role in determining provider service prices.
6.3.2. ACOs & Quality Outcomes
ACO providers performed better than non-ACO providers in the simulated market configurations.
The magnitude of the difference varied with market conditions. As described in section 4.7.1.1
and 4.7.1.2, the illustrative ACO is modeled to impact health outcomes via two pathways: increase
in effort and care continuity. The impact of care continuity was constant on provider effort markets,
and only increases when ACOs have higher market share. Meanwhile, ACO effort increase was a
percentage increase for each ACO provider.
In markets with fewer providers, more care was concentrated at ACO providers, which means the
impact of the continuity was higher than more diluted markets where an ACO has less density of
care. To clarify this tradeoff, Table 6-1 considers two ACO market configurations adapted from
Table 5-4.
Table 6-1. Two configurations of simulated ACO markets defined by the number of providers to illustrate
the separable impact of care continuity and increased effort.
# of
Providers
in Market
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
Provider Price
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
Provider Effort
Average
ACO
Market
Share
∆
%(𝑨𝑪𝑶 𝒏𝒐𝒏𝑨𝑪𝑶 )
In-Hospital
Mortality
Rates
Comments
3 +6.43%* +2.96% * 72.41% -5.26%* Improvements in mortality
rates were achieved
primarily through
continuity of care (high
market share)
6 +24.36%* +7.87% * 35.23% -8.94% * Improvement in mortality
rates achieved mainly
through higher effort (low
market share)
* p-value<0.05
The ACOs in the markets with 3 providers increased effort less than non-ACO providers compared
to the ACO provider in the market with 6 providers, as shown in column 3 of Table 6-1. However,
107
The ACOs in the markets with 3 providers had a higher average ACO market share, capitalizing
on the care continuity pathway to improve health outcomes. Meanwhile, ACO providers in the
market with 6 providers improved mortality rates by relying more on increased effort, given the
smaller ACO market share to realize continuity of care benefits. Therefore, this model allowed for
quantifying pathways to impact health outcomes, each pathway having a separable magnitude of
effect in dissimilar market conditions.
ACO Parameters
Changes in ACO parameters relating to ACO effort investment (as a percentage multiplier) and
the magnitude of adverse outcome mitigation due to care continuity had an expected impact on
ACO results. When ACO investments (ACOEffortIncrease) were set to a high level (7.5%), some
ACO providers responded by lowering their effort, increasing effort variability. This illustrated an
emergent behavior where ACO providers’ effort decisions increase in variability, as seen in Figure
5-20. This parameter represented the potential and cost of improvements achieved through ACOs.
Higher potential meant ACO providers would achieve better quality for a higher cost. The higher
ACO effort increase pushed some providers to reduce their overall effort. This was seen as some
providers reducing their overall effort to compensate for the variable effort increase associated
with participating in an ACO. The insight here is that if ACO formations require substantial
provider investments to achieve better outcomes, then it is likely that some ACO providers would
reduce their overall investments, increasing the variability in care. This may be translated into less
standardization of quality and care delivery.
When the ACO investment was modeled at the low setting (2.5%), there was no statistically
detectable difference in health outcomes due to ACO effort increase. Meanwhile, the impact of the
magnitude of adverse outcome mitigation due to care continuity was straightforward, and had
statistically significant impacts on ACO results. These findings shed light on the importance of
having accurate theoretical pathways and empirical estimates to model the impact of ACO
formations.
Provider Quality & Profit Orientation
In this model, the tension between provider profits and quality was captured in the provider’s
quality orientation parameter. The sensitivity analysis on this parameter revealed key relationships
between provider quality/profit attitudes and ACO effectiveness in driving providers to increase
effort. The emergent provider behavior observed for different provider orientations showed the
richness of this modeling technique and the insights gained from this mode. When providers were
more profit oriented, ACO formations were associated the highest effort increase. Results
illustrated the tendency to use ACO formations to incentive profit oriented providers to invest in
quality. On the other hand, quality orientated providers did not invest significantly more effort
with ACO formations. Providers with higher quality orientation already invested in higher effort
that ACO participation does not compel them to invest more. Providers with balanced tendencies
were motivated by ACO formations to invest effort, suggesting ACO worked best with profit
oriented providers among the tested provider profit/quality orientations. This highlighted the
importance of knowing provider attributes and quality/profit tendencies when evaluating ACOs.
It is also important to understand baseline market performance to estimate the magnitude and
direction of change ACO formations may propel a market to.
108
6.3.3. ACOs & Value
ACO formations shifted market power in private health care markets. ACO providers had higher
prices and insurers showed weakened power to exclude providers from their networks. Also, ACO
quality was consistently higher, but the extent of the difference depended on the simulated market
configuration. The focus in this subsection is on the value associated with ACO formations by
discussing how ACO prices and effort correlated and value is generated.
Marginal Value of ACOs
A marginal cost-benefit evaluation of ACO formations revealed that the ratio varies under different
market conditions. An analysis on ACO prices and mortality rates indicated ACO providers
generate higher value in markets with fewer providers. This is illustrated in Figure 6-2 where ACO
providers in markets with fewer providers, indicated by the number next to each line, have steeper
slopes that correspond to larger reduction in mortality rates achieved per unit change of price.
Figure 6-2. Marginal cost-benefit assessment of providers in ACO markets.
Figure Note: This analysis is performed by number of providers in market. Non-ACO providers
are shown with ‘o’ symbols, while ACO providers are indicated with ‘x’. Each line represents the
∆𝑴𝒐𝒓𝒕𝑹𝒂𝒕𝒆 𝒏𝒐𝒏𝑨𝑪𝑶 𝑨𝑪𝑶 /∆𝑷𝒓𝒊𝒄𝒆 𝒏𝒐𝒏𝑨𝑪𝑶 𝑨𝑪𝑶 for a market with number of providers (providers
in market = ACO + non-ACO providers). Steeper line indicated higher value, since higher
mortality rates reduction is achieved per unit change of price. As the number of provider in the
market increases, value of ACO providers decreased.
109
These findings challenged notions that ACO formations in unconcentrated markets are generally
safer from an anti-competitive perspective. In simulated health care markets with already high
prices, as is the case in markets with 3 providers, the value added by the ACO is higher than in
markets with lower prices. A similar pattern was present in other observable provider health
outcomes. This insight can be explained by the fact that prices are already higher, and thus the
price increases less in markets with fewer providers, and these providers had higher market shares,
which makes the continuity of care pathway effective in reducing adverse outcomes.
Correlation Between Prices and Quality
ACO formations facilitated value-based competition in all simulated markets. Value-based
competition is assumed to be proxied by the correlation between provider prices and provider
effort [179]. Higher correlation between provider prices and effort indicated value-based
competition. Effectively, when provider prices were not associated with effort, then variation in
provider prices was not explained by variation in effort. In more dispersed baseline markets with
larger population sizes, there seized to be significant correlations between provider price and
effort. This may be explained by the availability of enough volume to providers (larger
population). There was enough patient volume to providers that increasing quality was not
necessary to attract patients. A correlation between provider prices and effort were less likely to
emerge in these highly populated markets. This was more evident in markets of larger area, where
consumers were less likely to switch providers (blue dots are flatter in bottom right plot of Figure
5-15).
ACO formations retained the correlation between average provider prices and provide quality for
the various market conditions simulated. This indicated that while ACO formations tended to
increase the average market price, the increase was correlated with increases in quality.
Implications of these findings inform formations in rural settings with dispersed population over a
large market area. The findings from this model suggested that independent providers prices might
not be correlated with quality in less densely populated areas, ACO formations in these markets
would maintain a correlation between prices and quality. ACO formations may tie provider prices
and effort through the shared-risk mechanisms. Risk-based payments, as with the ACO shared
savings, better aligned incentives for providers when the competitive forces did not compel non-
ACO providers to invest in effort.
Patient preferences of provider quality also played a key role in the correlation between provider
prices and quality. When consumers had higher preferences for better provider quality, the
correlation between price and quality increased (noticeable albeit not statistically significant). This
correlation was additive with the increase in correlation from the ACO formation. Figure 6-3
illustrates how ACO formations combined with higher consumer preference for provider quality
translated into higher correlation between prices and effort.
110
Figure 6-3. Consumer quality preference & value-based competition in ACO and baseline markets.
Note: The plot depicts the visible additive impact of ACO formations and high consumer
preference for provider quality on the correlation between provider price and effort. This finding
indicates that increased consumer preferences for quality reduces the likelihood for ACO
formations to increase price without increasing effort.
This implies that ACO formations are less likely to increase provider prices while reducing
provider effort if consumers are aware of provider quality and prefer high quality providers. This
insight demonstrated the power of this modeling approach to separate the impact of ACO risk-
based payments and consumer quality preferences and their interaction in influencing the value-
based competition. Furthermore, this demonstrated a synergistic policy lever promoting ACO
formations along with initiatives to encourage patient choice and public provider quality reporting
could improve value-based competition more than either initiative alone.
ACO Savings
Another perspective to evaluate the value of ACO formations is to assess the savings they generate.
The illustrative ACOs failed to generate savings in the original ACO experiment. A closer look
revealed that ACO price increases were too high to be offset by the reduction in utilization to
generate savings at baseline utilization rates. The sensitivity analysis experiment that multiplied
patient utilization rates by a multiplicative factor confirmed that ACOs generate positive savings
when serving patients with higher utilization rates. Multiplying the utilization rates for the
consumer population mimics serving patients with higher needs. At 2x utilization rates, ACOs
started to probabilistically generate positive savings. At 2.5x, all simulated ACOs generated
positive savings.
These results suggested that serving entire patient population in an ACO model might not be cost
saving because the short term financial benefit, in the form of savings, does not offset the increase
in care costs at ACO providers. Generic patient needs did not benefit enough from utilization
.8 .9 1 1.1
1.5 2 2.5 1.5 2 2.5
No Quality Preference High Quality Preference
Baseline Market ACO Market
Provider Price
Average Provider Effort
Consumer Quality Preference & Value-Based Competition
111
reductions to make up for the ACO price difference. However, for patients with complex needs or
condition-specific ACOs were more likely to generate savings. Effectively, this model enabled
establishing market-specific expected utilization cut-off point after which ACOs were expected to
generate savings.
6.4. Policy & Research Implications
This subsection connects the insights gained from modeling the shift in competitive dynamics due
to ACO formations in private health care markets to relevant market policies and research efforts.
Illustrative ACO formations in the simulated markets challenged some existing notions regarding
their competitive effects, while other model findings were consistent with initiatives and other
research initiatives.
Anti-Trust Agencies Evaluation of ACO Formations
In less competitive markets, modeled ACO formations delivered better value represented by lower
marginal price increases for reduction in mortality. In more competitive, less concentrated markets,
ACO formations delivered lower value represented due to higher marginal price increase for
reduction in mortality (Figure 6-2). This challenges the ‘safety zone’ in DOJ and FTC review
processes to approve providers joining ACO Medicare Shared Savings Program [180]. The anti-
trust safety zone facilitates ACO formations in more competitive markets with dispersed market
shares because they are less likely to produce anti-competitive behavior [180]. However, model
results show that in more competitive markets, ACO formations produce higher marginal price
increases than less competitive markets (Figure 6-2) for the illustrative ACO structure and chosen
price determination process. The model demonstrated that the effects of ACO formations depended
largely on market baseline price and quality levels. Findings are especially relevant given hospitals
participating in ACO tend to be in urban areas and large [181]. From a competitive standpoint,
these hospitals might not pass the MSSP ACO formation review process given their significant
size. However, from a policy perspective, model findings suggest that a more comprehensive
assessment of an ACO formation should incorporate current baseline market provider prices and
quality, along with the conventional competitive indices.
Provider quality and profit attitudes were also key in examining the value of ACO formations.
Provider efforts in ACO markets depended largely on their quality/profit orientation. Illustrative
ACO experiment sensitivity analyses suggested that providers with balanced profit/quality
orientations were compelled to the highest increase in effort from the baseline compared to more
profit-seeking or quality-seeking providers. Effectively, high quality oriented providers invested
more effort without ACOs, while profit oriented providers responded to ACO arrangements by
increasing effort. This insight underscores the importance of further studying the profit-quality
tensions present in provider decisions, a subject that has made significant theoretical and practical
developments recently [182].
Illustrative ACO formations in this model improved outcomes through better continuous care
when ACOs have higher market share. The tension between the benefit of care concentration for
providing a continuum of care and increases in market power due to higher market shares were
observed in this model. The tradeoff differs under various market conditions tested and plays a
significant role in the value of ACO formations assessed by the anti-trust agencies. Hence, from a
research perspective, this model offers a tool to evaluate the possible competitive influence of
prospective ACO formations.
112
Initiatives to Enhance Value-Based Competition
This model and analysis uncovered two avenues to increase value-based competition. (1) The risk-
based payment component of the ACO agreement, shared savings, increased the observed
correlation between provider effort and prices at the aggregate market level. Shifting risks to
providers strongly incentivized changes in care delivery in the model and actual ACOs [183]. (2)
Increased consumer preference for provider quality had a similar observable effect. The interaction
of these two components produced the highest correlation between provider price and quality
(Figure 6-3). The two studied avenues were especially important to sustain value-based
competition when baseline market forces weakened (from less densely populated markets).
Increased patient quality preferences alone were not sufficient to ensure correlation between
provider prices and quality in the model. Ongoing efforts by Health and Human Services
Department to increase value-based health delivery may be improved by incorporating the insights
from this illustrative model findings [184]. From a policy perspective, these findings suggest that
risk-based ACO arrangements combined with quality-aware and actively choosing patients would
reduce the likelihood of anti-competitive ACOs, which increase prices without increasing effort.
This represents an effective synergistic policy lever; where educating patients, improving public
provider quality reports, and encouraging active choice of providers would increase the likelihood
of ACOs delivering better value. The importance of empowering patients is consistent with
research on patient choice and its role in promoting provider competition [179], [185].
This model can serve as a tool to further investigate the impact of better provider quality reporting
and initiatives to encourage active patient choice of providers on competition in health care
markets. The sensitivity analysis on consumer preferences demonstrated the model’s ability to
study varying preferences and capture the downstream impact on market trajectories and average
outcomes. As the sensitivity analysis showed, ACO formations were less likely to be anti-
competitive when consumers have stronger preferences. The impact of provider quality reporting
and choice have long been a subject of research and policy interest [116], [160], [186], and can be
explored using the model built for this research. Additionally, as the tension between value and
volume plays out in private health care markets, more payers are resorting to multiple payment
model designs and incentives, only one of which is the ACO. Researchers and analysts are turning
their attention to the different payment model designs, and how multiple designs work together
[187]. This model offers insights on ACO formations and can possibly harbor multiple payment
schemes and designs, providing insights on their composite value, as demonstrated by the case
study on ACO formations.
ACO Financial Risks & Savings
ACO providers did not produce savings in the simulated ACO markets until patient utilization was
inflated by a multiplicative factor. The insight here is that ACOs targeted at patients with complex
needs or specific health conditions are more likely to generate yearly financial savings through
better continuous care and quality investments. The elderly, individuals with comorbidities, and
disease-specific ACOs are more likely to have a substantial reduction of acute episodes and
adverse outcomes to offset price increases due to provider coordination. From a policy perspective,
this model enabled the assessment of a market-specific utilization threshold around which ACO
providers are likely to generate savings. As expected, this threshold varies for different simulated
markets. This threshold can be used to inform the prospects of potential ACO formations in
113
achieving financial savings for payers. This model presents a research tool on which markets can
be simulated to study varying patient utilization patterns and perform assessments on resulting
ACO savings.
Most private payer ACOs involve provider downside risk [188]. However, it is common that
private ACO contracts are upside-only contracts (approximately 41% of private ACOs in 2014)
[188]. In upside-only contracts, if a private ACO provider fails to generate savings there is no
provider risk; meaning ACO can only gain from an ACO arrangement. While such agreements
shield providers from bearing the risk of an unbalanced ACO patient panel and protect smaller
providers, this also means that payers bear the full burden of shifts in provider market power. The
findings of this model would advise against upside-only private arrangements, especially for
generic or average-utilization panels of patients. This model can help understand and quantify
ACO financial risks from a payer perspective given the utilization levels and market conditions of
the patient panel. Commercial ACO arrangements also involve quality bonuses, which are not
included in the model and the effect of which remains to be explored.
ACO Operational Attributes
Model findings were sensitive to ACO formations operational parameters. By design, the
illustrative ACOs modeled in this research involved simplified pathways to increasing quality
investments and improving health outcomes. From a research perspective, the model highlighted
the importance of understanding the operations and mechanisms involved in private ACO
arrangements to assess their prospective impact. Compared to public ACOs, little is known about
private ACOs’ characteristics and arrangements [188]. The sensitivity of the modeled ACOs to
key operational parameters support ongoing efforts to study different facets of private ACOs and
agreements [21], [26], [68]. The insights revealed from the ACOEffortIncrease parameter
represent the potential and cost of improvements achieved through ACOs. If ACO formations can
achieve higher quality improvements through higher cost (higher ACOEffortIncrease), then ACO
formations are less likely encourage additional provider investments, due to the cost and distance
of this potential. These studies reveal the interconnectedness of private and public ACOs in terms
of spillover and structure. With the presence of multiple payers, patient populations, and
specialties, various ACO pathways to improving care and efficiency need to be explored given
their importance in realizing ACO potential [72].
6.5. Research limitations & Future work
This subsection focuses on communicating the limitations and shortcomings of this model and
potential future directions to mitigate the impact of these limitations on drawing meaningful
conclusions.
Performing Structural Sensitivity Analysis
The current model assumed a certain structure for the bargaining problem and ACO effect. ACOs
combined market share and WTP, while some structural implications were made for the bargaining
problem. The sensitivity of model outcomes and conclusions to the structures as they were defined
here can be explored. For instance, instead of having an additive ACO where market shares and
WTP of the participating providers are added, it might be possible to investigate other methods of
combining their contributions to ACO bargaining. This might yield different ACO price patterns.
Another possible structural improvement would be incorporating insurer market share in the
market power parameter β in the bargaining problem and testing other insurer network formation
114
mechanisms, besides all-will-providers. These changes would assess the robustness of the model
findings to structurally different decisions and markets. A simple experiment relating to the
structural sensitivity of the model can be examining the impact of mergers on markets. This subject
has been extensively studied [43], [48], [189], [190]. In studying mergers, the ACO structure can
easily be repurposed such that two providers represent one entity to examine and cross-validate
the modeled impact of mergers on market prices.
Accounting for External Market Influences
The ABM developed in this research represented a “closed” market with no external influences.
Actual health care markets are affected by several different forces, including, but not limited to,
public health insurance (Medicare, Medicaid, CHIP) and HMO presences. Demand and provider
prices in this model are assumed to be separate from Medicare and Medicaid provider
reimbursement rates and demand. HMO penetration in health care markets have been shown to
be associated with private ACO formations, and therefore accounting for their market effects is
essential in studying the impact of ACO formations [59], [72], [77], [146]`. The insights drawn
from this model did not account for these exogenous market influences that may change the way
ACO formations impact private health care markets.
Published literature that focused on private health care markets prices and demand typically used
Medicare prices as references around which private market prices are anchored [52], [191].
Employing such an approach here is technically feasible by modeling a form of public health
insurance within the market. More theoretical work is needed to account for the impact of these
prices and demand in the bargaining arena. Therefore, this model might be considered a suitable
test bed to explore and test developments in bargaining and pricing theories in private health care
markets while accounting for a portion of the population covered through a public insurance
program.
Incorporating Relevant Data Sources and Models
The model exclusively used publicly available data sources and parameters to recreate consumer
utilization and calibrate provider and insurer decisions. Incorporating more suitable data sources
would enhance several different facets of the model. For instance, using national hospital data
would allow for accurately modeling provider costs, margins, and outcomes with distributions for
heterogeneity in those parameters. Utilization data would improve the accuracy of modeled patient
utilization, which proved to be closely tied to ACO’s ability to generate savings. For future work,
access to non-publicly available datasets opens avenues for model cross-validation. The
performance of this model can be cross-validated with models of other types, if the data inputs and
assumptions can be matched.
Extending Competitive Dimensions
Providers and insurers in the model were limited in the ways their services are differentiated.
Providers choose their effort levels, while insurers choose their network of providers. In actual
health care markets, providers differ in quality, capacity, services provided, and patient
experiences. Insurers in actual markets differ in benefits, cost sharing levels, breadth of provider
network, and customer service. The added dimensions of product differentiation from a provider
and insurer agent perspective would bring the modeled markets closer to actual private health care
115
markets. However, the theoretical and behavioral aspects of modeling these decisions are beyond
the scope of this model.
For future work, incorporating provider capacity constraints and service type differentiation are
theoretically and technically accessible avenues of differentiation that might reveal interesting
insights. Providers can be modeled to have a certain number of inpatient beds, and providers can
be of different types: primary care physician, outpatient facility, and hospitals. Exploring the
impact of methods of differentiation would unveil possibly interesting insights on the impact of
capacity on competition, and work towards improving model generalizability.
Improving Agent Decision-Making
The nature of consumer choice in the model were restricted to choosing among providers and
insurers. This limited the model’s ability to account for uninsured individuals who choose not to
purchase coverage, or those who choose to forgo care. In real markets, such individuals make up
a sizable portion of the market, and omitting that segment might compromise the generalizability
and applicability of model insights. Therefore, with the availability of more consumer choice data
and behaviors, it is possible to better calibrate the consumers’ WTP values such that they only
choose to purchase coverage when their WTP surpasses insurer premiums.
Provider and insurer agents were modeled to be myopic. These agents did not strategize with price,
provider network formation, or effort over several years. This model constitutes a fertile testbed
to model possible game-theoretic approaches to provider and insurer decisions. Possible
approaches include stochastic games framework to model strategic, concurrent agent decisions.
Such approaches would make agents more prudent, and enhance the model’s ability to capture
adaptive behaviors and market trajectories. This makes the model a suitable ground to reconcile
theory and application. The theory is used to conceptualize the model and agent decisions, and the
model outcomes would be validated to actual outcomes to evaluate the suitability of the theory to
explain actual agent behaviors and patterns.
7. Conclusion
The primary objective of this dissertation was to construct artificial health care markets and
investigate the possible impact of ACO formations on competitive dynamics. This was achieved
by outlining a structural framework which abstracts defining components of private health care
markets to recreate tractable, verifiable, and meaningful closed systems that mimic the health
market. The structural framework was operationalized with mathematical and theoretical rigor and
programmed as an ABM using Java on Anylogic. An assessment of the base model exemplified
its ability to recreate competitive subtleties and its comparability to empirical findings on the
aggregate level.
The model experimented with illustrative ACO formations that were induced in a wide range of
market structures and sizes. ACO formation makes two providers participate in an ACO with
context-specific impacts on bargaining and quality. Even with the illustrative nature of this
experiment and simplified ACO structure, highly relevant insights surfaced with the analysis of
ACO effects and sensitivity to model parameters. For instance, illustrative results showed that in
competitive markets, ACO providers had the most to gain from forming ACOs in terms of market
116
power and ACO price. Provider prices were lower in competitive markets due to the balanced
insurer/provider market power, and ACO formations tipped this balance in favor of the ACO
providers. While in less competitive markets with high baseline provider prices, ACO formations
induced a smaller marginal provider price increase. These insights challenged notions of
expediting the anti-trust agency’s review process for ACO formations in unconcentrated markets
(‘safety zone’) because ACO formations were deemed unlikely to be anti-competitive. Several
other insights relating to the synergistic impact of consumer quality preferences and risk-based
agreements on sustaining value-based competition were revealed. Emergent provider behaviors
were observed as provider quality orientation was varied and impact of ACO formations changed.
The case study on ACOs showed the benefit of having docile hypothetical markets on which a
multitude of interventions and payment models with competitive angles can be explored.
The key ACO insights and model developed here were obtained using complexity science
approaches and systems engineering methodologies. The research methodology was characterized
as a generative approach to explain system and agent behaviors compared to conventional
reductionist approaches. A ground-up, explicit model has been built to recreate health care market
dynamics with the potential to test and evaluate market policies and intervention, as shown with
the ACO case study. The tool developed to study health care markets as a CAS was an ABM that
operationalized the conceptual and mathematical relationships. The ABM exemplified the
advantages of using the agent-based technique and contributes by demonstrating the feasibility
and merits of using ABM to study complex markets. The model richness enabled extensive
analysis and explorations of emergent agent and system behaviors such as value-based competition
and decision tendencies. This research contributed to the field of systems engineering by extending
its methodologies and techniques to utilize theories and compliment methods from health
economics, decision analysis, behavioral sciences, and econometrics. Other contributions included
demonstrating the approach’s utility in the development of a decision tool to assess ACO formation
and uncover insights elusive to conventional methodologies.
The programmed model had an intuitive, functional front-end for users to experiment with the
artificial market and understand its trajectory and progress. ABM of markets were a more organic
representation of the involved agents where the decision maker can observe individual entities and
system level behaviors. The model included ways to assess various tradeoffs surrounding
competitive conditions and market share, prices, and health outcome. This tool offered a
comprehensive view into key identifiable aspects of health care markets to enable decision makers
to understand the model reach deep into the richness of those artificial markets and learn about the
dynamics of health care markets. The model, as a decision tool, handled illustrative ACO
formations enabling decision makers to run experiments and verify model relationships. In
summary, this research produced a decision tool designed to enable market experiments,
communicate vital aspects of the modeled market, and engage users to explore in the closed,
tractable, and comprehensive health care market.
The insights gained from this model and approach complement existing efforts in the field of
economics to explore the complex nature of competition in health care delivery. While the modeled
markets may fall short when compared to actual health markets, this model helped bridging a
disconnect between classical theoretical models on competition (monopolies, oligopolies, and
perfect competition) with mathematical and statistical econometric studies. Microeconomic,
117
behavioral economic, and utility theory were synthesized in the model while econometric methods
were used to study the outcomes. In addition, the model has potential in addressing limitations by
exploring the structural sensitivity, agent decision structures, and exogenous market influences. At
the fundamental level, the insights learned from this model spark ideas to challenge existing
notions and policy norms. The findings help identify possible policy levers and areas for empirical
efforts.
The artificial health care markets would serve as testbeds for a variety of different interventions
and delivery models. With the illustrative ACO experiment, several different facets were studied;
including patient choice, health outcomes, provider-insurer bargaining, insurer network formation,
and total health expenditures. This exemplifies the utility of such generative models and the future
value in such methodologies. As health challenges become more complex, risky, and costly,
developing innovative methods is imperative. Health care markets are ever-changing with
evolving health policies, provider and insurer behavior, and patient needs. This research brings
innovative methods a step closer to actualization. The model built here is made feasible with the
preponderance of individual level data (electronic medical records, utilization patterns, health
insurance choice data), maturity of theoretical work (decision analysis, behavioral economics) and
availability of computational power. The synthesis of all these resources and mechanisms was
enabled through complexity theory and systems engineering. Using systems engineering, methods,
theories, and data from multiple domains are harnessed to overcome disciplinary bounds and offer
a different perspective to understand exceedingly complex challenges, as demonstrated in this
dissertation.
118
References
[1] A. Dunn and A. H. Shapiro, “Physician Market Power and Medical-Care Expenditures,”
NBER Work. Pap., pp. 1–59, 2012.
[2] J. Davis and R. Pear, “Trump Issues Executive Order Scaling Back Parts of Obamacare -
The New York Times,” 2017. [Online]. Available:
https://www.nytimes.com/2017/01/20/us/politics/trump-executive-order-obamacare.html.
[Accessed: 03-Feb-2017].
[3] Jesssie Hellman, “Insurers confront big ObamaCare decision | TheHill,” The Hill.
[Online]. Available: http://thehill.com/policy/healthcare/337876-insurers-confront-big-
obamacare-decision. [Accessed: 19-Jun-2017].
[4] M. Gaynor, K. Ho, and R. J. Town, “The Industrial Organization of Health-Care
Markets,” J. Econ. Lit., vol. 53, no. 2, pp. 235–284, 2015.
[5] G. J. Bazzoli, S. M. Shortell, F. Ciliberto, P. D. Kralovec, and N. L. Dubbs, “Tracking the
changing provider landscape: Implications for health policy and practice,” Health Aff.,
vol. 20, no. 6, pp. 188–196, 2001.
[6] S. A. Schroeder, “Improving the Health of the American People,” N. Engl. J. Med., vol.
357, no. 12, pp. 1221–1228, 2007.
[7] US Department of Health and Human Services, “2013 National Healthcare Quality
Report,” 2014.
[8] D. R. Rittenhouse, S. M. Shortell, and E. S. Fisher, “Primary Care and Accountable Care -
Two Essential Elements of Delivery System Reform,” N. Engl. J. Med., vol. 361, no. 124,
2009.
[9] M. England, “From fee-for-service to Accountable Health Plans,” in Schreter R,
Sharfstein S, Schreter C (eds) Allies and adversaries, Washington DC: American
Psychiatric Association Press, 1994, pp. 3–5.
[10] Patient Protection and Affordable Care Act. 2010, pp. 1–906.
[11] A. C. Enthoven, “The History and Principles of Managed Competition,” Health Aff., vol.
119
12, pp. 24–48, 1993.
[12] P. M. Ellwood, A. C. Enthoven, and L. Etheredge, “The Jackson Hole initiatives for a
twenty-first century American health care system.,” Health Econ., vol. 1, pp. 149–168,
1992.
[13] Covered California, “Coverage Levels/Metal Teirs,” 2017. [Online]. Available:
http://www.coveredca.com/individuals-and-families/getting-covered/coverage-
basics/coverage-levels/. [Accessed: 06-Oct-2017].
[14] R. Kronick, D. Goodman, J. Wennberg, and E. Wagner, “The Marketplace in Health Care
Reform - The Demographic Limitations of Managed Competition,” N. Engl. J. Med., vol.
328, no. 2, 1993.
[15] A. C. Enthoven, “Market-Based Reform of U . S . Health Care Financing and Delivery :
Managed Care and Managed Competition,” pp. 195–214.
[16] L. Einav and J. Levin, “Managed competition in health insurance,” J. Eur. Econ. Assoc.,
vol. 13, no. 6, pp. 998–1021, 2015.
[17] H. Centers for Medicare and Medicaid Services (CMS), “CMS.gov Accountable Care
Organizations,” 2015. [Online]. Available: http://www.cms.gov/Medicare/Medicare-Fee-
for-Service-Payment/ACO/index.html?redirect=/aco.
[18] A. S. M. Shortell, R. Gillies, and F. Wu, “United States Innovations in Health Care
Delivery,” Public Health Rev., vol. 32, no. 1, pp. 190–212.
[19] D. R. Rittenhouse, S. M. Shortell, and E. S. Fisher, “Primary Care and Accountable Care
— Two Essential Elements of Delivery-System Reform,” Mew Engl. J. Med., vol. 361,
no. 24, pp. 2301–2303, 2009.
[20] D. J. Nyweide et al., “Association of Pioneer Accountable Care Organizations vs
Traditional Medicare Fee for Service With Spending, Utilization, and Patient Experience,”
J. Am. Med. Assoc., vol. 313, no. 21, pp. 2152–2161, 2015.
[21] D. Muhlestein and M. McCellan, “Accountable Care Organizations In 2016: Private And
Public-Sector Growth And Dispersion,” Health Affairs Blog, 2016. [Online]. Available:
http://healthaffairs.org/blog/2016/04/21/accountable-care-organizations-in-2016-private-
and-public-sector-growth-and-dispersion/. [Accessed: 06-Aug-2017].
120
[22] R. A. Berenson, P. B. Ginsburg, and N. Kemper, “Unchecked Provider Clout In California
Foreshadowes Challenges to Health Reform,” Health Aff., vol. 29, no. 4, pp. 699–705,
2010.
[23] M. Gaynor and R. J. Town, “Competition in Health Care Markets,” Handb. Heal. Econ.,
vol. 2, no. 12, pp. 499–637, 2011.
[24] J. C. Robinson and E. L. Nolan, “Accountable Care Organizations in California - Lessons
for the National Debate on Delivery System Reform,” Integr. Healthc. Assoc. White Pap.,
2010.
[25] J. Goldsmith, “Accountable Care Organizations: the Case for Flexible Partnerships
Between Health Plans and Providers,” Health Aff., vol. 30, pp. 32–40, 2011.
[26] R. Berenson and R. Burton, “Accountable Care Organizations in Medicare and the Private
Sector : A Status Update,” 2012.
[27] Federal Trade Commission and Department of Justice, “Statement of Antitrust
Enforcement Policy Regarding Accountable Care Organizations Participating in the
Medicare Shared Savings Program.”
[28] M. Gaynor, R. Town, and K. Ho, “The Industrial Organization of Health Care Markets,”
NBER Work. Pap., 2014.
[29] L. Tesfatsion, “Agent-based computational economics: modeling economies as complex
adaptive systems,” Inf. Sci. (Ny)., vol. 149, no. December 2002, pp. 262–268, 2003.
[30] G. F. Anderson, U. E. Reinhardt, P. S. Hussey, and V. Petrosyan, “It’s the prices, stupid:
Why the United States is so different from other countries,” Health Aff., vol. 22, no. 3, pp.
89–105, 2003.
[31] K. Arrow, “Uncertainty and the welfare economics of medical care,” Am. Econ. Rev., pp.
941–973, 1963.
[32] L. Dafny, K. Ho, and M. Varela, “Let them Have Choice: Gains from Shifting away from
Employer-Sponsored Health Insurance and Toward an Individual Exchange,” Am. Econ.
J. Econ. Policy, vol. 5, no. July, pp. 32–58, 2013.
121
[33] D. Polsky, J. Weiner, C. Colameco, and N. V. Becker, “Health Insurance Marketplace
Enrollment Rates by Type of Exchange,” Data Briefs, Mar. 2014.
[34] I. Graetz, C. M. Kaplan, E. K. Kaplan, J. E. Bailey, and T. M. Waters, “The U.S. Health
Insurance Marketplace: Are Premiums Truly Affordable?,” Ann. Intern. Med., vol. 161,
no. 8, p. 599, Oct. 2014.
[35] A. Burke, A. Misra, and S. Sheingold, “Premium Affordability, Competition, and Choice
in the Health Insurance Marketplace, 2014,” 2014.
[36] B. D. Sommers, T. Musco, K. Finegold, M. Z. Gunja, A. Burke, and A. M. McDowell,
“Health Reform and Changes in Health Insurance Coverage in 2014,” N. Engl. J. Med.,
vol. 371, no. 9, pp. 867–874, Aug. 2014.
[37] J. Church and R. Ware, Industrial organization: a strategic approach, no. January. 2000.
[38] D. W. Carlton and J. M. Perloff, Modern industrial organization. 2005.
[39] T. F. Bresnahan, “Empirical Studies of Industries with Market Power,” Handbook of
Industrial Organization, vol. II. pp. 1021–1957, 1989.
[40] D. P. Scanlon, M. Chernew, S. Swaminathan, and W. Lee, “Competition in health
insurance markets: limitations of current measures for policy analysis.,” Med. Care Res.
Rev., vol. 63, no. 6, p. 37S–55S, 2006.
[41] J. B. Baker and T. F. Bresnahan, “Empirical Methods of Identifying and Measuring
Market Power,” Antitrust Law J., vol. 61, no. 1, pp. 3–16, 1992.
[42] K. G. Elzinga and D. E. Mills, “The Lerner Index of monopoly power: Origins and uses,”
in American Economic Review, 2011, vol. 101, no. 3, pp. 558–564.
[43] G. Gowrisankaran, R. Town, and C. Capps, “Mergers When Prices are Negotiated:
Evidence from the Hospital Industry,” National, 2013.
[44] L. C. Baker, “Measuring Competition in Health Care Markets,” Health Serv. Res., vol. 36,
no. 1 Pt 2, pp. 223–251, 2001.
[45] E. Glen Weyl and J. Tirole, “Market power screens willingness-to-pay,” Q. J. Econ., vol.
122
127, no. 4, pp. 1971–2003, 2012.
[46] C. S. Capps, D. Dranove, and M. Satterthwaite, “Competition and market power in option
demand markets.,” Rand J. Econ., vol. 34, no. 4, pp. 737–763, 2003.
[47] R. Town, R. Feldman, L. R. Burns, and D. Wholey, “The Welfare Consequences of
Hospital Mergers,” Health Care (Don. Mills)., pp. 0–49, 2005.
[48] Federal Trade Commission, “Improving Health Care: A Dose of Competition,” 2004.
[49] S. Kleiner, S. Lyons, and W. D. White, “Provider Concentration in Markets for Physician
Services for Patients with Traditional Medicare,” Health Manage., vol. 1, no. 1, pp. 3–18,
2012.
[50] M. Gaynor and R. Town, “The Impact of Hospital Consolidation —Update,” Robert Wood
Johnson Found. Policy Br., vol. 9, no. June, pp. 1–8, 2012.
[51] J. E. Schneider, P. Li, D. G. Klepser, N. A. Peterson, T. T. Brown, and R. M. Scheffler,
“The effect of physician and health plan market concentration on prices in commercial
health insurance markets,” Int. J. Health Care Finance Econ., vol. 8, no. 1, pp. 13–26,
2008.
[52] C. White, A. M. Bond, and J. D. Reschovsky, “High and Varying Prices for Privately
Insured Patients Underscore Hospital Market Power,” Res. Br. Find. from Heal. Syst.
Chang., vol. 27, 2013.
[53] V. R. Fuchs, “Managed care and merger mania.,” JAMA, vol. 277, no. 11, pp. 920–921,
1997.
[54] R. J. Town, D. Wholey, R. Feldman, and L. R. Burns, “Revisiting the relationship
between managed care and hospital consolidation,” Health Serv. Res., vol. 42, no. 1 I, pp.
219–238, 2007.
[55] D. Dranove, C. J. Simon, and W. D. White, “Is managed care leading to consolidation in
health-care markets?,” Health Serv. Res., vol. 37, no. 3, pp. 573–594, 2002.
[56] L. Dafny, M. Duggan, and S. Ramanarayanan, “Paying a premium on your premium?
Consolidation in the US health insurance industry,” American Economic Review, vol. 102,
no. 2. pp. 1161–1185, 2012.
123
[57] M. R. McKellar, S. Naimer, M. B. Landrum, T. B. Gibson, A. Chandra, and M. Chernew,
“Insurer market structure and variation in commercial health care spending,” Health Serv.
Res., vol. 49, no. 3, pp. 878–892, 2014.
[58] G. a. Melnick, Y. C. Shen, and V. Y. Wu, “The Increased Concentration Of Health Plan
Markets Can Benefit Consumers Through Lower Hospital Prices,” Health Aff., vol. 30, no.
9, pp. 1728–1733, 2011.
[59] K. Ho and R. S. Lee, “Insurer Competition and Negotiated Hospital Prices,” Natl. Bur.
Econ. Res., vol. Working Pa, no. December, pp. 1–39, 2013.
[60] K. Swartz, M. A. Hall, and T. S. Jost, “Realizing Health Reform’s Potential How Insurers
Competed in the Affordable Care Act’s First Year,” 2015.
[61] Steven Findlay, “As Insurance Options Shrink, Families Are ‘Holding Our Breath’ |
Kaiser Health News,” Kaiser Health News. [Online]. Available: http://khn.org/news/as-
delaware-insurance-options-shrink-families-are-holding-our-breath/. [Accessed: 19-Jun-
2017].
[62] Will Stone, “Protected But Priced Out: Patients Worry About Health Law’s Future In
Arizona | Kaiser Health News,” Kaiser Health News. [Online]. Available:
http://khn.org/news/protected-but-priced-out-patients-worry-about-health-laws-future-in-
arizona/. [Accessed: 19-Jun-2017].
[63] M. Gaynor, “Competition and Quality in Health Care Markets,” Found. Trends
Microeconomics, vol. 2, no. 6, pp. 441–508, 2007.
[64] K. Ho, “Insurer-Provider Networks in the Medical Care Market,” no. 2000, pp. 1–58,
2009.
[65] J. H. Miller and S. E. Page, Complex Adaptive Systems: An Introduction to Computational
Models of Social Life, vol. 27. Princeton University Press, 2007.
[66] Centers for Medicare & Medicaid Services, Shared Savings Program, Section 1899 of the
Social Security Act (as amended by Affordable Care Act Section 3022 Medicare Shared
Saving Program). 2011.
[67] M. Petersen, D. Muhlestein, and P. Gardner, “Growth and Dispersion of Accountable Care
124
Organizations : August 2013 Update,” Cent. Accountable Care Intell., no. August, 2013.
[68] E. Shigekawa and K. Lausch, “Commercial ACO Products: Market Leaders and Trends,”
2014.
[69] J. Goldsmith, L. R. Burns, A. Sen, and T. Goldsmith, “Integrated Delivery Networks: In
Search of Benefits and Market Effects,” no. February, 2015.
[70] A. E. Cuellar and P. J. Gertler, “Strategic integration of hospitals and physicians,” J.
Health Econ., vol. 25, pp. 1–28, 2006.
[71] Centers for Medicare and Medicaid Services (CMS), HHS, “Medicare program; Medicare
Shared Savings Program: Accountable Care Organizations. Final rule.,” Fed. Regist., vol.
76, no. 212, pp. 67802–67990, 2011.
[72] California Healthcare Foundation, “Arranged Marriages : The Evolution of ACO
Partnerships in California,” Calif. Healthc. Found. Heal. Care Alm., no. september, 2013.
[73] E. Trompeter, E. Gomes, and J. Brown, “Case Study From Public and Private ACO -
What Works and Why.”
[74] A. Higgins, K. Stewart, K. Dawson, and C. Bocchino, “Early Lessons From Accountable
Care Models In The Private Sector: Partnerships Between Health Plans and Providers,”
Health Aff., vol. 30, pp. 1718–1727, 2011.
[75] D. Suzanne et al., “Promising Payment Reform: Risk-Sharing with Accountable Care
Organizations,” 2011.
[76] V. A. Lewis, A. B. McClurg, J. Smith, E. S. Fisher, and J. P. W. Bynum, “Attributing
patients to accountable care organizations: Performance year approach aligns
stakeholders’ interests,” Health Aff., vol. 32, no. 3, pp. 587–595, 2013.
[77] V. A. Lewis, C. H. Colla, K. L. Carluzzo, S. E. Kler, and E. S. Fisher, “Accountable care
organizations in the United States: Market and demographic factors associated with
formation,” Health Serv. Res., vol. 48, no. 6 PART1, pp. 1840–1858, 2013.
[78] B. J. L. Berry, L. D. Kiel, and E. Elliott, “Adaptive Agents, Intelligence, and Emergent
Human Organization: Capturing Complexity Through Agent-Based Modeling.,” Proc.
Natl. Acad. Sci. U. S. A., vol. 99 Suppl 3, pp. 7187–7188, 2002.
125
[79] W. K. V. Chan, Y.-J. Son, and C. M. Macal, “Agent-Based Simulation Tutorial -
Simulation of Emergent Behavior and Difference Between Agent-Based Simulation and
Discrete-Event Simulation,” Proc. 2010 Winter Simul. Conf., pp. 135–150, 2010.
[80] R. Axelrod, The Complexity of Cooperation - Agent Based Models of Competition and
Collaboration, vol. 1, no. 1. Princeton University Press, 1997.
[81] J. M. Epstein, Growing Artificial Societies - Social Science from the Bottom Up, vol. 3,
no. 1. 2012.
[82] J. B. Urban, N. Osgood, and P. Mabry, “Developmental Systems Science: Exploring the
Application of Systems Science Methods to Developmental Science Questions,” Res.
Hum. Dev., vol. 8, no. 1, pp. 1–25, 2011.
[83] P. W. Anderson, K. Arrow, and D. Pines, “The economy as an evolving complex system,”
1988.
[84] W. B. Arthur, S. N. Durlauf, and D. A. Lane, The economy as an evolving complex system
II, vol. 28. Addison-Wesley Reading, MA, 1997.
[85] L. E. Blume and S. N. Durlauf, The economy as an evolving complex system, III: Current
perspectives and future directions. Oxford University Press, 2005.
[86] D. Kernick, Complexity and healthcare organisation. 2002.
[87] P. Plsek, “Redesigning Health Care with Insights from Complexity Science,” in Crossing
the Quality Chasm: A New Health System for the 21st Century, 2001.
[88] A. V. D. Roux, “Complex systems thinking and current impasses in health disparities
research,” Am. J. Public Health, vol. 101, no. 9, pp. 1627–1634, 2011.
[89] A. H. Auchincloss and A. V. Diez Roux, “A new tool for epidemiology: The usefulness of
dynamic-agent models in understanding place effects on health,” Am. J. Epidemiol., vol.
168, no. 1, pp. 1–8, 2008.
[90] J. P. Sturmberg and C. Martin, Handbook of systems and complexity in health. Springer
Science & Business Media, 2013.
126
[91] P. Plsek, “Complexity and the Adoption of Innovation in Health Care Complexity and the
Adoption of Innovation in Health Care,” 2003.
[92] T. Wilson, T. Holt, and T. Greenhalgh, “Complexity science: Complexity and clinical
care,” BMJ Br. Med. J., vol. 323, no. 7314, pp. 685–688, 2001.
[93] W. L. Miller, B. F. Crabtree, R. Mcdaniel, and K. C. Stange, “Understanding change in
primary care practice using complexity theory,” J. Fam. Pract., vol. 46, no. 5, p. 396,
1998.
[94] C. D. Norman, A. Best, S. Mortimer, T. Huerta, and A. Buchan, “Evaluating the Science
of Discovery in Complex Health Systems,” Am. J. Eval., vol. 32, no. 1, pp. 70–84, 2011.
[95] J. M. Epstein, Generative Social Science: Studies in Agent-Based Computational
Modeling. 2006.
[96] R. A. Hammond, “Complex Systems Modeling for Obesity,” Prev. Chronic Dis., vol. 6,
no. 3, pp. 1–10, 2009.
[97] D. Rickles, P. Hawe, and A. Shiell, “A simple guide to chaos and complexity.,” J.
Epidemiol. Community Health, vol. 61, no. 11, pp. 933–7, 2007.
[98] G. L. Nell, “Competition as Market Progress: An Austrian Rationale for Agent-Based
Modeling,” Rev. Austrian Econ., vol. 23, pp. 127–145, 2010.
[99] D. Bunn and F. S. Oliveira, “Agent-based simulation - an application to the new electricity
trading arrangements of England and Wales. IEEE Transactions on,” Evol. Comput. 2001,
Vol, vol. 5, pp. 5p493–503, 2001.
[100] D. W. Bunn, “Evaluating Individual Market Power in Electricity Markets via Agent-based
Simulation,” Ann. Oper. Res., vol. 121, pp. 57–77, 2003.
[101] E. Guerci, M. A. Rastegar, and S. Cincotti, “Agent-based Modeling and Simulation of
Competitive Wholesale Electricity Markets,” Simulation, pp. 241–286, 2010.
[102] K. Cai, J. Niu, and S. Parsons, “On the Effects of Competition Between Agent-Based
Double Auction Markets,” Electron. Commer. Res. Appl., vol. 13, no. 4, pp. 229–242,
127
2014.
[103] I. Dogan, R. B. Chinnam, Y. Jia, and G. Vanteddu, “Design and Analysis of Agents for
Supply Chain Management: Experiences From the Trading Agent Competition,” Int. J.
Model. Simul., vol. 28, no. 4, 2008.
[104] L. Tesfatsion, Structure, behavior, and market power in an evolutionary labor market with
adaptive search, vol. 25. 2001.
[105] L. Feng, B. Li, B. Podobnik, T. Preis, and H. E. Stanley, “Linking agent-based models and
stochastic models of financial markets,” Proc Natl Acad Sci U S A, vol. 109, no. 22, pp.
8388–8393, 2012.
[106] G. Pegoretti, F. Rentocchini, and G. V. Marzetti, “An Agent-Based Model of Product
Competition: Network Structure and Coexistence Under Different Information Regimes,”
2010.
[107] A. P. Kirman and N. J. Vriend, “Evolving market structure: An ACE model of price
dispersion and loyalty,” J. Econ. Dyn. Control, vol. 25, pp. 459–502, 2001.
[108] S. Glied and N. Tilipman, “Simulation modeling of health care policy.,” Annu. Rev. Public
Health, vol. 31, pp. 439–455, 2010.
[109] D. M. Gaba and D. Raemer, “The tide is turning: organizational structures to embed
simulation in the fabric of healthcare.,” J. Soc. Simul. Healthc., vol. 2, no. 1, pp. 1–3,
2007.
[110] Institute of Medicine & Committee on Quality of Health Care in America, Crossing the
Quality Chasm: A New Health System for the 21st Century, vol. 323, no. 7322. 2001.
[111] P. Philippe and O. Mansi, “Nonlinearity in the epidemiology of complex health and
disease process,” Theor. Med. Bioeth., vol. 19, pp. 591–607, 1998.
[112] P. Liu and S. Wu, “An agent-based simulation model to study accountable care
organizations.,” Health Care Manag. Sci., Apr. 2014.
[113] A. Alibrahim and S. Wu, “An agent-based simulation model of patient choice of health
care providers in accountable care organizations,” Health Care Manag. Sci., pp. 1–13,
2016.
128
[114] T. G. Trucano, “Prediction and Uncertainty in Computational Modeling of Complex
Phenomena: A Whitepaper,” New Mexico, 1998.
[115] R. Andersen and J. F. Newman, “Societal and individual determinants of medical care
utilisation in the United States,” Milbank Mem. Fund Q., vol. 51, no. 1, pp. 95–124, 1973.
[116] J. T. Kolstad and M. E. Chernew, “Quality and Consumer Decision Making in the Market
for Health Insurance and Health Care Services,” Med. Care Res. Rev. Suppl. to, vol. 66,
no. 1, pp. 28–52, 2009.
[117] A. Dixon, R. Ruth, J. Appleby, P. Burge, N. Devlin, and H. Magee, “Patient choice: how
patients choose and how providers respond,” 2010.
[118] B. Gauthier and W. Wane, “Bypassing health providers: the quest for better price and
quality of health care in Chad.,” Soc. Sci. Med., vol. 73, no. 4, pp. 540–9, Aug. 2011.
[119] M. Varkevisser and S. a van der Geest, “Why do patients bypass the nearest hospital? An
empirical analysis for orthopaedic care and neurosurgery in the Netherlands.,” Eur. J.
Health Econ., vol. 8, no. 3, pp. 287–95, Sep. 2007.
[120] C. Saunders, G. R. Bellamy, N. Menachemi, A. S. Chukmaitov, and R. G. Brooks,
“Bypassing the local rural hospital for outpatient procedures.,” J. Rural Health, vol. 25,
no. 2, pp. 174–81, Jan. 2009.
[121] W.-T. C. Tai, F. W. Porell, and E. K. Adams, “Hospital choice of rural Medicare
beneficiaries: patient, hospital attributes, and the patient-physician relationship.,” Health
Serv. Res., vol. 39, no. 6 Pt 1, pp. 1903–22, Dec. 2004.
[122] M. McGuirk and F. W. Porell, “Spatial patterns of hospital utilization: the impact of
distance and time.,” Inquiry, vol. 21, no. 1, pp. 84–95, Jan. 1984.
[123] W. G. Manning, J. P. Newhouse, N. Duan, E. B. Keeler, A. Leibowitz, and M. S. Marquis,
“Health insurance and the demand for medical care: evidence from a randomized
experiment,” Am Econ Rev, vol. 77, no. 3, pp. 251–277, 1987.
[124] E. B. Keeler, J. L. Buchanan, J. E. Rolph, J. M. Hanley, and D. M. Reboussin, “Demand
for Episodes of Medical Treatment in the Health Insurance Experiment.” RAND, 1988.
129
[125] S. A. Flocke, K. C. Stange, and S. J. Zyzanski, “The impact of insurance type and forced
discontinuity on the delivery of primary care.,” J. Fam. Pract., vol. 45, no. 2, pp. 129–
135, 1997.
[126] Alain Enthoven, “Managed Competition 2014: Rescued By The Private Sector?” [Online].
Available: http://healthaffairs.org/blog/2014/05/12/managed-competition-2014-rescued-
by-the-private-sector/. [Accessed: 28-Jun-2017].
[127] J. R. Gabel, H. Whitmore, M. Green, S. T. Stromberg, D. S. Weinstein, and R. Oran, “In
second year of marketplaces, new entrants, ACA ‘Co-Ops,’ and medicaid plans restrain
average premium growth rates,” Health Aff., vol. 34, no. 12, pp. 2020–2026, 2015.
[128] Centers for Medicare & Medicaid Services, “Market Rating Reforms,” 2013. [Online].
Available: https://www.cms.gov/CCIIO/Programs-and-Initiatives/Health-Insurance-
Market-Reforms/Market-Rating-Reforms.html. [Accessed: 20-Jun-2017].
[129] H. Peters, Game Theory - A Multi-leveled Approach. Springer, 2008.
[130] R. Marks, “Breeding Hybrid Strategies: Optimal Behavior for Oligopolists,” J. Evol.
Econ., vol. 2, pp. 17–38, 1992.
[131] K. R. Brekke, L. Siciliani, and O. R. Straume, “Price and quality in spatial competition,”
Reg. Sci. Urban Econ., vol. 40, no. 6, pp. 471–480, Nov. 2010.
[132] D. Polsky and J. Weiner, “State Variation in Narrow Networks on the ACA
Marketplaces,” 2015.
[133] R. Cooper, “Health Plan Features: Implications of Narrow Networks and the Trade-Off
between Price and Choice,” 2014.
[134] E. Elliott and L. D. Kiel, “Exploring Cooperation and Competition Using Agent-Based
Modeling,” Proc. Natl. Acad. Sci. USA, vol. 99, no. 10, pp. 7193–7194, 2002.
[135] D. Challet and Y.-C. Zhang, “Emergence of Cooperation and Organization in an
Evolutionary Game,” p. 8, 1997.
[136] R. A. McCain, “Barriers and Bounds to Rationality: Essays on Economic Complexity and
Dynamics in Interactive Systems (Duncan K. Foley’s edition),” J. Artif. Soc. Soc. Simul.,
vol. 2, no. 3–4, 1999.
130
[137] E. W. Ford, R. Wells, and B. Bailey, “Sustainable network advantages: a game theoretic
approach to community-based health care coalitions.,” Health Care Manage. Rev., vol. 29,
no. 2, pp. 159–169, 2004.
[138] M. Tambe, “Implementing Agent Teams in Dynamic Multiagent Environments,” Appl.
Artif. Intell., vol. 12, pp. 189–210, 1998.
[139] R. Town and G. Vistnes, “Hospital Competition in HMO Networks,” J. Health Econ., vol.
20, pp. 733–753, 2001.
[140] G. Gowrisankaran, A. Nevo, and R. Town, “Mergers when prices are negotiated:
Evidence from the hospital industry,” Am. Econ. Rev., vol. 105, no. 1, pp. 172–203, 2015.
[141] M. S. Lewis and K. E. Pflum, “Diagnosing hospital system bargaining power in managed
care networks,” Am. Econ. J. Econ. Policy, vol. 7, no. 1, pp. 243–274, 2015.
[142] G. Rausser, J. Swinnen, and P. Zusman, “The Nash Solution to the Bargaining Problem,”
in Political Power and Economic Policy - Theory, Analysis, and Empirical Applications,
2012, pp. 30–49.
[143] M. C. Shields, P. H. Patel, M. Manning, and L. Sacks, “A model for integrating
independent physicians into accountable care organizations,” Health Aff., vol. 30, no. 1,
pp. 161–172, 2011.
[144] C. E. Pollack, G. E. Weissman, K. W. Lemke, P. S. Hussey, and J. P. Weiner, “Patient
sharing among physicians and costs of care: A network analytic approach to care
coordination using claims data,” J. Gen. Intern. Med., vol. 28, no. 3, pp. 459–465, 2013.
[145] S. H. Jee and M. D. Cabana, “Indices for continuity of care: a systematic review of the
literature.,” Med. Care Res. Rev., vol. 63, no. 2, pp. 158–88, 2006.
[146] A. C. Enthoven, “Integrated Delivery Systems: The Cure for Fragmentation,” Am. J.
Manag. Care, vol. 15, no. December, pp. S284–S290, 2009.
[147] J. McWilliams, L. BE, and C. ME, “Changes in health care spending and quality for
medicare beneficiaries associated with a commercial aco contract,” JAMA, vol. 310, no. 8,
pp. 829–836, Aug. 2013.
131
[148] The Anylogic Company, “Anylogic University Edition.” 2015.
[149] Centers for Disease Control and Prevention (CDC), National Center for Health Statistics
(NCHS), and National Health and Nutrition Examination Survey Questionnaire, “No
Title,” Hyattsville, MD, 2014.
[150] J. L. Buchanan, E. B. Keeler, J. E. Rolph, and M. R. Holmer, “Simulating Health
Expenditures Under Alternative Insurance Plans,” no. September 2016, 1991.
[151] D. K. K. Lee and S. a. Zenios, “An Evidence-Based Incentive System for Medicare’s End-
Stage Renal Disease Program,” Manage. Sci., vol. 58, no. April 2015, pp. 1092–1105,
2012.
[152] D. P. Ly, A. K. Jha, and A. M. Epstein, “The association between hospital margins,
quality of care, and closure or other change in operating status,” J. Gen. Intern. Med., vol.
26, no. 11, pp. 1291–1296, 2011.
[153] Centers for Disease Control and Prevention (CDC) and National Center for Health
Statistics (NCHS), “Compressed Mortality File 1999-2015 on CDC WONDER Online
Database.”
[154] Centers for Disease Control and Prevention (CDC) and National Center for Health
Statistics (NCHS), “Multiple Cause of Death 1999-2015.”
[155] Manatt, Phelps & Phillips, and Vericred Inc., “HIX Compare Datasets,” Robert Wood
Johnson Foundation, 2017. .
[156] W. Pohlmeier and V. Ulrich, “An Econometric-Model of the 2-Part Decision-Making
Process in the Demand for Health-Care,” J. Hum. Resour., vol. 30, no. 2, pp. 339–361,
1995.
[157] P. Deb and P. K. Trivedi, “The structure of demand for health care: latent class versus
two-part models.,” J. Health Econ., vol. 21, no. 4, pp. 601–625, 2002.
[158] M. Grossman, “The Human Capital Model of the Demand for Health,” J. Polit. Econ., no.
Working Paper No. 7078, p. 102, 1999.
[159] J. F. Fries, “Reducing Medical Need and Demand for Medical Care: Implications for
Quality Management and Outcome Improvement,” Qual. Manag. Health Care, vol. 6, no.
132
1, pp. 34–44, 1997.
[160] A. Victoor, D. M. J. Delnoij, R. D. Friele, and J. J. D. J. M. Rademakers, “Determinants of
patient choice of healthcare providers: a scoping review.,” BMC Health Serv. Res., vol.
12, no. 1, p. 272, Jan. 2012.
[161] A. Victoor, R. D. Friele, D. M. J. Delnoij, and J. J. D. J. M. Rademakers, “Free choice of
healthcare providers in the Netherlands is both a goal in itself and a precondition:
modelling the policy assumptions underlying the promotion of patient choice through
documentary analysis and interviews.,” BMC Health Serv. Res., vol. 12, no. 1, p. 441, Jan.
2012.
[162] J. J. Escarce and K. Kapur, “Do patients bypass rural hospitals? Determinants of inpatient
hospital choice in rural California.,” J. Health Care Poor Underserved, vol. 20, no. 3, pp.
625–44, Aug. 2009.
[163] J. Basu, “Severity of illness, race, and choice of local versus distant hospitals among the
elderly.,” J. Health Care Poor Underserved, vol. 16, no. 2, pp. 391–405, May 2005.
[164] C. Roh and K. Lee, “Hospital Choice By Rural Medicare Beneficiaries: Does Hospital
Ownership Matter? - A Colorado Case,” Jounral Heal. Hum. Serv. Adminstration, vol. 28,
no. 3, pp. 346–365, 2005.
[165] “Health Insurance Marketplace Premiums After Shopping, Switching, And Premium Tax
Credits , 2015–2016,” 2016.
[166] R. Netemeyer, M. Van Ryn, and I. Ajzen, “The theory of planned behavior,”
Orgnizational Behav. Hum. Decis. Process., vol. 50, pp. 179–211, 1991.
[167] Centers for Medicare and Medicaid Services (CMS), “Medicare.gov | Hospital Compare.”
[Online]. Available: https://data.medicare.gov/Hospital-Compare/Readmissions-and-
Deaths-Hospital/ynj2-r877. [Accessed: 01-Jan-2017].
[168] M. G. Vita, “Regulatory restrictions on selective contracting: An empirical analysis of
‘any-willing-provider’ regulations,” J. Health Econ., vol. 20, no. 6, pp. 955–966, 2001.
[169] McKinsey, “Hospital networks: Evolution of the configurations on the 2015 exchanges,”
2015.
133
[170] A. S. Moriya, W. B. Vogt, and M. Gaynor, “Hospital prices and market structure in the
hospital and insurance industries.,” Health Econ. Policy. Law, vol. 5, no. 2010, pp. 459–
479, 2010.
[171] E. E. Trish and B. J. Herring, “How do health insurer market concentration and bargaining
power with hospitals affect health insurance premiums?,” J. Health Econ., vol. 42, pp.
104–114, 2015.
[172] L. Dafny, J. Gruber, and C. Ody, “More Insurers Lower Premiums: Evidence from Initial
Pricing in the Health Insurance Marketplaces,” Am. J. Heal. Econ., vol. 1, no. 1, pp. 53–
81, 2015.
[173] L. Dafny and C. Ody, “New Health Care Symposium: No Evidence That Insurance
Market Consolidation Leads To Greater Innovation,” 2016. [Online]. Available:
http://healthaffairs.org/blog/2016/02/24/no-evidence-that-insurance-market-consolidation-
leads-to-greater-innovation/. [Accessed: 16-Jul-2017].
[174] S. Bennett and M. Smith, “2015 health insurance marketplace competitiveness study,” no.
February, 2015.
[175] J. Holahan, L. J. Blumberg, and E. Wengle, “What Characterizes the Marketplaces with
One or Two Insurers?,” 2017.
[176] M. Gaynor and W. B. Vogt, “Competition among hospitals,” Rand J. Econ., vol. 34, no. 4,
pp. 764–785, 2003.
[177] M. Gaynor and R. J. Town, “Competition in Health Care Markets,” Handb. Heal. Econ.,
no. 2, pp. 499–637, 2012.
[178] P. B. Ginsburg, “Wide Variation in Hospital and Physician Payment Rates Evidence of
Provider Market Power,” Cent. Stud. Heal. Syst. Chance Res. Br., no. 16, pp. 1–11, 2010.
[179] M. E. Porter and E. O. Teisberg, Redefining health care: creating value-based competition
on results. Harvard Business Press, 2006.
[180] Department of Justice and Federal Trade Commission, “Department of Justice/Federal
Trade Commission Issue Final Statement of Antitrust Policy Enforcement Regarding
Accountable Care Organizations,” 2011.
134
[181] C. H. Colla, V. A. Lewis, E. Tierney, and D. B. Muhlestein, “Hospitals participating in
ACOs tend to be large and urban, allowing access to capital and data,” Health Aff., vol.
35, no. 3, pp. 431–439, 2016.
[182] A. Chen and D. Lakdawalla, “Saving Lives or Saving Money? Understanding the Dual
Nature of Physician Preferences,” NBER Work. Pap., 2016.
[183] T. Tu, D. Muhlestein, S. L. Kocot, and R. White, “The Impact of Accountable Care
Origins and Future of Accountable Care Organizations,” p. 11, 2015.
[184] S. M. Burwell, “Setting Value-Based Payment Goals — HHS Efforts to Improve U.S.
Health Care,” N. Engl. J. Med., vol. 372, no. 10, pp. 897–899, 2015.
[185] Z. Cooper, S. Gibbons, S. Jones, and A. McGuire, “Does hospital competition save lives?
Evidence from the English NHS patient choice reforms,” Econ. J., vol. 121, no. 554,
2011.
[186] E. C. Schneider and T. Lieberman, “Publicly disclosed information about the quality of
health care: response of the US public.,” Qual. Health Care, vol. 10, pp. 96–103, 2001.
[187] R. A. Berenson, D. K. Upadhyay, S. F. Delbanco, and R. Murray, “Payment Methods and
Benefit Designs: How They Work and How They Work Together to Improve Health
Care.” 2016.
[188] V. A. Lewis, C. H. Colla, W. L. Schpero, S. M. Shortell, and E. S. Fisher, “ACO
contracting with private and public payers: A baseline comparative analysis,” Am. J.
Manag. Care, vol. 20, no. 12, 2014.
[189] G. Bernile, E. Lyandres, and A. Zhdanov, “A theory of strategic mergers,” Rev. Financ.,
vol. 16, pp. 517–575, 2012.
[190] L. Dafny, K. Ho, and R. Lee, “the Price Effects of Cross-Market Hospital Mergers,”
NBER Work. Pap. Ser., vol. 1, 2016.
[191] Z. Cooper, S. Craig, M. Gaynor, and J. Van Reenen, “The Price Ain’t Right? Hospital
Prices and Health Spending on the Privately Insured,” NBER Work. Pap., p. 37, 2015.
135
Appendices
Appendix A – Model Initialization Settings
This appendix refers to the settings for each agent in the baseline experiment. In this experiment,
parameters and rules were at the following settings:
Time Settings:
o Simulation Time: 5 years
o Observation Years: 3 years
o Bargaining procedures occur in month 11 of every year
o Consumers choose insurers at the beginning of every year
Insurers:
o Insurer quality parameter: uniform(0.3, 0.7)
o Insurers start with full network of providers
o Consumers are assigned to insurer at year 0
No insurer-switching cost incurred by consumers in year 0
o Insurer premiums are initialized at $400 (for a 27-year-old individual)
o Out-of-pocket costs are held at 30% for all plans
o A plan’s premium can only vary with consumer age (3:1 ratio)
$13.5 per additional year of age
Providers:
o Initial price factor: 1
o Effort levels ∈ [1,2,3]
o Initial effort level: 2
o Theory of Planned Behavior parameters
Weights (Sum up to 1)
Attitude weight: 0.6
Perceived control weight: 0.2
Subjective norm weight: 0.2
Current-year intention weight: 0.5
Choice smoother: 0.15
Consumers:
o Yearly Births = yearly deaths + yearly lost to Medicare (age>64)
o Utility-dollar conversion = $1/util
o Choice Smoother: 0.1
o Default Bias:
Insurer: Uniform(15,30)
Provider: Uniform(1,2)
o Health Needs: See 4.2.1.1
o Preferences and utility: See 4.2.1.2
o Choice of providers and insurers: See 4.2.1.2 and 4.2.1.3
136
Appendix B – Base Model Steady State Assessment
An assessment of the model steady state was performed by plotting the trajectories of the simulated
markets over the 5-year period for key agent outcomes. Outcomes relating to provider and insurer
HHIs along with average insurer premiums, provider prices, and provider effort are shown in the
figures below. Each line in the figures below corresponds to a simulated market. The x-axis
corresponds to the simulation year, and the y-axis to the outcome being assessed.
Insurer HHI
1 1.5 2 2.5 3 3.5 4 4.5 5
Year
100
200
300
400
500
600
700
800
900
1000
Insurer Premiums
137
Average Providers Price Factor
1 1.5 2 2.5 3 3.5 4 4.5 5
Year
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3
Provider Effort
138
Appendix C – Base Model Verification
This appendix covers the verification procedure outlined in 5.2.1. For each agent, a set of
procedures are performed to examine the programmed ABM’s consistency with the intended
relationships outlined in chapters 3 and 4. The following procedures are grouped by agent for a
visual inspection of the model’s behaviors.
Consumer Agent Verification
NHANES Sampling Verification:
Variable NHANES Distribution Model Distribution
Age
Gender
0 : Male
1 : Female
Race
1: Mexican American
2: Other Hispanic
3: Non-Hispanic White
4: Non-Hispanic Black
5: Other
Education Level
1: Less than 9
th
Grade
2: 9-11
th
Grade
3: High School Graduate
4: Some College
5: College or Above
139
Income Distribution
Annual income
[0 , 100,000+]
Disability
0: No disability
1: Disabled
Disease Count
Perceived Health Score
0: Poor Perceived Health
1: Excellent Perceived Health
Mental Health Score
0: Poor Perceived Health
1: Excellent Perceived Health
140
RAND Health Insurance Experiment Episodes:
Episodes in Keeler et al, 1988:
Study sample had the following proportions
Hospitalization Acute Chronic Wellness
Proportion of
Episodes
7.63% 40.24% 25.50% 26.63%
Proportion of
Overall cost 71.9% 9.5% 12.3% 6.3%
The proportions above were for a sample of different ages and attributes compared to the
sample used in this model. However, the proportions of episodes generated in the model
should have comparable proportions.
Episodes in a sample model run:
Provider Agent Verification
This verification inspected the relationship between provider effort levels, costs, and outcomes.
The following scatter plots indicate decreased returns in quality outcomes on effort. This is
apparent in the two scatter plots below that relate provider effort and mortality and readmissions
rates.
141
Provider costs increased quadratically with effort. The following scatter plot visibly confirms the
modeled relationships are consistent with the mathematical relations outlined in 4.2.2.
Provider costs in the model were a function of patient utilization. Looking at the per-patient costs
from a provider perspective when utilization rates are multiplied by a UtilizationMultiplier
reflected this increase in costs. For higher utilization rates, the per-patient costs from the provider’s
perspective increased.
200 300 400 500 600
Provider Cost Per Patient
1 1.5 2 2.5 3
Average Provider Effort
Provider Cost & Effort
.008 .01 .012 .014 .016 .018
Readmissions Rate (per Hospitalization)
1 1.5 2 2.5 3
Average Provider Effort
Provider Effort & Rehospitalization Rate
.4 .6 .8 1 1.2
Mortality Rate (per 100)
1 1.5 2 2.5 3
Average Provider Effort
Provider Effort & Mortality Rate
142
Insurer Verification
Insurer verification procedures ensured that insurer agents exhibit relationships consistent with
those expected from the defined relationships in sections 4.2.3 and 4.2.4. Specifically, insurer
premiums increased with higher enrollee utilization and provider inclusion in networks. The
following figures illustrate the observed model relationships and confirm the expected behaviors.
0 200 400 600 800
Provider Cost ($ per patient)
1.5 2 2.5 3
Provider Costs & Patient Utilization Rates
Patient Utilization Multiplier
.4 .6 .8 1
Proportion of Market Providers in Insurer Network
500 600 700 800 900
Insurer Premiums ($)
Insurer Premiums and Proportion of Provider Included in Networks
500 1,000 1,500
Insurer Premiums ($) for 27 yo
1.5 2 2.5 3 3.5
Insurer Premiums For Artifically Inflated Utilization
Utilization Multiplier
143
Appendix D – Base Model Face-Validity Assessment
Model face-validity assessment involved examining the relationships between competitive and
market outcomes. The relationships were then compared to the theorized relationships. The
appendix describes the following model relationships:
Provider concentration and provider prices
Insurer concentration and insurer premiums
Provider concentration and provider effort
Insurer concentration and health expenditures
Insurer concentration and provider prices
For each relationship, a linear fit on a scatter plot was fitted along with a simple linear regression.
If there was a statistically significant association between each pair of competitive and market
outcomes, then the association is said to be “strong”. If a statistically significant association is
observed in certain market configurations or if no association is observed, then the association is
“mixed”.
Provider Concentration & Provider Price
Provider Price Coefficient p-value 95% CI
Provider HHI 0.0000186 <0.001 [0.000013,0.0000243]
Constant 0.876 <0.001 [0.857, 0.894]
This represented a “strong” agreement between the modeled and theorized relationship since there
was a statistically significant relationship obtained from the regression model and observed in the
linear fit to the scatter plot.
Insurer Concentration & Insurer Premiums
.8 .9 1 1.1
Provider Price
2000 4000 6000 8000
Provider HHI
Provider Concentration & Provider Price
144
Insurer Premiums Coefficient p-value 95% CI
Insurer HHI 0.001404 0.382 [-.00175, 0.004562]
Constant 508.9 <0.001 [493.2, 524.6]
300 400 500 600 700
Insurer Premiums ($) for 27 YO
2000 4000 6000 8000 10000
Insurer HHI
Insurer Concentration & Premiums
300 400 500 600
2000 4000 6000 8000 100002000 4000 6000 8000 10000
Population: 30,000* Population: 45,000
Insurer HHI
* p-value < 0.05
Insurer Concentration & Premiums
Average Premiums ($ for 27 YO)
145
This represented a “mixed” agreement between the modeled and theorized relationship since there
was no statistically significant relationship obtained from the regression model and observed in
the linear fit to the scatter plot, but the direction of the observed relationship is consistent (albeit
no statistically significant). In markets with smaller populations, there is a statistically significant
association between insurer concentration and average insurer premiums (p-value = 0.026).
Provider Concentration & Provider Effort
Provider Effort Coefficient p-value 95% CI
Provider HHI -6.33e-06 0.461 [-.0000232, 0.0000105]
Constant 2.019 <0.001 [1.964, 2.074]
By Area:
Area = 200
Provider Effort Coefficient p-value 95% CI
Provider HHI 0.0000157 0.192 [-7.95e-06, 0.0000393]
Constant 1.950 <0.001 [1.873, 2.027]
Area = 400
Provider Effort Coefficient p-value 95% CI
Provider HHI -0.0000303 0.014 [-0.0000543, -6.31e-06]
Constant 2.095 <0.001 [2.017, 2.174]
1 1.5 2 2.5
Average Provider Effort
2000 4000 6000 8000
Provider HHI
Provider Concentration and Provider Effort
146
This represented a “mixed” agreement between the modeled and theorized relationship since a
statistically significant relationship obtained from the regression model and observed in the linear
fit to the scatter plot for some market configurations. Namely, the theorized relationship holds in
markets with larger areas.
Insurer Concentration & Health Expenditures
Yearly Expenditures Coefficient p-value 95% CI
Insurer HHI -.0003256 0.854 [-0.00380, 0.00315]
Constant 959.6 <0.001 [942.3, 976.8]
1 1.5 2 2.5
2000 4000 6000 8000 2000 4000 6000 8000
Area: 200x200 Area: 400x400*
Average Provider Effort
Provider HHI
* p-value<0.05
Provider Concentration and Provider Effort, by Market Area
700 800 900 1000 1100
Yearly Health Expenditures Per Person
2000 4000 6000 8000 10000
Insurer HHI
Insurer Concentration and Health Expenditures
147
By area & consumers in market:
This represented a “mixed” agreement between the modeled and theorized relationship since was
no statistically significant relationship in the regression model or observed in the linear fit to the
scatter plot. The direction of the coefficient is consistent with the theorized relationship direction,
albeit not statistically significant. Also, in markets with 45,000 consumers and area of 200, the
association was marginally significant (p-value = 0.66).
Insurer Concentration & Provider Prices
Provider Effort Coefficient p-value 95% CI
Insurer HHI 5.20e-07 0.787 [-3.27e-06, 4.31e-06]
Constant 2.095 <0.001 [2.017, 2.174]
700 800 900 1000 1100 700 800 900 1000 1100
2000 4000 6000 8000 100002000 4000 6000 8000 10000
Population: 30000, Area:200 Population: 30000, Area: 400
Population: 45000, Area: 200* Population: 45000, Area: 400
Insurer HHI
* p-value<0.1
Insurer Concentration & Health Expenditures
Yearly Expenditures ($ Per Person)
148
The findings represent a “mixed” agreement with the theorized relationship since there were no
statistically significant association between the insurer concentration and provider prices.
.8 .9 1 1.1
Provider Price
2000 4000 6000 8000 10000
Insurer HHI
Insurer Concentration & Provider Prices
149
Appendix E – ACO Continuity of Care Effect Verification
This appendix provides verification that the modeled ACO exhibited the expected effects on
quality of care through improved care continuity. Simple linear regressions were used to check for
a statistically significant association between ACO market share and adverse outcomes (morality
and readmissions rates). At 5% reduction in adverse outcomes, markets with higher concentration
of care delivered at ACOs had lower readmissions rates and mortality rates, consistent with the
5% reduction level. The two regression tables below are supported with a linear fit to a scatter plot
to observe the decrease in adverse outcomes as ACO market share of episodes increases.
The results from the regression models confirmed a significant relationship between ACO
market share and reductions in adverse outcomes (p-value<0.05). This confirmed the continuity
of care component of the modeled ACOs functioned as desired.
.5 .55 .6 .65 .7
Mortality Rates (Per 100 Individuals)
.2 .4 .6 .8 1
ACO Market Share
ACO Care Density & Market Mortality Rates
.0005 .001 .0015 .002
Readmissions Rates (per 100 individuals)
.2 .4 .6 .8 1
ACO Market Share
ACO Care Denisty & Market Readmissions Rates
150
Appendix F – ACO Model Sensitivity Analysis
ACO Parameters
ACOEffortIncrease
ACO Effort
Increase
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
Low – 2.5% +8.77%* +0.61% -3.34%*
Baseline – 5% +8.37%* +2.23%* -4.73%*
High – 7.5% +7.42%* +1.57% -3.84%*
* p-value<0.05
ACO ContinuityofCare
Care Continuity
Reduction in
Adverse Outcomes
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
Low – 2.5% +8.49%* +2.29%* -1.10%
Baseline – 5% +8.37%* +2.23%* -4.73%*
High – 7.5% +7.69%* +2.26%* -6.04%*
* p-value<0.05
Consumer Parameters
Quality Preference
Consumer Quality
Preference
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
None +6.38%* +1.96%* -4.18%*
Baseline +8.37%* +2.23%* -4.73%*
High +7.59%* +2.72%* -5.38%*
* p-value<0.05
Provider Quality Parameters
Profit-Quality Orientation
Provider Quality-
Profit Orientation
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
Profit Oriented +8.68%* +4.06%* -5.80%*
Balanced +8.37%* +2.23%* -4.73%*
Quality Oriented +7.87%* +0.91% -3.05%*
Insurer Parameters
Profit-Quality Orientation
151
Insurer Quality-
Profit Orientation
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Price
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Provider Effort
∆
%(𝑨𝑪𝑶 𝑩𝒂𝒔𝒆𝒍𝒊𝒏𝒆 )
Mortality Rates
Profit Oriented +7.59%* +2.00%* -4.08%*
Balanced +8.37%* +2.23%* -4.73%*
Quality Oriented +9.26%* +2.87%* -5.46%*
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
An agent-based model to study accountable care organizations
PDF
Designing health care provider payment systems to reduce potentially preventable medical needs and patient harm: a simulation study
PDF
A system framework for evidence based implementations in a health care organization
PDF
Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy
PDF
A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system
PDF
Using a human factors engineering perspective to design and evaluate communication and information technology tools to support depression care and physical activity behavior change among low-inco...
PDF
Modeling human bounded rationality in opportunistic security games
PDF
Investigation of health system performance: effects of integrated triple element method of high reliability, patient safety, and care coordination
PDF
Clinical prediction models to forecast depression in patients with diabetes and applications in depression screening policymaking
PDF
Total systems engineering evaluation of invasive pediatric medical therapies conducted in non-clinical environments
PDF
Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance
PDF
Evaluating city logistics using two-level location routing modeling and SCPM simulation
PDF
Network-based simulation of air pollution emissions associated with truck operations
PDF
Integration of behavioral health outcomes into electric health records to improve patient care
PDF
A risk analysis methodology to address human and organizational factors in offshore drilling safety: with an emphasis on negative pressure test
PDF
Assessment and analysis of direct care community worker training: addressing social determinants of health in the home care setting
PDF
Inter professional education and practice in the health care setting: an innovative model using human simulation learning
PDF
Integration of digital twin and generative models in model-based systems upgrade methodology
PDF
Asphaltene deposition during co₂ injection in enhanced oil recovery applications
PDF
Implicit provider bias in cardiovascular disease care of Black women: the lives of Black women with heart disease matter
Asset Metadata
Creator
Alibrahim, Abdullah Ibrahim
(author)
Core Title
Developing an agent-based simulation model to evaluate competition in private health care markets with an assessment of accountable care organizations
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Industrial and Systems Engineering
Publication Date
09/25/2017
Defense Date
08/18/2017
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
accountable care organizations,agent based simulation,health care competition,health systems,OAI-PMH Harvest
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Wu, Shinyi (
committee chair
), Romley, John (
committee member
), Von Winterfeldt, Detlof (
committee member
)
Creator Email
aalibrahim@outlook.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-433479
Unique identifier
UC11263177
Identifier
etd-AlibrahimA-5761.pdf (filename),usctheses-c40-433479 (legacy record id)
Legacy Identifier
etd-AlibrahimA-5761.pdf
Dmrecord
433479
Document Type
Dissertation
Rights
Alibrahim, Abdullah Ibrahim
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
accountable care organizations
agent based simulation
health care competition
health systems