Guaranteeing the Right to Health: The Role of Supply Chains and Access to Care 
 
 
A DISSERTATION 
SUBMITTED TO THE FACULTY OF THE 
UNIVERSITY OF MINNESOTA 
BY 
 
 
 
Eric Xu 
 
 
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS 
FOR THE DEGREE OF 
DOCTOR OF PHILOSOPHY 
 
 
Advisor: Kevin Linderman 
 
 
 
December 2022
© Eric Xu 2022
i 
 
Acknowledgements 
This dissertation was only possible because of all the help and support that I received from 
the wonderful people around me. Their support, positivity, and encouragement were essential 
throughout my PhD journey. 
First and foremost, I cannot thank my advisor, Kevin Linderman, enough for the constant 
support he provided me to pursue my research. Kevin is the wisest and most caring person that I 
closely interacted with during my PhD. Throughout my time as a student, I truly enjoyed our 
intellectual conversations, which made my PhD journey all the more worthwhile. He helped 
develop my research stream, skills, and approach throughout my time as a PhD student. I have 
learned how to navigate and succeed in the academic world under his guidance. In addition, he was 
my strongest advocate and staunchest supporter during a long and arduous job market process. 
In addition, I would like to thank each one of my PhD committee members. Anant Mishra, 
provided me with the guidance and insight I need to take the next step with my work with regards 
to analyses and writing. His dedication and effort pushed me to advance my research agenda and 
become an independent scholar. Russell Funk, encouraged me to pursue ideas that might be 
considered outside of the proverbial box. Through his guidance, I immersed myself in 
computational social science. During my time in his doctoral seminar, he helped cultivate my 
Python skills, which ultimately resulted in the computationally intensive approach applied to this 
dissertation. My external research committee member, Jean Abraham, from the School of Public 
Health, equipped me with the tools needed to pursue health policy research, provided strong 
questions related to my empirics, and provided me with guidance to navigate an academic career. 
Her constant support and encouragement gave me the confidence to pursue my work. Without her 
guidance and support, this dissertation would not have been possible.  
My committee chair, Susan Meyer Goldstein, has been a great source of my intellectual 
development throughout my PhD journey. I still remember the tour around campus she gave me 
when I visited campus as an applicant. Her feedback and strong questions always pushed me to 
think more deeply about the context of my work and the way I pursued my research. I am indebted 
to her for all the help she provided me with throughout my PhD. 
Thanks to all the tremendous Supply Chain and Operations faculty members helped and 
supported me through their seminars and feedback regarding my research: Kingshuk Sinha, Rachna 
Shah, Chris Nachtsheim, Karen Donohue, Karthik Natarajan, Natalie Huang, Necati Ertekin, and 
Hailong Cui. In addition, I also want to thank all the staff members at the Carlson School of 
Management for their day-to-day help: Deborah Brashear, Diane DeBoer, Melissa Grass, Sandy 
Herzan, Jill Johnson, and Vicki Lund. 
ii 
 
 In addition to these faculty members, I want to thank all the Supply Chain PhD students 
for their banter, positivity, and support throughout my time as a PhD student. I cannot imagine a 
more wonderful group of colleagues to have shared my PhD journey with these past six years. To 
the senior PhD students in the program, Sehwon Kang, Yi Tang, Vincent Yu, and Dwaipayan Roy 
helped me gain my footing during the first couple of years in the PhD program and continue to aid 
my development. In addition, the junior students Gautham Sunder, Alison Murphy, Hanu Tyagi, 
and Yeonjoo Lee provided me with warm support through our seminars and gatherings. Without 
my colleagues’ support and our fun times together, my PhD journey would have been much harder 
and bleaker. I hope that the SCO department continues to be a place where doctoral students lift 
each other up and live life to the fullest. 
 Finally, I give my greatest thanks to my family for their constant support supported and 
belief in me. My parents always provided endless support, respected my opinion, and cultivated an 
academic environment at home growing up. Finally, I also want to thank my amazing wife, Hyoju 
Jeong, who followed the ups and downs of the PhD journey with me from the very beginning. I am 
forever grateful for her selflessness and sacrifice throughout our time during PhD by helping me 
think through my ideas and providing me with feedback about my work. She always knew when 
to challenge me to push my work forward and when to encourage me through the toughest days of 
the program. I feel very lucky to have met her during my time in the PhD program, which made 
our PhD journey even more fruitful. With our Goldendoodle, Ming Ming, I find peace and joy at 
home every day, helping me to remain grounded and balanced through the stress of completing a 
PhD.
iii 
 
Abstract 
This dissertation investigates the effects of location of healthcare providers and patients 
within the healthcare supply chain on the delivery of healthcare. The dissertation consists of three 
essays that together examine the interplay between location, financing, public health 
interventions, and policymaking in the healthcare supply chain.  
Essay 1 investigates the impact of patients’ surrounding home environments on their 
health outcomes. Specifically, this chapter examines how access to specific forms of 
infrastructure impacts long term health outcomes. Patients may exhibit signs of a promising 
recovery while residing in inpatient care; however, when these patients return to their 
neighborhoods, the surrounding environment might trigger a pattern of behavior that may lead to 
higher chance of another inpatient stay. The analysis shows that accessibility to grocery stores 
within a half mile radius reduces the number of annual inpatient stays for heart failure patients.  
Essay 2 investigates on the impact of policy changes on healthcare supply chain 
utilization for insurance coverage expansion under the Patient Protection and Affordable Care Act 
(ACA). The empirical results show that patients make their decisions to access healthcare based 
on the distance to the nearest care delivery facility, whether it be a primary care clinic or an 
emergency department, and the hours of operation of the nearest primary care clinic reduces 
emergency department use. Our results provide a possible alternative explanation to the adage 
that insurance provision alone increases emergency department utilization.  
Essay 3 investigates the structural factors that impact the uptake of telehealth services 
under the expansion of broadband to primary care providers under the Rural Healthcare Program. 
Specifically, it focuses on physical access and broadband access to primary healthcare services. 
The empirical analysis shows that broadband coverage directly impacts synchronous telehealth 
visits in states with payment parity and service parity once the quality of providers broadband is 
improved through expanded funding under the Rural Healthcare Program (RHP). Notably, the 
effect of distance to the nearest provider is not impacted by the RHP expansion. With regards to 
asynchronous telehealth uptake, the analysis shows that the sole predictor of uptake is consumer 
broadband coverage regardless of a state’s payment parity or service parity laws related to the 
privately insured population.  
These papers collectively will contribute to the healthcare operations literature and 
policymakers by addressing ways to account for geographical location and the structural 
characteristics of the healthcare supply chain when delivering care to patients.  
iv 
 
Table of Contents 
Acknowledgements i 
Abstract iii 
List of Figures vii 
List of Tables viii 
Chapter 1: Introduction 1 
1.1 Motivation and Dissertation Contribution: 1 
1.2 Essays and Research Designs: 3 
1.2.1 Essay 1: Built Environment 3 
1.2.2 Essay 2: Physical Access and Emergency Department Usage 4 
1.2.3 Essay 3: E-Access versus Physical Access 5 
1.3 Dissertation Synopsis 6 
Chapter 2: Literature Review 8 
2.1 External Environment and Process Conformance 8 
2.2 Physical Access and the Healthcare Supply Chain 10 
2.3 E-Access Versus Physical Access 12 
Chapter 3: Built Environment and Health Outcomes 16 
3.1 Introduction 16 
3.2.2 Conformance Quality and Patient Inputs 19 
3.2.3 Population Health and the Built Environment 20 
3.2.4 Addressing the Impact of Built Environment 21 
3.3 Hypotheses 22 
3.3.2 Patient Context Versus Healthcare Delivery Context 24 
3.4 Data 25 
3.5 Empirical Strategy 28 
3.5.1 Difference-in-Differences Analysis 29 
3.6 Results 30 
3.6.1 Built Environment 31 
3.7.1 CMS Quality Conformance Guidelines 32 
3.8 Discussion 33 
3.9 Policy and Managerial Implications 34 
Chapter 4: How Does Physical Access Affect Emergency Department Use? Evidence From 
Insurance Coverage Expansion 36 
4.1 Introduction 36 
4.2 Study Context and Background 39 
v 
 
4.2.1 The Affordable Care Act 39 
4.2 Changes to Healthcare Demand and Supply following the ACA 41 
4.3 Hypotheses 42 
4.3.1 Where: Physical Access and Road Distance 42 
4.3.2 When: Physical Access and Hours of Operation 44 
4.3.3 How Much: Use of Primary Care 45 
4.4 Data and Methods 46 
4.4.1 Dependent and Independent Variables 48 
4.4.2 Control Variables 50 
4.4.3 Identification Strategy and Model Specification 52 
4.5 Results 54 
4.5.1 Effects of Physical Access on ED Discharges 54 
4.5.2 Effects of Nearest Primary Care Clinic Capacity on ED Discharges 57 
4.6 Robustness Checks and Alternative Explanations 60 
4.6.1 Examination of the Parallel Trends Assumption 61 
4.6.2 Controlling for Neighborhood Environment Effects 63 
4.6.3 Does the Provision of Health Insurance Create a Potential for Moral Hazard? 64 
4.6.4 Does Physical Access Affect Insurance Uptake? 64 
4.6.5 Does Transportation Modality or Household Vehicle Ownership Affect ED 
Discharges? 65 
4.7 Discussion 66 
4.7.1 Policy and Financial Implications 67 
4.7.2 Limitations and Future Research 69 
Chapter 5: E-Access versus Physical Access: An Examination of Telehealth 70 
5.1 Introduction 70 
5.2 Literature Review and Background 73 
5.2.2 The Evolution of Telemedicine 74 
5.2.3 Parity: Telehealth Reimbursement and Service 77 
5.2.4 Rural Health Care Program 78 
5.3.1 Physical Access and Synchronous Telehealth 80 
5.5.1 Difference-in-Differences Analysis 86 
5.6 Results 87 
5.6.1 All Telehealth Visits 87 
5.6.2 Synchronous Telehealth 90 
5.6.3 Asynchronous Telehealth 92 
vi 
 
5.7 Robustness Checks, Alternative Explanations, and Post Hoc Analysis 93 
5.7.1 Examination of the Parallel Trends Assumption 93 
5.8.2 Limitations and Future Research 95 
Chapter 6: Contribution 98 
Essay 1 Theoretical and Policy Implications 98 
Essay 2 Theoretical and Policy Implications 98 
Essay 3 Theoretical and Policy Implications 99 
Bibliography 101 
Appendix 112 
Appendix A1. Patient Matching Variables: Variable Descriptions and Data Sources 112 
Appendix B1. Constructing a Longitudinal Panel of Primary Care Clinics for Kentucky 
and North Carolina 113 
Appendix B2. Variable Descriptions and Data Sources 115 
Appendix B3. ZCTA Matching Variables: Variable Descriptions and Data Sources 116 
Appendix B4. Summary Statistics for Variables in DiD Specification 118 
Appendix B5. Summary Statistics for Variables in Fixed Effects Specification (California 
Subsample) 119 
Appendix B6. Effects of Physical Access on ED Discharges: Neighborhood Environment 
Controls Added 120 
Appendix B7. Effects of Nearest Primary Care Clinic Capacity on ED Discharges:                          
Neighborhood Environment Controls Added 121 
Appendix B8. Addressing Alternative Explanations Relating to Unintentional Injury 
Mortality and Number of Uninsured Individuals 122 
Appendix B9. Controlling for Transportation Modality and Vehicle Ownership 123 
Appendix B10. Using Alternative Measures of Distance Difference 124 
Figure B1: Higher Geographic Resolution of Census Blocks (compared to ZCTAs): 
Example of California 124 
Table B1: Number of Census Blocks Across States and Across Years 124 
Table B2: Using Alternative Measures of Distance Difference 125 
Appendix C1. Parity Laws by State 127 
Appendix C2. CPT Codes 131 
Appendix C3. Data Description 132 
Appendix C4. ZCTA3 Variable Summary Statistics 134 
Appendix C5. Leads and Lags Analysis 135 
 
  
vii 
 
List of Figures 
Figure 1. Proposed Access to Care Framework 1 
Figure 2: Visualization of Identification Strategy 28 
Figure 3: Timeline of Events Related to the Passage/Enforcement of the Affordable Care 
Act 41 
Figure 4: Distances to the Nearest Primary Care Clinic and Nearest ED from a ZCTA 
Centroid 49 
Figure 5: Trends in Medicaid Enrollees Across the Four States 54 
Figure 6: Hours of Operation Kernel Density Plots 57 
Figure 7: Primary Care Clinic Unique Encounters by Payer in California 58 
Figure 8: Unadjusted Trends in ED Discharges Across Treatment and Control Groups 61 
Figure 9: Total Healthcare Utilization Amongst Privately Insured Sample Population 79 
Figure 10: Road Service Area Versus Euclidean Distance 82 
Figure 11: Payment Parity Map 87 
Figure 12: Synchronous Telehealth Visits and the Rural Healthcare Program 87 
 
  
viii 
 
List of Tables 
Table 1: ZCTA Difference-in-Differences Variable Summary Statistics 30 
Table 2: Built Environment and Inpatient Stays 30 
Table 3: Built Environment and Inpatient Stays with CMS Metrics 32 
Table 4. Effects of Physical Access on ED Discharges 55 
Table 5: Marginal Effects Based on Table 4, Column 4 Results 56 
Table 6: Effects of Nearest Primary Care Clinic Capacity on ED Discharges 59 
Table 7: Marginal Effects Based On Table 7, Column 4 Results 60 
Table 8: Analysis of Leads and Lags to Adoption 62 
Table 9: Telemedicine Generations 74 
Table 10: ZCTA3 Model Specifications All Telehealth Visits 88 
Table 11: ZCTA3 Marginal Effects for All Telehealth Visits 89 
Table 12: ZCTA3 Model Specifications Synchronous Telehealth 90 
Table 13: ZCTA3 Marginal Effects for Synchronous Telehealth 90 
Table 14: ZCTA3 Model Specifications Asynchronous Telehealth 92 
Table 15: ZCTA3 Marginal Effects for Asynchronous Telehealth 93 
1 
 
Chapter 1: Introduction 
1.1 Motivation and Dissertation Contribution: 
 As the delivery of healthcare evolves, there is an increasing emphasis on the physical 
accessibility of care and the location of care delivery. Traditionally, firms made location decisions 
based on a narrow set of economic variables to serve specific customers subject to those service 
constraints. However, these location decisions can cause negative externalities on location 
populations related to their health outcomes. For example, hospitals in areas serving low-income 
patients would rather allow patient mortality to increase as opposed to addressing increased 
readmissions caused by the local environment (Wadhera 2018). To address this problem, 
researchers have suggested viewing healthcare delivery with an increased level of consideration for 
integratating location decisions amongst stakeholders, both providers and patients, in an effort to 
provide high quality, timely care, whether this integration comes in the form of an accountable care 
organization or patient centered medical homes (Dai and Tayur 2019, Shin 2019). In the healthcare 
operations management literature, the evolution of delivering healthcare inside a single care 
providing organization towards a healthcare supply chain perspective requires addressing the 
impact of multiple stakeholders simultaneously ranging from patients to providers to public health 
departments.  
Figure 1. Proposed Access to Care Framework 
 
Traditionally, healthcare operations management research has focused on how individual 
healthcare providers can deliver care at a single facility based on an input-process-output model 
(Dai and Tayur 2019). While this stream of research remains important, it is equally important to 
consider the way location impacts the delivery of care, whether it be physical access or patient 
characteristics. Past works often consider the location of providers and patients to be equivalent; 
2 
 
however, location plays a large role in health outcomes ranging from timely access to care to 
disease exposure (Coburn 2004). By understanding location effects, it is possible to improve access 
to care and quality of care by accounting for the heterogeneity of patients input into hospital 
processes. Therefore, this dissertation removes assumptions about geographic neutrality pertaining 
to care delivery and explores the broader context of physical access and location, both for patients 
and providers, to understand how these factors impact the uptake of care and care outcomes.  
This dissertation expands upon the tradition input-process-output model by addressing the 
impact of location related to both buyers (patients) and suppliers (healthcare providers) and how 
this ultimately impacts physical access. Figure 1 provides a framework for this dissertation that 
builds upon the ideas of an input-process-output model with increased consideration for the effects 
of location. From a population location standpoint, we examine the impact of insured populations 
and the effects of the surrounding built environment on patients’ utility. Instead of assuming 
geographic neutrality and that inputs are homogenous, we consider the impacts of heterogenous 
location effects for a population where process conformance guidelines may be unable to account 
for the variability of patient inputs (Coburn 2004). From a patient utility standpoint, we examine 
how location characteristics, provider location and provider hours of operation, impact when and 
where patients elect to receive care. Finally, we explore the impact of technology mediated access 
versus physical access on patients’ decisions regarding where they decide to seek care. We seek to 
understand the boundary conditions of when patients would elect to see a provider in person and 
when they would elect to see a provider virtually.   
Ultimately, this dissertation seeks to address the effects of location of entities within the 
healthcare supply chain and the impact of those factors on utility and quality of care outcomes. 
These essays together examine the interplay between location, financing, public health 
interventions, and policymaking. In the past, the healthcare operations management literature has 
focused heavily on the input-process-output model, particularly as it pertains to emergency 
department utilization. Instead, we delve deeper into the geographical context of this model in order 
to examine the impact of location input, process, and output. The three perspectives on location are 
presented as follows: a built environment perspective related to population location effects and 
process conformance, a physical accessibility perspective related to patient utility when finding a 
provider, and a technology mediated access perspective related to using physical or digital channels 
when seeking care. Following a survey of the healthcare operations literature, this dissertation 
expands the traditional input-process-output model to account for variations brought upon by 
location effects. 
3 
 
1.2 Essays and Research Designs: 
1.2.1 Essay 1: Built Environment 
The first essay entitled “The Impact of Place on Health: Built Environment and Healthcare 
Process Outcomes” focuses on the impact of the built environment on patient outcomes, 
specifically inpatient stays. In the healthcare operations management literature, researchers have 
acknowledged that the areas patients are drawn from can directly impact treatment outcomes. 
However, the healthcare operations management literature has focused on heavily on the process 
conformance within the delivery organization, e.g., a hospital setting, while placing less emphasis 
on the surroundings in which patients reside. We build upon previous research examining whether 
the surrounding environment can influence a firm’s processes (Muthulingam et al 2020). Patients 
may exhibit signs of a promising recovery while residing in inpatient care; however, when these 
patients return to their neighborhoods, the surrounding environment might trigger a pattern of 
behavior that may lead to higher chance of another inpatient stay. Therefore, the primary research 
question is how does neighborhood-level infrastructure, specifically those related to the social 
determinants of health, impact the frequency of inpatient stays? In addition, the secondary research 
question asks whether built environment characteristics outweigh the impact of hospital 
conformance? 
We test our predictions in the context of a natural disaster in the context of New York City 
following Hurricane Sandy. To estimate the causal impact of Hurricane Sandy, we find a credible 
comparison group within New York City. In our study, our treatment group consists of individuals 
living within predesignated flood zones that were damaged by the hurricane, while our control 
group consists of predesignated flood zones that were not damaged by the hurricane. By using 
ArcGIS, we construct a half mile radius across the road network surrounding each Census tract. 
Within each radius, we measure the change in five specific built environment characteristics: 
grocery stores, green space, recreational facilities, pharmacies, and primary care clinics. To ensure 
that geographies are similar, we only compare zones that were designated flood zones leading up 
to Hurricane Sandy. We measure the pre and post change of these characteristics in flood zones.  
We examine the healthcare ecosystem to understand how the environments that patients 
are drawn from paired with the care provided by the hospital may provide a more accurate 
representation of quality of care. This research seeks to account for the impact of macro-level 
factors on the success of micro-level interventions within patients. We provide a possible 
alternative method for accounting for patients’ surroundings to the current ICD-Z condition coding 
and other clinical screen methodologies using a series of publicly available data sets. Furthermore, 
4 
 
we provide a method for policymakers to better account for factors related to the surrounding 
environment that could impact hospital patient selection and ability to deliver optimal healthcare 
outcomes. 
1.2.2 Essay 2: Physical Access and Emergency Department Usage 
The second essay entitled “How Does Physical Access Affect Emergency Department Use? 
Evidence From Insurance Coverage Expansion” focuses on the impact of policy changes on 
healthcare supply chain utilization in insurance coverage expansion under the Patient Protection 
and Affordable Care Act (ACA). Previous healthcare operations management literature has focused 
heavily on reducing congestion at a single healthcare delivery organization, e.g., an emergency 
department, in relation to the ACA; however, our study expands the scope to the broader healthcare 
supply chain to understand how individuals optimize their utility based on the physical location 
and hours of operation of existing healthcare delivery organizations. Structural and infrastructural 
decisions related to the healthcare ecosystem will shift demand to alternate points of entry. 
Furthermore, we examine the impact of insurance delivery methods and whether it creates artificial 
physical access constraints. For example, Medicaid is often delivered in two formats: Fee for 
Service and managed care. These modalities place varying restrictions on the point of care from 
which individuals can receive care. 
We test our predictions in the context of the ACA’s enforcement across five states; Arizona, 
California, Florida, Kentucky, and North Carolina, using the United States Census Bureau data on 
ZIP Code Tabulation Areas (ZCTAs) from 2012 to 2016, consisting of 8,484 ZCTA-year 
observations. In addition, we use ArcGIS, a leading provider of geographic information data and a 
tool used in geospatial research, to dynamically model realistic road network conditions within 
ZCTAs and calculate the shortest road distance patients must travel to access the nearest ED or the 
nearest primary care clinic. Subsequently, considering the differential enforcement of the ACA 
across the five states as a natural experiment, wherein Arizona, California, and Kentucky 
experienced the full enforcement of ACA policy statutes—Medicaid expansion and Individual 
Mandate—in 2014 while Florida and North Carolina experienced partial enforcement through the 
Individual Mandate policy statute only, we examine the impact of physical access on the nearest 
ED use in the form of discharges. To develop a deeper understanding of the mechanism underlying 
the structural aspects of the healthcare supply chain, we examine if and to what extent do utilization 
levels at the nearest primary care clinic influence a patient’s use of the nearest ED. Our analysis 
differentiates between Medicaid patient encounters based on plans restricting access to a predefined 
network of providers (i.e., Medicaid managed care) and those using an unrestricted fee-for-service 
model (i.e., Medicaid fee-for-service). This differentiation of delivery systems allows us to examine 
5 
 
whether a constrained set of provider choices impacts patient utilization behavior when making 
nonurgent outpatient care decisions. 
We find that patients make their decisions to access healthcare based on the distance to the 
nearest care delivery facility, whether it be a primary care clinic or an emergency department, and 
the hours of operation of the nearest primary care clinic reduces emergency department use. Our 
results provide a possible alternative explanation to the adage that insurance provision alone 
increases emergency department utilization. Using a more nuanced data set of a Medicaid 
expansion state, we find evidence that capacity at the nearest primary care clinic plays a role in 
emergency department usage. Specifically, we note that as primary care providers at the nearest 
primary care clinic take on more providers the number of emergency department discharges from 
a ZCTA decreases. Moreover, patients utilizing the right service at the right time at the right place 
will contribute to lower costs for the federal government in relation to reimbursements. 
1.2.3 Essay 3: E-Access versus Physical Access 
In the third essay entitled “E-Access versus Physical Access: An Analysis of the Rural 
Health Program”, we examine whether telemedicine can address physical access problems related 
to primary care treatment. While our current conception of telemedicine has existed for half a 
century, beginning with the Space Technology Applied to Rural Papago Advanced Health Care 
(STARPAHC), recent telecommunication advancements have increased the adoption of telehealth 
services amongst patients and practitioners. However, fewer studies have explored the impact of 
broadband provision alongside physical access to primary care providers on telehealth uptake, 
either asynchronous or synchronous. Therefore, we examine the impact of telemedicine adoption 
on synchronous telehealth and asynchronous telehealth usage, specifically examining the 
conditions by which telehealth consultations become a substitute for in-person care or a 
complement to in-person care. Using a unique dataset of insurance claims for primary care visits, 
we examine in-person, asynchronous, and synchronous primary care visits. We examine telehealth 
utilization in the context of the Rural Healthcare Program to provide identification for our model. 
We test our predictions in a national sample on a sample consisting of 3-digit ZIP Code 
Tabulation Areas. We use ArcGIS to calculate the distance to nearby primary care clinics from 
Census tracts within each ZCTA. We then use the Federal Communications Commission (FCC) 
Fixed Broadband Deployment Data from Form 477 to account for the broadband availability within 
a market based on speed. Our control group consists of states without payment parity for telehealth 
services and our treatment group consists of states with payment parity for telehealth services. 
6 
 
Our results demonstrate the key impact the broadband access amongst patients has on their 
uptake of telehealth services. In addition, we find that in states with both payment and service parity 
that synchronous telehealth uptake rises as the associated broadband infrastructure improves for 
both patients and providers. Notably, in parity states, we see an additional 1,185 synchronous 
telehealth visits following the expansion of the Rural Healthcare Program. When physicians are 
reimbursed similarly to in-person visits and are required to provide similar treatment via telehealth 
or in-person, broadband quality becomes a key factor to that the real time nature of synchronous 
telehealth delivery experience is on par with in-person care delivery. With regards, to asynchronous 
telehealth services, we find that consumer broadband access is the primary factor determining 
uptake across our treatment and control groups. This may suggest that the lack of a face-to-face 
interaction reduces the need for both patients and providers to have adequate broadband as 
suggested by the FCC. Instead, patients’ experience is likely dictated by the upload speeds related 
to self-assessments/pictures/documents and download speeds for their respective physician’s 
diagnosis. Furthermore, as physicians do not need to make tradeoffs with in-person visits, it appears 
that parity laws do not dictate physicians’ propensity to provide asynchronous telehealth services.  
1.3 Dissertation Synopsis 
 Each of these essays extends the healthcare operations management literature on physical 
access and location by exploring a different aspect of location. Together, these studies advance our 
understanding of location effects on access to care and process outcomes. In addition, each essay 
leverages Geographic Information Systems (GIS) extensively to illustrate how location data can be 
used in the healthcare operations management literature to better understand physical access. We 
seek to move beyond traditional location dichotomies of rural versus urban by addressing the 
characteristics of different geographies and how location effects, both spatial and temporal, can 
impact the way patients uptake care. Furthermore, we introduce the possibility of new challenges 
and opportunities related to the digital access of care in relation to the physical access of care.  
Overall, these essays contribute to healthcare operations management literature, operations strategy 
literature, and to policymakers by removing assumptions about geographic neutrality in order to 
understand how location effects impact the input-process-output model. Furthermore, we provide 
practical implications for stakeholders across the healthcare supply chain responsible for delivering 
care. 
The remainder of this dissertation is structured as follows: Chapter 2 provides a broad 
overarching literature review of the three essays. Chapter 3 [Essay I], Chapter 4 [Essay 2], and 
Chapter 5 [Essay 3] represent the constituent essays of this study. Chapter 6 discusses the key 
7 
 
theoretical and policy implications for the broader view of location on healthcare delivery using a 
supply chain perspective. 
8 
 
Chapter 2: Literature Review 
 In this chapter, we provide an overview of the literature presented in each essay. For each 
essay, we provide a more in-depth nuanced literature review in their respective chapters. The intent 
is for our work to contribute to both the operations management literature and the health economics 
literature, as our work draws heavily from both literature streams, pertaining to the importance of 
location effects on access to care and care outcomes. Our approach examines the antecedents to 
health, whether it be the facility from which an individual receives care or the social determinants 
of health related to individuals on a day-to-day surroundings, to understand the impact of location 
and the physical accessibility to different resources related to health outcomes. Each essay is meant 
to contribute to the existing input-process-output framework by accounting for geographic 
heterogeneity related to care delivery across different locations. 
 In the first essay, we examine the inputs related to care delivery, specifically the location 
of where individuals reside. This examines how inputs to providers may vary based on the location 
from which they are drawn from. In past healthcare operations management literature, the impact 
of location has been noted, albeit not thoroughly discussed. The second essay breaks from 
the influential framework set forth by Asplin et al of input-process-output, towards a broader view 
of service provider locations (Asplin et al 2003). Instead, we examine components prior to the input 
phase of healthcare utilization related to nonurgent use, specifically the dichotomy of primary care 
providers versus emergency departments in the provision of outpatient use. Finally, the third essay 
examines the way patients chose to input themselves into the healthcare supply chain. Traditionally, 
the assumption is that individuals would physically uptake care by electing to input themselves at 
the nearest primary care provider to receive care. However, the impact of telemedicine on this input 
decision is less explored in the healthcare operations management literature, specifically the 
boundary conditions of physical access versus technologically mediated access.  
2.1 External Environment and Process Conformance 
In recent decades, the quality of care delivery has improved dramatically as scholars have 
focused on improving process conformance in the delivery of care (Coburn 2004, Ahn et al 2006a, 
Artiga and Hinton 2018). This biomedical model of disease reduced many ailments to molecular-
level pathogens that could be treated with strict adherence to a set of process conformance 
guidelines (Coburn 2004, Artiga and Hinton 2018). However, the improvement achieved via 
process conformance has slowed leading researchers to question whether this reductionist approach 
should be reconsidered. Past operations management research noted the impact of uncertainty 
related to the inputs of a process and how this might impact outcomes (Argote 1982). More recently, 
9 
 
authors have noted that discrepancies in value-based purchasing programs meant to improve 
process conformance within hospitals tend to exhibit outcome variation based on their location 
(Zhang et al 2016). While process conformance within the hospital remains important, the 
environments surrounding patients deserves equal consideration in the assessment of health 
outcomes amongst patients, especially given that patients’ locations can greatly impact their health 
outcomes. Notably, the Haussman Model of the early 20th century outlined a method of dividing 
specific economic functions within a geography and isolating functions determined to be 
deleterious towards health and restricting the physical contact individuals would have with those 
functions (Coburn 2004). In this case, the idea was to remove undesirable externalities impacting 
health to improve outcomes. By accounting for built environment, it is possible to incorporate 
contextual factors that may impact performance in terms of process conformance and account for 
the relation between operational performance to external environmental factors (Chandrasekaran, 
Linderman, Schroeder 2015). In the most recent series of health care reforms under the ACA, 
policymakers attempted to begin addressing the impact of neighborhood environments on health 
outcomes, most notably the National Prevention Strategy and ICD-Z condition coding. 
Even during the most recent series of healthcare reforms in the United States, policymakers 
and stakeholders acknowledge the impact of the social determinants of health on individual and 
population health. Recent healthcare reforms under the Affordable Care Act saw new policy 
guidelines meant to directly address the environments surrounding individuals ranging from the 
National Prevention Strategy to Accountable Care Organizations. With regards to Accountable 
Care Organizations, the use of built environment interventions was used to issues “park 
prescriptions” to encourage individuals to walk in local parks for 30 minutes per day (Zusman 
2014). Cities were also encouraged to consider zoning initiatives that increased accessibility to 
green space (Zusman 2014). On a larger scale, the Department of Health and Human Services 
created a National Prevention Strategy outlining ways for communities to improve population 
health. 
The National Prevention Strategy (NPS), the companion policy to the Patient Protection 
and Affordable Care Act (ACA), was meant to complement the insurance coverage expansion and 
quality of care improvement initiatives by addressing where patients live and work to improve 
preventative health measures resulting in better health outcomes. A lack of built environment 
infrastructure surrounding patients may encourage poor health behaviors and exacerbate prevalent 
health problems such as heart disease and depression (National Institutes of Health 2014). The 
National Prevention Strategy was meant to provide a comprehensive plan for the nation’s 
preventative health by identifying environmental factors that could exacerbate disease. The 
10 
 
measures outlined by the NPS were expected to improve health outcomes by addressing factors 
that were often not part of physicians’ purview of treatment. One specific area to note is the built 
environment characteristics as outlined by the social determinants of health. The Department of 
Health and Human Services notes that the built environment plays a significant role in the health 
and well-being of individuals living within a given geography.  
However, quantifying the effects of built environment is an equally important task which 
is being address through multiple initiatives from various stakeholders. In 2011, the Centers for 
Disease Control and Prevention/Agency for Toxic Substances and Disease Registry Social 
Vulnerability Index was introduced building upon previous work conducted by Geospatial 
Research, Analysis, and Services Program analyzing socially vulnerable populations and recovery 
efforts. This Social Vulnerability Index was meant to classify the impact of external stresses on 
human health, which include natural or human-caused disasters, or disease outbreaks. The CDC 
suggested that by reducing the social vulnerability of individuals that it would quantifiably reduce 
both human suffering and economic loss More recently, a new series of International Classification 
of Diseases (ICD) condition codes known as ICD-Z codes were introduced to help physicians 
account for their patients’ environments. While these efforts represent actions meant to better 
account for social determinants of health, we provide empirical evidence of their effects, 
specifically those related to surrounding structural characteristics, on health outcomes. With 
regards to medical education, the American College of Physicians recently took steps to reform 
medical education in an effort to help health care professionals identify and address social 
determinants of health that may negatively impact patient outcomes (Daniel et al 2018). Physicians 
are crucial for documenting the health impact of built environment on the population they serve. 
Unfortunately, the adoption of these codes by practitioners has been limited for a multitude of 
reasons ranging from a lack of familiarity to lack of clarity related to the type of provider who 
should document said information. In this study, we seek to examine whether publicly available 
Census data and zoning data could help providers account for the effects of built environment 
characteristics when treating patients. 
2.2 Physical Access and the Healthcare Supply Chain 
In recent years, prior studies in the healthcare management literature have increasingly 
focused on ED use and the performance of EDs (e.g., Song et al. 2015, Batt and Terwiesch 2017, 
Freeman et al. 2020). While efforts have been concentrated on a single point of care, i.e., EDs, 
researchers are increasingly recognizing the effects of healthcare policy shocks on the manner in 
which individuals access the broader healthcare supply chain (Dai and Tayur 2019). This essay 
analyzes the impact of provider location and hours of operation on the provision of nonurgent care 
11 
 
when a large-scale demand changing policy is enforced, specifically the Patient Protection and 
Affordable Care Act (ACA). One of the central provisions of this policy was the expansion of 
insurance through vehicles such as Medicaid expansion, which expanded eligibility for who could 
receive Medicaid while increasing reimbursement to physicians treating Medicaid patients.  
However, policymakers also realized that making health insurance available was not 
enough to ensure its uptake. Thus, the ACA-based Medicaid expansion was accompanied by the 
Individual Mandate serving a crucial role in compelling individuals to acquire health insurance.1 
At its core, the Individual Mandate served to regulate health insurance markets by influencing 
individual economic behavior—i.e., individuals either purchase health insurance or pay a tax 
penalty (Rosenbaum and Gruber 2010). Further, the tax penalty was set to increase each year to 
further persuade uninsured individuals to purchase insurance (Eibner and Saltzman 2015). 
However, in 2011, the Individual Mandate, in conjunction with Medicaid expansion, created 
significant controversy, resulting in litigation that reached the United States Supreme Court in a 
landmark case known as the National Federation of Independent Businesses v. Sebelius (Negrin et 
al. 2012). In 2012, the Supreme Court ruled that the Individual Mandate was legal for all 50 states 
under Congress’s taxing clause power (Article I, Section 8, Clause 1); however, the justices argued 
that the significant expansion of Medicaid was not within Congress’s spending power and was 
unconstitutionally coercive (Negrin et al. 2012). This ruling triggered a differential enforcement of 
the ACA, resulting in the current policy landscape, where some states have adopted Medicaid 
expansion (e.g., Arizona, California, Kentucky, Virginia) while other states have opted out of 
Medicaid expansion (e.g., Florida, Georgia, North Carolina, Texas). 
Primary care serves as the foundation of the American healthcare system with EDs playing 
a supporting role in outpatient care delivery (Chokshi et al. 2014). From 2010 to 2020, the ACA 
was expected to increase the use of primary care by an additional 15 million to 24 million 
individuals (Hofer 2011). This projected increase in primary care use was largely attributed to 
Medicaid expansion for two reasons: (i) uninsured individuals would gain financial access to care, 
and (ii) reimbursement for primary care physicians treating Medicaid patients would increase. Note 
that, in the past, low reimbursement rates might have caused primary care physicians to turn away 
Medicaid patients seeking treatment; therefore, Medicaid expansion alone was not expected to 
increase primary care use, given that physicians might continue to turn away Medicaid recipients 
 
1 The Individual Mandate generally requires that individuals purchase insurance if costs do not exceed 8% 
of household income, thereby ensuring financial access to healthcare. Select religious groups, American 
Indians, citizens not filing a tax return, incarcerated individuals, and those who are uninsured for a period of 
less than three months per calendar year can claim exemption from the Individual Mandate. 
12 
 
(Decker 2012). To address this issue, the ACA increased physician reimbursement rates for 
Medicaid patients, thus achieving payment parity with Medicare and private insurance (Klink 2015, 
Ladhania et al. 2019). This payment parity was expected to improve financial accessibility to 
primary care clinics for new Medicaid enrollees and create greater demand for outpatient care at 
these clinics.  
In addition, to address the expected increase in demand, the ACA implemented a series of 
infrastructural reforms meant to bolster the total supply of primary care physicians over time. These 
provisions, however, did not directly address the physical accessibility of primary care physicians, 
either in terms of physical distance or temporal availability. The absence of ACA’s focus on 
remedying the inherent heterogeneity of physical access to outpatient care across various 
geographic regions was notable, given that in the decades preceding the ACA, a free market 
approach to managing healthcare in the US rendered primary care clinics in rural areas financially 
unsustainable (due to the lower demand) and resulted in closures and consolidations of primary 
care clinics in rural areas. Unfortunately, this consolidation accelerated during the 2010s, creating 
an increasingly inequitable geographic distribution of primary care clinics (Millman 1993, Fulton 
2017, Kane 2017). Physician ownership of clinics also continued to fall steadily; in 1983, 75.8% 
of physicians owned their clinics, while in 2018, 45.9% of physicians owned their clinics (Kane 
2017). While the ACA provided $11 billion in funding for Federally Qualified Health Centers 
(FQHCs), these centers represent only a small portion of all primary care clinics; in addition, these 
FQHCs continued to report funding issues and problems with workforce recruitment/retention, 
hampering their effectiveness in providing care (Rosenbaum et al. 2017). As such, there continues 
to be a shortage of primary care clinics across the United States and the question of whether such 
care is accessible at the right place and at the right time remains understudied.  
2.3 E-Access Versus Physical Access 
In order to understand the proliferation of telemedicine, we examine the origins of live 
audio-visual based telemedicine going back to the Space Race beginning in 1955. With the advent 
of the human orbital spaceflight, it immediately became clear that it was infeasible to send all the 
necessary types of physicians into space with the astronauts; therefore, a new solution was required 
to supply medical care that did not require the immediate presence of physicians, in this case 
telepresence. Thus, a method to deliver healthcare under structurally dictated lean conditions was 
required. In 1971, NASA and the Papago Native American Tribe worked together to establish a 
project known as Space Technology Applied to Rural Papago Advanced Health Care, otherwise 
13 
 
known as STARPAHC, pioneering telemedicine and the remote delivery of healthcare (Henceroth 
1978).  
The concept of telemedicine has evolved over the decades as the care delivery technology 
has improved. As the technology has developed for multiple decades, telemedicine has come closer 
to providing technology mediated access to healthcare. In the past, limited technology may have 
made telemedicine a complementary good to in person treatment. However, with the increasing 
availability of broadband and other telepresence technologies, telemedicine in some cases can 
provide a substitutable service that can either be delivered entirely online or in-person at a distance 
mediated by a physician.  
Modern advancements in data transfer and communications made the delivery of 
telemedicine more viable from a cost and efficacy standpoint. Today, telemedicine generally refers 
to a concept surrounding interactive televideo with image and medical record transfer paired with 
remote monitoring (Strode, Gustke, and Allen 1999). Recent advances in secure data transmission, 
improved broadband speed, and user-friendly interfaces has increased the propensity of individuals 
to uptake telehealth (Hall and McGraw 2014, Agnisarman et al 2017, O’Shea et al 2022). However, 
while technological advances may make telemedicine more accessible to patients, laws governing 
reimbursement and service provision may still limit the adoption of telehealth amongst providers. 
In making the choice for telemedicine versus in-person care delivery, the structural 
characteristics of the broadband network and in-person care delivery play a large role. Rural areas 
often exhibit lower broadband penetration, which may deter the use of telehealth services given the 
current guidelines surrounding internet speed and telehealth (Drake et al 2019). Additionally, 
assuming broadband meets the requisite standards for both patient and providers, physical distance 
can play a role in the way patients uptake telehealth services. Reed et al note that individuals may 
be more willing to opt for telehealth when it requires 30 minutes or more to reach a primary care 
provider (Reed et al 2020). This essay seeks to expand on these lines of literature by investigating 
the interplay between physical access and technology mediated access in the delivery of primary 
care.  
In building upon previous telemedicine models, researchers have also examined how 
various disease states might impact outcomes. In the case of a recent Cochrane Review, Acute 
illnesses and chronic conditions do not have significant impact on adoption rates. Based on these 
findings, the type of illness is less important, e.g. chronic or acute; instead, the equipment available 
has more bearing on care delivery (Flogren et al 2015). Initially, there was some consideration that 
chronic illnesses might be easier to manage due to the slower pace of change in the disease state. 
For example, acute diseases were expected to pose a greater challenge in scheduling physicians due 
14 
 
to the less predictable nature of patient demand. On the other hand, chronic illnesses may be easier 
to monitor as rapid changes in condition are unlikely, which makes treatment planning more 
predictable. However, after examining the work of Flogren et al, it appears that acute and chronic 
conditions can be addressed through telemedicine initiatives (Flogren et al 2015). Given that 
telemedicine can handle a wide variety of conditions, it is important to understand how the delivery 
modality can impact uptake and outcomes.  
While synchronous and asynchronous modalities provide reliable options for delivering 
care, there are concerns related to care delivery. Specifically, the ease of use and validity of 
telemedicine interventions represents another cornerstone that is currently facing greater scrutiny. 
In certain cases of asynchronous interventions, mobile applications may generate automated 
recommendations for patients until they can see a physician, which could lead to problematic results. 
As the possible liability increases with telemedicine interventions, hospital systems will have a 
more difficult time in garnering support amongst care providers to use telemedicine-based 
interventions. In the case of mobile apps, the increasingly popular delivery via mobile devices 
serves as a method for doctors to keep up with their patients. However, in certain cases these apps 
may make recommendations without taking into consideration physician input or test scientific 
knowledge. In one specific case, direct to consumer (DTC) telemedicine technologies have been 
inadequate in diagnosing certain types of conditions (Resneck et al 2016).  
With regards to synchronous technologies, telephone visits represent one of the initial 
forays into telemedicine. In other cases, telemedicine is delivered via a physician-to-physician link, 
such as Project ECHO. While telemedicine is inherently dependent on the hardware, organizations 
are also crucial to the widespread dissemination and acceptance of the telemedicine concept 
assuming that all players are given equal access to accurate information. The concept of digital 
redlining also creates a different problem where even those seeking to utilize synchronous 
telehealth services may be limited by their accessibility to broadband internet (O’Shea 2022). The 
United States federal government has taken distinct steps to address this digital divide by increasing 
funding to providers to uptake telemedicine and expanding broadband access to consumers (Federal 
Communications Commission).  
As the importance of telemedicine has grown, the federal government has sought out 
methods to incentivize primary care providers to offer telehealth services to their patients. Started 
in 1997, the Rural Health Care Program provides funding to eligible healthcare providers for 
telecommunications and broadband services necessary for the provision of healthcare (Federal 
Communications Commission). Over the years, this program has expanded to include two 
components: the Healthcare Connect Fund Program and the Telecommunications Program. The 
15 
 
Telecommunication Program, which was established in 1997, was used to subsidize the difference 
between urban and rural rates for telecommunications services (Federal Communications 
Commission). The Healthcare Connect Fund Program was established 2012 with the aim of 
providing high-capacity broadband to eligible healthcare providers in an effort to improve 
connectivity, which could be used for the transmission of patient data or telehealth services (Federal 
Communications Commission). However, as time has gone on, these components of the Rural 
Health Care Program have been unable to keep pace. 
 Researchers have noted that rural hospitals were likely to benefit the most from telehealth, 
but least likely to have the ability to deliver these services (Zachrison et al 2020). Therefore, in 
2018, the Federal Communications Commission made two decisions to expand accessibility to 
these broadband accessibility funds. The first decision moved to increase the pre-existing program 
funding cap from $400 million to $571 million beginning in Funding Year 2017, which was also 
meant to serve as a reflection of inflation (Federal Communications Commission). Secondly, the 
program moved to carry-forward unused funds from the previous funding years in an effort to fund 
future funding years (Federal Communications Commission).  Taken together, these changes were 
meant to increase the availability of broadband in rural areas, which would likely directly impact 
physician willingness to provide telehealth services.  
While the Rural Healthcare Program denotes that coverage would be limited to rural areas, 
the eligibility requirements allow for providers in non-rural areas to uptake funds. These entities 
include the following: post-secondary educational institutions offering health care instruction, 
teaching hospitals, and medical schools; community health centers or health centers providing 
health care to migrants; local health departments or agencies; community mental health 
centers; not-for-profit hospitals; rural health clinics; skilled nursing facilities (as defined in section 
395i–3(a) of title 42; and consortium of health care providers consisting of one or more entities 
falling into the first seven categories. In this case, consortiums must be majority rural, which means 
that more than 50% of sites participating in this association must be rural; moreover, these 
consortiums must be majority rural within three years of obtaining their initial funding commitment.  
16 
 
Chapter 3: Built Environment and Health Outcomes 
3.1 Introduction 
 While process conformance in the American healthcare system has dramatically improved 
over the past several decades, these improvements have relied heavily on reductionist methods 
while failing to address the substantial disparities in care (Ahn, 2006a). Socioeconomic 
disadvantages, particularly aspects of patients’ surrounding physical environment, are known to 
lead to poor health outcomes (Berman et al., 2021). However, the environments surrounding 
patients have garnered increased attention over the last two decades. In 2010, the passage of the 
Patient Protection and Affordable Care Act (ACA) resulted in the development of the National 
Prevention Strategy (NPS), which prioritized understanding the human-made surroundings, such 
as grocery stores and sidewalks, and their impact on health outcomes for individuals. More recently, 
the International Classification of Diseases (ICD) has been expanded with ICD-Z condition coding 
to account for patient environmental factors impacting health outcomes.  
The idea that people’s environment impacts their health outcomes is not new; in fact, this 
concept dates back to Hippocrates, father of the Hippocratic oath, nearly 2,400 years ago (Jackson, 
2003). Public health researchers have long noted the impact of the surrounding environment on 
health behaviors, particularly in the case of built environment (Popkin, Duffey, & Gordon-Larsen, 
2005). Built environment is defined as the urban design, land use, and transportation system that 
encompass patterns of human activity within the physical environment (Handy, Boarnet, Ewing, & 
Killingsworth, 2002). In addition, operations management researchers have also noted the impact 
of the surrounding environment on health outcomes in relation to a firm’s processes (Zepeda & 
Sinha, 2016; Muthulingam et al., 2020). However, in the healthcare operations management 
literature, studies examining the impact of the surrounding environment on treatment process 
outcomes are scarce. Given that a patient’s neighborhood environment could impact their health 
outcomes, we seek to examine how the built environment might affect hospital processes and their 
related outcomes, specifically the number of annual inpatient stays.   
This study seeks to understand broader contextual factors surrounding patients that might 
impact treatment processes and the impact of those contextual factors on individual health 
outcomes. We build upon previous research examining whether the surrounding environment can 
influence a firm’s processes (Muthulingam et al., 2020). In the health economics literature, authors 
have examined changed to the surrounding environment in terms of the impact of gentrification on 
mental health (Tran et al., 2020). Our research seeks to expand the current literature regarding 
patients’ surrounding environments and process outcomes, particularly inpatient stays. Specifically, 
17 
 
we focus on key structural characteristics related to the social determinants of health. This study 
will also examine these environments in terms of the impact on readmissions, as this has been an 
outcome variable used by CMS to assess hospital systems. Therefore, we study the research 
question is how neighborhood-level infrastructure, specifically infrastructure related to the built 
environment under the social determinants of health, impacts the frequency of inpatient stays?  
We test our predictions in the context of a natural disaster – specifically, the impact of 
Hurricane Sandy on New York City. To estimate the causal impact of Hurricane Sandy, we find a 
credible treatment and control group within New York City consisting of areas designated as flood 
zones. In our study, the treatment group consists of individuals living within predesignated flood 
zones that were damaged by the hurricane, while our control group consists of predesignated flood 
zones that were not damaged by the hurricane. To ensure that geographies are similar, we only 
compare zones that were designated as flood zones leading up to Hurricane Sandy. We measure 
the pre-Hurricane Sandy and post-Hurricane Sandy change of these built environment 
characteristics in flood zones. Using ArcGIS, we construct a half-mile radius across the road 
network surrounding each Census tract to understand the acceptability of built environment 
resources. Within each radius, we measure the change in three specific built environment 
characteristics: grocery stores, green spaces, and recreation facilities. 
Our study addresses calls from both the healthcare operations literature and health policy 
literature to better understand how social determinants of health can impact patient outcomes, 
specifically by examining individual components. By understanding the effects of built 
environment characteristics, we provide evidence as to how these social determinants of health may 
alter the way individuals utilize inpatient services. This study makes an important contribution to 
the healthcare operations management literature by providing exploratory empirical evidence on 
the surroundings where patients live and their likelihood of being admitted for an inpatient stay 
resulting from the way patients interact with their surroundings. Our work provides an alternative 
method of accounting for the effects of patients’ built environments in relation to the recently 
introduced ICD-Z condition codes. However, this coding requires physicians to screen patients 
during the uptake phase to determine characteristics related to the social determinants of health 
before denoting a specific ICD-Z condition code. In contrast to the current ICD-Z and screening 
protocols, our study provides an alternative method of assessing the impact of the social 
determinants of health by instead using the patient’s home ZIP Code Tabulation Area to account 
for nearby surroundings using readily available public data sources.  
3.2 Study Context and Background 
18 
 
 Traditionally, efforts to improve healthcare in the United States have focused on the 
healthcare system and process conformance; however, recently, there have been increased efforts 
to use a broader approach accounting for social, economic, and environmental factors influencing 
health – factors often referred to as the social determinants of health (Artiga & Hinton, 2018). While 
the importance of social determinants of health is well recognized, the discussion quantifying the 
exact impact on patient health outcomes has accelerated over the past two decades, specifically in 
terms of work exploring the variance in health outcomes caused by these determinants (Silverstein 
et al., 2019; Whitman et al., 2022). In certain cases, the impact of these social determinants of 
health can explain a larger portion of the variation in health outcomes and disparities than 
traditional ideas about quality of care and access to care (Daniel et al., 2018; Silverstein et al., 2019). 
While social determinants of health may be viewed as a public health issue, these determinants 
often impact individual health. A failure to address these external factors to healthcare can be costly. 
In certain areas, the social determinants of health have likely resulted in more deaths than diseases 
such as cancer (Daniel et al., 2018). The resulting economic impact of the social determinants of 
health on health outcomes has been estimated to total nearly 309 billion dollars and 
disproportionately affects more disadvantaged populations (Daniel et al., 2018). By looking at the 
scarcity and availability of these factors, we seek to address the way upstream factors in the form 
of the social determinants of health may impact downstream healthcare process outcomes. 
Understanding how these factors impact process conformance could improve health outcomes as 
we move towards a more patient-centered approach to care delivery. 
3.2.1 Evolving Perspectives: Reductionist versus Systems Approaches to Healthcare 
Modern medicine has made great advances in deconstructing complex biological systems 
and isolating the cause-and-effect mechanisms of human diseases into distinct sets of factors. By 
reducing complex biological systems into smaller parts, physicians and scientists have been able to 
advance medicine substantially, an idea known as reductionism (Ahn et al., 2006a). By 
understanding these cause-and-effect mechanisms, the healthcare delivery system used reductionist 
thinking to construct the current process conformance paradigm of structure, process, and outcomes, 
placing the greatest emphasis on processes within facilities (Bell et al., 2002). While reductionism 
has led to rapid advances in medicine, this approach of narrowing illnesses down to singular causes 
has had less success in efforts to understand complex diseases. In the case of chronic illnesses, 
patients’ interaction with their surrounding environments, specifically factors related to the social 
determinants of health, often impact behavior and compliance with treatment regimens. 
Furthermore, these factors lie outside the healthcare delivery environment of a hospital or clinic, 
19 
 
which makes it difficult for healthcare providers to directly manage the social determinants of 
health. 
Conformance quality plays an important role in improving healthcare outcomes by 
providing guidelines and checkpoints for procedures. However, as processes for delivering care 
mature, improvements to these guidelines may yield diminishing returns (Johnson, 1991; Mold, 
2010). Ahn et al. (2006b) note that reductionist paradigm has allowed medicine to improve 
processes related to a single factor, but this approach is less effective in addressing the impact of 
patients’ surrounding environments and the effects of these environments on health outcomes 
(Silverstein et al., 2019). Healthcare operations have long focused on process improvement within 
the hospital; however, recent work suggests that conformance to process standards does not yield 
significant improvements in outcomes (Hawn et al., 2011; Nicholas et al., 2010). In certain cases, 
physicians have argued that an increased emphasis on process conformance often removes context, 
specifically surrounding the way patients interact with their respective environments (Mold, 2022). 
Therefore, it is important to step back from process conformance guidelines and examine the 
geographic markets served by hospitals and the effects of these markets on patients’ health 
outcomes.  
Conformance quality has focused greatly on the aspects of care delivery occurring within 
hospitals that might be associated with the spread of infections, patient stress levels, and timely 
patient transfers (Nelson, West, & Goodman, 2005). Contextual changes in the environment can 
directly alter health behaviors resulting in changes to disease progression (Pearson, 2011). However, 
quantifying the impact of the environment outside the hospital has been explored to a lesser degree.  
These environments can play a large role in shaping health outcomes over time, whether as a result 
of altered behaviors or through consistent access to care (e.g., the Framingham Heart Study; 
Christakis & Fowler, 2007).  
3.2.2 Conformance Quality and Patient Inputs 
Notably, medical and health economics researchers have shifted their attention towards 
examining the impact of the patient ecosystems surrounding care delivery by addressing the fact 
that patients’ health outcomes are not only the product of their interactions with care providers but 
also their day to day interactions with their residential environments. Past operations management 
literature has noted the effect of external factors, including geography, on patient outcomes 
(Goldstein, 2002; Terwiesch, 2011; Zhang et al., 2016). For example, Terwiesch (2011) noted that 
the gains made from focus (repetition and concentration on a single procedure) could be explained 
in large part by the geographies where a hospital draws its patients from. While these studies note 
20 
 
the significant impact of geographic context on health outcomes, they do not directly test the 
specific availability of the social determinants of health within these geographies on patient 
outcomes. A reductionist approach to care may be able to account for these factors to some degree, 
but its focus would likely not be on understanding the effects of geography and conformance. 
Conversely, through a systems lens, researchers can examine the interaction of patients with their 
broader geographic context to improve our understanding of how geographies impact health 
outcomes. Notably, this can allow for exploration of the heterogeneity in built environment, which 
may explain the variation in patient outcomes when hospitals uphold nearly ubiquitous 
conformance standards related to healthcare delivery. 
3.2.3 Population Health and the Built Environment 
To understand the environment in which patients live, it is important to explore the 
environments surrounding their homes, also known as the built environment. The built environment 
refers the urban design, land use, and transportation system that encompass patterns of human 
activity within the physical environment (Handy, Boarnet, Ewing, & Killingsworth, 2002). The 
built environment includes factors impacting a patient’s daily life, including those that contribute 
to behaviors that can impact disease progression or treatment outcomes. By accounting for the built 
environment, it is possible to incorporate contextual factors that may impact performance in terms 
of process conformance and account for the relation between operational performance and external 
environmental factors (Chandrasekaran, Linderman, & Schroeder, 2015).  
In the medical literature, four decades of research has yielded evidence related to the impact 
of built environment factors on process outcomes. This legacy begins with historical zoning policy 
within the United States. During the Great Depression, the United States government implemented 
a policy known as redlining that determined the credit worthiness of neighborhoods, which 
generally resulted in minority neighborhoods receiving the highest risk and being marked red on a 
color-coded scale (Perez et al., 2021). Even now, these neighborhoods often lack critical built 
environment infrastructure such as grocery stores, which results in food deserts caused by limited 
access to affordable and nutritious food. In such cases, the neighborhood built environment can 
increase the risk of adverse cardiovascular events (Kelli et al., 2019). Notably, in the case of cardiac 
surgery, preoperative malnutrition due to food scarcity in patients’ neighborhoods results in higher 
post-operative morbidity and mortality (Kelli et al., 2019). 
In the public health sphere, neighborhood quality audits represent a core aspect of 
addressing population health risks by understanding the underlying social, economic, and 
environmental conditions. In the past, local governments worked with public health agencies to 
21 
 
form a cohesive strategy of urban planning and public health initiatives in an effort to improve 
health outcomes (Putnam & Quinn, 2007; World Health Organization, 2020). However, as time 
has progressed, this collaboration has fallen by the wayside. Without accounting for the change in 
the social determinants of health, quality initiatives aimed at process conformance quality within 
hospitals, such as Value Based Programs (VBP), may be omitting a key variable related to patient 
context. As hospital systems are asked to implement these conformance quality-based 
improvements, the areas surrounding hospitals will play a larger role. Notably, many of these 
programs do not appear to consider the patient context, which could punish hospitals serving 
patients in areas that lack crucial built environment infrastructure. For example, in recent years, 
studies have suggested hospital systems are electing to make tradeoffs between readmissions and 
mortality when dealing with patient populations that are more likely to be readmitted (Wadhera, 
2018). Theses impacted patients often come from low-income areas where patients may suffer from 
poor access to components of the social determinants of health (Wadhera, 2018).  
3.2.4 Addressing the Impact of Built Environment 
Even during the most recent series of healthcare reforms in the United States, policymakers 
and stakeholders acknowledge the impact of the social determinants of health on individual and 
population health. Recent healthcare reforms under the Affordable Care Act saw new policy 
guidelines meant to directly address the environments surrounding individuals such as creating a 
National Prevention Strategy and encouraging the formation of Accountable Care Organizations. 
With regard to Accountable Care Organizations, the use of built environment interventions was 
used to issues “park prescriptions” to encourage individuals to walk in local parks for 30 minutes 
per day (Zusman, 2014). Cities were also encouraged to consider zoning initiatives that increased 
accessibility to green space (Zusman, 2014). On a larger scale, the Department of Health and 
Human Services created a National Prevention Strategy outlining ways for communities to improve 
population health. 
The National Prevention Strategy (NPS), the companion policy to the Patient Protection 
and Affordable Care Act (ACA), was meant to complement the insurance coverage expansion and 
quality of care improvement initiatives by addressing where patients live and work to improve 
preventative health measures, resulting in better health outcomes. A lack of built environment 
infrastructure surrounding patients may promote poor health behaviors and exacerbate prevalent 
health problems such as heart disease and depression (National Institutes of Health, 2014). The 
National Prevention Strategy was meant to provide a comprehensive plan for the nation’s 
preventative health by identifying environmental factors that could exacerbate disease. The 
measures outlined by the NPS were expected to improve health outcomes by addressing factors 
22 
 
that were often not part of physicians’ purview of treatment. One specific area of interest is the 
built environment characteristics as outlined by the social determinants of health. The Department 
of Health and Human Services notes that the built environment plays a significant role in the health 
and well-being of individuals living within a given geographical area. Therefore, we examine the 
specific components of the built environment that may impact health outcomes.  
To better address the social determinants of health, various stakeholders have created 
initiatives to quantify the effects of built environment. In 2011, the Centers for Disease Control and 
Prevention/Agency for Toxic Substances and Disease Registry Social Vulnerability Index was 
introduced, building upon previous work conducted by the Geospatial Research, Analysis, and 
Services Program analyzing socially vulnerable populations and recovery efforts. This Social 
Vulnerability Index was intended to classify the impact of external stressors on human health, 
including natural and human-caused disasters as well as disease outbreaks. The CDC suggested 
that reducing the social vulnerability of individuals would quantifiably reduce both human 
suffering and economic loss. More recently, a new series of International Classification of Diseases 
(ICD) condition codes known as ICD-Z codes were introduced to help physicians account for their 
patients’ environments. While these efforts represent actions meant to better account for social 
determinants of health, we provide empirical evidence of their effects, specifically those related to 
surrounding structural characteristics, on health outcomes. With regards to medical education, the 
American College of Physicians recently took steps to reform medical education in an effort to help 
healthcare professionals identify and address social determinants of health that may negatively 
impact patient outcomes (Daniel et al., 2018). Physicians are crucial for documenting the health 
impact of built environment on the population they serve. 
3.3 Hypotheses 
 We model the reconstruction of key built environment components following a natural 
disaster to understand the impact of the social determinants of health. In our study, we utilize the 
exogenous shock of Hurricane Sandy to understand how the sudden destruction of built 
environment characteristics impacts health outcomes, specifically the number of inpatient stays. 
3.3.1 Input Uncertainty 
In the world of healthcare operations management, a great deal of emphasis is placed on 
the process step related to care delivery, with fewer studies noting the heterogeneity of inputs for 
treatment. However, as Argote (1982) noted, the variation in environments can generate input 
uncertainty impacting health outcomes. Public health researchers have long examined the impact 
of varying patient contexts on disease development. However, in medicine, these contextual factors 
23 
 
are often secondary to reductionist approaches related to process conformance based on the initial 
patient encounter information. Physicians may simply look for comorbidities upon entry rather than 
as specific questions related to the built environment where a patient resides.  
The central role of the built environment is often related to the way it impacts individual 
health behaviors. For example, if only unhealthy food options exist, individuals will likely be forced 
to select those options based on their environments. The built environment is expected to contribute 
to health outcomes based on the clustering of available resources in residential areas related to 
patient context (Joost, 2016). Other research indicates that contextual changes often change disease 
progression by impacting behavior (Pearson, 2011). Therefore, the built environment can serve as 
a proxy for individual health behaviors that could impact process outcomes. We propose that the 
built environment impacts patient conditions by increasing input uncertainty that a standard set of 
biomedical exams may not be able to assess easily upon initial diagnosis. 
Therefore, we propose examining the characteristics related to the social determinants of 
health, specifically their built environment, to reduce the impact of input uncertainty related to 
patients. As Zhang et al. (2016) note, there are noticeable geographical differences in process 
outcomes. In recent years, a new set of condition codes, known as the ICD-Z codes, was introduced 
by the World Health Organization (WHO) to capture portions of the social determinants of health 
in an effort to improve outcomes. The introduction of these codes by the WHO was expected to 
enable healthcare delivery systems to identify disparities in population health, which might create 
a treatment strategy for a group of individuals based on their location. This would allow physicians 
to guide local policy to address environmental health factors. However, since the adoption of this 
new coding in 2016, the adoption rate has been lackluster amongst physicians (AHA, 2022). 
Nonetheless, these contextual factors provide critical information to providers by elucidating 
possible environmental factors impacting health behaviors that could result in health disparities. 
In a recent study conducted by the Agency for Healthcare Research and Quality, 
researchers noted that certain social determinants of health could predict the likelihood of opioid-
related inpatient stays (Owens et al., 2020). Whereas acute illness may require only considerations 
of healthcare delivery context, chronic illness requires a consideration of the patient context, which 
suggests that the features of these geographies will only increase patients’ suffering from chronic 
illness (Greenhalgh, 2009). Given our focus on the built environment and structural factors 
impacting health outcomes, we hypothesize as follows:  
Hypothesis 1a: Following the redevelopment after Hurricane Sandy, an increase in 
grocery stores reduces the number of inpatient stays. 
24 
 
Hypothesis 1b: Following the redevelopment after Hurricane Sandy, an increase in green 
space reduces the number of inpatient stays. 
Hypothesis 1c: Following the redevelopment after Hurricane Sandy, an increase in 
recreational facilities reduces the number of inpatient stays. 
3.3.2 Patient Context Versus Healthcare Delivery Context 
Procedures are often considered solely based on the healthcare delivery context of the 
process and physicians (Bell et al., 2002). However, patients are subject to interactions with their 
context (i.e., built environment) prior to treatment and during the recovery process. The impact of 
a patient’s external environment has been attributed to disease development. Health behaviors, such 
as nutrition, can be behavioral byproducts of the aforementioned built environment, which may 
provide additional considerations in terms of process and outcomes (Popkin, Duffey, and Gordon-
Larsen 2005). If inadequate resources exist, the patient may fall prey to behaviors that manifest in 
the form of disease and generate additional conditions. This context might play a greater role in the 
process outcomes by generating variability that cannot be immediately accounted for through basic 
physician patient encounter interview questions, thereby generating information asymmetry 
leading to input uncertainty.  
The social determinants of health, including the built environment, have long been 
considered the domain of public health, specifically with regard to programmatic interventions 
addressing how individuals interact with their surroundings; however, the operations management 
literature has placed less consideration on the social determinants of health when analyzing patient 
outcomes. However, the idea of controlling for patient context to improve process outcomes is not 
a revolutionary idea. Previously, doctors at St. Luke’s Roosevelt Hospital set out to find an 
alternative explanation detailing the issue of nutrition in advanced heart failure (Paccagnella, Calò, 
Caenaro, Salandin, Simini, & Heymsfield 1994). This study found that pre- and post-surgery 
nutrition in individuals with heart failure was crucial to achieving optimal baselines prior to surgery 
and could help sustain an individual post-surgery. The authors of this study show how the reduction 
in uncertainty generated by patient context with regard to nutrition prior to surgery can improve 
outcomes.  
By understanding how environments impact individuals, healthcare providers can 
understand how individuals’ health relates to their associated context (Pearson, 2011) and thus can 
account for factors related to the built environment (e.g., nutrition habits) ultimately leading to 
improved outcomes. An examination of interaction effects will reveal how the interaction of patient 
context and process conformance impacts individuals; furthermore, it is expected that the impact 
25 
 
of this patient context will matter more than their healthcare delivery context. Given the importance 
of the built environment in the pre- and post-treatment settings, we hypothesize as follows: 
Hypothesis 2: Following Hurricane Sandy induced redevelopment, hospital quality in the 
form of hospital acquired conditions will have a weaker negative impact on annual 
inpatient stays.  
3.4 Data 
To test the study’s hypotheses, we utilize data from the state of New York. We construct a 
longitudinal panel from 2011 to 2016 with ZCTA year observations. We use New York Statewide 
Planning and Research Cooperative System (SPARCS) patient data to longitudinally track patients 
with readmissions over the period from 2011 to 2016. SPARCS is a database that provides various 
healthcare stakeholders with financial planning and monitoring. Every hospital within the state of 
New York is required to report patient-level data such as diagnoses and treatments, services 
provided, and type of visit. In addition, these data allow us to examine whether individuals moved 
across geographic regions or hospitals in the given time frame.  
Our sample focuses on noninfectious diseases resulting in inpatient stays. Specifically, we 
focus on patients with heart disease. Each patient is tracked longitudinally over time to examine 
the impact of environmental change on inpatient hospitalizations. Our design examines the changes 
in the built environment related to the social determinants of health following a natural disaster. 
Dependent Variable: Our variable of interest is the number of inpatient stays per patient 
within a calendar year. While previous studies have examined the social determinants of health on 
all cause readmissions (Barnett et al., 2015; Sill et al., 2016; Truong et al., 2020), we expand our 
dependent variable to include all inpatient visits during a given year. However, we limit our sample 
to heart disease. We build on prior research examining the impact of the social determinants of 
health on inpatient stays (Owens et al., 2020).  
Independent Variables: To assess the built environment, we use data provided by the 
United States Census Bureau and the North American Industry Classification System. These key 
performance indicators of built environment were derived from the United Nations initiative United 
for Smart Sustainable Cities. These measures include unemployment, travel time, air pollution, 
green areas, recreational facilities, physicians, hospitals, health insurance coverage, housing 
expenditures, poverty, and violent crime rate.  
In addition, the selected variables related to infrastructure came from the following NAICS 
codes: grocery stores (445110), farmers markets (445210, 445220, 445230), recreational facilities 
(713940), and parks (712190). However, these data can only provide for the general availability of 
26 
 
services. We utilize density measures to examine the impact of these various types of infrastructure. 
This involves examining how much square area within a ZCTA is covered by a single entity within 
a category. To control for healthcare-related built environment variables, we include nearby 
pharmacies and clinics. We use the NAICS codes to denote these facilities as follows: primary care 
clinics (621111) and pharmacies (446110). These variables were all collected from the InfoGroup 
Historical Business Database to create a longitudinal panel of built environment factors while also 
acquiring the precise geocoded location.  
Instead of using a density measure constructed by dividing land mass by service type, we 
construct service areas at the Census tract level to understand how individuals might leverage 
services. To do this, we use ArcGIS Pro 2.7 Network Analyst to understand how individuals use 
specific services. We build a longitudinal dataset with exact location coordinates for each of the 
aforementioned services. Next, we calculate the distance from the centroid of each census tract 
within a ZCTA to understand how far on average individuals living in a ZCTA must travel to access 
a given service. We provide the average distance to each service in our model to improve our 
understanding of how closely services are located to the centroid of each Census tract. 
With regards to hospital conformance quality, we utilize the number of hospital acquired 
conditions related to annual heart failure inpatient stays. This variable accounts for possible 
variations in hospital processes that results in patients becoming sicker during an inpatient stay. 
With regard to hospital conformance quality, we utilize the number of hospital-acquired 
conditions related to annual heart failure inpatient stays. This variable accounts for possible 
variations in hospital processes that results in patients becoming sicker during an inpatient stay. 
Patient Control Variables: The SPARCS data also provide the relevant patient controls, 
including sex, age, procedure group, number of conditions upon entry, hospital-acquired conditions, 
type of payment method, and risk of mortality. With regard to payment, we include the following 
categories: Workers’ Compensation, Medicare, Medicaid, Other Federal Program, Blue Cross, 
CHAMPUS, Other Non-Federal Program, Disability, Title V, and Unknown. These factors allow 
us to build a more cohesive view of the patient being assessed by understanding their characteristics 
upon entry. In addition, it allows us to control for possible changes in access to care (e.g., insurance 
coverage) following the Affordable Care Act. 
Hospital Control Variables: This category includes the number of beds, trauma center level, 
and whether the hospital contains an emergency department. These variables address the hospital 
built environment in order to understand the capabilities and services delivered at a specific location. 
To account for possible physician individual effects, we generate a unique identifier for each 
physician–patient dyad. Next, we account for the number of various dyads that exist annually 
27 
 
within our sample. The goal is to account for consistency of care with the same physician. For 
example, if this value is equal to 1, this suggests that the patient saw the same physician each time 
they were admitted for an inpatient stay within a given setting. 
To adequately identify the hospital characteristics, we use the audited version of New York 
State Hospital Cost Report Data, which provides financial statements released by each hospital. 
These data indicate the number of beds and type of ownership for each hospital. To cross-validate 
the system to which a given hospital belongs, we utilize the data provided by the Dartmouth Atlas 
related to ownership and capacity. To match these disparate data sets, we match according to the 
location address. For hospitals that report as a single entity, we match existing financial accounting 
data reported to the state of New York to understand which hospitals report as a single entity. 
Closures of hospitals were also monitored using historical news stories.  
Neighborhood Control Variables: We control for the following additional neighborhood 
characteristics to ensure the effect observed is related to the built environment characteristics. 
These controls include the following variables: Earned Income Tax Credit, Alternative Minimum 
Tax, Form 1040s, Urgent Cares, General Hospitals, Mental Hospitals, and Median Household 
Income. To account for housing stability, we utilize the Department of Housing and Urban 
Development Aggregated USPS Administrative Data to control for vacancies and shifts in 
occupants. 
In addition, we utilize Property Land Use Tax Lot Output to control for the zoning 
characteristics of buildings within a ZCTA. This data set provides information on the average 
occupancy of units, zoning rules impacting residents, and property taxes paid to the city. By 
controlling for zoning rules, we control for the types of changes to the built environment that can 
occur given current city guidelines.  
Neighborhood Care Delivery Control Variables: We control for the availability of 
pharmacies and primary care clinics surrounding the Census tracts in each ZCTA. Recent literature 
has noted that pharmacies often close in urban areas serving publicly insured individuals 
(Guadamuz et al., 2019); furthermore, these closures often result in reduced adherence to 
medications given the lack of access to drugs (Qato et al., 2019). With regard to lack of primary 
care, the Health Resources and Services Administration defines Health Professional Shortage Areas 
to identify population groups within the United States lacking an adequate number of health 
professionals to provide data to the local population. Therefore, we seek to account for these factors 
surrounding patients to better isolate the effect of the surrounding built environment. 
28 
 
3.5 Empirical Strategy 
 To understand the impact of built environment characteristics, we identify an event that 
would eliminate or severely hinder access to our selected structural resources. We build upon 
previous literature utilizing Hurricane Sandy as an exogenous shock on New York City (Rehse et 
al 2019). Furthermore, we examine the change in new building permitting following Hurricane 
Sandy to account for local improvements to the built environment. This allows us to consider 
changes in a patient’s surrounding building environment and the impact on health outcomes.  
Therefore, our sample consists of individuals living within flood zones within New York City. 
These zones are well published in the literature. Our treatment group consists of flood zones that 
were damaged during Hurricane Sandy, while our control group consists of flood zones that were 
damaged during Hurricane Sandy that maintain a steady state with regards to building permits.  
Figure 2: Visualization of Identification Strategy 
 
Description: 
The blue areas represent flood zones within New York City as of 2012. The hashed lines represent areas 
damaged according to FEMA. 
 In our study, we utilize Hurricane Sandy as an exogenous shock impacting the built 
environment in New York City. By tracking individuals who remain in ZCTAs over time, we 
examine how the surrounding built environment impacts these individuals’ health outcomes, 
29 
 
specifically the number of inpatient stays over time for those choosing to remain. The vector of 
time-varying control variables is denoted by 𝑿 , 𝜆  represents the ZCTA fixed effect, and 𝜇 
represents year fixed effect. 
Inpatient Stays = α + β1Grocery Store Distance + β2Green Space Distance
+ β3Recreational Facility Distance + β4Pharmacy Distance
+ β5Primary Care Distance + β6(Redevelopment × Grocery Store Distance)
+ β7(Redevelopment × Green Space Distance)
+ β8(Redevelopment × Recreational Facility Distance) + 𝛄𝐗 + λ + μ + ε 
3.5.1 Difference-in-Differences Analysis 
To understand the impact of the social determinants of health related to the built 
environment, we examine the impact of a natural disaster and the subsequent built environment 
change related to the shock. In our study, we examine the impact of Hurricane Sandy on New York 
City where we denote damaged flood zones witness a change in building permits as our treatment 
group. To assess damage, we use data from the Small Business Administration disaster assistance 
loans related to Hurricane Sandy to denote areas with high levels of damage. Flood zones that are 
damaged with no change in building permits related to built environment characteristics serve as 
our control group. 
 To ensure our sample consists of patients who did not move during the period of our study, 
we create unique ZCTA identifiers. When individuals’ identifiers remain consistent, we retain them 
in our sample. This strategy allows us to understand what happens to individuals as the environment 
changes around them instead of self-selecting into a different built environment. 
To reduce concerns of selection bias affecting our results, we utilize a Coarsened Exact 
Matching (CEM) strategy to generate a matched sample of treated and control patient-year 
observations that do not exhibit systematic or significant differences in the pre-intervention period. 
Specifically, CEM allows for multi-dimensional exact matching consistent with a two-stage 
matching technique, which controls for selection on observable differences by eliminating non-
analogous observations in the treatment and control populations (Iacus et al., 2012). We use the 
following covariates at the patient level in the pre-intervention period: age, race, gender, percentage 
of population within a ZCTA receiving a Supplementary Security Income, ZCTA Inward Migration, 
ZCTA Population Density, and ZCTA Median Household Income. Our study seeks to understand 
the impact of the social determinants of health on individual health outcomes. We match on two 
30 
 
major categories of variables: individual characteristics and built environment characteristics. 
Appendix A1 provides a description of the matching variables.   
3.6 Results 
 In this section, we report the results from the estimation of the DiD specification 
highlighted in equation (1). Table 1 provides the summary statistics for the variables used in the 
model specification. In Table 2, we provide the estimation results.  
Table 1: ZCTA Difference-in-Differences Variable Summary Statistics 
 Mean SD Min Max 
Dependent/Independent 
Variables 
    
Annual Inpatient Stays [Heart 
Failure] 
0.810 0.956 0.000 12.000 
Grocery Stores Half Mile 0.838 0.946 0.000 6.250 
Green Space Half Mile 0.194 0.344 0.000 5.000 
Recreation Facilities Half Mile 1.021 2.294 0.000 29.000 
     
Neighborhood Care Delivery Characteristics    
Pharmacies Half Mile 3.634 4.026 0.000 20.000 
Primary Care Clinics Half Mile 49.351 106.753 0.000 1147.333 
     
Hospital Characteristics     
Number of Beds 511.538 250.998 80.000 1526.000 
LVS Function Evaluation Score 99.512 0.731 95.000 100.000 
ACE Inhibitor/ARB Score 95.410 5.030 75.000 100.000 
     
Patient Characteristics     
Patient Age 73.885 14.285 16.000 110.000 
LOS Average [Heart Failure] 2.236 3.413 0.000 74.000 
Average Copayment [Heart 
Failure] 
1.597 3.027 0.000 74.000 
Severity of Illness [Heart Failure] 0.881 1.124 0.000 4.000 
Average Comorbidities 0.752 1.020 0.000 7.000 
Average Hospital Acquired 
Conditions 
0.104 0.301 0.000 3.000 
Inpatient Stays [Other Illnesses] 1.418 1.541 0.000 17.000 
     
Neighborhood Characteristics     
Population Density: Residential 
Area 
0.004 0.010 0.001 0.465 
Population: Black % 0.311 0.302 0.001 0.934 
Population: Native % 0.003 0.003 0.000 0.014 
Population: Asian % 0.091 0.088 0.000 0.460 
Population: Other % 0.092 0.093 0.000 0.410 
Median Household Income 55114.423 19026.475 27011.000 113973.000 
Observations 3814 
 
Column (1) in Table 2 provides our baseline main effects model that includes a complete 
set of controls. The main effects of Grocery Stores Half Mile, Green Space Half Mile, Recreation 
31 
 
Facilities Half Mile, and Hospital-Acquired Conditions (Heart Failure Visits). Columns (2) and (3) 
build upon the main effects model hierarchically through the addition of interaction terms for built 
environment characteristics and hospital conformance. Finally, column (4) represents the full 
model that includes all interaction terms. We use the results from column (4) for interpreting our 
hypotheses tests. 
Table 2: Built Environment and Inpatient Stays 
 OLS 
 (1) (2) (3) (4) 
 β(SE) β(SE) β(SE) β(SE) 
Built Environment     
Grocery Stores Half Mile -0.18209 (0.098)* -0.19618 (0.106)* -0.18326 (0.099)* -0.19734 (0.107)* 
Green Space Half Mile -0.11373 (0.200) -0.09769 (0.246) -0.11396 (0.200) -0.09745 (0.247) 
Recreation Facilities Half Mile -0.00097 (0.051) -0.00525 (0.062) -0.00093 (0.051) -0.00550 (0.062) 
Pharmacies Half Mile 0.08898 
(0.024)*** 
0.09889 
(0.028)*** 
0.08913 
(0.024)*** 
0.09908 
(0.028)*** 
Primary Care Half Mile -0.00337 
(0.001)** 
-0.00367 
(0.001)*** 
-0.00336 
(0.001)** 
-0.00366 
(0.001)*** 
     
Hospital Characteristics     
Emergency Department -1.83313 
(0.377)*** 
-1.78991 
(0.376)*** 
-1.82940 
(0.386)*** 
-1.78651 
(0.385)*** 
Teaching Hospital -0.12232 (0.245) -0.13162 (0.247) -0.12485 (0.245) -0.13406 (0.247) 
Number of Beds 0.00039 (0.000) 0.00039 (0.000) 0.00039 (0.000) 0.00040 (0.000) 
Hospital Acquired Conditions [Heart Failure Visits] 1.12844 
(0.071)*** 
1.12722 
(0.071)*** 
1.12376 
(0.083)*** 
1.12279 
(0.083)*** 
     
Treatment     
Hurricane Sandy Induced Redevelopment -0.21428 (0.114)* -0.09917 (0.130) -0.21987 (0.117)* -0.10485 (0.130) 
     
Interactions     
Redevelopment X Grocery Stores   -0.25016 
(0.095)** 
  -0.25033 
(0.096)** 
Redevelopment X Green Space   0.29087 (0.308)   0.29098 (0.307) 
Redevelopment X Recreation Facilities   0.04625 (0.037)   0.04656 (0.038) 
Redevelopment X Hospital Acquired Conditions [Heart Failure 
Visits] 
    0.01470 (0.098) 0.01394 (0.098) 
     
Constant 3.39030 (9.585) 2.79420 (8.893) 4.22927 (9.572) 2.07194 (8.933) 
Patient Controls Yes Yes Yes Yes 
Insurance Controls Yes Yes Yes Yes 
Hospital Ownership Controls Yes Yes Yes Yes 
Demographic Controls Yes Yes Yes Yes 
Hospital System FE Yes Yes Yes Yes 
County FE Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes 
Clustered SE Yes Yes Yes Yes 
Number of Observations 2111 2111 2111 2111 
* p<0.10, ** p<0.05, *** p<0.01 
3.6.1 Built Environment 
 Examining column (4) of Table 2, we find that the coefficient for Redevelopment × 
Grocery Stores exhibits a negative significant effect on annual inpatient stays related to heart failure 
(β = -0.25033, p<0.05) suggesting that an increasing number of grocery stores following Hurricane 
Sandy redevelopment reduces the average number of inpatient stays for heart failure patients living 
within a ZCTA. Thus, H1a is supported. With regard to green space, we find that the coefficient 
for Redevelopment × Green Space is statistically insignificant (β = 0.29098, p>0.10). Therefore, 
we find support for H1b. Next, we consider the impact of Recreation Facilities on the average 
number of annual inpatients stays related to heart failure. We find that Redevelopment × Recreation 
32 
 
Facilities exhibits a positive and insignificant effect on the average number of annual inpatient 
stays (β = 0.04656, p<0.10).   
3.6.2 Hospital Conformance 
 Finally, we examine the impact of hospital quality on the annual number of inpatient stays 
related to heart failure within a ZCTA. To examine hospital quality, we utilize the total number of 
hospital-acquired conditions annually. We find that while Hospital-Acquired Conditions (Heart 
Failure Visits) exhibits a significant and positive effect prior to redevelopment (β = 1.12279, 
p<0.01), this effect becomes insignificant following the redevelopment of built environment in 
flood zones (β = 0.01394, p>0.10). This result may suggest that the redevelopment of patients’ 
home surroundings may reduce aspects of input uncertainty resulting in increased inpatient stays. 
3.7 Robustness Checks, Alternative Explanations, and Post Hoc Analysis 
3.7.1 CMS Quality Conformance Guidelines 
Table 3: Built Environment and Inpatient Stays with CMS Metrics 
 (1) (2) 
 β(SE) β(SE) 
Built Environment   
Grocery Stores Half Mile -0.17740 (0.099)* -0.19591 (0.106)* 
Green Space Half Mile -0.28612 (0.146)* -0.29693 (0.168)* 
Recreation Facilities Half Mile 0.03295 (0.047) 0.04092 (0.050) 
Pharmacies Half Mile 0.09154 (0.025)*** 0.10093 (0.026)*** 
Primary Care Half Mile -0.00437 (0.001)*** -0.00473 (0.002)*** 
   
Hospital Characteristics   
Emergency Department -1.81923 (0.380)*** -1.75843 (0.336)*** 
Teaching Hospital -0.12954 (0.233) -0.11980 (0.233) 
Number of Beds 0.00046 (0.000) 0.00045 (0.000) 
LVS Function Evaluation Score 0.06936 (0.094) 0.09587 (0.100) 
ACE Inhibitor/ARB Score 0.00543 (0.009) 0.00435 (0.009) 
   
Treatment   
Hurricane Sandy Induced Redevelopment -0.18767 (0.116) 10.75051 (12.879) 
   
Interactions   
Redevelopment X Grocery Stores   -0.19838 (0.088)** 
Redevelopment X Green Space   0.29599 (0.363) 
Redevelopment X Recreation Facilities   0.02553 (0.025) 
Redevelopment X LVS Function Evaluation   -0.10448 (0.133) 
Redevelopment X ACE Inhibitor/ARB   -0.00433 (0.010) 
   
Constant -5.74945 (11.901) -7.74771 (11.877) 
Patient Controls Yes Yes 
Insurance Controls Yes Yes 
Hospital Ownership Controls Yes Yes 
Demographic Controls Yes Yes 
Hospital System FE Yes Yes 
County FE Yes Yes 
Year FE Yes Yes 
Clustered SE Yes Yes 
Number of Observations 2092 2092 
* p<0.10, ** p<0.05, *** p<0.01 
Additionally, we examine the impact of hospital conformance quality using Center for 
Medicaid and Medicare Services Hospital Compare-based metrics for heart failure on the average 
33 
 
number of inpatient stays related to heart failure within a ZCTA. With regard to conformance, we 
utilize two measures: Evaluation of Left Ventricular Systolic (LVS) Function (LVS Function 
Evaluation) and ACE Inhibitor or Angiotensin Receptor Blocker (ARB) for Left Ventricular 
Systolic Dysfunction (LVSD) (ACE Inhibitor/ARB). Evaluation of Left Ventricular Systolic (LVS) 
Function denotes that a left ventricular systolic (LVS) function evaluation was conducted before 
arrival, conducted during hospitalization, or planned for after discharge. ACE Inhibitor or 
Angiotensin Receptor Blocker (ARB) for Left Ventricular Systolic Dysfunction (LVSD) refers to 
heart failure patients with left ventricular systolic dysfunction (LVSD) and without both 
angiotensin converting enzyme inhibitor (ACE inhibitor) and Angiotensin Receptor Blocker (ARB) 
contraindications who were prescribed an ACE inhibitor or an ARB at hospital discharge.  
In Table 3 column (2), we find that the coefficients for both Redevelopment × LVS 
Function Evaluation (β = -0.10448, p>0.10) and Redevelopment × ACE Inhibitor/ARB (β = -
0.00433, p>0.10) are insignificant. One possible explanation for these results is that many hospitals 
have been able to achieve CMS mandated guidelines, leaving little variation between hospitals 
related to these quality metrics. 
3.8 Discussion 
While we acknowledge the important of process conformance standards in healthcare 
outcomes, we offer a systems-based view that accounts for the impact of the built environment 
surrounding the places where patients live and work. Our study seeks to examine the economic 
impact of a specific set of built environment characteristics as they pertain to the social 
determinants of health. In the case of heart failure, we find support for the impact of the patients’ 
built environment in relation to the number of inpatient stays for patients. Our analysis reveals that 
grocery stores play a critical role in reducing the number of inpatient stays for heart failure patients.  
Secondly, our study provides a possible alternative to the use of ICD-Z codes when 
accounting for patients’ built environment. Using data that are mostly publicly available, we show 
that the effects of built environment can be capture adequately in relation to health outcomes. 
Readily available American Community Survey and zoning data could be used to help physicians 
account for variations in the patients being treated. This information could allow physicians to 
adjust pre- and post-operative procedures in an effort to reduce the number of inpatient stays. While 
our study examines only a narrow subset of conditions, our results provide initial empirical 
evidence that this method of accounting for built environment effects can be effective.  
34 
 
3.9 Policy and Managerial Implications 
 In relation to conformance quality programs, we note that accounting for the social 
determinants of health through publicly available data could help to identify areas for public health 
interventions. For hospital systems, publicly available data could be used to determine which 
patients require pre-operative interventions to minimize the effect of specific components of the 
social determinants of health related to built environment. Furthermore, government value-based 
purchasing programs may consider placing greater emphasis on the geographies from which 
hospitals draw their patients.  
In addition, our results suggest that the effects of process conformance and the built 
environment may impact patient outcomes independently. While reductionist methods have 
empirically been shown to improve health outcomes, the built environment does impact health 
outcomes by changing the input parameters of the care delivery process. This suggests that current 
Valued Based Purchasing programs may be penalizing hospitals even if said hospitals conformed 
rigorously to process conformance guidelines outlined by CMS. As our results suggest, providers 
conforming to these process guidelines may still be unable to overcome the input heterogeneity for 
hospitals. Future Value Based Purchasing programs should thus consider how to account for the 
variability in built environment characteristics of hospitals’ patient populations. 
3.10 Limitations and Future Research 
As with any study, our work has several limitations. Given the exploratory nature of our 
study, our analyses focus on the New York City metropolitan area, and thus our findings may not 
be generalizable to more rural areas. Future studies can expand upon our findings by examining 
non-urban areas. Furthermore, while we do have patient data, we do not have in-depth records 
related to individual physician notes and test results, which could make epigenetic factors difficult 
to account for in our study sample. Lastly, we also note that data on hospital staffing levels is not 
available in our study, which makes it difficult to address capacity constraints directly. 
Future work might include a more nuanced sample with more in-depth individual measures. 
Biomarkers would be the next logical step to understand whether the impact of patient context is 
impacting the immune system in a manner that negatively impacts health outcomes. Epigenetic 
inheritance, the idea that environments can impact the expression of certain immune factors and 
overall genetic expression, provides a promising method to analyze the interaction of patients and 
the broader system of contextual variables (Chen et al 2017). Furthermore, the use of biomarkers 
provides a more nuanced method of measuring the impact of the social determinants of health and 
quality-based outcome measures. A longitudinal process study could follow patients and doctors 
35 
 
over time to better understand whether hospitals can adapt current processes to account for the 
patient context to a greater extent than standardized guidelines. 
With regard to healthcare structural characteristics, future research could examine the 
impact of mail-order prescription drug services and telehealth. For areas lacking pharmacies and 
primary care clinics, mail-order prescriptions and telehealth visits could bridge the gap in care 
continuity and adherence to medications. In addition, it may be possible to understand whether 
these resources serve as substitutes or complements to traditional brick-and-mortar facilities.  
 
36 
 
Chapter 4: How Does Physical Access Affect Emergency Department Use? Evidence From 
Insurance Coverage Expansion 
4.1 Introduction 
A critical challenge for the healthcare supply chain is bridging the gap between patient 
demand and the supply of care (Finkelstein et al. 2012, Gruber and Sommers 2019). Over the past 
half century, American policymakers have made various efforts to improve the affordability and 
accessibility of healthcare. Most notably, the United States Social Security Amendments of 1965 
established Medicaid and Medicare to provide health insurance for elderly (Medicare) and low-
income (Medicaid) individuals. However, physical limitations to healthcare access continue to 
prevent individuals from receiving the right care, at the right time, and at the right place (Benitez 
and Seiber 2018). In 2010, policymakers passed the Affordable Care Act (ACA) to improve access 
to healthcare and reduce the cost of care through the expansion of health insurance coverage.  
The ACA’s provision to enhance Medicaid represents the largest expansion of the program 
since its inception in 1965. This expansion of Medicaid, paired with the mandatory insurance 
uptake statute, more commonly known as the Individual Mandate, was expected to increase demand 
for outpatient care at primary care clinics (Sommers and Epstein 2010, Sonier 2013). Policymakers 
made an assumption that previously uninsured patients requiring nonurgent outpatient care would 
reduce their reliance on emergency departments (EDs) and shift their utilization toward primary 
care clinics (Obama 2016). Assuming this demand shift toward primary care providers occurred, 
the cost savings were expected to be sizable. For example, studies conducted by the Health Care 
Cost Institute showed that in 2016, the average visit to a primary care clinic cost $106, while the 
average visit to an ED cost $894 (Hargraves and Kennedy 2018).  
However, recent research lacks consensus regarding the ACA’s impact on reducing 
preventable ED visits, with results ranging from marginal or no effects to an actual increase in ED 
visits (e.g., Pines et al 2016, Nikpay 2017, McConville 2018, Sabbatini and Dugan 2022). 
Healthcare and policy researchers have identified potential factors leading to these mixed results, 
including insurance coverage expansion to disadvantaged areas with limited capacity for absorbing 
increased demand for care (Gruber and Sommers 2019). Moreover, scholars have begun to examine 
factors that impact preventable ED visits other than financial access in the form of insurance 
coverage, specifically exploring the underlying healthcare supply chain (Benitez and Seiber 2018, 
Basu et al. 2019). This study examines whether and to what extent the variation in ACA-related 
preventable ED visits can be explained by the characteristics of the underlying healthcare supply 
chain.  
37 
 
Adopting a supply chain perspective can not only provide clues to explain the mixed results, 
but can also better inform policymakers on the actionable steps they can take to improve the 
effectiveness of the legislation (Betcheva et al. 2021, Dai and Tayur 2020). In particular, 
policymakers must examine the interplay between structure and infrastructure as it relates to the 
healthcare supply chain with respect to demand. Structure refers to “the organization’s physical 
brick-and-mortar attributes, such as the amount of production (or service delivery) provided” 
(Hayes et al. 2005, p. 41). Infrastructure refers to “systems, policies and practices that determine 
how the structural aspects are to be managed,” thereby allowing a structure to achieve its full 
potential (Hayes et al. 2005, p. 41). Within the context of this study, the structural characteristics 
we focus on are as follows: (i) the physical location of the EDs and primary care clinics, and (ii) 
the hours of operation of these facilities. On the other hand, the ACA represents an infrastructural 
change to the healthcare supply chain that establishes new policies geared toward improving 
financial access to outpatient care: in this case, the manner in which patients use primary care 
clinics and EDs. As policymakers work to make healthcare more affordable and more accessible 
through infrastructural reforms, we predict that newly insured individuals seeking outpatient 
services are likely to make accessibility decisions based on the relatively understudied structural 
characteristics of the healthcare supply chain that influence physical access (Katz et al. 2012).  
We test our predictions in the context of the ACA’s enforcement across four states—
California, Florida, Kentucky, and North Carolina—using granular data on ZIP Code Tabulation 
Areas (ZCTAs)2 from 2012 to 2016, consisting of 6,725 ZCTA-year observations. In addition, we 
use ArcGIS, a leading provider of geographic informational data and a tool used in geospatial 
research, to dynamically model realistic road network conditions within ZCTAs and calculate the 
shortest road distance patients must travel to access the nearest ED or the nearest primary care 
clinic. Subsequently, considering the differential enforcement of the ACA across the four states as 
a natural experiment, wherein California and Kentucky experienced the full enforcement of ACA 
policy statutes—Medicaid expansion and the Individual Mandate—in 2014, while Florida and 
North Carolina experienced partial enforcement through the Individual Mandate policy statute only, 
we examine the impact of physical access on preventable visits at the nearest ED in the form of ED 
discharges (i.e., visits to the ED that did not result in inpatient admissions or death). To develop a 
 
2 The term “Zip Code Tabulation Area” differs from the more widely known concept of ZIP code. ZIP codes 
were created by the United States Postal Service to facilitate mail delivery. They do not denote fixed 
geographic areas. In contrast, ZIP Code Tabulation Areas are generalized representations of ZIP codes built 
from existing census blocks. In creating ZCTAs, the Census Bureau considers the most frequently occurring 
ZIP code in an area as the ZCTA code.    
38 
 
deeper understanding of the mechanism underlying the structural aspects of the healthcare supply 
chain, we examine if and to what extent utilization levels at the nearest primary care clinic influence 
ED discharges. Herein, our analysis examines provider patient panel size at the nearest primary 
care clinic and also differentiates between Medicaid patient encounters based on plans restricting 
access to a predefined network of providers (i.e., Medicaid managed care) and those using an 
unrestricted fee-for-service model (i.e., Medicaid fee-for-service). This differentiation of delivery 
systems allows us to examine whether a constrained set of provider choices impacts patient 
utilization behavior when making nonurgent outpatient care decisions.   
While our results demonstrate that physical access is a key driver of ED discharges, the 
effect of distance is attenuated by the extent of the ACA’s enforcement. In other words, following 
the full enforcement of the ACA, newly insured individuals are less likely to continue using the 
nearest ED, even if it is closer than the nearest primary care clinic. During the post enforcement 
era, a one mile increase in distance difference—i.e., the difference in the road distance from a 
ZCTA centroid to the nearest primary care clinic, and from the ZCTA centroid to the nearest ED—
results in 17 fewer annual ED discharges from the ZCTA. Using figures provided by Hargraves 
and Kennedy (2018) regarding the approximate cost of an average visit to an ED vis-à-vis an 
average visit to a primary care clinic, this difference would potentially result in a reduction of 
US$42.73 million in annual ED costs across the two states under the full enforcement of the ACA 
in our sample, California and Kentucky. Moreover, we find that primary care clinics operating with 
additional weekend hours reduce annual ED discharges from a ZCTA. That is, following the full 
enforcement of the ACA, an additional hour of operation at the nearest primary care clinic during 
the weekend results in 22 fewer annual ED discharges. Across California and Kentucky, this one 
hour increase in the weekend hours of operation at the nearest primary care clinic would potentially 
result in a reduction of US$56.18 million in annual ED costs. 
With regard to utilization levels, we find a negative relationship between the average 
patient panel size for providers at the nearest primary care clinic and ED discharges, suggesting 
that when providers see more unique patients, it increases timely access to care at primary care 
clinics and reduces the potential for newly insured individuals to use the ED. Finally, our results 
reveal that an increase in Medicaid managed care patients at the nearest primary care clinic is 
associated with an increase in annual ED discharges, while an increase in Medicaid fee-for-service 
patients is associated with a decrease in annual ED discharges. These results suggest that following 
the full enforcement of the ACA, predetermined provider networks may have resulted in congestion 
at primary care clinics, pushing newly insured patients toward the nearest ED for timely access to 
care.  
39 
 
This study makes an important contribution to the healthcare management literature by 
providing concrete empirical evidence of the interplay between physical access to primary care 
clinics—both in terms of physical distance and hours of operation—and insurance coverage 
expansion, as well as how this interplay influences patients’ outpatient care decisions. Relatedly, 
by simultaneously analyzing primary care clinics and EDs, our study highlights how individuals’ 
outpatient usage decisions are impacted by the design of the healthcare supply chain. While the 
implications of the interplay between policy-driven infrastructural decisions and structural 
decisions of healthcare entities (e.g., primary care clinics and EDs) on healthcare accessibility 
continue to receive a great deal of attention within academic circles (e.g., Fulton 2017, Benitez and 
Seiber 2018, Basu et al. 2019, Dai and Tayur 2019, KC et al. 2020) and among policymakers (e.g., 
US Congress 2018, US Congress 2019) and the popular press (e.g., Marks 2020, Rapaport 2020), 
our study contributes to this line of research by highlighting the importance of healthcare supply 
chain design based on physical location and the temporal availability of healthcare entities within 
a geographical area (Green 2012, Dai and Tayur 2019, KC et al. 2020). Finally, our study’s findings 
highlight the role of capacity planning in primary care clinics as an important supply-side factor 
for policymakers to consider when implementing healthcare reforms aimed at altering demand. 
Specifically, our findings suggest that Medicaid managed care provider networks, which cover 
nearly 70% of Medicaid beneficiaries in the US (Garfield et al. 2018), may lack the capacity to 
meet recent demand changes and should be reassessed in an effort to reduce the gap between 
healthcare supply and demand.  
4.2 Study Context and Background  
4.2.1 The Affordable Care Act 
Recent studies in the healthcare management literature have increasingly focused on ED 
utilization and the performance of EDs (e.g., Song et al. 2015, Batt and Terwiesch 2017, Freeman 
et al. 2021). While efforts have been concentrated on a single point of care, i.e., EDs, researchers 
are increasingly recognizing the effects of policy shocks on the manner in which individuals access 
the broader healthcare supply chain (Dai and Tayur 2019). This study analyzes the enforcement of 
the ACA as a policy affecting ED discharges.  
One of the legislation’s central goals was to encourage primary care use for nonurgent 
healthcare needs, especially for health screenings and preventative services (Koh and Sebelius 
2010). This increase in primary care use was expected to reduce nonurgent and outpatient use of 
EDs (Obama 2016). The goal was to be achieved through the ACA’s expansion of health insurance 
coverage for low-income individuals, which was addressed in part through the expansion of 
Medicaid, a government insurance program. Under ACA-based Medicaid expansion, nonelderly 
40 
 
Americans between the ages of 18 to 64 with incomes at or below 138% of the federal poverty line 
would be eligible for Medicaid coverage (Centers for Medicaid and Medicaid Services 2015). Thus, 
the ACA focused on a major segment of the uninsured population with incomes exceeding 
historical Medicaid eligibility requirements, and with jobs that may not qualify for employer-based 
insurance. 
However, policymakers also realized that making health insurance available was not 
enough to ensure its uptake. Thus, the ACA-based Medicaid expansion was accompanied by the 
Individual Mandate, serving a crucial role by compelling individuals to acquire health insurance.3 
The Individual Mandate served to regulate health insurance markets by influencing individual 
economic behavior—i.e., individuals either purchase health insurance or pay a tax penalty 
(Rosenbaum and Gruber 2010). Further, the tax penalty was set to increase each year to further 
persuade uninsured individuals to purchase insurance. However, in 2011, the Individual Mandate, 
in conjunction with Medicaid expansion, created significant controversy, resulting in litigation that 
reached the United States Supreme Court in a landmark case known as the National Federation of 
Independent Businesses v. Sebelius (Negrin et al. 2012). In 2012, the Supreme Court ruled that the 
Individual Mandate was legal for all 50 states under Congress’s taxing and spending clause power; 
however, the justices argued that the significant expansion of Medicaid was not within Congress’s 
spending powers and was unconstitutionally coercive (Negrin et al. 2012). This ruling triggered a 
differential enforcement of the ACA, resulting in the current policy landscape, where some states 
have adopted Medicaid expansion (e.g., Arizona, California, Kentucky), while other states have 
opted out of Medicaid expansion (e.g., Florida, Georgia, North Carolina). Below, we provide a 
brief overview of changes to healthcare demand and supply in the United States following the 
ACA’s passage. 
  
 
3 The Individual Mandate generally requires that individuals purchase insurance if costs do not exceed 8% 
of household income. Select religious groups, American Indians, citizens not filing a tax return, incarcerated 
individuals, and those who are uninsured for a period of less than three months per calendar year can claim 
an exemption. 
41 
 
Figure 3: Timeline of Events Related to the Passage/Enforcement of the Affordable Care 
Act 
 
4.2 Changes to Healthcare Demand and Supply following the ACA 
Primary care serves as the foundation of the American healthcare system, with EDs playing a 
supporting role in outpatient care delivery. From 2010 to 2020, the ACA was expected to increase 
the use of primary care by an additional 15 million to 24 million individuals (Hofer 2011). This 
projected increase in primary care use was largely attributable to Medicaid expansion for two 
reasons: (i) uninsured individuals would gain financial access to care, and (ii) reimbursement for 
primary care physicians treating Medicaid patients would increase. Note that in the past, low 
reimbursement rates might have caused primary care physicians to turn away Medicaid patients 
seeking treatment; therefore, Medicaid expansion alone was not expected to increase primary care 
use, given that physicians might continue to turn away Medicaid recipients (Decker 2012). To 
address this issue, the ACA increased physician reimbursement rates for Medicaid patients, thus 
achieving payment parity with Medicare and private insurance (Klink 2015). This payment parity 
was expected to improve financial accessibility to primary care clinics for new Medicaid enrollees 
and create greater demand for outpatient care at these clinics.  
In addition, to address the expected increase in demand, the ACA implemented a series of 
infrastructural reforms meant to bolster the total supply of primary care physicians over time. 
42 
 
However, these provisions did not directly address the physical accessibility of primary care 
physicians, either in terms of physical distance or temporal availability. The absence of the ACA’s 
focus on remedying the inherent heterogeneity of physical access to outpatient care across various 
geographic regions was notable, given that in the decades preceding the ACA, a free market 
approach to managing healthcare in the US rendered primary care clinics in rural areas financially 
unsustainable (due to lower demand) and resulted in closures and consolidations of primary care 
clinics in these areas. Unfortunately, this consolidation accelerated during the 2010s, creating an 
increasingly inequitable geographic distribution of primary care clinics (Millman 1993, Fulton 
2017, Kane 2017). From 2010 to 2016, the percentage of US Metropolitan Statistical Areas with a 
Herfindahl-Hirschman Index (HHI) above 2,500 (representing highly concentrated markets) rose 
85.2% for primary care clinics, highlighting a significant increase in the inequitable geographic 
distribution of such clinics (Fulton 2017). Physician ownership of clinics also continued to fall 
steadily; in 1983, 75.8% of physicians owned their clinics, while in 2018, 45.9% of physicians 
owned their clinics (Kane 2017). While the ACA provided $11 billion in funding for Federally 
Qualified Health Centers (FQHCs), these centers represent only a small portion of all primary care 
clinics. In addition, these FQHCs continued to report funding issues and problems with workforce 
recruitment/retention, hampering their effectiveness in providing care (Rosenbaum et al. 2017). As 
such, there continues to be a shortage of primary care clinics across the United States, and the 
question of whether such care is accessible at the right place and at the right time remains 
understudied.  
4.3 Hypotheses 
We model the differential enforcement of the ACA, coupled with physical access and primary care 
utilization, to study its effects on ED discharges. The two ACA policy statutes—Medicaid 
expansion and the Individual Mandate—were concurrently deployed in 2014. We consider the full 
enforcement of the ACA to encompass both Medicaid expansion and the Individual Mandate, and 
partial enforcement to exclusively encompass the Individual Mandate.  
4.3.1 Where: Physical Access and Road Distance 
For individuals, insurance represents an important step toward improving financial access to care. 
Newly insured individuals may elect to receive first-contact care from either a primary care clinic 
or an ED, which can be considered substitutable, as the services rendered by either entity are 
delivered at a very low cost to the individual (Abraham 2014). Absent the cost constraint, 
individuals are more likely to place a greater emphasis on receiving care at the right place and at 
43 
 
the right time. In our study, the right place for nonurgent outpatient care is a primary care clinic, 
and the right time is when the patient requires care.  
Prior studies have examined individual usage decisions based on an individual’s distance 
to the nearest healthcare provider, in most cases the ED, through the lens of maximum coverage 
models (e.g., Ahmadi-Javid et al. 2017, Güneş et al. 2019). These models examine physical access 
from an individual’s perspective, and all else remaining fixed, point to the ED facility with 
minimum expected travel time or a minimum distance from the origin as the most convenient access 
option for an individual. However, these models do not examine the interplay between primary care 
clinics and EDs in delivering nonurgent outpatient care. The difference in the road distance an 
individual must travel to the nearest primary care clinic and to the nearest ED, or the distance 
difference, is important, as patients are likely to make decisions based on the nearest facility from 
which they can receive nonurgent services. Therefore, in our study, we examine how the distance 
difference might influence individual access decisions.  
For newly insured individuals, the cost prior to the full enforcement of the ACA would 
have consisted of the cost of treatment and the cost of travel to a provider (which is directly affected 
by physical distance to the healthcare provider). Because Medicaid eliminates a critical portion of 
total costs related to the provision of care, patients are likely to have greater financial leeway to 
spend on transportation in order to access the right care based on their circumstances. Additionally, 
newly insured Medicaid patients gain the ability to receive a referral from primary care providers 
for additional treatment needs, which may direct patients to another facility. Of note, it has been 
reported that new Medicaid insurees often have more complicated health conditions requiring more 
advanced treatment (Raven and Steiner 2018). In light of these complex conditions, primary care 
providers may refer individuals to specialists (and, in extreme cases, even EDs) for treatment that 
cannot be provided at their clinics (Finkelstein et al. 2016, Raven and Steiner 2018). In other words, 
following the full enforcement of the ACA, newly insured individuals are more likely to gain 
information from their primary care providers about where they should receive further treatment 
based on their condition, rather than mostly based on travel distance. As a result, these individuals 
are expected to make decisions to access care based not only on considerations surrounding 
transportation cost (and therefore travel distance), but, more importantly, on an expert’s opinion of 
their health condition. Building on the aforementioned arguments, and representing a positive value 
of distance difference as indicative of the nearest primary care clinic being farther away than the 
nearest ED, we predict that following the full enforcement of the ACA, distance difference will 
have a weaker positive impact on ED discharges as individuals place less emphasis on distance.  
 
44 
 
HYPOTHESIS 1. Distance difference will have a weaker positive impact on ED discharges 
following the full enforcement of the ACA. 
4.3.2 When: Physical Access and Hours of Operation 
Our study expands on the concept of physical access by incorporating the temporal 
characteristics of the healthcare supply chain. In comparison to EDs, which generally operate 24 
hours per day year-round, primary care clinics typically operate within a far narrower spectrum of 
hours, limiting when individuals can access primary care clinics. We argue that the temporal 
accessibility of the nearest primary care clinic is a salient factor for newly insured individuals, who 
are typically low-income individuals, and who are less likely to have the flexibility to reschedule 
their work time to meet their healthcare needs. These individuals may gravitate toward using the 
ED, even in non-emergency situations, if they cannot access a primary care clinic at the right time 
(Green 2012, Kangovi et al. 2013).  
Toward this end, recent studies have shown that Medicaid patients in expansion states have 
not only experienced increased wait times for primary care, but have also encountered increased 
difficulty in making appointments (Miller and Wherry 2017). For newly insured Medicaid patients, 
the convenience of “on-demand care” and the immediate temporal accessibility of EDs may 
outweigh the benefits of preventative care at a primary care clinic (Kangovi et al. 2013). Therefore, 
we examine the effect of nearest primary care clinics’ temporal accessibility on ED discharges, 
specifically, primary care clinics operating outside the standard working hours of 8:00 a.m. to 6:00 
p.m., i.e., between 12:00 a.m. and 8:00 a.m. or between 6:00 p.m. and 12:00 a.m. Monday through 
Friday, as well as weekend hours. All else remaining equal, we expect that following the full 
enforcement of the ACA, an increase in the temporal accessibility of primary care clinics through 
extended hours of operation beyond standard working hours—i.e., weekday out of hours and 
weekend hours—is likely to increase primary care use among newly insured patients, resulting in 
lower ED discharge levels.  
 
HYPOTHESIS 2a. An increase in weekday out-of-hours operation at the nearest primary 
care clinic will have a stronger negative impact on ED discharges following the full 
enforcement of the ACA. 
 
HYPOTHESIS 2b. An increase in weekend hours of operation at the nearest primary care 
clinic will have a stronger negative impact on ED discharges following the full enforcement 
of the ACA. 
 
45 
 
4.3.3 How Much: Use of Primary Care 
The ACA’s provision establishing insurance payment parity encouraged primary care 
physicians to take on new Medicaid patients (Polsky 2017). However, the historic consolidation of 
primary care capacity (as discussed in Section 2.2) has meant that patients have a limited selection 
of primary care clinics from which to receive nonurgent outpatient care. To this end, while the 
ACA by establishing insurance payment parity served to attenuate the provider shortage problem 
faced previously by Medicaid patients, the policy did not directly address the geographic mismatch 
between patient demand and the supply of new physicians (Edlin 2012). Taking the inherent 
capacity limitations of primary care clinics within a geographical region into consideration, we 
argue that as newly insured patients in a state under the full enforcement of the ACA seek outpatient 
care, congestion at primary care clinics is likely to increase the difficulty faced by patients in 
making an appointment, most notably among Medicaid recipients (Miller and Wherry 2017).  
 In this regard, patient panel size at the nearest primary care clinic is an important metric, 
because it is indicative of the number of unique patients a provider can see without compromising 
the timeliness or quality of care delivered to patients. We suggest that as patient panel size increases 
at the nearest primary care clinic, newly insured patients are more likely to access outpatient care 
at the clinic instead of seeking such care at the nearest ED. Moreover, we suggest that the methods 
of delivering Medicaid may also affect capacity utilization and congestion. In the case of Medicaid, 
the two predominant forms of delivery are Medicaid fee-for-service and Medicaid managed care. 
Under a fee-for-service delivery system, the state pays directly for any services provided to 
Medicaid enrollees. This delivery system allows individuals to receive treatment from any provider, 
assuming the provider accepts Medicaid. In addition, physicians earn more money as the volume 
of patient visits increases. In managed care delivery systems, a state contracts with a managed care 
organization (e.g., Kaiser Permanente or United Healthcare) at a set fee based on a per-person per-
month basis. These managed care organizations in turn select a group of providers to form a 
network that delivers care for their enrollees, and providers receive a flat fee from the managed 
care organizations for treating a set number of patients. While the primary reason for implementing 
managed care is cost containment, notably by preventing physicians from overtreating patients, the 
system has been criticized for reducing provider incentives to provide timely and necessary services, 
as their compensation is not directly tied to the amount of services rendered (Kongstvedt 2013).  
Therefore, with the advent of payment parity, we expect that physicians will be 
incentivized to treat Medicaid fee-for-service patients. Under a fee-for-service plan, physicians will 
receive payment for services rendered, thereby encouraging physicians to provide additional 
treatment to patients. By providing timely treatment to Medicaid fee-for-service patients when 
46 
 
required, providers at the nearest primary care clinic are expected to absorb the increased demand 
for nonurgent outpatient care, reducing ED discharges. On the other hand, under the requirements 
for treatment by a narrower, pre-defined network of providers, Medicaid managed care patients 
seeking nonurgent outpatient care are more likely to compete for provider capacity with other 
Medicaid patients seeking specialty care referrals at the nearest primary care clinic (Gold and 
Paradise 2012, Melnikow et al. 2020). Given the reimbursement scheme of managed care providers, 
we expect that the limited incentive to provide patients with timely treatment and the congestion at 
the primary care clinic, coupled with the increased demand following the full enforcement of the 
ACA, will push Medicaid managed care patients toward the nearest ED (Kangovi et al. 2013). In 
summary, in a state under the full enforcement of the ACA, as congestion increases at the nearest 
primary care clinic, patients seeking timely access to nonurgent outpatient care will venture to the 
nearest ED for treatment. Therefore, we posit the following: 
 
HYPOTHESIS 3a. In a state under the full enforcement of the ACA, increasing provider 
patient panel size at the nearest primary care clinic will decrease ED discharges.  
 
HYPOTHESIS 3b. In a state under the full enforcement of the ACA, increasing Medicaid 
fee-for-service encounters at the nearest primary care clinic will decrease ED discharges. 
 
HYPOTHESIS 3c. In a state under the full enforcement of the ACA, increasing Medicaid 
managed care encounters at the nearest primary care clinic will increase ED discharges. 
4.4 Data and Methods 
We test the study’s hypotheses relating to physical access using data from four states with 
differential enforcement of the ACA: California and Kentucky, which experienced the full 
enforcement of the ACA; and Florida and North Carolina, which experienced the partial 
enforcement of the ACA. Specifically, we collect data at the ZCTA level in California, Florida, 
Kentucky, and North Carolina from 2012-2016, collectively comprising 8,484 ZCTA-year 
observations, to examine how physical access affects ED discharges following the full enforcement 
of the ACA. Subsequently, to examine the effects of congestion on ED discharges at the nearest 
primary care clinic in a state under the full enforcement of the ACA, we analyze data at the ZCTA 
level in California, consisting of 3,995 ZCTA-year observations from 2012-2016. It is worth noting 
that prior studies on healthcare delivery or accessibility across geographic areas have frequently 
carried out inferences at a more aggregate hospital service area (HSA) level or the county level 
(e.g., Atasoy et al. 2018, Kim and KC 2020). However, inferences at the HSA or the county level 
may mask the heterogeneity in care delivery or access across different neighborhoods, or they may 
47 
 
overlook system-wide tradeoffs individuals may consider (e.g., temporal availability of clinics and 
travel distance to clinics) (Gentili et al. 2015, Gentili et al. 2018). Thus, our focus on the ZCTA 
level allows for greater geographic resolution, which in turn enables us to study the predictors of 
ED discharges with greater precision. 
The data on ED discharges in a given ZCTA year were obtained from California’s Office 
of Statewide Health and Planning (OSHPD),  Florida’s Agency for Health Care Administration 
(AHCA), and the Agency for Healthcare Research and Quality (AHRQ). While the OSHPD data 
identifies EDs using addresses, the AHCA and AHRQ data require additional information to 
identify EDs. Specifically, for the AHCA and AHRQ data, we utilize the National Plan and 
Provider Enumeration System (NPPES) to identify the facilities where patients are treated. Each 
facility carries a National Provider Identifier (NPI) code, which is used to identify the exact ED 
used by an individual in a given ZCTA year. The use of patient origin-ED discharge data is 
consistent with prior studies relating to emergency department accessibility (e.g., Hsia et al 2012, 
Brown et al 2015). We set 2012 as the starting year for data collection because the data provided 
by OSHPD began in that year. 
To construct our primary care clinic data, we utilize two approaches based on data 
availability. With regards to California and Florida, we collect data on the location of primary care 
clinics and their hours of operation from open data portals on the OSHPD and AHCA websites 
respectively. However, Kentucky and North Carolina do not openly provide longitudinal panel data 
on primary care clinic location and operational status. Therefore, to build a longitudinal panel of 
primary care clinics, we construct a Selenium-based web scraper using Python to collect the data 
from the public domain, specifically state licensing boards. All entries were validated against the 
NPPES to ensure the physicians at these facilities were licensed during the study period. The details 
regarding the construction of a longitudinal panel of primary care clinics for KY and NC using the 
Selenium-based web scraper are discussed in detail in Appendix 1. Subsequently, to capture 
measures relating to physical access, namely, road distance and hours of operation of primary care 
clinics, we use ArcGIS Pro 2.8—a leading provider of geographic information data—and Google 
Places Application Programming Interface (API). Finally, our extensive review of the data sources 
for collecting primary care clinic capacity (e.g., provider patient panel size, Medicaid fee-for-
service encounters) indicates that while California provides detailed publicly available data on 
primary care clinic capacity measures at the ZCTA level, such data remains largely inaccessible 
from other states. We therefore collect data from OSHPD to construct measures of primary care 
clinic capacity for California.  
We control for time-varying contextual factors (e.g., insurance penetration and coverage 
48 
 
type in a ZCTA, demography, etc.) that could influence such visits or affect patient access behavior. 
These variables, which are discussed in greater detail in Section 4.2, were collected from the 
following sources: the United States Census American Community Survey, the Department of 
Homeland Security’s Homeland Infrastructure Foundation-Level Data, and the Google Places API. 
In summary, our analyses are based on a unique and granular dataset aggregated from multiple 
public and proprietary sources to account for the heterogeneity in factors affecting ED discharges, 
while also estimating the effects of physical access and primary care utilization. 
4.4.1 Dependent and Independent Variables 
Dependent Variable: Our dependent variable of interest is ED Discharges, which represents the 
annual number of discharges at the nearest ED from a ZCTA centroid. ED visits can result in three 
outcomes: admission to the hospital, discharge, or death. By using discharges, we capture the 
number of individuals who visit the ED but are not admitted to the hospital for an inpatient stay 
(Kinderman et al. 2013). These discharges are also referred to as treat and release visits and account 
for nearly 90% of all ED visits (Center for Medicaid and Medicare Services 2021, Office of the 
Assistant Secretary for Planning and Evaluation 2021). A focus on ED discharges allows us to 
account for conditions that would most likely be treatable by a primary care provider. In addition, 
we note that ED discharges are a commonly used metric to understand the impact of health 
insurance, with regard to the ACA as well as other contexts (e.g., Card et al 2008, Anderson et al 
2010, Finkelstein et al 2012). 
Independent Variables: The independent variable of interest relating to the enforcement of 
the ACA is ACA Full Enforcement, which is coded as ‘1’ for all ZCTAs in California and Kentucky 
from 2014 onward, when the two states experienced the full enforcement of the ACA, and ‘0’ for 
all ZCTAs in these states prior to 2014. Given that Florida and North Carolina experienced only 
partial enforcement of the ACA, we code ACA Full Enforcement as ‘0’ for all ZCTAs in these two 
states. Thus, all ZCTAs in California and Kentucky from 2014 onward serve as the treatment group, 
whereas all ZCTAs in California and Kentucky prior to 2014 and all ZCTAs in North Carolina and 
Florida serve as the control group. 
The variable Distance Difference represents the difference in the road distance from a 
ZCTA centroid to the nearest primary care clinic, and from the ZCTA centroid to the nearest ED 
(i.e., DistancePrimary Care –  DistanceED). Thus, the greater the Distance Difference, the farther the 
location of the primary care clinic compared to the nearest ED from the ZCTA centroid. The 
latitude and longitude coordinates for a ZCTA’s centroid were acquired from the US Census 
49 
 
TIGER files to calculate the distance to the nearest primary care clinic and the nearest ED.4 While 
Euclidean distance has been used in prior studies to examine the accessibility of care facilities (e.g., 
Buchmueller et al. 2006, Berman et al. 2007), road distance represents a more realistic measure of  
physical accessibility as it accounts for topographical differences and geographical impediments 
(e.g., deserts, large bodies of water, mountain ranges, national parks) across ZCTAs, which 
Euclidean distance may overlook.  
The steps used to calculate the road distance measures are as follows. First, an origin-
destination distance matrix is constructed to find the road distance from a ZCTA centroid to the 
nearest primary care clinic and the nearest ED. We use the Network Analyst package in ArcGIS 
Pro 2.8 to calculate the shortest road distance from an origin point to a destination point while 
traveling across the existing American road network. We digitally construct the road network using 
the ESRI North America StreetMap package, which is derived from the TIGER/Line files provided 
by the United States Census. Upon completion of the origin-destination distance matrix, we sort 
the road distances to find the nearest primary care clinic (DistancePrimary Care) and the nearest ED 
(DistanceED) in relation to a ZCTA centroid. Figure 4 provides a visual representation of these 
distances. Finally, we calculate the difference in these two distances to obtain the measure for 
Distance Difference.  
Figure 4: Distances to the Nearest Primary Care Clinic and Nearest ED from a ZCTA 
Centroid 
 
Note: We use the above ZCTA located in Fresno county, California as an example to provide a visual representation 
of the distances from the ZCTA centroid. 
 
4 In our analysis, we drop ZCTAs neighboring other states to account for individuals crossing over into 
neighboring states to receive outpatient care. For example, a person living in Tahoe, California may cross 
over state lines into Reno, Nevada to receive care that it is closer than the nearest facility within the state of 
California. 
50 
 
To assess the temporal aspect of physical access, we construct two variables related to 
hours of operation for the nearest primary care clinic, (i) Average Weekday Out of Hours, i.e., 
average hours of operation outside of 8:00 a.m. and 6:00 p.m. from Monday to Friday, and (ii) 
Average Weekend Hours, i.e., average hours of operation on Saturday and Sunday. Our approach 
utilizes Google Places API paired with a Python script to collect the information needed to calculate 
the nearest primary care clinic hours of operation and export the results. Specifically, the Google 
Places API query uses the geo-coordinates, names, and street addresses of clinics in our sample to 
obtain their hours of operation for each day of the week, which are then used to calculate the 
average weekday out-of-hours operation and the average weekend hours of operation. The minutes 
in the hours of operation are converted into decimal values. For example, if a facility operated for 
8 hours and 30 minutes, our data would reflect the hours of operation as 8.5 hours.  
Note that for a limited number of clinics for which the Google Places API query could not 
determine hours of operation, we instead utilize Yelp and the Yellow Pages. Additionally, for 
clinics for which hours of operation were unavailable through either Google Places API, Yelp, or 
the Yellow Pages, we impute the average weekday out-of-hours operation and average weekend 
hours of operation for all clinics in the same primary care service area (PCSA)—i.e., a geographic 
approximation of markets for primary care service (Goodman et al. 2003, KC and Terwiesch 
2011)—as the nearest primary care clinic.  
Finally, to study the effects of primary care utilization on ED discharges, we include 
independent variables related to the nearest primary care clinic’s capacity. The variable Primary 
Care Provider Patient Panel refers to the average number of unique patients (in 1000s) that were 
seen by a primary care provider in a given primary care clinic in a given year. This variable is 
constructed by dividing the number of unique patients who visited a primary care clinic in a given 
year by the number of primary care provider full-time equivalents (FTEs) for that clinic. In relation 
to Medicaid encounters at the nearest primary care clinic, we differentiate between the two unique 
Medicaid delivery systems: Medicaid Fee-for-Service encounters (in 1000s of unique patients) and 
Medicaid Managed Care encounters (in 1000s of unique patients). As previously mentioned, the 
fee-for-service system allows individuals to receive care from any provider that accepts Medicaid, 
and the state pays the provider directly for a covered service. In contrast, the managed care system 
requires individuals to enroll in a plan that allows them to access a predefined network of providers 
who accept Medicaid and are associated with the plan. The state pays a fixed fee to the plan for 
each enrolled individual, and the plan subsequently pays providers for covered services.  
4.4.2 Control Variables 
Physical Access Control Variables: Given that the challenges associated with physical access to 
51 
 
primary care clinics and EDs may vary across ZCTAs, we include a number of time-varying control 
variables to account for differences in physical access across ZCTAs: Base Distance, Relative 
Distance, and PCSA Clinic Density, Urgent Care Clinic Density, and HSA Clinic Density.  
 Specifically, the variable Base Distance refers to the minimum distance an individual must 
travel from a ZCTA centroid to receive outpatient care from either an ED or a primary care clinic, 
i.e., Min (DistancePrimary Care, DistanceED).  Relative Distance refers to the road distance 
between the nearest primary care clinic and the nearest ED. The variable PCSA Clinic Density 
refers to the number of primary care clinics per square mile within a PCSA. Additionally, given 
that individual usage decisions may be affected by the level of access to urgent care clinics, we 
include the variable Urgent Care Clinic Density, which represents the number of urgent care clinics 
per square mile within a ZCTA. Finally, the variable HSA Density refers to the number of hospitals 
with emergency departments within a Hospital Service Area.5 
Insurance Control Variables: To account for the type of insurance individuals use, we 
include a number of time-varying control variables for the following payer groups as a percentage 
of the population within a ZCTA: uninsured/self-pay (Uninsured %), direct purchase insurance 
(Direct Purchase Insurance %), employer-based insurance (Employer Based Insurance %), 
Medicaid (Medicaid %), Medicare (Medicare %), TRICARE (TRICARE %), and Veterans Affairs 
insurance (VA Insurance %). In addition, we include a variable, Medicaid Expansion Status, to 
account for the gradual county-level expansion of Medicaid leading to statewide expansion in 
California by 2014. Consistent with the work of Golberstein et al. (2015), this variable is coded as 
‘0’ for no changes in eligibility requirements and ‘1’ for expansion of eligibility requirements. The 
inclusion of this variable accounts for instances when a county in which a ZCTA is located expands 
Medicaid eligibility. 
 Demographic Control Variables: The demographic control variables account for time-
varying characteristics of the general population in a ZCTA. The population characteristics include 
the following: the natural log of median annual household income (ln Median Annual Household 
Income), racial composition in a given year (Hispanic %, Black %, Native %, Asian %, Other %), 
level of annual inward migration from other states (Inward Migration [Other States]), number of 
individuals annually below 138% of the Federal Poverty Line [Below FPL138 %], sex ratio in a 
given year (Sex Ratio [Males to Females]), percentage of population by age group in a given year 
(Age: Under 25 Years %, Age: 25-34 Years %, Age: 35-44 Years %, Age: 45-64 Years %, Age: 
 
5 In our study sample, nearly 85% of the ZCTAs had a clear choice of a primary care clinic located within its 
corresponding PCSA and nearly 92% of ZCTAs had a clear choice of ED located within its corresponding 
HSA. 
52 
 
Over 65 Years %), percentage of the population in a given year that is married (Married  %), 
percentage of the population in a given year that is either separated or divorced 
(Separated/Divorced %), percentage of the population in a given year that is widowed 
(Widowed %), percentage of the population in a given year with US citizenship (Citizens %), 
average family size in a given year (Average Family Size), educational level of the population age 
25 and over (Education [25 and Over]: High School,  Education [25 and Over]: Some College, 
Education [25 and Over]: College), percentage of the population with full-time, full-year 
employment in a given year (Full Time Full Year %), percentage of the population with full-time, 
partial-year employment in a given year (Full Time Part Year %), and percentage of the population 
with part-time employment in a given year (Part Time %).  
 Primary Care Clinic Control Variables: To account for staffing variations in the nearest 
primary care clinic across ZCTAs, we control for the full-time equivalents related to primary care 
physicians within a clinic using the variable Physician FTEs. In certain cases, nurse practitioners 
may also be able to provide primary care services. Therefore, we include another variable, Other 
FTEs, to account for these providers.  
Appendix B2 provides a listing of all of the dependent, independent, and control variables, 
including their variable descriptions and the data sources. 
4.4.3 Identification Strategy and Model Specification 
The differential enforcement of the ACA across the treatment group states (California and 
Kentucky) and the control group states (Florida and North Carolina) serves as a natural experiment, 
allowing us to use a difference-in-differences (DiD) specification at the ZCTA level to estimate the 
heterogeneous impact of the ACA with regard to physical access on ED discharges. In this 
specification, as shown in equation (1), the coefficients of interest are the interaction of ACA Full 
Enforcement with the physical access measures, Distance Difference, Average Weekday Out of 
Hours, and Average Weekend Hours, for a ZCTA in a given year. The vector of time-varying 
control variables is denoted by 𝑿, 𝜆 represents the ZCTA fixed effect, θ represents the state fixed 
effect, and 𝜇 represents year fixed effect. Finally, we also account for state-time trends by including 
the interaction between state and time fixed effects, θ × μ.  
 
 ED Discharges 
= α + β1Distance Difference + β2Average Weekday Out − of − Hours +
 + β3Average Weekend Hours + β4ACA Full Enforcement +
 + β5ACA Full Enforcement ×
 Distance Difference  + β6ACA Full Enforcement × Average Weekday Out −
of − Hours + β7ACA Full Enforcement ×  Average Weekend Hours +  𝛄𝐗 + λ +
θ + μ + θ × μ + ε                           (1) 
53 
 
 
To reduce concerns that selection bias may affect our results, we use the Coarsened Exact 
Matching (CEM) procedure to generate a matched sample of treated and control ZCTA-year 
observations that do not exhibit systematic or significant differences in the pre-intervention period. 
Specifically, CEM allows for multi-dimensional exact matching consistent with a two-stage 
matching technique, which controls for selection on observable differences by eliminating non-
analogous observations in the treatment and control populations (Iacus et al. 2012). We use the 
following covariates at the ZCTA-level in the pre-intervention period: population density, median 
household income, median age, percentage of individuals between 100% and 149% of the Federal 
Poverty Line, and percentage of individuals receiving Supplemental Nutrition Assistance Program 
(SNAP) benefits. Because our study seeks to understand the impact of physical access, we use 
population density as a proxy for physical access; furthermore, the use of this covariate allows us 
to account for the urban/rural divide among the ZCTAs in our sample. We match on socioeconomic 
characteristics reflecting eligibility (median household income, median age, and percentage of 
individuals between 100% and 149% of the Federal Poverty Line) to ensure that ZCTAs contain 
similar Medicaid expansion-eligible populations. Finally, SNAP benefits signal the proportion of 
the population that is likely to be eligible for Medicaid under the full enforcement of the ACA 
(Dorn et al. 2013). The inclusion of this covariate also serves as a proxy for the manner in which 
individuals uptake government services among the eligible population. In addition, our matching 
specification accounts for three years for each variable prior to the Individual Mandate: 2011, 2012, 
and 2013 (e.g., Median Household Income in 2011, in 2012, and in 2013). Appendix 3 provides a 
description of the matching variables. 
Next, we examine how the capacity of the nearest primary care clinic affects ED discharges 
following the full enforcement of the ACA. As noted in Section 4, among the treatment group states 
(California and Kentucky) and unlike other states, California provides detailed publicly available 
data on primary care clinic capacity at the ZCTA level. We therefore use the California-only 
subsample of matched treatment and control ZCTA-year observations to address this question. In 
the model specification presented in equation (2), the coefficients of interest are the interaction of 
ACA Full Enforcement with Primary Care Provider Patient Panel and Medicaid patient encounters, 
Medicaid Fee-for-Service and Medicaid Managed Care for a ZCTA, in a given year. The vector of 
time-varying control variables in equation (2) is denoted by 𝑿, 𝜆 represents the ZCTA fixed effect, 
and 𝜇 represents year fixed effect.  
 
𝐸D Discharges 
54 
 
= α + β1Distance Difference + β2Average Weekday Out − of − Hours  +
+ β3Average Weekend Hours + β4ACA Full Enforcement +
+ β5Primary Care Provider Patient Panel + β6Medicaid Fee-for-Service +
+ β7Medicaid Managed Care +  β8ACA Full Enforcement × Distance Difference +
+ β9ACA Full Enforcement × Average Weekday Out − of − Hours +
+ β10ACA Full Enforcement ×
Average Weekend Hours +  β11ACA Full Enforcement ×
 Primary Care Provider Patient Panel + β12ACA Full Enforcement ×
Medicaid Fee-for-Service + β13ACA Full Enforcement × Medicaid Managed Care +
 𝛄𝐗 + λ + μ + ε                (2) 
4.5 Results 
4.5.1 Effects of Physical Access on ED Discharges 
A key assumption underlying the effects of physical access on ED discharges in relation to 
the differential enforcement of the ACA is that while the individual mandate went into effect across 
all states, only the treatment group states that also experienced Medicaid expansion were likely to 
see a surge in Medicaid enrollees compared to the control group states, which did not experience 
Medicaid expansion. In other words, while the Individual Mandate was expected to push some part 
of the population in the control group states to acquire Medicaid insurance or private insurance in 
lieu of a tax penalty, it was unlikely to motivate a larger percentage of the uninsured population to 
do so.  To examine this assumption, we provide graphical evidence in Figure 5 illustrating the 
change in Medicaid enrollees across the four states in our sample during the 2012-2016 time period. 
As expected, while we see a sharp rise in the percentage of Medicaid enrollees in California and 
Kentucky from 2014 onwards, the trend is relatively flat for Florida and North Carolina. 
Figure 5: Trends in Medicaid Enrollees Across the Four States 
 
  
55 
 
Table 4. Effects of Physical Access on ED Discharges 
 Dependent Variable: ED Discharges   
 
Negative Binomial  
Regression 
OLS 
Regression 
 (1) (2) (3) (4) (5) 
Main Effects      
Distance Difference 
0.00681 
(0.002)*** 
0.00745 
(0.002)*** 
0.00677 
(0.002)*** 
0.00740 
(0.002)*** 
0.00708 
(0.002)*** 
Average Weekday Out of Hours 
0.00282 
(0.002) 
0.00286 
(0.002) 
0.00270 
(0.002) 
0.00273 
(0.002) 
0.00231 
(0.002) 
Average Weekend Hours 
-0.00113 
(0.001) 
-0.00110 
(0.001) 
-0.00072 
(0.001) 
-0.00072 
(0.001) 
-0.00069 
(0.001) 
ACA Full Enforcement [ACA Full] 
0.05696 
(0.028)** 
0.05405 
(0.028)* 
0.05944 
(0.028)** 
0.05645 
(0.028)** 
0.06017 
(0.033)* 
      
Interactions      
ACA Full X Distance Difference   
-0.00219 
(0.001)*** 
  
-0.00212 
(0.001)*** 
-0.00192 
(0.001)** 
ACA Full X Average Weekday Out of 
Hours 
    
0.00260 
(0.003) 
0.00249 
(0.003) 
0.00303 
(0.003) 
ACA Full X Average Weekend Hours     
-0.00233 
(0.001)** 
-0.00218 
(0.001)** 
-0.00229 
(0.001)** 
      
Constant 
9.76756 
(0.597)*** 
9.70872 
(0.597)*** 
9.80270 
(0.595)*** 
9.74362 
(0.595)*** 
10.06241 
(0.631)*** 
Physical Access Controls Yes Yes Yes Yes Yes 
Insurance Controls  Yes Yes Yes Yes Yes 
Demographic Controls Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes Yes 
State FE Yes Yes Yes Yes Yes 
State X Year FE Yes Yes Yes Yes Yes 
Number of Observations 6725 6725 6725 6725 6725 
* p<0.10, ** p<0.05, *** p<0.01. Heteroskedastic-robust standard errors clustered at the ZCTA level are included in 
parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations obtained using the 
Coarsened Exact Matching (CEM) procedure. 
Having examined this assumption, we report the estimation results for the DiD 
specification highlighted in equation (1), which tests Hypotheses 1 and 2. Appendix B4 provides 
the summary statistics for the variables used in this specification (based on the matched sample), 
whereas Table 1 presents the estimation results. Given that our dependent variable, ED Discharges, 
is a count variable that exhibits overdispersion—that is, its variance is significantly greater than its 
mean—we use an unconditional negative binomial regression in estimating the DiD specification. 
Column (1) in Table 4 provides a baseline main effects model that includes the complete set of 
controls, the main effects of physical access, Distance Difference, Average Weekday Out of Hours, 
and Average Weekend Hours, and the policy measure, ACA Full Enforcement. Columns (2) and (3) 
hierarchically build upon the main effects model by including the interaction terms of the physical 
access measures with the policy measure, whereas column (4) represents the full model, which 
56 
 
includes all of the interaction terms. We use the results from column (4) to interpret our hypotheses 
tests. In addition, for comparison, we also estimate the DiD specification in column (5) using OLS 
regression with a log transformed measure of the dependent variable. 
Focusing on column (4), we find that the coefficient for ACA Full Enforcement X Distance 
Difference is negative and significant (β = -0.00212, p<0.01), suggesting that the effect of distance 
difference on ED discharges is attenuated following the full enforcement of the ACA. H1 is 
therefore supported. To better understand this result, we examine the average marginal effects in 
Table 5. We find that prior to the ACA’s full enforcement, a ZCTA could expect to see an increase 
of about 71 ED discharges annually on average for every mile increase in distance difference. 
However, following the ACA’s full enforcement, a one-mile increase in distance difference 
corresponds to an increase of approximately 54 ED discharges annually on average, highlighting a 
reduction by about 17 ED discharges annually on average.  
Table 5: Marginal Effects Based on Table 4, Column 4 Results 
 Distance Difference Average Weekend Hours 
Pre 71.31081 (23.487)*** -6.94853 (10.500) 
Post 54.00111 (26.038)** -29.70897 (13.422)** 
Contrast [Post – Pre] -17.30970 (8.045)** -22.76043 (10.946)** 
* p<0.10, ** p<0.05, *** p<0.01 
Next, we consider the impact of the nearest primary care clinic’s hours of operation on ED 
discharges. As seen in column (4), we find that the coefficient of ACA Full Enforcement X Average 
Weekday Out of Hours is statistically insignificant (β = 0.00249, p>0.10). Therefore, we do not find 
support for H2a. Interestingly, this result appears to be at odds with the results from prior research 
(e.g., O’Malley 2012, Berchet and Nader 2016). To better understand this result, we examine the 
kernel density plots for the Average Weekday Out-of-Hours variable across the treatment and 
control groups in Figure 6 (for reference, we also include the kernel density plots for the Average 
Weekend Hours variable across the treatment and control groups). We find that the distribution of 
Average Weekday Out-of-Hours variable for the treatment group primary care clinics is extremely 
right-skewed compared to the control group primary care clinics, highlighting the presence of very 
limited out-of-hours operation for the treatment group primary care clinics. This plot suggests that 
the absence of a statistically significant interaction between ACA Full Enforcement X Average 
Weekday Out of Hours may be attributable to the limited number of primary care clinics in 
California and Kentucky providing weekday out-of-hours operation compared to those from 
Florida and North Carolina.   
With regard to our results pertaining to weekend hours of operation at the nearest primary 
care clinic, we find that the coefficient of ACA Full Enforcement X Average Weekend Hours is 
57 
 
negative and statistically significant (β = -0.00218, p<0.05). This result suggests that an increase in 
weekend hours of operation at the nearest primary care clinic has a stronger negative effect on ED 
discharges following the full enforcement of the ACA. H2b is therefore supported. An examination 
of the average marginal effects, as shown in Table 5, indicates that prior to the full enforcement of 
the ACA, every additional hour of operation at the nearest primary care clinics during the weekend 
would result in about seven fewer ED discharges annually on average. However, following the full 
enforcement of the ACA, this effect corresponds to about 29 fewer ED discharges annually on 
average for every additional hour of operation at the nearest primary care clinic during the weekend, 
thereby highlighting a reduction of about 22 ED discharges annually on average.  
Figure 6: Hours of Operation Kernel Density Plots 
   
Overall, the results suggest that, although physical access to the nearest primary care clinic 
in relation to the nearest ED is an important factor that affects ED discharges, the effect of distance 
difference and weekend hours of operation on ED discharges is attenuated following the ACA’s 
full enforcement.  
4.5.2 Effects of Nearest Primary Care Clinic Capacity on ED Discharges  
Prior to estimating the effects of the nearest primary care clinic capacity on ED discharges in 
California, we first descriptively examine the trends in primary care usage in the state during the 
study time period. Figure 4 examines trends in unique primary care clinic encounters across 
Medicaid enrollees and private insurance enrollees. As seen from this figure, we observe a 
substantial increase in the number of unique primary care clinic encounters for managed care 
enrollees, while the trends for fee-for-service and privately insured enrollees remain muted.  
 Given these trends, we now present the estimation results for the specification highlighted 
in equation (2). Appendix B5 provides the summary statistics for the variables used in this 
specification based on the California subsample. In Table 6, we present the estimation results. 
58 
 
Column (1) represents the baseline model, which includes all of the control variables as well as the 
independent variables relating to physical access. Columns (2) and (3) build upon the baseline 
model to assess the impact of the panel size for primary care provider and Medicaid patient 
encounters by delivery system at the nearest primary care clinic following the full enforcement of 
the ACA. Finally, column (4) represents the full model, including all of the interaction terms. We 
use the full model results for interpretation.  
Figure 7: Primary Care Clinic Unique Encounters by Payer in California 
 
  
59 
 
Table 6: Effects of Nearest Primary Care Clinic Capacity on ED Discharges 
 Dependent Variable: ED Discharges  
 
Negative Binomial  
Regression 
OLS 
Regression 
 (1) (2) (3) (4) (5) 
Main Effects      
Distance Difference 
0.00910 
(0.003)*** 
0.00890 
(0.003)*** 
0.00858 
(0.003)*** 
0.00855 
(0.003)*** 
0.00782 
(0.003)*** 
Average Weekday Out of Hours 
0.00459  
(0.004) 
0.00394 
(0.004) 
0.00487  
(0.004) 
0.00429  
(0.004) 
0.00464  
(0.004) 
Average Weekend Hours 
0.00037  
(0.002) 
0.00043 
(0.002) 
0.00063  
(0.002) 
0.00071  
(0.002) 
0.00012  
(0.002) 
ACA Full Enforcement 
0.13431 
(0.010)*** 
0.13666 
(0.010)*** 
0.13210 
(0.010)*** 
0.13440 
(0.010)*** 
0.13786 
(0.010)*** 
      
Primary Care Clinic Capacity      
Primary Care Provider Patient Panel   
0.00136 
(0.001)* 
  
0.00102 
(0.001) 
0.00107  
(0.001) 
Medicaid Fee for Service    
0.00065  
(0.001) 
0.00072  
(0.001) 
0.00088  
(0.001) 
Medicaid Managed Care    
-0.00084  
(0.000)* 
-0.00080  
(0.000) 
-0.00083  
(0.001) 
      
Interactions      
ACA Full X Distance Difference 
-0.00213 
(0.001)*** 
-0.00205 
(0.001)*** 
-0.00210 
(0.001)*** 
-0.00205 
(0.001)*** 
-0.00172 
(0.001)** 
ACA Full X Average Weekday Out of 
Hours 
-0.00102  
(0.004) 
-0.00074 
(0.004) 
-0.00115  
(0.004) 
-0.00090 
(0.004) 
-0.00042  
(0.004) 
ACA Full X Average Weekend Hours 
-0.00189  
(0.001)* 
-0.00186 
(0.001)* 
-0.00214 
(0.001)* 
-0.00208 
(0.001)* 
-0.00206 
(0.001)* 
      
ACA Full X Primary Care Provider 
Patient Panel 
  
-0.00143 
(0.001)** 
  
-0.00125 
(0.001)** 
-0.00137 
(0.001)** 
ACA Full X Medicaid Fee for Service    
-0.00099  
(0.001) 
-0.00105 
(0.001)* 
-0.00114 
(0.001)* 
ACA Full X Medicaid Managed Care    
0.00157 
(0.000)*** 
0.00150 
(0.000)*** 
0.00155 
(0.000)*** 
      
Constant 
9.86108 
(0.578)*** 
9.84529  
(0.576)*** 
9.87997 
(0.574)*** 
9.87247 
(0.573)*** 
8.75834 
(0.617)*** 
Primary Care Clinic Controls  Yes Yes Yes Yes Yes 
Physical Access Controls Yes Yes Yes Yes Yes 
Insurance Controls Yes Yes Yes Yes Yes 
Demographic Controls Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes Yes 
Number of Observations 3995 3995 3995 3995 3995 
*p<0.10, ** p<0.05, *** p<0.01. Heteroskedastic-robust standard errors clustered at the ZCTA level are included in 
parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations (from California) 
obtained using the Coarsened Exact Matching (CEM) procedure. 
Focusing on column (4), we find that the coefficient of ACA Full Enforcement X Primary 
Care Provider Patient Panel is negative and statistically significant (β = -0.00125, p<0.05). An 
examination of the average marginal effects in Table 7 suggests that following the full enforcement 
of the ACA, for a given ZCTA, an increase in the primary care provider panel by 1000 unique 
60 
 
patients at the nearest clinic results in two less ED discharges annually. However, prior to the full 
enforcement of the ACA, an increase in the primary care provider panel by 1000 unique patients 
would result in about nine additional ED discharges annually. This result illustrates the potential 
implications of payment parity in the form of reimbursement changes, which incentivize a primary 
care provider to treat more Medicaid patients, in turn alleviating outpatient demand at the nearest 
ED for each respective ZCTA. H3a is therefore supported. 
With regard to the effects of Medicaid patient encounters by delivery system, we find that 
the coefficient of ACA Full Enforcement X Medicaid Fee for Service is negative and statistically 
significant (β = -0.00105, p<0.10), whereas the coefficient of ACA Full Enforcement X Medicaid 
Managed Care is positive and statistically significant (β = 0.00150, p<0.01). Note that the fee-for-
service delivery system allows access to a greater number of primary care providers as long as said 
providers accept Medicaid compared to the managed care delivery system, which imposes 
structural constraints on patients in terms of where care can be received. To that end, our results 
suggest that following the enforcement of the ACA, an increase in the number of patient encounters 
through the Medicaid fee-for-service delivery system helps to alleviate the use of the nearest ED, 
whereas more restricted Medicaid service in the form of managed care networks tends to increase 
the use of the nearest ED. An examination of the average marginal effects in Table 7 reveals that 
following the enforcement of the ACA, for every 1,000 Medicaid managed care encounters, there 
will be an increase of approximately 15 ED discharges annually on average (compared to the pre-
intervention period). In contrast, for every 1,000 Medicaid fee-for-service encounters following the 
enforcement of the ACA, there will be approximately 10 fewer ED discharges annually on average 
(compared to the pre-intervention period). Hypotheses 3b and 3c are therefore supported.  
Table 7: Marginal Effects Based On Table 7, Column 4 Results 
 
Primary Care Provider 
Patient Panel 
Medicaid  
Fee for Service 
Medicaid  
Managed Care 
Pre 9.28949 (6.749) 6.60851 (5.876) -7.30487 (4.536) 
Post -2.48314 (6.657) -3.70020 (3.157) 7.47486 (3.536)** 
Contrast -11.77263 (5.637)** -10.30871 (5.721)* 14.77973 (4.299)*** 
Observations 3995 3995 3995 
* p<0.10, ** p<0.05, *** p<0.01 
These results, when taken together, suggest that the ACA’s financial access policies bolster 
primary care providers’ willingness to take on Medicaid patients. However, the delivery system 
can create structural constraints on patients’ access to primary care clinics, likely determining 
whether patients seek nonurgent outpatient care from such clinics or EDs.  
4.6 Robustness Checks and Alternative Explanations 
In this section, we document the results of additional analyses carried out to test the 
61 
 
robustness of the observed effects and investigate the possibility of alternative explanations.  
4.6.1 Examination of the Parallel Trends Assumption 
The baseline identifying assumption for the model specification examining the effects of 
the ACA with regard to physical access on ED discharges is the “parallel trends” assumption, which 
requires that in the absence of the intervention, the differences in the outcomes across the treatment 
and the control groups should be constant over time (Bertrand et al. 2004). To this end, the use of 
a DiD specification with matching, as we have done in this study, presents two advantages: (i) it 
lowers the mean squared error of the DID estimator and increases the likelihood that the true 
intervention effect is contained through the confidence interval of the estimator, and (ii) it renders 
the DID estimator less sensitive to deviations from the parallel trends assumption. Nonetheless, we 
follow prior research (e.g., Hydari et al. 2019, Song et al. 2018) in carrying out a visual examination 
of the trends in the pre-intervention period, as well as a leads-and-lags analysis to evaluate the 
parallel trends assumption. Figure 8 plots the trends in ED discharges across the treatment and 
control groups before adjusting for covariates or fixed effects. We see that prior to the full 
enforcement of the ACA, the unadjusted trends in ED discharges across the treatment and control 
groups appear to be parallel. 
Figure 8: Unadjusted Trends in ED Discharges Across Treatment and Control Groups 
 
  
62 
 
Table 8: Analysis of Leads and Lags to Adoption 
 Dependent Variable: ED Discharges 
 Negative Binomial Regression 
 (1) (2) 
Pre ACA Full Enforcement -0.01330 (0.019)   
2 Years Before   -0.01273 (0.019) 
1 Year Before   -0.02419 (0.021) 
   
ACA Full Enforcement   
   
Post ACA Full Enforcement 0.04212 (0.019)**   
1 Year After   0.03794 (0.009)*** 
2 Years After   0.04211 (0.019)** 
   
Constant 9.80386 (0.599)*** 9.80100 (0.599)*** 
Physical Access Controls  Yes Yes 
Insurance Controls Yes Yes 
Demographic Controls Yes Yes 
Year FE Yes Yes 
ZCTA FE Yes Yes 
State FE Yes Yes 
State X Year FE Yes Yes 
Number of Observations 6725 6725 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard errors clustered at 
the ZCTA level in parentheses.  For estimation, we use a matched sample of treated and control ZCTA-year 
observations obtained using the Coarsened Exact Matching (CEM) procedure. 
 
To formally test whether these trends differed prior to the full enforcement of the ACA, we 
estimate an econometric model that considers the year of intervention (i.e., 2014) as the base 
category and includes several dummy variables (i.e., leads and lags) with respect to the year of 
intervention as well as the control variables. Such a model enables us to compare and test for 
differences in the slope of ED discharges across the treatment and control groups before the 
intervention. Table 8 documents the analyses results. In column (1), we examine the pre-treatment 
and post-treatment periods. The coefficient Pre ACA Full Enforcement is insignificant, suggesting 
the absence of pre-trends on ED discharges. In column (2), we deconstruct the periods by the 
respective years. Consistent with our results in column (1), we do not find any significant effect 
prior to the full enforcement of the ACA in column (2). Collectively, the results from the visual 
examination of the trends and the leads and lags analyses suggest that the identifying assumption 
of parallel trends is less likely to be violated in our sample.  
Beyond the parallel trend assumption, it is plausible that our results may be driven by 
systematic trends in physical access measures across the treatment and control group states across 
years during the study period. To examine this possibility, we undertake a pairwise comparison of 
the means of the physical access measures using 95% Bonferroni-adjusted confidence intervals. In 
the results from this analysis, the 95% Bonferroni-adjusted confidence intervals for each pairwise 
comparison for the physical access variables across the treatment and control group states do not 
63 
 
indicate any systematic time-varying differences for the period of the study sample. Collectively, 
these results suggest that the physical access measures did not systematically change over time 
during the period of the study, suggesting that endogeneity of physical access is less likely to be a 
concern in our analysis. 
4.6.2 Controlling for Neighborhood Environment Effects  
Although we control for the respective ZCTA fixed effect, it is possible that the time-
varying characteristics of the neighborhood environment associated with the ZCTA may affect 
health outcomes in the community and influence ED discharges. Therefore, we re-estimate our 
main models by including the following variables at the ZCTA-year level as controls for 
neighborhood environment in our analysis: AQI Good %, Community Health Educators, Work 
Travel Time [Minutes], and Violent Crime Rate Per 1,000 People. The variable AQI Good % 
represents the number of good air quality days over the total days of measurement as recorded by 
the US Environmental Protection Agency. We note the air quality could potentially exacerbate 
respiratory problems in certain individuals, thereby affecting ED discharges. The variable 
Community Health Educators, as reported by the United States Bureau of Labor Statistics, accounts 
for the number of healthcare workers who provide informal counseling to people about health and 
wellness, including helping people understand insurance options and coverage details. The variable 
Work Travel Time [Minutes], as reported by the United States Census American Community 
Survey, accounts for the amount of time it takes individuals to travel to work. Finally, using data 
from the FBI Uniform Crime report, we include the variable Violent Crime Rate Per 1,000 People 
to account for the impact of violent crime on ED discharges. Prior research has noted that about 
one in nine individuals sent to the ED for injuries caused by intentional acts of violence will end 
up being violently injured again within a two-year window (Kaufman et al. 2016).  
 The results from the estimation of our main models, including these neighborhood 
environment control variables, are presented in the Appendix. Specifically, as seen from column 
(4) in Appendix B6, the coefficients of ACA Full Enforcement X Distance Difference and ACA Full 
Enforcement X Average Weekend Hours are qualitatively consistent with the main model results in 
terms of their statistical significance, direction, and magnitude, as presented in Appendix B6. 
Furthermore, we note that the coefficient of ACA Full Enforcement X Average Weekday Out of 
Hours continues to remain statistically insignificant. Similarly, as seen from column (4) in 
Appendix B7, the coefficients of ACA Full Enforcement X Primary Care Provider Patient Panel, 
ACA Full Enforcement X Medicaid Fee for Service, and ACA Full Enforcement X Medicaid 
Managed Care continue to remain qualitatively consistent with the main model results presented 
64 
 
in Table 4.  
4.6.3 Does the Provision of Health Insurance Create a Potential for Moral Hazard? 
Prior research has suggested that the provision of health insurance may create a potential 
for moral hazard, wherein individuals exhibit increasingly risky behavior knowing that they no 
longer shoulder the large cost burden related to care (Einav and Finklestein 2018, Lee 2018). In 
other words, the dependent variable measure of ED discharges may be capturing outcomes of this 
risky behavior on the part of individuals rather than capturing nonurgent outpatient demand. To 
examine whether this potential for moral hazard may be affecting our results, we construct an 
alternative dependent variable, Unintentional Injuries Mortality, which represents the percentage 
of mortality as a share of the total mortality that occurs due to unintentional injuries. According to 
the US Centers for Diseases Control and Prevention, an unintentional injury is defined as “physical 
damage that results when a human body is suddenly subjected to energy in amounts that exceed 
the threshold of physiologic tolerance-or else the result of a lack of one or more vital elements, 
such as oxygen” (US CDC 2012, p.15). Common types of unintentional injuries include motor 
vehicle accidents, suffocation, drowning, poisoning, fire/burns, falls, and sports and recreation 
injuries.   Appendix B8, column (1) presents the results of this analysis. We note that 
the availability of mortality data at the ZCTA level is limited to two states within our sample, 
California and Florida. Therefore, we report results for these two states in Appendix 8. We find 
that the interactions of the ACA Full Enforcement with the physical access variables (Distance 
Difference, Average Weekday Out of Hours, and Average Weekend Hours) do not significantly 
predict Unintentional Injuries Mortality. These results imply that the provision of insurance and 
access to primary care clinics does not significantly alter individual risk-taking behavior, which 
could lead to increased ED discharges and confound our results. In additional analyses, available 
upon request, our results remain unchanged when Unintentional Injuries Mortality is included as a 
control variable in the main model specifications.  
4.6.4 Does Physical Access Affect Insurance Uptake? 
We explore another possible alternative explanation for our results where individuals 
uptake insurance only when they have greater physical access to primary care clinics. In other 
words, one may argue that individuals living farther away from primary care clinics may simply 
remain uninsured and pay the tax penalty knowing they do not have physical access to primary care. 
This may confound the effects of the interaction between ACA Full Enforcement and physical 
access variables on ED Discharges. Therefore, we examine the relationship between physical 
access and the extent of the uninsured population in a ZCTA in a given year. The dependent variable 
65 
 
in this model, Number of Uninsured Individuals, as reported by the United States Census American 
Community Survey, represents the number of uninsured individuals for a given ZCTA in a given 
year. Furthermore, given that the Individual Mandate requires uninsured individuals to pay a tax 
penalty in lieu of insurance, and that some individuals may choose to pay the tax penalty, we use 
data from the Internal Revenue Service as reported on Form 1040, US Individual Tax Return, to 
control for the number of individuals paying the tax penalty (Number of Tax Returns with Penalty) 
and the average tax penalty (Average Tax Penalty in US Dollars) paid by these individuals within 
a given ZCTA in a given year, in addition to all of the other control variables used in our main 
analysis.  
 We further note that given the enforcement of the Individual Mandate tax penalty 
beginning in 2014, the sample reflects the period under which individuals were subjected to the 
Individual Mandate. Appendix B8, column (2) presents the results of this analysis. We find that 
none of the physical access variables of interest significantly predict the Number of Uninsured 
Individuals, indicating that, ceteris paribus, individuals are unlikely to make decisions to uptake 
insurance or remain uninsured based on the physical access of primary care clinics. This result 
highlights the reduced potential for endogeneity between the insured population after the full 
enforcement of the ACA and physical access to primary care clinics, thereby affecting our results.  
4.6.5 Does Transportation Modality or Household Vehicle Ownership Affect ED 
Discharges? 
Given the heterogeneity in geographies in our study, it is plausible that different modalities 
of transportation could impact the extent to which individuals also access healthcare services at the 
nearest primary care facility or ED. To account for the possible effects of transportation modalities, 
we utilize data from the United States Census American Community Survey related to 
transportation modality employed for commuting to work as a proxy for transportation modalities 
used to access healthcare services. These modalities include Driving Alone %, Carpooling %, Mass 
Transit %, Walking %, and Other Transit Modality %. In addition, the number of vehicles per 
household may dictate the time and ease of accessing a provider. Using data from the same source 
as the transportation modalities, we account for household vehicle ownership as a percentage of 
total households within a ZCTA. Specifically, we include the following controls: Households: One 
Vehicle %, Households: Two Vehicles %, Households: Three Vehicles %, and Households: Four 
or More Vehicles %.  
Appendix B9 lists the estimation results following the inclusion of transportation modality 
in column (1) and household vehicle ownership in column (2) as control variables. Across both of 
66 
 
the columns, the coefficients of ACA Full Enforcement X Distance Difference and ACA Full 
Enforcement X Average Weekend Hours remain qualitatively consistent with the main model 
results presented in Table 2 in terms of their statistical significance, direction, and magnitude. 
Collectively, these results suggest that the effects of physical access on ED discharges remain 
robust to the inclusion of transportation modality and vehicle ownership controls.  
4.6.6.  Using Alternative Measures of Distance Difference 
Our study builds upon previous literature using centroids as a proxy for individual location 
when calculating the approximate distance between two points (Gentili et al. 2015, Gentili et al. 
2018). We acknowledge that this measurement approach may not result in the most accurate 
measurements for certain scenarios where, at the individual level, the distance difference metric 
may understate or overstate the actual distance an individual may need to travel to the ED compared 
to the primary care clinic. Therefore, we run our analysis by re-estimating the measure of Distance 
Difference in two distinct ways: (i) using census block levels, and (ii) using population-weighted 
centroid at the ZCTA level. We explain the analyses using these approaches in greater detail in 
Appendix B10. As shown in the Appendix, we find that our results pertaining to the coefficient of 
ACA Full Enforcement X Distance Difference following the full enforcement of the ACA continue 
to remain consistent even when using different specifications of Distance Difference.  
4.7 Discussion 
The effort to understand the mixed findings regarding the relationship between the 
provision of insurance and ED discharges, specifically as it pertains to the ACA, has yielded calls 
for research to uncover the underlying factors that may better explain this relationship (Sommers 
and Simon 2017). Our study undertakes a nuanced examination of the relationship between 
insurance provision and ED discharges by examining the effects of physical access, specifically 
physical access to primary care clinics (relative to EDs) and their capacity characteristics, on this 
relationship. We focused on three questions from a patient’s perspective: where the nearest primary 
care clinic is in terms of road distance compared to the nearest ED, when the nearest primary care 
clinic is open, and the capacity of the nearest primary care clinic to provide timely care. Our 
analysis, first and foremost, reveals that while distance continues to be an important factor for 
newly insured individuals when it comes to making care usage decisions, efforts to improve 
financial access to healthcare in the form of insurance coverage expansion and mandatory uptake 
can attenuate the extent to which distance affects care usage decisions for such individuals.  
Second, and relatedly, our study explores the temporal accessibility of the healthcare 
supply chain and the impact of this physical access factor following a demand-altering policy 
intervention. By employing a healthcare supply chain perspective, we account for both the broader 
67 
 
characteristics of primary care clinic placement, while also accounting for an additional structural 
characteristic of each clinic—namely, hours of operation—in order to understand the varying usage 
outcomes following the full enforcement of the ACA. If the difference in distance is negligible, 
patients may place greater emphasis on the hours of operation when making a usage decision. In 
this regard, our results indicate that, holding all else constant, patients are less likely to utilize EDs 
as the number of weekend hours of operation at the nearest primary care clinic increase. Noting the 
importance of temporal accessibility to primary care clinics, other developed countries (e.g., the 
United Kingdom, Netherlands, Germany) have financially incentivized primary care clinics to 
expand their hours of operation beyond traditional working hours, thereby shifting nonurgent 
demand away from the ED (Berchet and Nader 2016).  
Third, our results also highlight the impact of the nearest primary care clinic’s capacity as 
an important consideration that may affect newly insured individuals’ decisions to use the nearest 
ED. In particular, we find a negative relationship between the average patient panel size for 
providers at the nearest primary care clinic and ED discharges from a ZCTA, suggesting that when 
providers see more unique patients, it increases timely access to care at primary care clinics and 
reduces the potential for newly insured individuals to use the ED. Furthermore, our results indicate 
that the delivery system for Medicaid matters—i.e., while the Medicaid fee-for-service delivery 
system reduces constraints on patients in terms of their access to timely care, a Medicaid managed 
care delivery system can create artificial bottlenecks for patients, limiting access to timely care. 
Patients with Medicaid managed care often use primary care clinics as an access point for both 
nonurgent outpatient care, and also to acquire a referral to specialty care (Melnikow et al. 2020). 
By funneling patients into a limited number of in-network clinics, Medicaid managed care 
recipients may be unable to receive timely care, increasing the potential for said individuals to 
instead seek treatment at the nearest ED. 
4.7.1 Policy and Financial Implications 
The Affordable Care Act provides a natural experiment to disentangle financial and 
physical access, allowing the legislation’s impact on healthcare utilization to be examined. 
Furthermore, it provides important lessons for policymakers and healthcare providers. In the past, 
research may have advocated for markets to determine location and capacity characteristics for 
healthcare delivery based on market efficiency. However, we note that policies creating large 
demand spikes require supply-chain-based policy solutions to adequately address demand. For 
example, in vertically integrated health systems, primary care clinic hours could be extended to 
alleviate ED discharges. Additionally, state and local governments could use incentives and zoning 
68 
 
to encourage independent physician groups to place their facilities along roadways leading to the 
nearest ED as a means of diverting nonurgent visits.  
Financial incentives can also provide an alternative lever to improve physical access to 
outpatient services. However, this is largely dependent on the manner in which financial levers are 
implemented. In some states, such as Oregon and Tennessee, administrators have set copayments 
as a method to discourage Medicaid patients from using the ED. While the efficacy of this approach 
has been debated in prior research (Mortensen 2010), it is important to note that the approach does 
not address the structural features of the current healthcare supply chain. In light of our findings 
relating to the important role played by differences in road distance and weekend hours of operation, 
hospital systems could consider locating primary care clinic capacity with provisions for increased 
levels of weekend hours of operation closer to EDs, allowing clinics to absorb nonurgent outpatient 
care. Alternatively, EDs could divert patients to a primary care clinic, assuming the hospital 
operates said facilities on its premises during the time of the encounter. Our results also suggest 
that additional Medicaid managed care patients at the nearest clinic result in increased ED 
discharges. To ensure the adequacy of the healthcare supply chain, policymakers could incentivize 
managed care organizations to increase access to primary care, either through the provision of new 
clinics within managed care networks, or by increasing the number of primary care providers within 
such networks.  
These aforementioned suggestions to increase access to primary care can have significant 
consequences for the financial health of states that have implemented Medicaid expansion. Recent 
studies suggest that the current projected impact of Medicaid on state budgets is unsustainable, 
accounting for nearly a quarter of state spending (National Governors Association 2021). Therefore, 
patients using the right service at the right time may contribute to lowering these projected cost 
increases. Using figures provided by Hargraves and Kennedy (2018), paired with an adjustment for 
inflation, the current approximate costs for an average ED visit in 2022 US dollars is $1,102.98, 
while the average primary care physician cost in 2022 US dollars is $130.78. Using the average 
marginal effects from Table 4, we find that every mile closer a primary care clinic is situated to the 
nearest ED results in an approximate annual reduction of $16,828.49 [i.e., (71.31 – 54.00) × 
(1,102.98 − 130.78) = 16,828.49] in ED costs. Additionally, we find that an increase of one hour 
in the average weekend hours of operation of the nearest clinic to a ZCTA results in an approximate 
annual reduction of $22,127.69 [i.e., ((−6.95) −  ( − 29.71)) × (1,102.98 − 130.78) = 22,127.69] in 
ED costs. Across the treatment group states of California and Kentucky, which include 1,769 
ZCTAs for California and 770 ZCTAs for Kentucky, this would mean that every mile closer a 
primary care clinic is situated to the nearest ED results in a savings of $42.73 million ($16,828.49 
69 
 
× 2,539 = $42,727,536.11) over the course of a year. In addition, an increase of one hour in the 
average weekend hours of operation for clinics across the two states would yield savings of $56.18 
million ($22,127.69 × 2,539 = $56,182,204.91) in ED costs.  
4.7.2 Limitations and Future Research 
As with any study, our work has some limitations. In particular, our analyses use patient 
data aggregated at the ZCTA level. This may not represent the heterogeneity related to patient 
behavior in terms of use patterns, which may be dictated by geography at the neighborhood level. 
Additionally, patient-level data with Protected Health Information (PHI) may provide a more 
nuanced view of how people with acute or chronic conditions use EDs for nonurgent outpatient 
care following Medicaid expansion. These patient records would allow us to analyze the distance 
travelled by each patient and control for specific aspects of individual health characteristics. 
Furthermore, this micro view could provide a greater understanding of how access to different 
transit modes (e.g., mass transit versus private transport) or health condition types (e.g., chronic 
versus acute) may impact the use of EDs for nonurgent outpatient care. However, in the absence of 
negotiations with each individual hospital for patient records, our data represents some of the most 
granular patient data available according to the expert determination method outlined by the Health 
Insurance Portability and Accountability Act of 1996 (HIPAA).  
Future work might include an examination of the implementation of Medicaid expansion 
following the Individual Mandate to understand whether the effects observed in this paper hold. 
Alternatives to the current healthcare supply chain design may also be examined to address the 
current supply gap. Notably, telemedicine represents a rapidly scalable technology that could 
address the current gaps in primary care delivery. The consolidation of primary care clinics has 
revolved heavily around the fixed costs of physical infrastructure. However, telemedicine provides 
a way to use concentrated capacity in a distributive fashion, which in many cases reduces the need 
for brick-and-mortar facilities in shortage areas (Garfield et al. 2018). Finally, reservation systems 
through omni-channels could provide an additional means by which to examine the problem by 
providing patients with a method to route themselves within the primary care portion of the 
healthcare supply chain. Individuals may use digital channels to first triage their condition and 
understand where availability exists before electing to receive telemedicine-based treatment or in-
person treatment at the nearest primary care clinic. This could create a more predictable patient 
flow, which could in turn be managed accordingly through the optimal points within the healthcare 
system. 
70 
 
Chapter 5: E-Access versus Physical Access: An Examination of Telehealth 
5.1 Introduction 
 Telemedicine has often been touted as a technology that can bridge the gap in access to 
healthcare, especially in underserved areas (Medicare Payment Advisory Commission, 2016). Prior 
to the COVID-19 pandemic, telehealth use skyrocketed 53% from 2016 to 2017 alone (American 
Medical Association, 2020). Today, approximately 50% of hospitals engage in some form of 
telehealth delivery (Mechanic et al., 2022). With its promise of improving health equity, telehealth 
has been a topic of great interest amongst policymakers, healthcare systems, and researchers alike. 
However, policies and related infrastructure are not always in place to allow for access to telehealth 
services. Therefore, it is important to understand the infrastructure that underpins telehealth access 
and the legal environment surrounding reimbursement and uptake. 
In recent years, telemedicine has gained substantial attention from scholars and 
policymakers, since the service can provide a myriad of benefits related to care equity by expanding 
healthcare accessibility and improving healthcare quality. A recent study examining the economic 
returns of telehealth found that telehealth in the United States reduces hospital admissions by 25 
percent while reducing the overall length of stay by 59 percent (Schadelbauer, 2017). Additionally, 
as patients can remotely access providers via telemedicine, patients save on travel expenses and 
lost wages associated with seeing a provider in person (Schadelbauer, 2017). Despite the numerous 
benefits of telemedicine, however, its adoption by patients and providers has been slower than 
expected (Barton, 2007). The major reasons suggested for this slow adoption are the lack of 
infrastructure and technology supporting telemedicine (e.g., broadband access) and the lack of 
supporting policies (Adler-Milstein, 2014).  
Recent advancements in hardware, software, and data transmission technologies have given 
more people access to telehealth. However, the accessibility of telehealth services has remained 
inequitable. For example, the digital divide in broadband accessibility still determines who has 
access to care through this new medium (Barton, 2018). While technology is critical for the delivery 
of telemedicine, the policies surrounding telehealth deployment are important for the uptake of 
services for both patients and providers. Different states have adopted different rules related to the 
reimbursement and requirements of telemedicine. 
Beyond the technology itself, physician adoption of telemedicine has been an issue in 
making telehealth services more readily available. In many cases, the current legal environment 
has not been conducive to supporting the uptake of services by physicians. For example, service 
71 
 
parity laws ensure that insurance companies are required to cover the same services to patients, 
whether through synchronous telehealth services or in-person visits. Specifically, laws surrounding 
the payment and uptake of telehealth represent an important aspect related to telemedicine adoption. 
Reimbursement remains a critical hurdle to primary care provider telehealth adoption (Dorsey & 
Topol, 2016; Jonk, 2022). In addition, providers may be disincentivized from providing real-time 
telehealth coverage to patients if the reimbursement structure rewards these visits differently than 
in-person visits. Therefore, to better understand antecedents of telemedicine uptake, we need to 
investigate the legal environment surrounding the uptake and delivery of telehealth services as well 
as broadband access. 
 While understanding the legal environment surrounding telehealth and broadband access 
is critical in explaining telemedicine uptake, it is also important to understand factors related to 
accessibility impacting the uptake of telehealth. In previous studies, the analysis was often 
restricted to a single facility with populations that might exhibit similar access to broadband or 
primary care providers (Powell et al., 2017; Reed et al., 2020, Bavafa, Hitt, & Terwiesch, 2018). 
While these studies provide evidence of factors impacting telehealth usage, it is difficult to 
generalize how these factors might impact telehealth uptake more broadly. In many cases, these 
studies occur at large academic medical centers located in large metropolitan areas that are likely 
to have better broadband access. In addition, previous studies focus on payers such as Medicare 
and Medicaid; however, fewer studies have focused on the impact of parity laws on privately 
insured individuals. In this study, we focus specifically on patient-to-provider telehealth services. 
Therefore, to systematically understand the effect of broadband access and legal 
environment, we utilize a sample with coverage over the Continental United States (CONUS) and 
Outside the Continental United States (OCONUS). We utilize a sample of 19 states including: 
Alabama, Arizona, Arkansas, California, Colorado, Delaware, Florida, Georgia, Hawaii, Iowa, 
Idaho, Massachusetts, North Carolina, Ohio, Pennsylvania, South Carolina, Virginia, Wisconsin, 
and Wyoming. To explore patients’ motivation for up taking telehealth services, we explore the 
technology-mediated access to healthcare, specifically pertaining to the broadband access for 
telemedicine and the physical access to providers for in-person care. This requires understanding 
whether telehealth use is a byproduct of physical accessibility to providers and/or a result of 
increased broadband accessibility. In addition, we account for the variations in law related to 
payment parity and coverage parity when delivering telehealth services to ensure that physicians 
are not disincentivized to provide services. By understanding the digital divide and physical 
accessibility of primary care providers, we explore the ways patients uptake healthcare through 
72 
 
different channels. Specifically, we ask: How do structural factors related to distance and 
broadband internet access impact the technology-mediated access of healthcare services?  
We predict that synchronous telehealth requires both parties to have adequate broadband 
to ensure uptake given the coupled nature of patients and providers. If synchronous telehealth is 
meant to serve as a substitute for in-person visits, the experience required likely needs to be on par 
with in-person visits in terms of response time and diagnosis quality (O’Shea, 2022). In addition, 
we note that parity laws will be a key component in determining the effect of broadband on 
synchronous visits. Next, we predict that, in terms of asynchronous care, consumer broadband 
access remains the most important factor. In addition, asynchronous visits do not require a fixed 
schedule as synchronous telehealth visits do. This decoupling of diagnosis and treatment 
recommendations makes scheduling easier. In addition, these visits will not take away from in-
person visits, which may reduce any hesitancy caused by variations in reimbursement.  
We test our predictions in the context of the Rural Health Care Program Funding Cap Order 
of 2018 with a national sample of 3-digit ZIP Code Tabulation Areas from 2016 to 2019. To 
measure physical accessibility to providers, we use ArcGIS Pro to dynamically model realistic road 
network conditions by deconstructing 3-digit ZCTAs to 5-digit ZCTAs and examining the number 
of providers within a 30-mile radius. These 5-digit ZCTAs are then aggregated to 3-digit ZCTAs. 
This allows us to construct a service area across the road network where patients would be able to 
seek treatment from a primary care provider. To measure broadband internet access, we utilize the 
Federal Communications Commission (FCC) Fixed Broadband Deployment Data Form 477 to 
analyze the speed and availability of broadband internet services at the block level. All broadband 
providers maintaining any form of facility related to the deployment of internet are required to file 
data biannually on the areas where they provide internet access (Federal Communications 
Commission, 2018). Specifically, we examine the availability of fixed broadband in the form of 
fiber to the end user given the future-proofed nature of the technology (e.g., limited only by speed 
of light transmission). Our control group consists of states without payment and service parity for 
telehealth services, and our treatment group consists of states with payment and service parity for 
telehealth services. In addition, we utilize Census American Community Survey data and County 
Business Patterns data to control for the insurance characteristics of local markets.  
Our results demonstrate the key impact of broadband access on patients’ uptake of 
telehealth services. In addition, in states with both payment and service parity, we find that 
synchronous telehealth uptake rises as the associated broadband infrastructure improves for both 
patients and providers. When physicians are reimbursed similarly to in-person visits and are 
insurance companies are required to pay for similar treatment via telehealth or in person, broadband 
73 
 
quality becomes a key factor in ensuring that the real time nature of synchronous telehealth delivery 
experience is on par with in-person care delivery. With regard to asynchronous telehealth services, 
we find that consumer broadband access is the primary factor determining uptake across our 
treatment and control groups. This may suggest that the lack of a face-to-face interaction reduces 
the need for both patients and providers to have the FCC mandated broadband speeds. Instead, 
patients’ experience is likely dictated by the upload speeds related to self-
assessments/pictures/documents and download speeds for their respective physicians’ diagnoses. 
Furthermore, as physicians do not need to make tradeoffs with in-person visits, it appears that parity 
laws do not dictate physicians’ propensity to provide asynchronous telehealth services.  
This study contributes to the healthcare operations management literature by providing a 
nuanced examination of the boundary conditions related to telehealth uptake, specifically 
pertaining to the various infrastructural and structural characteristics of telehealth deployment that 
directly impact the technology-mediated access provided by telehealth. Furthermore, we explore 
the conditions pertaining to both synchronous and asynchronous care for direct-to-consumer 
delivery of telehealth. Notably, if recent policy changes relaxing telemedicine data privacy 
restrictions, implementation of service parity laws, and the implementation of payment parity laws 
hold, our results represent a conservative estimate regarding the impact of broadband access on 
telehealth uptake amongst privately insured individuals. While we note that telemedicine is a small 
proportion of total healthcare usage compared to in-person visits, recent increases in usage and an 
increasing number of providers delivering telehealth warrant a better understanding of the 
infrastructure to deploy telehealth services specifically broadband deployment.  
5.2 Literature Review and Background 
 The concept of telemedicine has evolved over the decades as the care delivery technology 
has improved. As the technology has developed, telemedicine has come closer to providing 
technology-mediated access to healthcare on par with in-person primary care visits. In the past, 
limited technology may have made telemedicine a complementary good to in-person treatment. 
However, with the increasing availability of broadband and other telepresence technologies, 
telemedicine in some cases can provide a substitutable service that can be delivered either entirely 
online or in person at a distance mediated by a physician.  
Modern advancements in data transfer and communications have made the delivery of 
telemedicine more viable from a cost and efficacy standpoint. Today, telemedicine generally refers 
to a concept surrounding interactive televideo with image and medical record transfer paired with 
remote monitoring (Strode, Gustke, & Allen 1999). Recent advances in secure data transmission, 
74 
 
improved broadband speed, and user-friendly interfaces have increased the propensity of 
individuals to uptake telehealth (Hall & McGraw, 2014; Agnisarman et al., 2017; O’Shea et al., 
2022). However, while technological advances may make telemedicine more accessible to patients, 
laws governing reimbursement and service provision may still limit the adoption of telehealth 
amongst providers. In this section, we provide a comparison of the existing literature examining 
the technological underpinnings of telehealth and the operational components to formulate how 
different forms of telemedicine may be impacted by the physical accessibility to primary care 
providers and the availability of broadband. In addition, we illustrate the challenges in delivering 
telemedicine and the existing boundaries that may impact patient uptake of telehealth services and 
physicians’ willingness to offer telehealth.  
5.2.1 Genesis of Modern Telemedicine 
With the advent of the space age, telehealth became a vital part of delivering care. It was 
infeasible to send all the necessary types of physicians into space with the astronauts; therefore, a 
new solution was required to supply medical care that did not require the immediate presence of 
physicians. Thus, a method to deliver healthcare under structurally dictated lean conditions was 
required. In 1971, NASA and the Papago Native American Tribe worked together to establish a 
project known as Space Technology Applied to Rural Papago Advanced Health Care 
(STARPAHC), pioneering telemedicine and the remote delivery of healthcare (Henceroth, 1978). 
In this model, primary servers, such as community health medics, provided the initial phase of care, 
with physicians located at a major hospital serving as secondary servers; furthermore, the system 
utilized a computerized health record system to support care by coordinating information amongst 
care providers (Henceroth, 1978). This initial model involved both mobile and fixed clinics to 
understand how the two models might vary in their service delivery. The fixed-site clinics in this 
initial setup were able to accommodate twice as many patients per hour; in addition, the fixed-site 
clinics remained open for longer hours (Henceroth, 1978). This initial trial revealed that 
telemedicine could serve as a complement to existing infrastructure, but further research was 
required to understand the appropriate resource allocation levels. 
5.2.2 The Evolution of Telemedicine 
 From these origins, telemedicine has undergone multiple generations of evolution in terms 
of capability and delivery. Notably, with the advent of the internet, telemedicine can be delivered 
directly to consumers with the appropriate technology. In the case of heart failure, four distinct 
generations of telemedicine now exist (Anker et al., 2011). This model details the evolution of 
technical advances that have created distinct generations within telemedicine implementation. 
Table 9 details the various generations of telemedicine based on Anker et al.’s findings. As 
75 
 
telehealth has evolved, the ability to treat a wider host of ailments has also evolved. We note that 
asynchronous telehealth services are more in line with the first and second generations of 
telemedicine, while synchronous telehealth services move toward the third- and fourth-generation 
conceptions of telehealth. With the expansion in capabilities, the efficacy of telemedicine has been 
examined to understand how well this format of treatment delivery can account for an input 
involving a wider range of diseases. 
In building upon previous telemedicine models, researchers have also examined how 
various disease states might impact outcomes. A recent Cochrane Review found that acute illnesses 
and chronic conditions do not have a significant impact on adoption rates of telehealth based care 
delivery. Based on these findings, the type of illness (i.e., chronic or acute) is less important; instead, 
the equipment available has more bearing on care delivery (Flogren et al., 2015). Initially, there 
was some consideration of the possibility that chronic illnesses might be easier to manage due to 
the slower pace of change in the disease state. For example, acute diseases were expected to pose 
a greater challenge in scheduling physicians due to the less predictable nature of patient demand. 
On the other hand, chronic illnesses may be easier to monitor as rapid changes in condition are 
unlikely, which makes treatment planning more predictable. However, the work of Flogren et al. 
(2015) indicates that both acute and chronic conditions can be addressed through telemedicine 
initiatives. Given that telemedicine can handle a wide variety of conditions, it is important to 
understand how the delivery modality can impact uptake and outcomes.  
Table 9: Telemedicine Generations 
Telemedicine Generation  Generational Traits 
First  ▪ Non-reactive data collection and analysis 
▪ Measurements are collected and forwarded to a care 
provider 
▪ Providers cannot respond immediately 
Second  ▪ Non-immediate analytical or decision making 
structure 
▪ Real time processing of patient data, e.g. automated 
algorithms 
▪ Care providers will be able to recognize issues; 
however, delays may occur if information is process 
during non-office hours 
Third  ▪ Constant analytical and decision support 
▪ Monitoring centers house physicians and nurses 
where these individuals can provide care constantly 
Fourth  ▪ Extends on third generation technology but 
incorporates invasive and noninvasive procedure 
delivery through telemedical devices 
▪ Requires constant physician presence 
76 
 
While synchronous and asynchronous modalities provide reliable options for delivering 
care, providers have noted concerns related to care delivery. Specifically, the ease of use and 
validity of telemedicine interventions are currently under scrutiny. In certain cases of asynchronous 
interventions, mobile applications may generate automated recommendations for patients until they 
can see a physician, which could lead to problematic results. As the possible liability increases with 
telemedicine interventions, hospital systems will have a more difficult time garnering support 
amongst care providers to use telemedicine-based interventions. In the case of mobile apps, the 
increasingly popular delivery via mobile devices serves as a method for doctors to keep up with 
their patients. However, in certain cases these apps may make recommendations without 
considering physician input or testing scientific knowledge. In one specific case, direct-to-
consumer (DTC) telemedicine technologies were found to be inadequate in diagnosing certain 
types of conditions (Resneck et al., 2016).  
With regard to synchronous technologies, telephone visits represent one of the initial forays 
into telemedicine. In other cases, telemedicine is delivered via a physician-to-physician link, such 
as Project ECHO. While telemedicine is inherently dependent on the hardware, organizations are 
also crucial to the widespread dissemination and acceptance of the telemedicine concept, assuming 
that all players are given equal access to accurate information. The concept of digital redlining also 
creates a different problem wherein even those seeking to utilize synchronous telehealth services 
may be limited by the accessibility of broadband internet (O’Shea, 2022).  
Recently, the National Quality Forum (NQF) outlined a framework offering a 
comprehensive guide for developing measures to assess telehealth (Chuo et al 2020). The NQF 
framework notes that, in terms of access to care, travel, timeliness of care, actionable information, 
added value of telehealth to provide evidence-based best practices, patient empowerment, and care 
coordination are relevant measurement areas for assessing telehealth (Hollander et al., 2017). As 
telehealth delivery becomes more prevalent, the need to understand how individuals make usage 
decisions becomes more important in deploying the technology. Providers have noted that 
telehealth allows for better retention of patients through improved contact, which reduces the need 
to transfer them to other institutions or providers (Jonk et al., 2021). 
 However, current telemedicine deployment is often the result of historical factors related 
to each type of technology. Certain models of telehealth have been developed to overcome 
structural hurdles related to the deployment of telehealth. Notably, asynchronous telehealth services 
require less bandwidth than synchronous telehealth visits, which allows for easier deployment of 
these types of services without high-quality or up-to-date broadband access (Wilson & Maeder, 
77 
 
2015). Modalities such as asynchronous telehealth can allow for time savings, flexibility, and 
efficiency that synchronous modalities may not offer. However, asynchronous modalities lack the 
ability for patients and providers to engage in more in-depth interactions where certain information 
could lead to an improved ability to make real-time clinical decisions (AAHA).  
Previously, operations management research has explored how telehealth services alter the 
way individuals utilize in-person services in relation to telehealth services (Bavafa, Hitt, & 
Terwiesch, 2018). Researchers have noted that telemedicine in certain cases does not seem to 
replace in-person visits; instead, telemedicine appears to be a complement to in-person visits that 
increases the overall amount of care used by patients. 
5.2.3 Parity: Telehealth Reimbursement and Service 
In the wake of the COVID-19 pandemic, quarantining and mandatory social distancing 
increased interest in a hybrid model of patient care amongst industry providers (Roth, 2021). 
However, telemedicine uptake may be strongly influenced by the legal environment surrounding 
provider reimbursement for services and whether patients can receive similar care through either 
physical or digital channels of care. For example, if providing similar services to patients through 
in-person care and digital channels takes a similar amount of time but the reimbursement rate for 
telemedicine is lower, providers are unlikely to adopt telemedicine given the misaligned financial 
incentives. Private insurers often base their reimbursement policies, specifically reimbursement 
rates to providers, using Medicare as a benchmark. Medicare first began to reimburse for telehealth 
services after the passage of the Balanced Budget Act of 1997 (Rhueban & Krupinski, 2018). The 
conditions for reimbursement were expanded for Medicare under the Benefits Improvement and 
Protection Act of 2000 (Rhueban & Krupinski, 2018). However, since 2000, there has been very 
little change to the conditions for Medicare reimbursement. Notably, private insurers have followed 
suit and are beginning to increase the number of plans that include telehealth services for their 
enrollees (Gumpert, 2015). 
With regard to private payers, as of 2018, there were 40 jurisdictions enacting laws that 
govern private payer telehealth reimbursement (Center for Connected Health Policy, 2019). These 
laws require coverage parity for services that are delivered either in person or via telehealth 
assuming that the same standard of care is met. During the recent COVID-19 pandemic, many laws 
were adjusted to establish payment parity and coverage parity, thereby incentivizing physicians to 
provide similar services to patients, whether in person or virtually, while ensuring that physicians 
were reimbursed equally for equivalent services.  
78 
 
 However, prior to COVID-19, states were at different stages of adopting payment parity 
and service parity. With regard to private insurance companies, payment parity and service parity 
were crucial components for ensuring that patients would not be denied telehealth coverage simply 
for the delivery modality and that physicians would be reimbursed equally for the delivered 
telehealth services. We note that beyond parity considerations, there are structural considerations 
underpinning the delivery of telehealth, specifically as it pertains to broadband connectivity for 
providers. 
5.2.4 Rural Health Care Program 
 As the importance of telemedicine has grown, the federal government has sought out 
methods to incentivize primary care providers to offer telehealth services to their patients. Started 
in 1997, the Rural Health Care Program provides funding to eligible healthcare providers for 
telecommunication and broadband services necessary for the provision of healthcare (Federal 
Communications Commission). Over the years, this program has expanded to include two 
components: the Healthcare Connect Fund Program and the Telecommunications Program. The 
Telecommunication Program was established in 1997 to subsidize rural healthcare providers by 
addressing the difference between urban and rural rates for telecommunications services (Federal 
Communications Commission). The Healthcare Connect Fund Program was established in 2012 
with the aim of providing high-capacity broadband to eligible healthcare providers in an effort to 
improve connectivity, which could be used for the transmission of patient data or telehealth services 
(Federal Communications Commission). However, these components of the Rural Health Care 
Program have been unable to keep pace with the demand for telecommunication services over time. 
 Researchers have noted that rural hospitals are the most likely to benefit from telehealth 
but the least likely to be able to deliver these services (Zachrison et al., 2020). Therefore, in 2018, 
the Federal Communications Commission made two decisions to expand accessibility to these 
broadband accessibility funds. The first decision increased the pre-existing program funding cap 
from $400 million to $571 million beginning in Funding Year 2017, which was also meant to 
account for inflation (Federal Communications Commission). Second, the program moved to carry 
forward unused funds from the previous funding years to fund future funding years (Federal 
Communications Commission). Together, these changes were meant to increase the availability of 
broadband in rural areas, which would likely directly impact physician willingness to provide 
telehealth services.  
While the Rural Healthcare Program generally specifies that coverage would be limited to 
rural areas, the eligibility requirements allow for providers in non-rural areas to uptake funds. These 
entities include post-secondary educational institutions offering health care instruction, teaching 
79 
 
hospitals, and medical schools; community health centers or health centers providing health care 
to migrants; local health departments or agencies; community mental health centers; not-for-profit 
hospitals; rural health clinics; skilled nursing facilities (as defined in section 395i–3(a) of title 42); 
and consortiums of health care providers consisting of one or more entities falling into the first 
seven categories. In this case, consortiums must be majority rural (i.e., 50% of participating sites 
must be rural) after three years of obtaining their initial funding commitment.  
With regard to non-rural hospital sites, support for any site with 400 or more licensed beds 
is capped at $30,000 per year in support for recurring charges and $70,000 in support for non-
recurring charges every five years, exclusive of any costs shared by the network. 
5.3 Theoretical Framework 
We model the delivery of telehealth in direct-to-consumer (DTC) -based technologies, 
specifically examining patient-to-provider visits. We utilize the expansion of the Rural Healthcare 
Program under the Rural Healthcare Program Funding Cap Order of 2018 to explore the structural 
characteristics surrounding telehealth delivery. In the case of care delivery, we seek to understand 
the structural and infrastructural components surrounding telehealth visits and whether a specific 
type of telehealth serves as a substitutable good to in-person care. To ensure comparability, our 
treatment group contains states that have adopted both payment parity and coverage parity 
throughout the study period. 
Figure 9: Total Healthcare Utilization Amongst Privately Insured Sample Population 
 
0
10,000,000
20,000,000
30,000,000
40,000,000
50,000,000
60,000,000
70,000,000
80,000,000
90,000,000
100,000,000
2016 2017 2018 2019
N
u
m
b
er
 o
f 
V
is
it
s
Year
Control Treatment
80 
 
5.3.1 Physical Access and Synchronous Telehealth 
 Telemedicine as a value proposition assumes that patients have access to broadband 
internet and providers have the ability to deliver care utilizing broadband internet. Without 
adequate broadband penetration, telemedicine is not a viable complement or substitute for in-person 
care. Rural areas often exhibit lower broadband penetration that would limit access to telehealth-
based primary care services (Drake et al., 2019). In this case, we focus on broadband penetration 
in the form of fiber optic infrastructure. Fiber is considered to be future-proof given that data 
transmission is bound only be the speed of light. Einstein’s special theory of relativity effectively 
notes that no object, including the photons transmitting data, can travel faster than the speed of 
light6 (Barton, 2017). That said, broadband quality for both patients and providers dictates the 
quality of the examination and the quality of care, which ultimately impact patients’ willingness to 
consistently uptake telehealth services (Dorsey & Topol, 2016; O’Shea et al., 2022). However, it 
is likely that patients will attempt to make an in-person visit to a primary care provider prior to 
utilizing a synchronous telehealth visit.  
In addition to broadband penetration, the actual distance individuals travel to primary care 
providers may impact their considerations to uptake telehealth. Especially in rural areas, 
infrastructure related to primary care is likely limited, which would effectively limit the number of 
providers available within an easily travelable distance. Researchers note that individuals may be 
more willing to opt for telehealth when it requires 30 minutes or longer to find a primary care 
provider (Powell et al., 2017; Reed et al., 2020). This suggests that there is a threshold travel time 
at which individuals may consider opting for one service over another. In this case, we consider 30 
miles to be the largest radius that an individual would likely be able to travel during a 30-minute 
period. However, we note that we do not expect the effect of physical access to providers to be 
impacted greatly by the expanded availability of synchronous telehealth services. For example, an 
in-person visit may take an average of 20 days to secure a simple 20-minute appointment (Dorsey 
& Topol, 2016). Broadband itself is unlikely to shift the dynamics at these providers, e.g., the way 
individuals use in-person care versus telehealth services (Bavafa, Hitt, & Terwiesch, 2018). 
Therefore, this study seeks to understand whether synchronous telehealth visits serve as 
complements or substitutes for in-person care. We hypothesize: 
 
6 We acknowledge that physicists have postulated that superluminal (i.e., faster-than-light) communication 
is possible, through either tachyons or wormholes (Butcher, 2014; Feinberg, 1967; Serway et al., 2004). 
However, in the context of this paper, these proposed mechanisms currently cannot be functionalized at scale 
to provide superluminal communication.  
81 
 
Hypothesis 1a: Following the expansion of the Rural Health Care Program in states with 
service and payment parity, synchronous telehealth uptake increases for areas with higher 
broadband penetration for both patients and providers. 
Hypothesis 1b: Following the expansion of the Rural Health Care Program in states with 
service and payment parity, synchronous telehealth uptake decreases for patients with 
primary care access within a 30-mile radius. 
5.3.2 Asynchronous Telehealth 
 While most individuals think of synchronous telehealth services when considering 
telehealth, there has been an increased interest in asynchronous telehealth delivery given the 
increased flexibility offered by the modality. Asynchronous telehealth provision was considered by 
stakeholders to be a way to bridge the digital divide (Wilson & Maeder, 2015). In addition, it allows 
physicians to flexibly adjust their schedules to treat patients instead of blocking time as required 
for synchronous telehealth services. By decoupling the interaction between the patient and provider, 
components of a checkup or diagnosis can occur at different times, allowing for greater flexibility 
in the patient’s and provider’s respective schedules (Wilson & Maeder, 2015). Therefore, these 
visits do not carry the same tradeoffs that would be seen with synchronous telehealth visits, 
especially in terms of weighing whether a telehealth visit will result in equivalent pay as an in-
person visit given the same time block. Notably, while there are concerns regarding the lack of 
direct contact between patients and providers, users appear to be highly satisfied with asynchronous 
telehealth (Kruse et al., 2017).  
 Recent studies have shown that the efficacy of asynchronous programs mirrors that of in-
person consultations (Armstrong et al., 2015; Armstrong et al., 2018; Boothe et al., 2022). Given 
that asynchronous telehealth does not require real-time interaction, individuals making uptake 
decisions are likely to base their decisions on the time it takes to upload relevant documents, images, 
or videos related to their conditions and downloading the physician diagnosis once it has been 
provided. Thus, patients’ satisfaction and continued use likely depends upon the speed with which 
associated health information can be uploaded and how quickly diagnoses can be downloaded from 
the designated telehealth interface (O’Shea et al., 2022). Uploads may include self-captured clinical 
images and documents related to patient history, while downloads include physician diagnoses and 
digital prescriptions (Armstrong et al., 2015). Given that provider broadband speeds are unobserved 
by the patient, we believe that the expansion of the Rural Healthcare Program will not impact 
patients’ decisions to uptake asynchronous telehealth services. Therefore, we hypothesize that only 
a patient’s broadband connection will determine their uptake of asynchronous telemedicine. 
82 
 
Hypothesis 2: Asynchronous telehealth uptake increases as consumer broadband access 
increases. 
5.4 Data 
 To test our study’s hypotheses, we utilize data from FairHealth, an independent nonprofit 
managing the nation’s largest database of privately billed health insurance claims, encompassing 
private payers from 2016 to 2019. Our selection of these years avoids the switch in condition coding 
from ICD-9 to ICD-10 and reduces any possible confounding effects related to the COVID-19 
pandemic, where the relaxation of restrictions related to the provision of care may have altered 
uptake. However, due to the restrictions related to the data, our sample utilizes 3-digit ZIP Code 
Tabulation Areas (ZCTA).  
In addition, we conduct an environmental scan related to parity laws surrounding telehealth. 
States can institute two forms of parity related to the delivery of telehealth services: service parity 
and payment parity. In our study, our treatment group consists of states instituting both service and 
payment parity. This allows us to ensure that services provided via telehealth are not unique to 
those provided in person based on the CPT code. In addition, we also ensure all treatment states 
within our sample institute payment parity for fully insured insurance plans. Payment parity ensures 
that there is no disincentive for providers in terms of providing in-person versus telehealth-based 
care. These forms of parity have been noted to impact providers’ willingness to uptake telehealth 
services (Zachrison et al., 2021). 
Dependent Variables: Our three main dependent variables of interest are the total number 
of telehealth visits, the number of synchronous visits, and the number of asynchronous visits from 
a 3-digit ZCTA. Specifically, we functionalize the dependent variables as count variables utilizing 
the raw number of visits from each 3-digit ZCTA for visits involving patient-to-provider contact. 
These dependent variables allow us to understand the changes in utilization for different formats 
of telemedicine. 
We utilize Current Procedural Terminology (CPT) codes from our sample to understand 
whether service was provided via in-person consultation, synchronous telehealth service, or 
asynchronous telehealth service. In the case of our telehealth visits, our study focuses only on 
patient-to-physician interactions where the provider is directly providing care. The CPT codes used 
to designate asynchronous and synchronous visits are listed in Appendix B.  
Independent Variables: While the United States Census stopped providing 3-digit ZCTA 
data with the 2010 Census, we utilize ArcGIS Pro to construct 3-digit ZCTAs from 5-digit ZCTAs. 
In order to do this, we first construct the 3-digit ZCTA identifier, which consists of the first 3 digits 
of each 5-digit ZCTA. Next, we utilize the Dissolve feature within the Data Management Tools. 
83 
 
This feature allows us to dissolve the boundaries between ZCTAs that share the same first 3 digits. 
This allows us to build the relevant shapefiles related to each 3-digit ZCTA. 
Figure 10: Road Service Area Versus Euclidean Distance 
 
Description: In this figure, we examine the difference between using a 30-mile radius based solely on 
Euclidean distance versus one using the road network. In the figure above, we examine the ZCTA 22904. 
The blue circle represents a 30-mile radius based solely on Euclidean distance. The gold shape illustrates a 
30-mile distance across the road network.  
To assess the impact of physical access to primary care on telehealth uptake, we utilize the 
InfoGroup Historical Business Database and the Center for Medicare and Medicaid Services (CMS) 
National Plan and Provider Enumeration System (NPPES) National Provider Identifier (NPI) 
Registry.  Within the InfoGroup Historical Business Database, we utilize the 8-digit NAICS code 
(62111107) for primary care clinics. To cross-validate these locations as primary care clinics, we 
match the geocoded practice addresses in the NPPES NPI registry. We limit our definition to the 
following healthcare provider taxonomy codes: 207Q00000X, 207QG0300X, 207R00000X, 
208000000X, and 208D00000X.  
We construct a panel of primary care providers through a multistep process. First, we 
geocode the practice addresses within the NPPES NPI Registry using the Geocode Addresses in 
ArcGIS Pro. Next, we utilize the geocoded NPI practice addresses and spatially join these with the 
InfoGroup Historical Business Database. This allows us to construct a longitudinal panel of 
providers for our sample.  
84 
 
By geolocating the sample of primary care clinics, we calculate the distance to the nearest 
primary care clinic from each component 5-digit ZCTA centroid of a 3-digit ZCTA. This allows us 
to approximate the general availability of primary care services from brick-and-mortar facilities. 
Specifically, we utilize ArcGIS Pro and the Service Area function under Network Analyst tools. 
For clarity, Figure 10 provides an illustration of how the Service Area function differs from simply 
drawing a 30-mile radius from a 5-digit ZCTA centroid. In this case, we bound the availability of 
primary care clinics by state borders. 
Additionally, we note that broadband access is an important factor related to telehealth 
uptake. To account for the availability of broadband access, we use the Federal Communications 
Commission (FCC) Fixed Broadband Deployment Data from Form 477 to account for the 
availability of fixed broadband within a market based on speed. All facilities-based broadband 
providers are required to file data with the FCC twice a year utilizing Form 477 to show where they 
offer broadband service (FCC 2018). In this case, we avoid satellite internet given the shortcomings 
related to delivery in terms of reliability, signal latency, and speed. Notably, satellite services, such 
as StarLink, suffer from a phenomenon known as rain fade given the frequency bands used (e.g., 
Ku and Ka; Safaai-Jazi et al 1995).  
By limiting our sample to fixed broadband access, we can control for a possible factor that 
may limit individual uptake of telemedicine. We note that the federally suggested guidelines for 
telehealth provision are 10 Mbps for basic services and 25 Mbps for real-time care (FCC 2015). 
Hospitals need to have 100 Mbps to multitask multiple services. We utilize the Census tabulation 
blocks to construct the general availability of services by speed available to patients. This results 
in a variable illustrating the percentage of tabulation blocks within a 3-digit ZCTA covered by a 
broadband provider meeting federal guidelines for the provision of telehealth. 
Insurance Control Variables: Payment parity laws generally apply only to fully insured 
plans. Given this constraint, we control for the types of insurance plans (either self-insured or fully 
insured, and either employer-provided or direct purchase) at the 3-digit ZCTA level. We calculate 
the number of firms within a 3-digit ZCTA with 1,000 or more employees to proxy the propensity 
for individuals to be self-insured. Large employers often use self-insured plans as a method of cost 
cutting with regards to health insurance costs. Additionally, we control for the insured population 
by using the American Community Survey data on individuals who purchase insurance directly 
and those who receive employer-based health insurance. This allows us to understand the total 
number of individuals insured utilizing private insurance. 
Parity Variables: With regards to the determination of parity, we utilize a multitude of 
sources. In addition, we rely on Justia, a firm providing open access to longitudinal legal data, to 
85 
 
examine statues from years past. We utilize Justia to examine the policy statutes outlining service 
parity and payment parity across our study period. Our sample consists only of geographies that 
either adopted both payment and service parity or adopted neither. We provide the specific legal 
statutes in Appendix C1. 
Rural Healthcare Program Control Variables: While the Rural Healthcare Program 
provides funding for providers to adopt telehealth, this funding can vary depending on the purpose 
of the funding in relation to types of improvements and dollar values. Therefore, we utilize FCC 
and Universal Services Administrative Company Form 462/From 466/Form 471 data to account 
for the deployment characteristics of funding expansion to the Rural Healthcare Program. In our 
model, we account for the percentage of recipients that are individuals, the total amount of funding 
provided to a 3-digit ZCTA, and the percentage of primary care clinics receiving the Rural 
Healthcare Program funding.  
To control for the funding characteristics of the RHP in our sample, we construct a series 
of control variables related to program funding. Given that an application could contain multiple 
funding requests for a single clinic, we collapse all the funding requests for each clinic by year. 
Therefore, if we had a clinic with 5 funding requests, we would account for that clinic and add all 
the funding received into a single item. In addition, we ensure to account for the differences 
between Individual and Consortium applicants. This variable is constructed as a percentage of 
applicants within a 3-digit ZCTA that were filed by an individual. Finally, we account for the total 
funding allocated to a specific 3-digit ZCTA by the RHP.  
Geographic Control Variables: To control for the characteristics of each 3-digit ZCTA, we 
utilize the US Census American Community Survey (ACS). These controls include sex ratio, age 
groups (Age: Under 18, Age: 18 to 24, Age: 25 to 34, Age: 35 to 44, Age: 45 to 54, Age: 55 to 64), 
racial composition (Black, Native, Asian, Other), inward migration, unemployment, individuals 
with private insurance that have a high school education or equivalent, married population, 
separated or divorced population, widowed population, and average household income.  
We use average household income instead of median household income given that median 
household income is unavailable at the 3-digit ZCTA level beginning with the 2010 Census. The 
average household income is calculated by aggregating the total income for the 5-digit ZCTAs 
within a 3-digit ZCTA and dividing this value by the total number of households within a 3-digit 
ZCTA. With regard to computer fluency, we believe that the controls for age groups should allow 
us to account for the fluency related to digital interfaces observed within our population of privately 
insured individuals.  
5.5 Empirical Strategy 
86 
 
To understand the impact of telehealth availability and uptake, we identify a policy shock 
that changes accessibility, the Rural Health Care Program Funding Cap Order. This shock allows 
us to examine the impact of structural changes resulting from improved funding for critical 
telehealth infrastructure. Specifically, we focus on the Healthcare Connect Fund Program that 
provides funding for broadband services and equipment. We use a sample provided by FairHealth 
that encompasses private payers across CONUS and OCONUS.  
In addition, our treatment group consists of states that enacted both payment parity and 
service parity during our study period. Our control group consists of states that have adopted neither 
payment parity nor service parity. This allows us to compare the effects of changes in physical and 
digital access to telehealth services with and without the presence of parity. Furthermore, our 
sample focuses solely on patient-to-provider visits. 
5.5.1 Difference-in-Differences Analysis 
 To understand the impact of telehealth, we examine the impact of a policy shock aiding 
providers to provide telehealth services, the Rural Health Care Program Funding Cap Order. In our 
study, we examine the impact of this policy on the availability of telehealth in various geographies. 
We denote our control group as 3-digit ZCTAs in states with no payment parity laws for telehealth 
and in-person services. For our treatment group, we use 3-digit ZCTAs in states with payment 
parity laws for synchronous and asynchronous telehealth visits. The vector of time-varying control 
variables in the equation is denoted by 𝑿, 𝜆 represents the ZCTA fixed effect, and 𝜇 represents year 
fixed effect. 
Telehealth Visits
= α + β1Broadband Penetration + β2Primary Care Distance
+ β3Rural Healthcare Program Expansion
+ β4(Rural Healthcare Program Expansion × Broadband Penetration)
+ β5(Rural Healthcare Program Expansion × Primary Care Providers) + 𝛄𝐗
+ λ + μ + ε (1) 
To better delineate the effects of broadband and physical access on telehealth uptake, we 
construct two additional models that delineates between synchronous telehealth visits and 
asynchronous telehealth visits.  
87 
 
Synchronous Telehealth Visits
= α + β1Broadband Penetration + β2Primary Care Distance
+ β3Rural Healthcare Program Expansion
+ β4(Rural Healthcare Program Expansion × Broadband Penetration)
+ β5(Rural Healthcare Program Expansion × Primary Care Providers) + 𝛄𝐗
+ λ + μ + ε (2) 
Asynchronous Telehealth Visits
= α + β1Broadband Penetration + β2Primary Care Distance
+ β3Rural Healthcare Program Expansion
+ β4(Rural Healthcare Program Expansion × Broadband Penetration)
+ β5(Rural Healthcare Program Expansion × Primary Care Providers) + 𝛄𝐗
+ λ + μ + ε (3) 
Figure 11: Payment Parity Map 
 
Note: 
The green states represent states with payment parity and service parity. The gray states represent states 
with neither payment parity nor service parity.  
Figure 11 presents an illustration of the states with both service and payment parity and the 
states with neither. Prior to the Rural Health Care Funding Cap Order, we find that the states exhibit 
parallel trends in amount of telehealth usage. However, following the expansion of the RHP, we 
find a divergence in the uptake of patient-to-provider telehealth services. To understand how the 
policy impacts telehealth uptake by delivery type, we use the aforementioned model specification 
with two alternative dependent variables. In this case, one model specification examines 
synchronous visits only while another examines asynchronous visits only.  
5.6 Results 
5.6.1 All Telehealth Visits 
A key component of effective telehealth delivery is the availability of broadband internet 
for both patients and providers when both service and payment parity are present. Therefore, we 
88 
 
examine the patient-to-provider usage of telehealth across our sample of 19 states with varying 
payment and service parity legal statutes. 
We report the estimation results for the DiD specification utilizing all telehealth visits 
between a patient and provider. This model specification allows us to broadly denote the effects of 
broadband access and physical access on telehealth uptake. Column (1) in Table 10 provides a 
baseline main effects model that includes the complete set of controls, the main effects of 
broadband and primary care accessibility, Consumer Broadband Coverage and Service Area 
Providers, and the policy measure, RHP Expansion. Columns (2) and (3) hierarchically build upon 
the main effects model by including the interaction terms of the physical access measures with the 
policy measure, whereas column (4) represents the full model, which includes all of the interaction 
terms. We use the results from column (4) to interpret our hypotheses tests. 
Figure 12: Synchronous Telehealth Visits and the Rural Healthcare Program 
 
 
  
0
50,000
100,000
150,000
200,000
250,000
300,000
350,000
2016 2017 2018 2019
T
o
ta
l 
V
is
it
s
Years
Control Treatment Funding Cap Order
89 
 
Table 10: ZCTA3 Model Specifications All Telehealth Visits 
 Negative Binomial Regression 
 (1) (2) (3) (4) 
 β(SE) β(SE) β(SE) β(SE) 
Main Effects     
Consumer Broadband % 0.40509 (0.343) 0.38283 (0.341) 0.40144 (0.343) 0.38957 (0.340) 
Providers in Service Area -0.00027 (0.000) -0.00034 (0.000)* -0.00030 (0.000) -0.00027 (0.000) 
Rural Healthcare Program [RHP 
Expansion] 
-0.02463 (0.087) -0.11650 (0.105) -0.03002 (0.095) -0.11104 (0.107) 
     
Interactions     
RHP Expansion X Broadband Access   0.50973 (0.253)**   0.54951 (0.262)** 
RHP Expansion X Providers in 
Service Area 
    0.00001 (0.000) -0.00002 (0.000) 
     
RHP Controls     
RHP Applicants: Individual % 0.36992 (0.222)* 0.35614 (0.228) 0.36861 (0.224)* 0.35810 (0.228) 
Total RHP Funding -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) 
RHP Clinics % -0.43496 (0.500) -0.47719 (0.502) -0.42804 (0.497) -0.49671 (0.502) 
     
Constant 0.78582 (4.529) 1.39315 (4.457) 0.80941 (4.548) 1.38603 (4.473) 
ln Alpha [Dispersion Parameter] -1.94452 
(0.086)*** 
-1.95073 
(0.087)*** 
-1.94460 
(0.086)*** 
-1.95104 
(0.087)*** 
Demographic Controls Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes 
State FE Yes Yes Yes Yes 
Number of Observations 904 904 904 904 
* p<0.10, ** p<0.05, *** p<0.01 
Table 11: ZCTA3 Marginal Effects for All Telehealth Visits 
 Broadband Coverage 
 β(SE) 
Pre-Rural Healthcare Program 694.750 (611.658) 
Post-Rural Healthcare Program 1671.511 (800.997)** 
  
Contrast 976.761 (478.152)** 
Observations 904 
* p<0.10, ** p<0.05, *** p<0.01 
Focusing on column (4) of Table 10, we find that the coefficient for RHP Expansion × 
Consumer Broadband % is positive and significant (β = 0.54951, p<0.05), suggesting that the effect 
of consumer broadband coverage paired with improved primary care provider broadband coverage 
increases telehealth visits following the Rural Healthcare Program expansion in states with both 
payment and service parity. To better understand this result, we examine the average marginal 
effects in Table 11. Our average marginal effects suggest that for every 1% increase in the number 
of Census blocks covered by broadband within a 3-digit ZCTA, there is a subsequent increase of 
approximately 977 annual telehealth visits. However, we note that the technology-mediated access 
to healthcare provided by synchronous and asynchronous telehealth services is dependent on 
different aspects of broadband access and physical access to primary care providers.  
90 
 
5.6.2 Synchronous Telehealth 
In the synchronous telehealth model, we specifically examine the effects broadband 
access and physical access on synchronous telehealth uptake. Broadband quality directly 
determines individuals’ propensity to uptake synchronous telehealth services (O’Shea et al., 
2022). 
Table 12: ZCTA3 Model Specifications Synchronous Telehealth 
 Negative Binomial Regression 
 (1) (2) (3) (4) 
 β(SE) β(SE) β(SE) β(SE) 
Main Effects     
Consumer Broadband % 0.23885 (0.322) 0.21247 (0.325) 0.21058 (0.322) 0.19847 (0.324) 
Providers in Service Area -0.00021 (0.000) -0.00030 (0.000) -0.00051 (0.000)* -0.00048 (0.000)* 
Rural Healthcare Program [RHP 
Expansion] 
0.02457 (0.089) -0.09477 (0.109) -0.02486 (0.098) -0.10881 (0.112) 
     
Interactions     
RHP Expansion X Broadband Access   0.66369 (0.284)**   0.56698 (0.278)** 
RHP Expansion X Providers in 
Service Area 
    0.00008 (0.000) 0.00005 (0.000) 
     
RHP Controls     
RHP Applicants: Individual % 0.40005 (0.217)* 0.38389 (0.223)* 0.39003 (0.220)* 0.37994 (0.224)* 
Total RHP Funding -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) 
RHP Clinics % -0.69373 (0.474) -0.74386 (0.478) -0.62493 (0.469) -0.69293 (0.475) 
     
Constant 4.81533 (4.473) 5.68251 (4.460) 5.02475 (4.359) 5.68886 (4.397) 
ln Alpha [Dispersion Parameter] -1.83838 
(0.085)*** 
-1.84752 
(0.086)*** 
-1.84312 
(0.085)*** 
-1.84922 
(0.086)*** 
Demographic Controls Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes 
State FE Yes Yes Yes Yes 
Number of Observations 904 904 904 904 
* p<0.10, ** p<0.05, *** p<0.01 
We report the estimation results for the DiD specification highlighted in equation 2, which 
tests Hypothesis 1a and Hypothesis 1b. Given that our dependent variable, Synchronous Telehealth 
Visits, is a count variable that exhibits overdispersion—that is, its variance is significantly greater 
than its mean—we use an unconditional negative binomial regression in estimating the DiD 
specification. Column (1) in Table 12 provides a baseline main effects model that includes the 
complete set of controls, the main effects of broadband and primary care accessibility, Consumer 
Broadband Coverage and Service Area Providers, and the policy measure, RHP Expansion. 
Columns (2) and (3) hierarchically build upon the main effects model by including the interaction 
terms of the physical access measures with the policy measure, whereas column (4) represents the 
full model, which includes all of the interaction terms. We use the results from column (4) to 
interpret our hypotheses tests. 
Focusing on column (4), we find that the coefficient for RHP Expansion X Consumer 
Broadband % is positive and significant (β = 0.56698, p<0.05), suggesting that the effect of 
91 
 
consumer broadband coverage paired with improved primary care provider broadband coverage 
increases synchronous telehealth visits following the Rural Healthcare Program expansion in states 
with both payment and service parity. To better understand this result, we examine the average 
marginal effects in Table 13. This suggests that for every 1% increase in the number of Census 
blocks covered within a 3-digit ZCTA, there is a subsequent increase of approximately 1,185 
annual synchronous telehealth visits.  
We find that in states with payment parity, a combination of the Rural Healthcare Program 
and Consumer Broadband coverage increases the number of synchronous telehealth visits. Given 
that synchronous telehealth requires both parties to converse in real time, it becomes apparent that 
broadband is a critical component for ensuring the success of these interactions by simulating in-
person type experience.  
Table 13: ZCTA3 Marginal Effects for Synchronous Telehealth 
 Broadband Coverage Service Area Providers 
 β(SE) β(SE) 
Pre-Rural Healthcare Program 327.421 (534.830) -0.799 (0.438)* 
Post-Rural Healthcare Program 1512.450 (855.155)* -0.860 (0.531) 
   
Contrast 1185.028 (585.961)** -0.061 (0.120) 
Observations 904 904 
* p<0.10, ** p<0.05, *** p<0.01 
Next, we consider the impact of the availability of in-person primary care providers within 
a 30-mile radius. As seen in column (4), we find that the coefficient of RHP Expansion X Service 
Area Providers is statistically insignificant (β = 0.00005, p>0.10). Therefore, we do not find 
support for the H1b. However, previous studies suggest that the number of providers within a 
certain radius impacts patients’ willingness to uptake telehealth. However, we note that the main 
effect of Service Area Providers (β = -0.00048, p<0.10) Based on the marginal effects, this suggests 
that an increase of one primary care provider within a 30-mile radius would reduce synchronous 
telehealth visits by -0.799 prior to the expansion of the Rural Healthcare Program.  
 Overall, this may suggest that patients do not treat these services as substitutes, but rather 
as complements in most situations allowing for more frequent, consistent monitoring of health 
conditions amongst the patient population. With the increased accessibility brought about by the 
Rural Healthcare Expansion program, individuals are more willing to uptake synchronous 
telehealth services in conjunction with in-person visits, which is in line with previous studies 
(Bavafa, Hitt, and Terwiesch 2018). 
92 
 
5.6.3 Asynchronous Telehealth 
 In the asynchronous telehealth model, we specifically examine the effects broadband 
access and physical access on asynchronous telehealth uptake. Our analysis estimates DiD 
specification highlighted in equation 3. We note that asynchronous telehealth visits account for a 
much smaller portion of visits in our sample; however, it is important form of technology mediated 
access to healthcare given the less restrictive requirements related to broadband access. Focusing 
on column (4) in Table 14, we find that the coefficient for Consumer Broadband % is positive and 
significant (β = 2.44732, p<0.05), suggesting that the number of blocks within a 3-digit ZCTA with 
broadband is the primary factor impacting asynchronous telehealth uptake. To better understand 
this phenomenon, we examine the average marginal effects in Table 15. Prior to the Rural 
Healthcare Program, a 1% increase in blocks covered resulted in approximately 269 annual 
asynchronous telehealth visits. Following the expansion of the Rural Healthcare Program, we find 
that a 1% increase in blocks covered results in approximately 189 annual asynchronous telehealth 
visits. When examining the contrast, we find that the difference between the pre and post period is 
not significant. This suggests that changes in provider broadband do not appear to significantly 
impact the way patients uptake asynchronous telehealth services.  
Table 14: ZCTA3 Model Specifications Asynchronous Telehealth 
 Negative Binomial Regression 
 (1) (2) (3) (4) 
 β(SE) β(SE) β(SE) β(SE) 
Main Effects     
Consumer Broadband % 2.36492 (0.981)** 2.36525 (0.980)** 2.44171 (0.980)** 2.44732 (0.986)** 
Providers in Service Area -0.00053 (0.001) -0.00052 (0.001) 0.00002 (0.001) 0.00005 (0.001) 
Rural Healthcare Program [RHP 
Expansion] 
-0.30139 (0.267) -0.28578 (0.321) -0.22285 (0.287) -0.25212 (0.325) 
     
Interactions     
RHP Expansion X Broadband Access   -0.08237 (0.682)   0.19185 (0.755) 
RHP Expansion X Providers in 
Service Area 
    -0.00012 (0.000) -0.00013 (0.000) 
     
RHP Controls     
RHP Applicants: Individual % 0.78366 (0.492) 0.78814 (0.496) 0.81270 (0.492)* 0.80507 (0.494) 
Total RHP Funding -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) -0.00000 (0.000) 
RHP Clinics % 1.16877 (1.641) 1.17479 (1.640) 1.06883 (1.642) 1.04563 (1.638) 
     
Constant -3.45915 (12.443) -3.50209 (12.404) -4.09138 (12.769) -4.04691 (12.758) 
ln Alpha [Dispersion Parameter] -0.27082 
(0.098)*** 
-0.27091 
(0.098)*** 
-0.27300 
(0.098)*** 
-0.27301 
(0.098)*** 
Demographic Controls Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes 
State FE Yes Yes Yes Yes 
Number of Observations 904 904 904 904 
* p<0.10, ** p<0.05, *** p<0.01 
  
93 
 
Table 15: ZCTA3 Marginal Effects for Asynchronous Telehealth 
 Broadband Coverage 
 β(SE) 
Pre-Rural Healthcare Program 269.365 (134.737)** 
Post-Rural Healthcare Program 188.636 (105.539)* 
  
Contrast -80.729 (85.699) 
Observations 904 
* p<0.10, ** p<0.05, *** p<0.01 
We find that asynchronous telehealth is not dependent on parity or the Rural Healthcare 
Program expansion. As we hypothesized, asynchronous telehealth services do not require 
physicians to block time for their consultations with patients. Without a blocked time period, 
asynchronous telehealth visits are not disincentivized as there is no time trade off that exists with 
synchronous visits in comparison to in-person visits. In addition, by decoupling interactions 
between patients and providers in asynchronous care delivery, patients may not be dissatisfied with 
a provider having poor internet quality, as the uploading and downloading of associated documents 
will depend solely on the patient’s internet quality.  
5.7 Robustness Checks, Alternative Explanations, and Post Hoc Analysis 
5.7.1 Examination of the Parallel Trends Assumption 
 A key identifying assumption for the DiD specification examining the effects of consumer 
broadband coverage on telehealth uptake is the “parallel trends” assumption, which requires that, 
in the absence of the intervention, the differences in the outcomes across the treatment and control 
groups should be constant over time (Bertrand et al., 2004). To this end, Ryan et al. (2019, p. 3706) 
demonstrate that the use of a DiD specification with matching, as in this study, presents two 
advantages: (i) it lowers the mean squared error of the DID estimator and increases the likelihood 
that the true intervention effect is contained through the confidence interval of the estimator, and 
(ii) it renders the DID estimator less sensitive to deviations from the parallel trends assumption.  
 Nonetheless, we follow prior research (e.g., Hydari et al., 2019, Song et al., 2018) in 
carrying out a visual examination of the trends in the pre-intervention period, as well as a leads-
and-lags analysis to evaluate the parallel trends assumption. Figure 4 plots the trends in telehealth 
visits across the treatment and control groups before adjusting for covariates or fixed effects. We 
see that prior to the full expansion of the Rural Healthcare Program, the unadjusted trends in 
synchronous visits across the treatment and control groups appear to be parallel. To formally test 
whether these trends differed prior to the expansion of the Rural Healthcare Program, we estimate 
an econometric model that considers the year of intervention (i.e., 2018) as the base category and 
includes several dummy variables (i.e., leads and lags) with respect to the year of intervention as 
94 
 
well as the control variables. Such a model enables us to compare and test for differences in the 
slope of telehealth visits across the treatment and control groups before the intervention.  
Appendix C5 documents the analyses results. In column (1) of Appendix C5, we examine 
the pre-treatment and post-treatment periods. The coefficient Pre RHP Expansion is insignificant, 
suggesting the absence of pre-trends on telehealth visits. In column (2), we deconstruct the periods 
by the respective years. Consistent with our results in column (1), we do not find any significant 
effect prior to the expansion of the Rural Healthcare Program in column (2). Collectively, the 
results from the visual examination of the trends and the leads and lags analyses suggest that the 
identifying assumption of parallel trends is less likely to be violated in our sample.  
5.8 Discussion and Limitations 
 The need to understand the boundary conditions surrounding telehealth uptake has become 
increasingly important, especially considering the recent COVID-19 pandemic. Our study builds 
upon previous work and explores the impact of broadband coverage and physical availability of 
providers on telehealth usage. We have focused on the patient uptake of synchronous telehealth 
services based on the broadband coverage for both patients and providers alongside patients’ ability 
to physically access primary care providers within a timely fashion. While technologies like 
asynchronous services were developed to deal with bandwidth availability, the uptake of 
asynchronous services is still largely determined by the availability of fixed broadband services 
(Wilson & Maeder, 2015).  
Our results show that the operations strategy will play an important role in the expansion 
of telehealth. A core factor will be understanding how technology-mediated access occurs based 
on the modality and availability of the underlying broadband infrastructure and primary care 
availability. Technology-mediated access will be highly dependent on the way broadband is 
deployed to users, specifically technologies such as fiber. However, the availability of broadband 
alone will not be sufficient without a conducive legal environment ensuring payment parity and 
service parity for telehealth services. In the case of synchronous services, service parity and legal 
parity will be core components of ensuring the adoption of telehealth amongst patients and 
providers.  
5.8.1 Policy Implications 
From an operations strategy perspective, the relevant policy changes may depend on the 
type of telemedicine intervention being deployed to a specific geography. In the case of 
synchronous telehealth, the use of subsidies for providers and increased broadband access for 
patients may provide a suitable path forward. Given the coupled nature of synchronous and 
asynchronous services, broadband for both providers and patients is crucial to the deployment of 
95 
 
the technology. For the use of asynchronous services, structural policies related to increasing the 
deployment of fixed broadband may increase uptake. In effect, this direct-to-consumer modality 
depends on the broadband present for patients where they uptake asynchronous telehealth services. 
From a physician’s perspective, asynchronous telehealth decouples the need for in-person visits, 
which allows for greater flexibility to see patients.  
Given the recent pandemic, we note that reimbursement policies, coverage requirements, 
and encryption requirements have changed. As the government moved to make telemedicine more 
available, certain requirements were relaxed in order to increase uptake. However, when the 
COVID-19 pandemic ends, it is possible that the currently relaxed restrictions surrounding 
telehealth will be reinstated. If these restrictions remain permanently relaxed, it is likely that our 
study provides a conservative estimate of synchronous telehealth usage amongst patients. With 
regard to the Rural Healthcare Program, the funding cap has increased in the pandemic. Notably, 
in 2021, the funding cap increased by USD 612 million, indicating a clear desire by the federal 
government to expand uptake of telehealth by providers (American Hospital Association 2021). 
Furthermore, under the USD 45 billion Internet for All initiative, customers can be expected to 
have improved access to both synchronous and asynchronous telehealth-based care. 
In addition, we note that the results from this paper represent a conservative estimate given 
that the legal landscape has changed for telemedicine. Due to the pandemic, many states opted to 
relax restrictions related to the provision of telehealth services, and it remains to be seen whether 
states will continue to relax restrictions or return to a pre-pandemic legal environment. Notably, 
the recent announcement by the Biden-Harris administration regarding the USD 25 billion funding 
for broadband in the American Rescue Plan could dramatically change consumer access to fiber-
based broadband (The White House, 2022). However, the implementation of this infrastructure 
plan may take years, and therefore any impact is not likely to be immediately felt in terms of uptake. 
In addition, improvements to wireless broadband technology with later generations of 5G and early 
generations of 6G could make synchronous and asynchronous telehealth services more readily 
accessible. With 5G Advanced, the precursor to 6G, we may see an increased ability to deploy 
synchronous telehealth allowing for similar uptake to fixed broadband (Wang, 2022). From a 
broader supply chain perspective, large-scale structural and infrastructural policy changes related 
to telemedicine may explore alternative avenues for encouraging uptake by modality.  
5.8.2 Limitations and Future Research 
Like any study, this work has some limitations. First, we acknowledge that the geographic 
granularity of the data may constrain more nuanced analysis of individual usage patterns. 
96 
 
Furthermore, we acknowledge that the state-level controls for self-insured versus fully insured 
individuals would be better served with more nuanced controls of plan types, which might be 
achieved through the use of IRS Form 5500 requiring firms of a certain size to disclose benefits 
provided to employees. While we utilize a series of methods to control for the self-insured versus 
fully insured populations, improvements related to the type of insurance covering individuals may 
help to parse the differences between synchronous and asynchronous telehealth usage with regards 
to states with parity laws. However, we note that many states are now creating all-payer databases, 
which would allow for researchers in the future to better control for these plan types when 
examining telehealth uptake and usage. Furthermore, a notable limitation is the inability to account 
for digital literacy amongst the insured population. While we account for various age groups as a 
proxy, this may not capture the tech savviness of individuals within each of these age brackets.  
 Moving forward, new data on the horizon not only will be able to address the shortcomings 
of this paper but also may provide avenues for future research. Notably, as all-payer databases 
become available, improved data sets may allow for the exploration of additional modalities of 
telemedicine. Specifically, the use of remote monitoring could be examined to see whether it 
reduces the propensity for individuals to experience catastrophic health events by allowing for 
improved monitoring and maintenance of health conditions. In addition, remote monitoring may 
be considered alongside current queuing research to possibly address ways to account for arrivals 
at hospitals, which could improve scheduling and utilization of current healthcare resources. For 
example, the Department of Health and Human Services has provided guidelines for tele-triaging 
patients (Department of Health and Human Services, 2021). This would allow patients to enter their 
symptoms through a phone application or a kiosk at the hospital. A provider than would examine 
the input and make a recommendation regarding whether the patient should go to the nearest 
emergency department for care.  
 Future work might also build upon Sun et al.’s (2020) work related to telemedicine uptake 
and length of stay. This type of work could be extended to include an examination of payment and 
service parity alongside condition types. As all-payer databases become more readily available, it 
may also be possible to conduct a future study with higher geographic resolution, which represents 
a limitation of the current paper. Furthermore, the substitutable or complementary nature of 
telehealth services to in-person visits can be expanded upon as individual usage data for patients 
become available, allowing for the examination of condition types and timing of visits. This may 
help hospital systems and primary care providers to better understand the nature of these treatment 
modalities while helping policy makers understand the structural and infrastructural requirements 
related to the technology-mediated nature of telehealth services. In addition, widespread adoption 
97 
 
will likely come from studies focused on showing that in-person visits and telehealth visits, whether 
synchronous or asynchronous, yield similar results for individual patients. Future studies might 
also examine the cost savings associated with both synchronous and asynchronous telehealth visits 
in relation to in-person visits.    
98 
 
Chapter 6: Contribution 
Essay 1 Theoretical and Policy Implications 
 In Essay 1, I explore how the effects of built environment can impact patient outcomes in 
the form of inpatient stays. From a theoretical standpoint, we provide a possible explanation for the 
diminishing returns related to conformance quality (Johnson 1991, Mold 2010). Our results suggest 
that patient outcomes are dependent on both process conformance and the patients’ built 
environment. While reductionist methods have been empirically shown to improve process 
outcomes, accounting for the built environment may improve process outcomes by addressing the 
underlying heterogeneity in patients coming into the hospital for treatment (Ahn 2006a, Mold 2022). 
Our study builds on previous work related to the effects of surrounding environments on process 
outcomes (Zepeda and Sinha 2016, Muthulingam et al 2020). Furthermore, our work provides the 
foundation for future work related to pre-operative process conformance by pursuing a patient 
centric regimen of care accounting for the impact of built environment.  
 From a policy perspective, health care providers should consider how to address the 
impacts of built environment on hospital performance related to quality conformance programs. In 
relation to conformance quality programs, we note that publicly available data can help providers 
account for the built environment and deliver more patient centric care for their patient population. 
Publicly available data could also be used to determine which patients require a pre-operative 
intervention strategy to address for the effects of built environment similar to Paccagnella et al’s 
strategy (Paccagnella, Calò, Caenaro, Salandin, Simini, & Heymsfield 1994).  
To address the effects of built environment from a reimbursement perspective, 
policymakers instituting government value-based purchasing programs should consider weighting 
hospital performance based on the geographic areas served by hospital. Our results suggest that 
current conformance programs may currently be penalizing poorly performing hospitals that have 
conformed rigorously to process conformance guidelines by failing to account for the input 
heterogeneity related to the patients being served by said hospitals. Therefore, future programs 
related to process conformance should account for the variability in built environment 
characteristics of the patient populations served by hospitals. 
Essay 2 Theoretical and Policy Implications 
 In Essay 2, I disentangle financial and physical access following an expansion of healthcare 
coverage. In the past, research may have advocated for the market to determine the location and 
capacity characteristics for healthcare delivery based on market efficiency. While the markets may 
eventually achieve equilibrium, we note that policies creating large demand spikes require supply-
99 
 
chain-based policy solutions to adequately address demand in a timely fashion. 
Policymakers and stakeholders may consider possible levers to address the changes in 
demand resulting from a rapid expansion of healthcare coverage. For example, in vertically 
integrated health systems, primary care clinic hours could be extended to alleviate ED discharges. 
Additionally, state and local governments could use incentives and zoning to encourage 
independent physician groups to place their facilities along roadways leading to the nearest ED as 
a means of diverting nonurgent visits.  
Financial incentives can also provide an alternative lever to improve physical access to 
outpatient services. However, this is largely dependent on the manner in which financial levers are 
implemented. In some states, such as Oregon and Tennessee, administrators have set copayments 
as a method to discourage Medicaid patients from using the ED. While the efficacy of this approach 
has been debated in prior research (Mortensen 2010), it is important to note that the approach does 
not address the structural features of the current healthcare supply chain.  
In light of our findings relating to the important role played by differences in road distance 
and weekend hours of operation, hospital systems could consider locating primary care clinic 
capacity with provisions for increased levels of weekend hours of operation closer to EDs, allowing 
clinics to absorb nonurgent outpatient care. Alternatively, EDs could divert patients to a primary 
care clinic, assuming the hospital operates said facilities on its premises during the time of the 
encounter. Our results also suggest that additional Medicaid managed care patients at the nearest 
clinic result in increased ED discharges. To ensure the adequacy of the healthcare supply chain, 
policymakers could incentivize managed care organizations to increase access to primary care, 
either through the provision of new clinics within managed care networks, or by increasing the 
number of primary care providers within such networks.  
Essay 3 Theoretical and Policy Implications 
In Essay 3, I explore the effects of technology mediated access on telehealth uptake. Our 
study expands on the previous work by examining the tradeoffs individuals consider when making 
telehealth utilization decisions, specifically between distance and broadband access. While 
previous work has focused heavily on physical access related to travel, fewer studies have explored 
the impact of technology mediated access in the form of broadband internet accessibility. From a 
theoretical standpoint, we provide additional evidence that indicates telehealth services may serve 
as a complement rather than substitute to in-person visits, providing support for previous work 
related to the complementary effect of telemedicine (Bavafa, Hitt, & Terwiesch, 2018). This essay 
provides an initial framework for understanding the boundary conditions of physical access and 
100 
 
broadband access related to telehealth deployment when services are provided directly to the 
consumer. 
In the case of synchronous telehealth, the use of subsidies for providers and increasing 
broadband access for patients may provide a suitable path forward. Given the coupled nature of 
synchronous services, broadband is crucial for both providers and patients when deploying this 
technology. For the use of asynchronous services, structural policies related to increasing the 
deployment of fixed broadband may increase uptake. In effect, this direct-to-consumer modality 
depends on the broadband present for patients where they uptake asynchronous telehealth services. 
From a physician’s perspective, asynchronous telehealth decouples the need for in-person visits, 
which allows for greater flexibility to see patients.  
With regards to the future deployment of telemedicine, policymakers and providers will 
need to consider the structural and infrastructural characteristics related to the type of telemedicine 
modality (synchronous or asynchronous) being deployed. With regards to synchronous service, 
physicians may consider their access to broadband as well as their patient population’s broadband 
access before offering synchronous telehealth services.  With regards to asynchronous service, 
physicians may place greater weight on their patients’ access to broadband internet. The decoupled 
nature of asynchronous service means that patients’ experience will be largely dependent on their 
own broadband speed.  
In addition, this essay provides policymakers with insights into the deployment of 
telehealth following the Rural Healthcare Program Funding Cap Order. Overall, programs related 
to increasing broadband access to consumers is important for both the uptake of synchronous 
telehealth services and asynchronous telehealth services. When the goal is to expand synchronous 
telehealth services, our study shows that policymakers might focus on expanding Rural Healthcare 
Program funding to providers in states with both payment parity and service parity laws in place. 
101 
 
Bibliography 
2016 National Healthcare Quality and Disparities Report. Content last reviewed July 2017. 
Agency for Healthcare Research and Quality, Rockville, MD. 
http://www.ahrq.gov/research/findings/nhqrdr/nhqdr16/index.html 
Aboolian R, Berman O,  Verter V (2015) Maximal accessibility network design in the public 
sector. Transportation Science 50(1): 336-347. 
Abraham JM (2014) How might the Affordable Care Act's coverage expansion provisions influence 
demand for medical care? The Milbank Quarterly 92(1): 63-87. 
Adler-Milstein, J., Kvedar, J. and Bates, D.W., 2014. Telehealth among US hospitals: several 
factors, including state reimbursement and licensure policies, influence adoption. Health 
Affairs, 33(2), pp.207-215. 
Agnisarman, S., Narasimha, S., Madathil, K.C., Welch, B., Brinda, F.N.U., Ashok, A. and 
McElligott, J., 2017. Toward a more usable home-based video telemedicine system: a 
heuristic evaluation of the clinician user interfaces of home-based video telemedicine 
systems. JMIR human factors, 4(2), p.e7293. 
Ahn, A. C., Tewari, M., Poon, C. S., & Phillips, R. S. (2006). The limits of reductionism in 
medicine: could systems biology offer an alternative?. PLoS Med, 3(6), e208. 
Ahn, A. C., Tewari, M., Poon, C. S., & Phillips, R. S. (2006). The clinical applications of a systems 
approach. PLoS Med, 3(7), e209. 
Ahmadi-Javid A, Seyedi P, Syam SS (2017) A survey of healthcare facility location. Computers & 
Operations Research 79: 223-263. 
Allison PD, Waterman RP (2002) Fixed–effects negative binomial regression models. Sociological 
Methodology 32(1): 247-265. 
American Animal Hospital Association. Asynchronous vs. Synchronous. 
https://www.aaha.org/aaha-guidelines/telehealth-guidelines/considerations-for-choosing-
technology/asynchronous-vs.-synchronous/.  
American Hospital Association, 2021. FCC makes additional funding available for Rural Health 
Care Program | AHA News. (2021, June 24). American Hospital Association | AHA News. 
https://www.aha.org/news/headline/2021-06-24-fcc-makes-additional-funding-available-
rural-health-care-program 
American Medical Association, 2020. Physicians’ motivations and requirements for adopting 
digital health Adoption and attitudinal shifts from 2016 to 2019. 
Anderson M, Dobkin C, Gross T (2012) The effect of health insurance coverage on the use of medical 
services. American Economic Journal: Economic Policy 4(1): 1-27. 
Anderson, J. M., MacDonald, J. M., Bluthenthal, R., & Ashwood, J. S. (2013). Reducing crime 
by shaping the built environment with zoning: An empirical study of Los Angeles. 
University of Pennsylvania Law Review, 699-756. 
Anker, S. D., Koehler, F., & Abraham, W. T. (2011). Telemedicine and remote management of 
patients with heart failure. The Lancet, 378(9792), 731-739. 
Argote, L. (1982). Input uncertainty and organizational coordination in hospital emergency 
units. Administrative science quarterly, 420-434. 
Arora, S., Kalishman, S., Thornton, K., Dion, D., Murata, G., Deming, P., Parish, B., Brown, J., 
Komaromy, M., Colleran, K. and Bankhurst, A. (2010). Expanding access to hepatitis C 
virus treatment—Extension for Community Healthcare Outcomes (ECHO) project: 
disruptive innovation in specialty care. Hepatology, 52(3), pp.1124-1133. 
Artiga, S. and Hinton, E., 2019. Beyond health care: the role of social determinants in promoting 
health and health equity. Health, 20(10), pp.1-13. 
Asplin, B.R., Magid, D.J., Rhodes, K.V., Solberg, L.I., Lurie, N. and Camargo Jr, C.A., 2003. A 
conceptual model of emergency department crowding. Annals of emergency 
medicine, 42(2), pp.173-180. 
102 
 
Atasoy H, Chen P yu, Ganju K (2018) The Spillover Effects of Health IT Investments on Regional 
Healthcare Costs. Management Science 64(6):2515–2534. 
Barnett, M.L., Hsu, J. and McWilliams, J.M., 2015. Patient characteristics and differences in 
hospital readmission rates. JAMA internal medicine, 175(11), pp.1803-1812. 
Barton, J. (2018, December). Promising Telehealth Initiatives Highlight the Need to Close Digital 
Divide - Dallasfed.org. https://www.dallasfed.org/cd/pubs/telehealth.aspx#n2 
Barton, P.L., Brega, A.G., Devore, P.A., Mueller, K., Paulich, M.J., Floersch, N.R., Goodrich, G.K., 
Talkington, S.G., Bontrager, J., Grigsby, B. and Hrincevich, C. (2007). Specialist 
physicians' knowledge and beliefs about telemedicine: a comparison of users and nonusers 
of the technology. Telemedicine and e-Health, 13(5), pp.487-500. 
Barton, J. (2018, December). Promising Telehealth Initiatives Highlight the Need to Close Digital 
Divide - Dallasfed.org. https://www.dallasfed.org/cd/pubs/telehealth.aspx#n2 
Basu S, Berkowitz SA, Phillips RL, Bitton A, Landon BE, Phillips RS (2019) Association of 
primary care physician supply with population mortality in the United States, 2005-
2015. JAMA Internal Medicine 179(4): 506-514. 
Batt RJ, Terwiesch C (2017) Early task initiation and other load-adaptive mechanisms in the 
emergency department. Management Science 63(11): 3531-3551. 
Bavafa, H., Hitt, L.M. and Terwiesch, C., 2018. The impact of e-visits on visit frequencies and 
patient health: Evidence from primary care. Management Science, 64(12), pp.5461-5480. 
Bell, I.R., Caspi, O., Schwartz, G.E., Grant, K.L., Gaudet, T.W., Rychener, D., Maizes, V. and 
Weil, A., 2002. Integrative medicine and systemic outcomes research: issues in the 
emergence of a new model for primary health care. Archives of internal medicine, 162(2), 
pp.133-140. 
Benitez JA, Seiber EE (2018) US Health Care Reform and Rural America: Results From the ACA's 
Medicaid Expansions. The Journal of Rural Health 34(2): 213-222. 
Berman, A. N., Biery, D. W., Ginder, C., Singh, A., Baek, J., Wadhera, R. K., Wu, W. Y., 
Divakaran, S., DeFilippis, E. M., Hainer, J., Cannon, C. P., Plutzky, J., Polk, D. M., 
Nasir, K., Di Carli, M. F., Ash, A. S., Bhatt, D. L., & Blankstein, R. (2021). Association 
of Socioeconomic Disadvantage With Long-term Mortality After Myocardial Infarction: 
The Mass General Brigham YOUNG-MI Registry. JAMA cardiology, 6(8), 880–888. 
https://doi.org/10.1001/jamacardio.2021.0487 
Berchet C, Nader C (2016) The organisation of out-of-hours primary care in OECD countries. 
(89). 
Betcheva L, Erhun F, Jiang H (2020) Supply Chain Thinking in Healthcare: Lessons and 
Outlooks. Manufacturing & Service Operations Management. 
Berman O, Krass D, Menezes MB (2007) Facility reliability issues in network p-median 
problems: Strategic centralization and co-location effects. Operations Research 55(2): 
332-350. 
Bertrand M, Duflo E, Mullainathan S (2004) How Much Should We Trust Differences-In-
Differences Estimates?. The Quarterly Journal of Economics 119(1): 249–275. 
Bobrick, A. and Martire, G., 2021. Introducing physical warp drives. Classical and Quantum 
Gravity, 38(10), p.105009. 
Boothe, E., West, B., Hendon, L.G., Kaplan, J.D. and Kirmse, B., 2022. Asynchronous 
telemedicine for clinical genetics consultations in the NICU: A single center’s 
solution. Journal of Perinatology, 42(2), pp.262-268. 
Brown AM, Decker SL, Selck FW (2015) Emergency department visits and proximity to patients' 
residences, 2009-2010. US Department of Health and Human Services, Centers for Disease 
Control and Prevention, National Center for Health Statistics. 
103 
 
Brownson, R. C., Haire-Joshu, D., & Luke, D. A. (2006). Shaping the context of health: a review 
of environmental and policy approaches in the prevention of chronic diseases. Annu. 
Rev. Public Health, 27, 341-370. 
Buchmueller TC, Jacobson M, Wold C (2006) How far to the hospital?: The effect of hospital 
closures on access to care. Journal of Health Economics 25(4): 740-761. 
Butcher, L.M., 2014. Casimir energy of a long wormhole throat. Physical Review D, 90(2), 
p.024019. 
Card D, Dobkin C, Maestas N (2008) The impact of nearly universal coverage on health care utilization: 
Evidence from Medicare. American Economic Review 98: 2242-2258. 
Center for Connected Health Policy. (2019, February). Telehealth Reimbursement. Public Health 
Institute. Retrieved May 18, 2022, from https://cdn.cchpca.org/files/2019-
03/TELEHEALTH%20REIMBURSEMENT%202019%20FINAL.pdf 
Centers for Medicare and Medicaid Services. Baltimore, MD. Available at: 
https://www.medicaid.gov/medicaid/eligibility/index.html. Accessed November 16, 
2018. 
Chandrasekaran, A., Linderman, K., & Schroeder, R. (2015). The role of project and organizational 
context in managing high‐tech R&D projects. Production and Operations Management, 
24(4), 560-586. 
Chen C, Scheffler G, Chandra A (2011). Massachusetts' health care reform and emergency 
department utilization. New England Journal of Medicine 365(12): e25. 
Chen, H., Kwong, J.C., Copes, R., Tu, K., Villeneuve, P.J., Van Donkelaar, A., Hystad, P., Martin, 
R.V., Murray, B.J., Jessiman, B. and Wilton, A.S., 2017. Living near major roads and the 
incidence of dementia, Parkinson's disease, and multiple sclerosis: a population-based 
cohort study. The Lancet, 389(10070), pp.718-726. 
Chokshi DA, Rugge J, Shah NR (2014). Redesigning the regulatory framework for ambulatory care 
services in New York. The Milbank Quarterly 92(4): 776-795. 
Christakis, N.A. and Fowler, J.H., 2007. The spread of obesity in a large social network over 32 
years. New England journal of medicine, 357(4), pp.370-379. 
Chuo, J., Macy, M.L. and Lorch, S.A., 2020. Strategies for evaluating telehealth. Pediatrics, 146(5). 
Corburn, J., 2004. Confronting the challenges in reconnecting urban planning and public 
health. American journal of public health, 94(4), pp.541-546. 
Cram, P., G. Rosenthal, M. Vaughn-Sarrazin. 2005. Cardiac revascularization in specialty and 
general hospitals. New England J. Medicine 352(14) 1454–1462. 
Dai T, Tayur S (2019). Healthcare operations management: A snapshot of emerging 
research. Manufacturing & Service Operations Management 22(5): 869-887. 
Decker SL (2012). In 2011 nearly one-third of physicians said they would not accept new Medicaid 
patients, but rising fees may help. Health Affairs 31(8): 1673-1679. 
Department of Health and Human Services. (2021, February 9). Telehealth.HHS.gov. Retrieved 
December 13, 2022, from https://telehealth.hhs.gov/providers/telehealth-for-emergency-
departments/tele-triage/ 
Di, Q., Wang, Y., Zanobetti, A., Wang, Y., Koutrakis, P., Choirat, C., Dominici, F. and Schwartz, 
J.D., 2017. Air pollution and mortality in the Medicare population. New England Journal 
of Medicine, 376(26), pp.2513-2522. 
Dorn S, Wheaton L, Johnson P, Dubay L (2013) Using SNAP Receipt to Establish, Verify, and 
Renew Medicaid Eligibility. Washington, DC Urban Institute. http//www. urban. 
org/publications/412808. html. 
Dorsey, E.R. and Topol, E.J., 2016. State of telehealth. New England journal of medicine, 375(2), 
pp.154-161. 
104 
 
Drake, C., Zhang, Y., Chaiyachati, K.H. and Polsky, D., 2019. The limitations of poor broadband 
internet access for telemedicine use in rural America: an observational study. Annals of 
internal medicine, 171(5), pp.382-384. 
Edlin M (2012). Is California Ready for Millions of Newly Insured? | California Healthline. 
Retrieved (June 16, 2020), https://californiahealthline.org/news/is-california-ready-for-
millions-of-newly-insured/. 
Eibner C, Saltzman E (2015) How Does the ACA Individual Mandate Affect Enrollment and 
Premiums in the Individual Insurance Market. St. Monica, CA RAND Corporation. 
Einav L, Finkelstein A (2018) Moral hazard in health insurance: What we know and how we 
know it. Journal of the European Economic Association 16(4): 957-982. 
Federal Communications Commission, Connect2HealthFCC. Data. Accessed at 
www.fcc.gov/reports-research/maps/connect2health/data.html 
Federal Communications Commission, 2018. Fixed broadband deployment data from FCC form 
477. 
Federal Communications Commission. Rural Health Care Program. Retrieved May 18, 2022, 
from https://www.fcc.gov/general/rural-health-care-program 
Feinberg, G., 1967. Possibility of faster-than-light particles. Physical Review, 159(5), p.1089. 
Finkelstein A, Taubman SL, Wright B, Bernstein M, Gruber J, Newhouse JP, Allen H, Baicker 
K., Oregon Health Study Group (2012) The Oregon health insurance experiment: 
evidence from the first year. The Quarterly Journal of Economics, 127(3): 1057-1106. 
Finkelstein AN, Taubman SL, Allen HL, Wright BJ, Baicker K (2016) Effect of Medicaid 
coverage on ED use—further evidence from Oregon’s experiment. New England Journal 
of Medicine 375(16): 1505-1507. 
Flodgren G, Rachas A, Farmer AJ, Inzitari M, Shepperd S. Interactive telemedicine: effects on 
professional practice and health care outcomes. Cochrane Database Syst Rev. 2015 Sep 
7;2015(9):CD002098. doi: 10.1002/14651858.CD002098.pub2. PMID: 26343551; 
PMCID: PMC6473731. 
Freeman M, Robinson S, Scholtes S (2020) Gatekeeping, fast and slow: An empirical study of 
referral errors in the emergency department. Management Science. Articles in Advance. 
Fulton BD (2017) Health care market concentration trends in the United States: evidence and policy 
responses. Health Affairs 36(9):1530-1538. 
Garfield R, Hinton E, Cornachlone E, Hall C (2018) Medicaid Managed Care Plans and Access to 
Care: Results from the Kaiser Family Foundation 2017 Survey of Medicaid Managed Care 
Plans. Retrieved (May 16, 2020), https://www.kff.org/report-section/medicaid-managed-
care-plans-and-access-to-care-introduction/. 
Gentili M, Isett K, Serban N, Swann J (2015) Small-area estimation of spatial access to care and 
its implications for policy. Journal of Urban Health 92(5): 864-909. 
Gentili M, Harati P, Serban N, O'Connor J, Swann J (2018) Quantifying disparities in accessibility 
and availability of pediatric primary care across multiple states with implications for 
targeted interventions. Health Services Research 53(3): 1458-1477. 
Golberstein E, Gonzales G, Sommers BD (2015) California’s early ACA expansion increased 
coverage and reduced out-of-pocket spending for the state’s low-income population. 
Health Affairs 34(10): 1688-1694. 
Gold M, Paradise J (2012) Current and Emerging Issues in Medicaid Risk-Based Managed Care: 
Insights from an Expert Roundtable (No. 386449cbe42f40519071f6124fcf714b). 
Mathematica Policy Research. 
105 
 
Goodman DC, Mick SS, Bott D, Stukel T, Chang CH, Marth N, Poage J, Carretta HJ (2003) 
Primary care service areas: a new tool for the evaluation of primary care services. Health 
Services Research 38(1p1): 287-309. 
Green LV (2012) OM forum—The vital role of operations analysis in improving healthcare 
delivery. Manufacturing & Service Operations Management 14(4): 488-494. 
Gruber J, Sommers BD (2019) The Affordable Care Act's effects on patients, providers, and the 
economy: What we've learned so far. Journal of Policy Analysis and Management 38(4): 
1028-1052. 
Guadamuz, J.S., Alexander, G.C., Zenk, S.N. and Qato, D.M., 2020. Assessment of pharmacy 
closures in the United States from 2009 through 2015. JAMA internal medicine, 180(1), 
pp.157-160. 
Gumpert, K., 2015. Telehealth services becoming popular with US consumers and 
insurers. Reuters, 23, p. 2015. 
Güneş ED, Melo T, Nickel S (2019) Location problems in healthcare. Location Science (Springer 
International Publishing, Cham): 657–686 
Hall, J.L. and McGraw, D., 2014. For telehealth to succeed, privacy and security risks must be 
identified and addressed. Health Affairs, 33(2), pp.216-221. 
Handy, S. L., Boarnet, M. G., Ewing, R., & Killingsworth, R. E. (2002). How the built environment 
affects physical activity: views from urban planning. American journal of preventive 
medicine, 23(2), 64-73. 
Hansson, E., Mattisson, K., Björk, J., Östergren, P. O., & Jakobsson, K. (2011). Relationship 
between commuting and health outcomes in a cross-sectional population survey in 
southern Sweden. BMC public health, 11(1), 834. 
Hargraves J, Kennedy K (2018) ER facility prices grew in tandem with faster-growing charges 
from 2009-2016. Retrieved (July 20, 2020), https://healthcostinstitute.org/emergency-
room/er-facility-prices-charges-2009-2016. 
Hargraves J, Frost A (2018) Trends In Primary Care Visits - HCCI. Retrieved (July 20, 2020), 
https://healthcostinstitute.org/hcci-research/trends-in-primary-care-visits. 
Hayes RH, Pisano GP, Upton DM, Wheelwright SC (2005) Operations, Strategy, and 
Technology: Pursuing the Competitive Edge  
Hawn, M.T., Vick, C.C., Richman, J., Holman, W., Deierhoi, R.J., Graham, L.A., Henderson, 
W.G. and Itani, K.M., 2011. Surgical site infection prevention: time to move beyond the 
surgical care improvement program. Annals of surgery, 254(3), pp.494-501. 
Henceroth, S. (1978). An application of decision modeling to Indian health care. Interfaces, 9(1), 
18-24. 
Hoehner, C. M., Barlow, C. E., Allen, P., & Schootman, M. (2012). Commuting distance, 
cardiorespiratory fitness, and metabolic risk. American journal of preventive medicine, 
42(6), 571-578. 
Hofer AN, Abraham JM, Moscovice I (2011) Expansion of coverage under the Patient Protection 
and Affordable Care Act and primary care utilization. The Milbank Quarterly, 89(1), 
pp.69-89. 
Hollander, J., Ward, M., Alverson, D., Bashshur, R., Darkins, A. and DePhillips, H., 2017. Creating 
a framework to support measure development for telehealth. In Washington, DC: National 
Quality Forum. 
Hood, K.K. and Wong, J.J., 2022. Telehealth for people with diabetes: poised for a new 
approach. The Lancet Diabetes & Endocrinology, 10(1), pp.8-10. 
Hsia RY, Kanzaria HK, Srebotnjak T, Maselli J, McCulloch C, Auerbach AD (2012) Is emergency 
department closure resulting in increased distance to the nearest emergency department 
associated with increased inpatient mortality? Annals of Emergency Medicine 60(6): 707-715. 
106 
 
Hydari MZ, Telang R, Marella WM (2019) Saving Patient Ryan-Can Advanced Electronic Medical 
Records Make Patient Care Safer? Management Science 65(5):2041–2059. 
Iacus SM, King G, Porro G (2012) Causal inference without balance checking: Coarsened exact 
matching. Political Analysis: 1-24. 
Jackson, R.J., 2003. The impact of the built environment on health: an emerging field. American 
journal of public health, 93(9), pp.1382-1384. 
Johnson, H.A., 1991. Diminishing returns on the road to diagnostic certainty. Jama, 265(17), 
pp.2229-2231. 
Jonk, Y.C., Burgess, A., Williamson, M.E., Thayer, D., MacKenzie, J., McGuire, C., Fox, K. and 
Coburn, A.F., 2021. Telehealth Use in a Rural State: A Mixed‐Methods Study Using 
Maine's All‐Payer Claims Database. The Journal of Rural Health, 37(4), pp.769-779. 
Kane CK (2017) Policy Research Perspectives: Updated Data on Physician Practice Arrangements: 
Physician Ownership Drops Below 50 Percent. Chicago, American Medical Association. 
Kangovi S, Barg FK, Carter T, Long JA, Shannon R, Grande D (2013) Understanding why patients 
of low socioeconomic status prefer hospitals over ambulatory care. Health Affairs, 32(7), 
pp.1196-1203. 
Katz EB, Carrier ER, Umscheid CA, Pines JM (2012) Comparative effectiveness of care 
coordination interventions in the emergency department: a systematic review. Annals of 
emergency medicine 60(1): 12-23. 
Kaufman E, Rising K, Wiebe DJ, Ebler DJ, Crandall ML, Delgado MK (2016). Recurrent violent 
injury: magnitude, risk factors, and opportunities for intervention from a statewide 
analysis. The American Journal of Emergency Medicine 34(9): 1823-1830. 
KC, D. S., & Terwiesch, C. (2011). The effects of focus on performance: Evidence from California 
hospitals. Management Science, 57(11), 1897-1912. 
KC DS, Scholtes S, Terwiesch C (2020). Empirical research in healthcare operations: past research, 
present understanding, and future opportunities. Manufacturing & Service Operations 
Management 22(1): 73-83. 
Kelli, H.M., Kim, J.H., Samman Tahhan, A., Liu, C., Ko, Y.A., Hammadah, M., Sullivan, S., 
Sandesara, P., Alkhoder, A.A., Choudhary, F.K. and Gafeer, M.M., 2019. Living in food 
deserts and adverse cardiovascular outcomes in patients with cardiovascular 
disease. Journal of the American Heart Association, 8(4), p.e010694. 
Kim T, Diwas KC (2020) The impact of hospital advertising on patient demand and health 
outcomes. Marketing Science 39(3):612–635. 
Kindermann D, Mutter R, Pines JM (2013) Emergency department transfers to acute care facilities, 
2009: Statistical Brief# 155. 
Klink K (2015). Incentives for physicians to pursue primary care in the ACA era. AMA Journal of 
Ethics 17(7): 637. 
Koh HK, Sebelius KG (2010). Promoting prevention through the affordable care act. New England 
Journal of Medicine 363(14): 1296-1299. 
Kongstvedt PR (2013) Essentials of Managed Health Care. Jones & Bartlett Publishers. 
Kruse, C.S., Krowski, N., Rodriguez, B., Tran, L., Vela, J. and Brooks, M., 2017. Telehealth and 
patient satisfaction: a systematic review and narrative analysis. BMJ open, 7(8), p.e016242. 
Ladhania R, Haviland AM, Venkat A, Telang R, Pines JM (2019) The effect of Medicaid expansion 
on the nature of new enrollees’ emergency department use. Medical Care Research and 
Review 78(1): 24–35. 
Lee J (2018) Effects of health insurance coverage on risky behaviors. Health Economics 27(4): 
762-777. 
107 
 
Lentz, E.W., 2021. Breaking the warp barrier: hyper-fast solitons in Einstein–Maxwell-plasma 
theory. Classical and Quantum Gravity, 38(7), p.075015. 
Liddicoat, S., Badcock, P. and Killackey, E., 2020. Principles for designing the built environment 
of mental health services. The Lancet Psychiatry, 7(10), pp.915-920. 
Lopez-Zetina, J., Lee, H., & Friis, R. (2006). The link between obesity and the built environment. 
Evidence from an ecological analysis of obesity and vehicle miles of travel in California. 
Health & place, 12(4), 656-664. 
Marks C (2020) America’s Looming Primary-Care Crisis | The New Yorker. New Yorker. 
Retrieved (July 27, 2020), https://www.newyorker.com/science/medical-
dispatch/americas-looming-primary-care-crisis. 
McConville S, Raven MC, Sabbagh SH, Hsia RY (2018) Frequent emergency department users: a 
statewide comparison before and after affordable care act implementation. Health 
Affairs 37(6): 881-889. 
Mechanic, O.J., Persaud, Y. and Kimball, A.B., 2021. Telehealth systems. [Updated 2022 Sep 
12]. StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing. 
Medicare Payment Advisory Commission, 2016. Report to the Congress: reforming the delivery 
system. Medicare Payment Advisory Commission. 
Melnikow J, Evans E, Xing G, Durbin S, Ritley D, Daniels B, Woodworth L (2020) Primary Care 
Access to New Patient Appointments for California Medicaid Enrollees: A Simulated 
Patient Study. The Annals of Family Medicine 18(3): 210-217. 
Miller S (2012). The effect of insurance on emergency room visits: an analysis of the 2006 
Massachusetts health reform. Journal of Public Economics 96(11-12): 893-908. 
Miller S, Wherry LR (2017) Health and access to care during the first 2 years of the ACA Medicaid 
expansions. New England Journal of Medicine 376(10): 947-956. 
Millman, M. (Ed.). (1993). Access to health care in America. National Academies Press. 
Mold, J.W., Hamm, R.M. and McCarthy, L.H., 2010. The law of diminishing returns in clinical 
medicine: how much risk reduction is enough?. The Journal of the American Board of 
Family Medicine, 23(3), pp.371-375. 
Mold, J.W., 2022. Failure of the Problem-Oriented Medical Paradigm and a Person-Centered 
Alternative. The Annals of Family Medicine, 20(2), pp.145-148. 
Mortensen, K., 2010. Copayments did not reduce Medicaid enrollees’ nonemergency use of 
emergency departments. Health Affairs, 29(9), pp.1643-1650. 
Muthulingam, S., Dhanorkar, S. and Corbett, C.J., 2021. Does Water Scarcity Affect 
Environmental Performance? Evidence from Manufacturing Facilities in 
Texas. Management Science. 
National Institutes of Health. (2014, November 04). The National Prevention Strategy: 
Prioritizing Prevention to Improve the Nation's Health. Retrieved January 12, 2021, from 
https://prevention.nih.gov/education-training/methods-mind-gap/national-prevention-
strategy-prioritizing-prevention-improve-nations-health 
Negrin, M., & Ariane, D. V. (2012). Supreme Court Health Care Ruling: The Mandate Can 
Stay. ABC News. 
Nelson, C., West, T., & Goodman, C. (2005). The hospital built environment: what role might 
funders of health services research play. Rockville, MD: Agency for Healthcare Research 
and Quality. 
Nicholas, L.H., Osborne, N.H., Birkmeyer, J.D. and Dimick, J.B., 2010. Hospital process 
compliance and surgical outcomes in medicare beneficiaries. Archives of 
surgery, 145(10), pp.999-1004. 
108 
 
Nikpay S, Freedman S, Levy H, Buchmueller T (2017) Effect of the Affordable Care Act 
Medicaid expansion on emergency department visits: evidence from state-level 
emergency department databases. Annals of Emergency Medicine 70(2): 215-225. 
Obama B (2016) United States health care reform: progress to date and next steps. JAMA 316(5): 
525-532. 
O’Malley AS (2012) After-hours access to primary care practices linked with lower emergency 
department use and less unmet medical need. Health Affairs 32(1): 175-183. 
O’Shea, A.M., Baum, A., Haraldsson, B., Shahnazi, A., Augustine, M.R., Mulligan, K. and 
Kaboli, P.J., 2022. Association of Adequacy of Broadband Internet Service With Access 
to Primary Care in the Veterans Health Administration Before and During the COVID-19 
Pandemic. JAMA Network Open, 5(10), pp.e2236524-e2236524. 
Owens PL (AHRQ), Weiss AJ (IBM Watson Health), Barrett ML (M.L. Barrett, Inc.), Reid LD 
(AHRQ). Social Determinants of Health and County Population Rates of Opioid-Related 
Inpatient Stays and Emergency Department Visits, 2016. HCUP Statistical Brief #260. 
June 2020. Agency for Healthcare Research and Quality, Rockville, MD. www.hcup-
us.ahrq.gov/reports/statbriefs/sb260-SocialDeterminants-County-Opioid-Rates-Hospital-
Use-2016.pdf. 
Paccagnella, A., Calò, M. A., Caenaro, G., Salandin, V., Jus, P., Simini, G., & Heymsfield, S. B. 
(1994). Cardiac cachexia: preoperative and postoperative nutrition management. Journal 
of Parenteral and Enteral Nutrition, 18(5), 409-416. 
Pearson, T. A. (2011). Public policy approaches to the prevention of heart disease and stroke. 
Circulation, 124(23), 2560-2571. 
Perez, N.P., Ahmad, H., Alemayehu, H., Newman, E.A. and Reyes-Ferral, C., 2021. The impact of 
social determinants of health on the overall wellbeing of children: A review for the 
pediatric surgeon. Journal of pediatric surgery. 
Pines JM, Zocchi M, Moghtaderi A, Black B, Farmer SA, Hufstetler G, Klauer K, Pilgrim R 
(2016) Medicaid expansion in 2014 did not increase emergency department use but did 
change insurance payer mix. Health Affairs 35(8): 1480-1486. 
Polsky D, Candon M, Saloner B, Wissoker D, Hempstead K, Kenney GM, Rhodes K (2017) 
Changes in primary care access between 2012 and 2016 for new patients with Medicaid 
and private coverage. JAMA Internal Medicine 177(4): 588-590. 
Powell, R.E., Henstenburg, J.M., Cooper, G., Hollander, J.E. and Rising, K.L., 2017. Patient 
perceptions of telehealth primary care video visits. The Annals of Family 
Medicine, 15(3), pp.225-229. 
Rapaport L (2020) Even with insurance, fewer Americans seeing primary care providers. Reuters. 
Retrieved (July 27, 2020), https://www.reuters.com/article/us-health-access-primary-
care/even-with-insurance-fewer-americans-seeing-primary-care-providers-
idUSKBN1ZX2U5. 
Rauh, V. A., Landrigan, P. J., & Claudio, L. (2008). Housing and health. Annals of the New York 
Academy of Sciences, 1136(1), 276-288. 
Raven MC, Steiner F (2018) A national study of outpatient health care providers’ effect on 
emergency department visit acuity and likelihood of hospitalization. Annals of 
Emergency Medicine 71(6): 728-736. 
Reed, M.E., Huang, J., Graetz, I., Lee, C., Muelly, E., Kennedy, C. and Kim, E., 2020. Patient 
characteristics associated with choosing a telemedicine visit vs office visit with the same 
primary care clinicians. JAMA network open, 3(6), pp.e205873-e205873. 
109 
 
Rehse, D., Riordan, R., Rottke, N. and Zietz, J., 2019. The effects of uncertainty on market 
liquidity: Evidence from Hurricane Sandy. Journal of Financial Economics, 134(2), 
pp.318-332. 
Resneck, J.S., Abrouk, M., Steuer, M., Tam, A., Yen, A., Lee, I., Kovarik, C.L. and Edison, K.E. 
(2016). Choice, transparency, coordination, and quality among direct-to-consumer 
telemedicine websites and apps treating skin disease. JAMA dermatology.  
Rheuban, K., & Krupinski, E. (2018). Understanding Telehealth (3rd ed., McGraw-Hill's 
AccessMedicine). New York, N.Y: McGraw-Hill Education LLC. 
Rosenbaum S, Gruber J (2010) Buying Health Care, the Individual Mandate, and the Constitution. 
New England Journal of Medicine 363(5):401–403. 
Rosenbaum S, Paradise J, Markus A, Sharac J, Tran C, Reynolds D, Shin P (2017) Community 
Health Centers: Recent Growth and the Role of the ACA. Retrieved (May 14, 2020), 
http://files.kff.org/attachment/Issue-Brief-Community-Health-Centers-Recent-Growth-
and-the-Role-of-the-ACA. 
Roth, J. (2021, September 27). Omnichannel care delivery is becoming the next chapter of 
healthcare delivery. Microsoft Industry Blogs. https://cloudblogs.microsoft.com/industry-
blog/health/2021/09/28/omnichannel-care-delivery-is-becoming-the-next-chapter-of-
healthcare-delivery/ 
Ryan AM, Kontopantelis E, Linden A, Burgess Jr JF (2019) Now trending: Coping with non-
parallel trends in difference-in-differences analysis. Statistical Methods in Medical 
Research 28(12): 3697-3711. 
Sabbatini AK, Dugan J (2022) Medicaid expansion and avoidable emergency department use—
implications for US national and state government spending. JAMA Network Open 
5(6):e2216917. doi:10.1001/jamanetworkopen.2022.16917. 
Schadelbauer, R., 2017. Anticipating economic returns of rural telehealth. Arlington, VA: NTCA-
The Rural Broadband Association. 
Serway, R.A., Moses, C.J. and Moyer, C.A., 2004. Modern physics. Cengage Learning. 
Shin, Andy. “Why ‘ecosystems’ Will Be the 2020 Leading Health Care Buzzword | AHA 
News.” American Hospital Association, 7 Nov. 2019, 
www.aha.org/news/healthcareinnovation-thursday-blog/2019-11-07-why-ecosystems-
will-be-2020-leading-health-care. 
Sills, M.R., Hall, M., Colvin, J.D., Macy, M.L., Cutler, G.J., Bettenhausen, J.L., Morse, R.B., 
Auger, K.A., Raphael, J.L., Gottlieb, L.M. and Fieldston, E.S., 2016. Association of social 
determinants with children’s hospitals’ preventable readmissions performance. JAMA 
pediatrics, 170(4), pp.350-358 
Silverstein, M., Hsu, H.E. and Bell, A., 2019. Addressing social determinants to improve 
population health: the balance between clinical care and public health. Jama, 322(24), 
pp.2379-2380. 
Simpson, A. T. (2013). A brief history of NASA’s contributions to telemedicine. NASA 
https://www.nasa.gov/content/a-brief-history-of-nasa-s-contributions-to-
telemedicine/#_edn1 (Accessed December 20 2021).  
Sommers BD, Epstein AM (2010) Medicaid expansion—the soft underbelly of health care 
reform?. New England Journal of Medicine. 
Sommers BD, Simon K (2017) Health insurance and emergency department use-a complex 
relationship. The New England Journal of Medicine 376(18): 1708. 
Song H, Tucker AL, Murrell KL (2015) The Diseconomies of Queue Pooling: An Empirical 
Investigation of Emergency Department Length of Stay. Management Science 
61(12):3032–3053. 
110 
 
Song H, Tucker AL, Murrell KL, Vinsonc DR (2018) Closing the Productivity Gap: Improving 
Worker Productivity Through Public Relative Performance Feedback and Validation of 
Best Practices. Management Science 64(6):2628–2649. 
Sonier J, Boudreaux MH, Blewett LA (2013) Medicaid ‘welcome-mat’ effect of Affordable Care 
Act implementation could be substantial. Health Affairs 32(7): 1319-1325. 
Sun, S., Lu, S.F. and Rui, H., 2020. Does telemedicine reduce emergency room congestion? 
Evidence from New York State. Information Systems Research, 31(3), pp.972-986. 
Tater M, Paradise J, Garfield R (2016) Medi-Cal Managed Care: An Overview and Key Issues. 
Retrieved (March 14, 2020), https://www.kff.org/medicaid/issue-brief/medi-cal-
managed-care-an-overview-and-key-issues/. 
Tran, L.D., Rice, T.H., Ong, P.M., Banerjee, S., Liou, J. and Ponce, N.A., 2020. Impact of 
gentrification on adult mental health. Health services research, 55(3), pp.432-444. 
Truong, H.P., Luke, A.A., Hammond, G., Wadhera, R.K., Reidhead, M. and Maddox, K.E.J., 
2020. Utilization of Social Determinants of Health ICD-10 Z-Codes Among Hospitalized 
Patients in the United States, 2016–2017. Medical care, 58(12), pp.1037-1043. 
United Health Group (2019) 18 Million Avoidable Hospital Emergency Department Visits Add 
$32 Billion in Costs to the Health Care System Each Year. (July):2019. 
Centers for Disease Control and Prevention, National Center for Injury Prevention and Control 
(2012) National Action Plan for Child Injury Prevention: An Agenda to Prevent Injuries 
and Promote the Safety of Children and Adolescents in the United States  
U.S.  Congress, 2018. House Subcommittee on Oversight and Investigations; Committee on 
Energy and Commerce. Examining the Impact of Healthcare Consolidation. (HRG-2018-
HEC-191104; Date: Feb. 14, 2018) Text in: ProQuest® Congressional Hearings Digital 
Collection; Accessed: July 21, 2020. 
U.S. Congress, 2019. Senate Committee on Health, Education, Labor, and Pensions. 
How Primary Care Affects Health Care Costs and Outcomes. (S43-20190205-19995: Feb 
5, 2019) Text in: ProQuest® Congressional Hearings Digital Collection; Accessed: July 
27, 2020. 
Wadhera, R. K., Maddox, K. E. J., Wasfy, J. H., Haneuse, S., Shen, C., & Yeh, R. W. (2018). 
Association of the Hospital Readmissions Reduction Program with mortality among 
Medicare beneficiaries hospitalized for heart failure, acute myocardial infarction, and 
pneumonia. Jama, 320(24), 2542-2552. 
Wang, O. (2022, October 28). How 6G Can Transform The World and Technology. IEEE 
Standards Association. https://standards.ieee.org/beyond-standards/how-6g-can-
transform-the-world-and-technology/ 
White D, Crane S (2015) Crowded out: The outlook for state higher education spending. Moody’s 
Analytics. 
Whitman, A., De Lew, N., Chappel, A., Aysola, V., Zuckerman, R. and Sommers, B.D., 2022. 
Addressing Social Determinants of Health: Examples of Successful Evidence-Based 
Strategies and Current Federal Efforts. 
Wilson, L.S. and Maeder, A.J., 2015. Recent directions in telemedicine: review of trends in 
research and practice. Healthcare informatics research, 21(4), pp.213-222. 
The White House. Fact Sheet: Biden-Harris Administration Announces over $25 Billion in 
American Rescue Plan Funding to Help Ensure Every American Has Access to High 
Speed, Affordable Internet. 7 June 2022, Retrieved December 13, 2022 from 
https://www.whitehouse.gov/briefing-room/statements-releases/2022/06/07/fact-sheet-
111 
 
biden-harris-administration-announces-over-25-billion-in-american-rescue-plan-funding-
to-help-ensure-every-american-has-access-to-high-speed-affordable-internet/.  
Zachrison, K.S., Boggs, K.M., M Hayden, E., Espinola, J.A. and Camargo, C.A., 2020. A 
national survey of telemedicine use by US emergency departments. Journal of 
telemedicine and telecare, 26(5), pp.278-284. 
Zachrison, K.S., Richard, J.V. and Mehrotra, A., 2021, August. Paying for telemedicine in 
smaller rural hospitals: extending the technology to those who benefit most. In JAMA 
Health Forum (Vol. 2, No. 8, pp. e211570-e211570). American Medical Association. 
Zhang, D. J., Gurvich, I., Van Mieghem, J. A., Park, E., Young, R. S., & Williams, M. V. (2016). 
Hospital readmissions reduction program: An economic and operational analysis. 
Management Science, 62(11), 3351-3371. 
Zusman, E.E., Carr, S.J., Robinson, J., Kasirye, O., Zell, B., Miller, W.J., Duarte, T., Engel, A.B., 
Hernandez, M., Horton, M.B. and Williams, F., 2014. Moving toward implementation: 
The potential for accountable care organizations and private–public partnerships to 
advance active neighborhood design. Preventive medicine, 69, pp.S98-S101.
112 
Appendix 
Appendix A1. Patient Matching Variables: Variable Descriptions and Data Sources 
 
Variable Description Data Source 
Age Age of patient SPARCS 
Race Race of patient SPARCS 
Gender Gender of patient SPARCS 
% Receiving SSI % of families receiving Supplementary Security Income US Census ACS 
Inward Migration Number of individuals moving into a given ZCTA US Census ACS 
Population Density Population Density taken by dividing the ACS 3 Year estimate 
of the population by a county’s land area 
US Census ACS 
Median Household Income Median household income within a given ZCTA US Census ACS 
113 
Appendix B1. Constructing a Longitudinal Panel of Primary Care Clinics for Kentucky and 
North Carolina 
In the case of California and Florida, the state provides a comprehensive list of primary care clinics 
that were opened and closed by year. However, Kentucky and North Carolina do not provide similar 
lists. Therefore, we utilize a combination of web scrapping and cross validation to build our panel 
of clinics. We provide a step-by-step walkthrough of this process in greater detail below.  
 
Step 1. Selenium Base Web Scrapping of State Licensing Databases 
To begin, we need to know the physicians that are licensed to practice in Kentucky and North 
Carolina respectively. Therefore, we acquire state-specific information on physician licensing that 
includes licensing status and practice location. This information allows us to know when an 
individual was licensed by the state to practice and when their license was renewed or allowed to 
expire. Furthermore, both states require physicians to provide an address for their practice location. 
 
To acquire this information, we use the Python package Selenium, a package for automating web 
applications, paired with Chromium, an open-source web browser project developed by Google, to 
write and execute our web scraping code.7 Our code is written specifically for each state to account 
for the way information is displayed by state websites. The output is a CSV file that can easily be 
cleaned and updated should the database change at any time.  
 
With regards to the data collected, we gather data on physicians considered to be primary care 
providers as outlined by the Institute of Medicine.8 These specialties include family medicine, 
general internal medicine, general pediatrics, and obstetrics and gynecology. Our sample is reduced 
to physicians who were licensed to operate during our study period of interest, e.g., 2012 to 2016. 
We remove individuals who have MDs but are in the process of residency training that are presented 
in the system.  
 
Step 2. Validation of Locations and Operation 
One shortcoming of state licensing databases is that physicians may provide home addresses as 
opposed to their practice addresses. Therefore, we use a multistep approach to validate the 
addresses collected from the initial web scrapping step. This involves standardizing addresses, 
validating locations, imputing missing addresses, and confirming that the listed practice location is 
not a home address.  
 
Standardizing Addresses: To ensure that all our collected addresses are in the same machine 
readable format, we utilize the PostGrid API system. PostGrid is a United States Postal Services 
(USPS) Coding Accuracy Support System (CASS) certified third-party service. This system is used 
by the USPS and firms to ensure that the final delivery point for an item is a valid deliverable 
address. The PostGrid API leverages the USPS address database to standardize the collected 
addresses to USPS delivery format and return a zoning code for each address, e.g., residential or 
commercial. 
 
Home Address versus Practice Address: In order to identify the type of address provided, we use 
a combination of manual validation and the PostGrid API. This allows to quickly identify whether 
location addresses are commercially zoned or potentially a home address. However, we note that 
this database can struggle with areas that are zoned for mixed use, e.g. a building that has both 
 
7 Given that the information is considered to be in the public domain, our web scrapping process is legal 
under the United States Ninth Circuit’s decision in the case of hiQ Labs, Inc. versus LinkedIn Corp, 938F.3d 
985 of 2019. 
8Source: https://www.ncbi.nlm.nih.gov/books/NBK232631/ 
114 
commercial space and residential space. Therefore, we manually validate addresses that are zoned 
residential to ensure that primary care clinics located in buildings zoned for mixed use are not 
overlooked. This manual validation process involves the use of Google Maps, Google Street View 
Time Machine, Zillow, Redfin, and Realtor.com. If the home is residential; Zillow, Redfin, and 
Realtor.com will generally acknowledge that the building is for residential use. Alternatively, if the 
zoning has changed, these websites will provide an indicator that a property is zoned to allow for 
commercial use.  
 
If the building appears to be zoned for mixed use, e.g. a new apartment complex with commercial 
space on the first floor, we will validate whether the space has been listed for commercial use. 
Overall, these websites allow us to determine whether an address belongs to a mixed use 
development, has been zoned commercial, or is simply a home address.  
 
Missing Addresses: In some cases, physicians may not provide addresses to the state licensing 
boards. If an address of practice is not provided, we utilize the Center for Medicaid and Medicare 
Services (CMS) National Plan and Provider Enumeration System (NPPES) National Provider 
Identifier (NPI) database to find a possible address of practice. Given that we know providers’ 
identification numbers in the form of NPIs, we can then use the NPPES to find their listed practice 
address. 
 
This address is then cross validated against WebMD, Yellow Pages, Doximity, and Vitals using 
both currently available versions of the site and past archived version of webpages through the 
Wayback Machine Internet Archive. As an added step, we utilize business postings for clinics on 
Facebook to validate the location and hours of clinics. For physicians who have retired or passed 
away during our period of interest, we utilize local public records, such as obituaries, to identify 
and extract their final year of practice and the location of their practice. 
 
Step 3. Cross Validation of Primary Care Clinics 
 
Given that licensing databases rely on physicians to self-report their practice locations, we use 
historical business data to cross validate our sample of clinics and the years for which they operate. 
For the purpose of cross-validation, we utilize the DataAxle Places API built off the InfoGroup 
Historical Business dataset. The InfoGroup data set consists of millions of US businesses and other 
organizations by calendar year.9 For our sample of addresses, we retain all addresses that return the 
8-digit NAICS code 62111107 indicating that the aforementioned business is operated by a 
physician or surgeon. This cross validation step allows us to ensure that the addresses provided 
remain consistent through our panel duration. 
 
9  Harvard Dataverse. “Infogroup Us Historical Business Data.” Harvard Dataverse, 17 Apr. 2020, 
https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi%3A10.7910%2FDVN%2FPNOFKI.  
115 
Appendix B2. Variable Descriptions and Data Sources 
Variable Description Data Source(s) 
Dependent Variable   
ED Discharges Number of ED discharges in a ZCTA in a given year OSHPD, AHCA, 
AHRQ 
Independent Variables   
ACA Full Enforcement Coded as ‘1’ if a ZCTA in a given year implemented both Medicaid 
expansion and the Individual Mandate 
 
Distance Difference Difference in distances between a ZCTA centroid and the nearest 
primary care clinic and a ZCTA centroid and the nearest ED in a given 
year 
OSHPD, AHCA, 
US Census, 
ArcGIS Pro 2.8 
Network Analyst, 
Selenium-based 
web scrapping 
Average Weekday Out-of-
Hours 
Average weekday hours of operation outside of 8:00 AM and 6:00 PM 
on Monday to Friday for the nearest primary care clinic in a ZCTA in a 
given year. 
Google Places 
API, Selenium-
based web 
scrapping 
Average Weekend Hours Average hours of operation on weekends for the nearest primary care 
clinic in a ZCTA in a given year. 
Google Places 
API, Selenium-
based web 
scrapping 
Primary Care Provider Patient 
Panel 
Average number of unique patients (in 1000s) that were seen by a 
primary care provider at the nearest primary care clinic in a ZCTA in a 
given year. 
OSHPD 
Medicaid Fee-for-Service  Number of encounters (unique patients in 1000s) at the nearest primary 
care clinic in a ZCTA in a given year made by Medicaid patients 
enrolled in fee-for-service delivery. 
OSHPD 
Medicaid Managed Care  Number of encounters (unique patients in 1000s) at the nearest primary 
care clinic in a ZCTA in a given year made by Medicaid patients 
enrolled in managed care delivery. 
OSHPD 
   
Primary Care Clinic Control 
Variables 
  
Physician FTE Number of physician full time of equivalents at the nearest primary care 
clinic to a ZCTA in a given year  
OSHPD 
Other FTE Number of provider full time of equivalents that are not physicians at the 
nearest primary care clinic to a ZCTA in a given year 
OSHPD 
   
Physical Access Control 
Variables 
  
Base Distance Minimum distance traveled from a ZCTA centroid in a given year to 
receive outpatient care from either the nearest ED or the nearest primary 
care clinic.  
OSHPD, AHCA, 
Selenium-based 
web scrapping, 
ArcGIS Pro 2.8 
Network Analyst 
Relative Distance Travel distance between the nearest primary care clinic and the nearest 
ED in a ZCTA in a given year. 
OSHPD, AHCA, 
Selenium-based 
web scrapping, 
Google Maps API 
PCSA Clinic Density Number of primary care clinics per square mile in a Primary Care 
Service Area for a ZCTA in a given year. 
Dartmouth Atlas 
of Health Care, 
OSHPD, AHCA, 
HCUP, NPPES, 
Selenium-based 
webscrapping 
Urgent Care Clinic Density Number of urgent care facilities per square mile in a ZCTA in a given 
year. 
  
Department of 
Homeland 
Security, US 
Census ACS 
HSA Density Number of emergency departments per square mile in a Hospital Service 
Area for a ZCTA in a given year 
Dartmouth Atlas 
of Health Care, 
OSHPD, AHCA, 
HCUP, NPPES 
   
Insurance Control Variables   
Uninsured % % of the population that is uninsured in a ZCTA in a given year. US Census ACS 
Direct Purchase Insurance % % of the population insured through directly purchased insurance in a 
ZCTA in a given year. 
US Census ACS 
116 
Employer Based Insurance % % of the population insured by their employer in a ZCTA in a given 
year. 
US Census ACS 
Medicaid % % of the population insured by Medicaid in a ZCTA in a given year.  US Census ACS 
Medicare % % of the population insured by Medicare in a ZCTA in a given year. US Census ACS 
TRICARE % % of the population insured by TRICARE in a ZCTA in a given year. 
The TRICARE, administered by the US Department of Defense,  
provides healthcare benefits to active-duty service members, National 
Guard and Reserve members, and their families when services cannot be 
provided at a military treatment facility. 
US Census ACS 
VA % 
 
 
 
% of the population insured by Veterans Affairs in a ZCTA in a given 
year. The VA, administered by the US Department of Veterans Affairs, 
covers only veterans—those who once served in the military and are no 
longer on active duty or are retired—meeting certain eligibility and 
health criteria. 
US Census ACS 
Demographic Control 
Variables 
  
ln Median Household Income The natural log of  the median household income in a ZCTA in a given 
year. 
US Census ACS 
Hispanic % % of the population in a ZCTA in a given year that is Hispanic US Census ACS 
Black % % of the population in a ZCTA in a given year that is Black US Census ACS 
Native American % % of the population in a ZCTA in a given year that is Native American US Census ACS 
Asian % % of the population in a ZCTA in a given year that is Asian US Census ACS 
Other % % of the population in a ZCTA in a given year that belongs to racial 
category other than Hispanic, Black, Native American, or Asian. 
US Census ACS 
Inward Migration [Other 
States] 
Migration of individuals into a ZCTA in a given year from other states US Census ACS 
Below FPL 138% % of population living below 138% of the Federal Poverty Line in a 
ZCTA in a given year. 
US Census ACS 
Sex Ratio [Males to Females] Ratio of males to females in a ZCTA in a given year. US Census ACS 
Age: Under 25 Years % % of the population under 25 in a ZCTA in a given year. US Census ACS 
Age: 25-34 Years % % of the population between 25 and 34 in a ZCTA in a given year. US Census ACS 
Age: 35-44 Years % % of the population between 25 and 34 in a ZCTA in a given year. US Census ACS 
Age: 45-54 Years % % of the population between 25 and 34 in a ZCTA in a given year. US Census ACS 
Age: 55-64 Years % % of the population between 55 and 64 in a ZCTA in a given year. US Census ACS 
Age: Over 65 Years % % of the population over 65 in a ZCTA in a given year. US Census ACS 
Married % % of the population that is married in a ZCTA in a given year.  US Census ACS 
Separated/Divorced % % of the population in a ZCTA in a given year that is either separated or 
divorced. 
US Census ACS 
Widowed % % of the population that is widowed in a ZCTA in a given year. US Census ACS 
Citizens % % of the population with citizenship in a ZCTA in a given year. US Census ACS 
Average Family Size Average number of family members in a ZCTA in a given year. US Census ACS 
Education [25 and Over]: 
High School 
% of the population in a ZCTA in a given year whose highest level of 
education is a high school diploma or GED equivalent 
US Census ACS 
Education [25 and Over]: 
Some College 
% of the population in a ZCTA in a given year whose highest level of 
education is some college or an associate’s degree 
US Census ACS 
Education [25 and Over]: 
College 
% of the population in a ZCTA in a given year whose highest level of 
education is bachelor’s or greater 
US Census ACS 
Full Time Full Year % % of the population in a ZCTA in a given year working 35 hours or 
more year-round 
US Census ACS 
Full Time Part Year % % of the population in a ZCTA in a given year working 35 hours or 
more for less than 50 weeks 
US Census ACS 
Part Time % % of the population in a ZCTA in a given year working 1 to 34 hours 
per week 
US Census ACS 
 Note: OSHPD refers to California’s Office of Statewide Health and Planning; AHCA refers to Florida’s Agency for Health Care 
Administration; US Census ACS refers to the United States Census American Community Survey. 
Appendix B3. ZCTA Matching Variables: Variable Descriptions and Data Sources 
 
Variable Years Description Data Source 
Population Density 2011, 2012, 
2013 
Population Density taken by dividing the ACS 3 Year estimate 
of the population by a county’s land area 
US Census 
ACS 
Median Household 
Income 
2011, 2012, 
2013 
Median household income within a given county US Census 
ACS 
Median Age 2011, 2012, 
2013 
Median age within a given county US Census 
ACS 
117 
% Living in Poverty 2011, 2012, 
2013 
% of families living in poverty within a given county US Census 
ACS 
% Receiving SNAP 
Benefits 
2011, 2012, 
2013 
% of families receiving SNAP benefits within a given county US Census 
ACS 
 
  
118 
Appendix B4. Summary Statistics for Variables in DiD Specification 
 
 Mean SD Min Max 
Dependent and Independent Variables     
ED Discharges 9860.954 7492.513 114.000 49246.000 
Distance Difference -2.110 3.614 -37.974 40.209 
Average Weekday Out of Hours 0.930 1.612 0.000 14.200 
Average Weekend Hours 1.722 3.208 0.000 24.000 
     
Physical Access Control Variables     
Base Distance 3.864 4.590 0.002 62.139 
Distance Relative 4.974 6.574 0.000 95.728 
PCSA Density 1.719 14.135 0.000 555.202 
Urgent Care Clinic Density 6.617 28.220 0.000 371.043 
HSA Density 0.021 0.036 0.000 0.308 
     
Insurance Control Variables     
Uninsured % 0.150 0.069 0.000 0.415 
Direct Purchase Insurance % 0.134 0.058 0.000 0.436 
Employer Based Insurance % 0.509 0.130 0.000 0.837 
Medicaid % 0.174 0.104 0.000 0.681 
Medicare % 0.146 0.055 0.000 0.569 
TRICARE % 0.026 0.049 0.000 0.700 
VA Insurance % 0.019 0.013 0.000 0.107 
Medicaid Expansion Status 0.581 0.494 0.000 1.000 
     
Demographic Control Variables     
ln Median Household Income 10.975 0.373 9.217 12.228 
Population: Hispanic % 0.272 0.219 0.000 0.981 
Population: Black % 0.098 0.132 0.000 1.000 
Population: Native % 0.007 0.016 0.000 0.587 
Population: Asian % 0.083 0.110 0.000 0.710 
Population: Other % 0.078 0.096 0.000 0.686 
Inward Migration [Other States] 1063.867 2011.100 0.000 17591.000 
Below FPL138 % 0.219 0.115 0.001 1.000 
Sex Ratio [Females to Males] 99.069 47.054 28.100 3721.800 
Age: Under 25 % 0.328 0.074 0.045 0.829 
Age: 25 to 34 % 0.139 0.045 0.007 0.472 
Age: 35 to 44 % 0.133 0.025 0.000 0.273 
Age: 45 to 54 % 0.142 0.026 0.000 0.294 
Age: 55 to 64 % 0.121 0.031 0.000 0.299 
Widowed % 0.045 0.016 0.000 0.154 
Separated/Divorced % 0.106 0.029 0.000 0.273 
Married % 0.361 0.084 0.000 0.663 
Population: Citizens % 0.895 0.076 0.547 1.000 
Average Family Size 3.354 0.440 2.140 5.040 
Education [25 and Over]: High School 0.227 0.081 0.013 0.598 
Education [25 and Over]: Some College 0.216 0.052 0.032 0.508 
Education [25 and Over]: College 0.407 0.178 0.030 0.915 
Full Time Full Year % 0.456 0.079 0.000 0.816 
Full Time Part Year % 0.096 0.028 0.000 0.747 
Part Time % 0.216 0.141 0.000 0.896 
Observations 6725 
 
  
119 
Appendix B5. Summary Statistics for Variables in Fixed Effects Specification (California 
Subsample) 
 
 Mean SD Min Max 
Dependent and Independent 
Variables 
    
ED Discharges 9966.890 7543.620 398.000 49246.000 
Distance Difference -1.761 3.925 -37.974 40.209 
Average Weekday Out of Hours 0.496 1.303 0.000 14.000 
Average Weekend Hours 1.603 3.008 0.000 24.000 
Primary Care Provider Patient Panel 1.684 2.961 0.004 37.680 
Medicaid Fee for Service 3.108 6.202 0.000 105.737 
Medicaid Managed Care 4.830 7.537 0.000 70.190 
     
Physical Access Control Variables     
Base Distance 4.130 5.192 0.019 62.139 
Distance Relative 4.694 6.074 0.000 95.728 
PCSA Density 0.281 0.255 0.000 1.342 
Urgent Care Clinic Density 0.104 0.397 0.000 8.571 
HAS Density 0.025 0.039 0.000 0.263 
     
Insurance Control Variables     
Uninsured % 0.149 0.070 0.002 0.414 
Direct Purchase Insurance % 0.126 0.060 0.005 0.382 
Employer Based Insurance % 0.503 0.137 0.072 0.837 
Medicaid % 0.200 0.114 0.015 0.681 
Medicare % 0.139 0.052 0.008 0.503 
TRICARE % 0.018 0.032 0.000 0.700 
VA Insurance % 0.016 0.011 0.000 0.093 
Medicaid Expansion Status 0.960 0.196 0.000 1.000 
     
Demographic Control Variables     
ln Median Household Income 11.013 0.383 9.386 12.228 
Population: Hispanic % 0.344 0.229 0.011 0.981 
Population: Black % 0.059 0.088 0.000 0.863 
Population: Native % 0.009 0.012 0.000 0.182 
Population: Asian % 0.117 0.130 0.000 0.710 
Population: Other % 0.115 0.108 0.000 0.686 
Inward Migration [Other States] 401.930 402.204 0.000 4779.000 
Below FPL138 % 0.230 0.122 0.023 0.771 
Sex Ratio [Females to Males] 99.304 12.352 28.100 243.500 
Age: Under 25 % 0.334 0.075 0.078 0.690 
Age: 25 to 34 % 0.144 0.046 0.026 0.472 
Age: 35 to 44 % 0.134 0.025 0.031 0.265 
Age: 45 to 54 % 0.140 0.025 0.011 0.294 
Age: 55 to 64 % 0.119 0.033 0.015 0.299 
Widowed % 0.043 0.015 0.002 0.150 
Separated/Divorced % 0.101 0.028 0.014 0.273 
Married % 0.350 0.077 0.061 0.578 
Population: Citizens % 0.873 0.080 0.547 1.000 
Average Family Size 3.460 0.487 2.140 5.040 
Education [25 and Over]: High School 0.208 0.069 0.013 0.415 
Education [25 and Over]: Some 
College 
0.225 0.056 0.046 0.440 
Education [25 and Over]: College 0.389 0.188 0.030 0.915 
Full Time Full Year % 0.434 0.074 0.140 0.657 
Full Time Part Year % 0.099 0.031 0.000 0.327 
Part Time % 0.174 0.041 0.020 0.352 
Observations 3995 
120 
Appendix B6. Effects of Physical Access on ED Discharges: Neighborhood Environment 
Controls Added 
 
 Dependent Variable: ED Discharges 
 Negative Binomial Regression 
OLS 
Regression 
 (1) (2) (3) (4) (5) 
Main Effects      
Distance Difference 
0.00687 
(0.002)*** 
0.00747 
(0.002)*** 
0.00681 
(0.002)*** 
0.00740 
(0.002)*** 
0.00708 
(0.002)*** 
Average Weekday Out of Hours 
0.00278 
(0.002) 
0.00282 
(0.002) 
0.00270 
(0.002) 
0.00273 
(0.002) 
0.00232 
(0.002) 
Average Weekend Hours 
-0.00112 
(0.001) 
-0.00109 
(0.001) 
-0.00075 
(0.001) 
-0.00074 
(0.001) 
-0.00070 
(0.001) 
ACA Full Enforcement [ACA Full] 
0.05960 
(0.028)** 
0.05612 
(0.028)** 
0.06144 
(0.028)** 
0.05794 
(0.028)** 
0.06223 
(0.033)* 
      
Interactions      
ACA Full X Distance Difference   
-0.00215 
(0.001)*** 
  
-0.00210 
(0.001)*** 
-0.00189 
(0.001)** 
ACA Full X Average Weekday Out of 
Hours 
    
0.00229 
(0.003) 
0.00220 
(0.003) 
0.00272 
(0.003) 
ACA Full X Average Weekend Hours     
-0.00213 
(0.001)* 
-0.00203 
(0.001)* 
-0.00211 
(0.001)* 
      
Constant 
9.78942 
(0.596)*** 
9.72554 
(0.596)*** 
9.81645 
(0.595)*** 
9.75267 
(0.595)*** 
10.05800 
(0.628)*** 
Physical Access Controls Yes Yes Yes Yes Yes 
Insurance Controls Yes Yes Yes Yes Yes 
Demographic Controls Yes Yes Yes Yes Yes 
Neighborhood Environment Controls Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes Yes 
State FE Yes Yes Yes Yes Yes 
State X Year FE Yes Yes Yes Yes Yes 
Number of Observations 6725 6725 6725 6725 6725 
 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard errors clustered at the ZCTA level 
in parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations obtained using the Coarsened 
Exact Matching (CEM) procedure. 
  
121 
Appendix B7. Effects of Nearest Primary Care Clinic Capacity on ED Discharges:                          
Neighborhood Environment Controls Added 
 Dependent Variable: ED Discharges 
 Negative Binomial Regression 
OLS 
Regression 
 (1) (2) (3) (4) (5) 
Main Effects      
Distance Difference 
0.00687 
(0.002)*** 
0.00747 
(0.002)*** 
0.00681 
(0.002)*** 
0.00740 
(0.002)*** 
0.00708 
(0.002)*** 
Average Weekday Out of Hours 
0.00278  
(0.002) 
0.00282 
(0.002) 
0.00270 
(0.002) 
0.00273 
(0.002) 
0.00232 
(0.002) 
Average Weekend Hours 
-0.00112  
(0.001) 
-0.00109 
(0.001) 
-0.00075 
(0.001) 
-0.00074 
(0.001) 
-0.00070 
(0.001) 
ACA Full Enforcement [ACA Full] 
0.05960  
(0.028)** 
0.05612 
(0.028)** 
0.06144 
(0.028)** 
0.05794 
(0.028)** 
0.06223 
(0.033)* 
      
Interactions      
ACA Full X Distance Difference   
-0.00215 
(0.001)*** 
  
-0.00210 
(0.001)*** 
-0.00189 
(0.001)** 
ACA Full X Average Weekday Out of 
Hours 
    
0.00229 
(0.003) 
0.00220 
(0.003) 
0.00272 
(0.003) 
ACA Full X Average Weekend Hours     
-0.00213 
(0.001)* 
-0.00203 
(0.001)* 
-0.00211 
(0.001)* 
      
Constant 
9.78942 
(0.596)*** 
9.72554 
(0.596)*** 
9.81645 
(0.595)*** 
9.75267 
(0.595)*** 
10.05800 
(0.628)*** 
Physical Access Controls Yes Yes Yes Yes Yes 
Insurance Controls Yes Yes Yes Yes Yes 
Demographic Controls Yes Yes Yes Yes Yes 
Neighborhood Environment Controls Yes Yes Yes Yes Yes 
Year FE Yes Yes Yes Yes Yes 
ZCTA FE Yes Yes Yes Yes Yes 
State FE Yes Yes Yes Yes Yes 
State X Year FE Yes Yes Yes Yes Yes 
Number of Observations 6725 6725 6725 6725 6725 
 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard errors clustered at the ZCTA level in 
parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations (from California) obtained using the 
Coarsened Exact Matching (CEM) procedure. 
  
122 
Appendix B8. Addressing Alternative Explanations Relating to Unintentional Injury 
Mortality and Number of Uninsured Individuals 
 Dependent Variable:  
Unintentional Injury Mortality 
Dependent Variable: 
Number of Uninsured 
Individuals 
 OLS Regression Negative Binomial Regression 
 (1) (2) 
Main Effects   
Distance Difference -0.00065 (0.001) -0.00279 (0.002) 
Average Weekday Out of Hours 0.00059 (0.001) -0.00137 (0.005) 
Average Weekend Hours -0.00098 (0.001)* 0.00149 (0.002) 
ACA Full Enforcement [ACA Full] -0.01891 (0.004)***   
   
Interactions   
ACA Full X Distance Difference -0.00016 (0.000)   
ACA Full X Average Weekday Out of Hours 0.00042 (0.001)   
ACA Full X Average Weekend Hours 0.00005 (0.000)   
Number of Tax Returns with IM Penalty   -0.00004 (0.000)*** 
Average Tax Penalty in Dollars   0.03447 (0.062) 
   
Constant 0.43709 (0.396) 8.66133 (1.017)*** 
Physical Access Controls Yes Yes 
Insurance Controls Yes Yes 
Demographic Controls Yes Yes 
Year FE Yes Yes 
ZCTA FE Yes Yes 
State FE Yes Yes 
State X Year FE Yes Yes 
Number of Observations 6048 4089 
 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard errors clustered at the ZCTA level 
in parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations obtained using the Coarsened 
Exact Matching (CEM) procedure. The number of observations in column (1) are lower because of small cell suppression (small cell 
suppression is done when a ZCTA in a given year observes a value relating to mortality, admissions, discharges, or patient services 
between 1 and 10. This is done to protect the confidentiality of individuals as smaller values could potentially lead to identification of 
such individuals). The number of observations in column  (2) is lower because we  use  the sample beginning from 2014 when the 
Individual Mandate took effect.
123 
Appendix B9. Controlling for Transportation Modality and Vehicle Ownership 
 Dependent Variable: ED Discharges 
 (1) (2) 
Main Effects   
Distance Difference 0.00751 (0.002)*** 0.00736 (0.002)*** 
Average Weekday Out of Hours 0.00259 (0.002) 0.00283 (0.002) 
Average Weekend Hours -0.00068 (0.001) -0.00075 (0.001) 
ACA Full Enforcement [ACA Full] 0.05826 (0.028)** 0.05624 (0.028)** 
   
Interactions   
ACA Full X Distance Difference -0.00213 (0.001)*** -0.00214 (0.001)*** 
ACA Full X Average Weekday Out of 
Hours 
0.00277 (0.003) 0.00238 (0.003) 
ACA Full X Average Weekend Hours -0.00235 (0.001)** -0.00213 (0.001)** 
   
Mobility Controls   
Car/Truck/Van: Driving Alone% 0.07709 (0.101)   
Car/Truck/Van: Carpooling % 0.08939 (0.125)   
Mass Transit % -0.43258 (0.192)**   
Walking % 0.28652 (0.171)*   
Other Transit Modality % 0.11230 (0.248)   
Households: One Vehicle %   -0.03069 (0.168) 
Households: Two Vehicles %   -0.12869 (0.161) 
Households: Three Vehicles %   -0.05337 (0.173) 
Households: Four or More Vehicles %   -0.04047 (0.209) 
   
Constant 9.60138 (0.609)*** 9.70621 (0.591)*** 
Physical Access Controls Yes Yes 
Insurance Controls Yes Yes 
Demographic Controls Yes Yes 
Year FE Yes Yes 
ZCTA FE Yes Yes 
State FE Yes Yes 
State X Year FE Yes Yes 
Number of Observations 6725 6725 
 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard 
errors clustered at the ZCTA level in parentheses. For estimation, we use a matched sample of treated 
and control ZCTA-year observations (from California) obtained using the Coarsened Exact Matching 
(CEM) procedure. 
  
124 
Appendix B10. Using Alternative Measures of Distance Difference 
Our study builds upon previous literature using centroids as a proxy for individual location when 
calculating the approximate distance between two points (Gentili et al. 2015, Gentili et al. 2018). 
We acknowledge that this measurement approach may not result in the most accurate measurements 
for certain scenarios where, at the individual level, the distance difference metric may understate 
or overstate the actual distance an individual may need to travel to the ED compared to the primary 
care clinic. Therefore, we run our analysis by re-estimating the measure of Distance Difference in 
two distinct ways: 
 
• Using census block levels, and  
• Using population-weighted centroid at the ZCTA level 
 
1) Calculating Distance Difference Using Census Blocks 
First, we use a more granular approach to calculating distance at the census block level and 
subsequently aggregating it to the ZCTA level. Census blocks, also known as tabulation blocks, 
are statistical areas bounded by both visible and non-visible features (Rossiter 2011). Visible 
features include roads, streams, and railroad tracks.  Non-visible boundaries include property lines, 
city, township, school district, county limits, and short line-of-sight extensions of roads. To provide 
further context, we provide the hierarchy of census geographic entities below from Rossiter 2014 
in Figure A1 below. As seen from the figure, census Blocks represent the lowest level of 
geographical resolution. A visual comparison of the geographical resolution of census blocks vs 
that of ZCTAs in California in Figure A2 further illustrates this and this is confirmed from the 
number of census blocks across states, as shown in Table A1.   
 
Figure B1: Higher Geographic Resolution of Census Blocks (compared to ZCTAs): 
Example of California 
  
 
Table B1: Number of Census Blocks Across States and Across Years 
 
 2012 2013 2014 2015 2016 
California 713,009 714,845 710,145 710,145 710,145 
Florida 488,553 490,815 484,481 484,481 484,481 
Kentucky 163,811 164,704 161,672 161,672 161,672 
North Carolina 292,784 296,602 288,987 288,987 288,987 
 
ZCTAs  Census Blocks 
125 
We measure the distance from the centroid of each census block within a ZCTA to the nearest 
provider, whether that may be the nearest primary care clinic or the nearest ED. We utilize these 
distances to calculate the Distance Difference measured at each census block level. These values 
are then aggregated to the ZCTA level where we rerun our main model and compare the results. 
Overall, this analysis was an intensive exercise requiring the distance calculation involving a total 
of 16,488,718 routes (i.e., routes to the nearest ED and nearest primary care clinic from the centroid 
of each census block) to reconstruct our distances across our study period from 2012 to 2016, and 
then subsequently aggregate this back up to the ZCTA level.  
 
2) Calculating Distance Difference Using Population-Weighted Centroid at ZCTA Level 
Second, given that the population within a given ZCTA may be unevenly distributed and that the 
geographical centroid of a ZCTA may not accurately represent physical access in such cases, we 
use a population-weighted centroid at the ZCTA level as an alternative to calculate Distance 
Difference.  As Hall et al. (2019) note, population-weighted centroids can enable greater precision 
in the calculation of distances as they capture the average access distances from the center of the 
population within a ZCTA instead of the geographic center, where the population may not be 
located.  
 
We measure distance from the population weighted centroid of each ZCTA to the nearest provider, 
whether that might be nearest primary care clinic or the nearest ED. We utilize these distances to 
calculate the Distance Difference measured from each population weighted centroid. The 
population weighted centroid is constructed using the population estimates for census tabulation 
blocks. For this procedure, we utilize the Mean Center function located inside the ArcGIS Pro 
Spatial Statistics Tools toolbox. While population data at the census block level is available for 
each Decennial Census level, i.e., 2000, 2010, and 2020, there are no American Community Survey 
5-year estimates for population available at the block level. Therefore, we construct the population 
weighted centroids using the 2010 census block population estimates.  
As Table A2 indicates, we find that our results remain consistent even when using these two 
alternative ways for calculating Distance Difference.  
Table B2: Using Alternative Measures of Distance Difference 
 
 Dependent Variable: ED Discharges 
 
Using Census Block 
(1)  
Using Population Weighted 
ZCTA Centroid 
(2) 
Main Effects   
Distance Difference 0.00223 (0.001)***  0.00626 (0.002)*** 
Average Weekday Out of Hours 0.00283 (0.002) 0.00274 (0.002) 
Average Weekend Hours -0.00072 (0.001) -0.00072 (0.001) 
ACA Full Enforcement [ACA Full] 0.05582 (0.028)** 0.05631 (0.028)** 
   
Interactions   
ACA Full X Distance Difference -0.00175 (0.001)* -0.00200 (0.001)***  
ACA Full X Average Weekday Out of Hours 0.00249 (0.003) 0.00255 (0.003) 
ACA Full X Average Weekend Hours -0.00228 (0.001)** -0.00224 (0.001)** 
   
Physical Access Controls Yes Yes 
Insurance Controls Yes Yes 
Demographic Controls Yes Yes 
Year FE Yes Yes 
ZCTA FE Yes Yes 
126 
State FE Yes Yes 
State X Year FE Yes Yes 
Number of Observations 6725 6725 
* p<0.10, ** p<0.05, *** p<0.01. All model specifications included heteroskedastic-robust standard errors clustered at 
the ZCTA level in parentheses. For estimation, we use a matched sample of treated and control ZCTA-year observations 
obtained using the Coarsened Exact Matching (CEM) procedure. 
127 
Appendix C1. Parity Laws by State 
State Payment Parity Law Service Parity Law 
Alabama No legal statute No legal statute 
Arizona 
2013 Arizona Revised Statutes 
Title 20 - Insurance 
§ 20-841.09  
 
A. All contracts issued, delivered or renewed on or after 
January 1, 2015 must provide coverage for health care 
services that are provided through telemedicine if the health 
care service would be covered were it provided through in-
person consultation between the subscriber and a health 
care provider and provided to a subscriber receiving the 
service in a rural region of this state. The contract may limit 
the coverage to those health care providers who are 
members of the corporation's provider network. 
 
2015 Arizona Revised Statutes 
Title 20 - Insurance 
§ 20-1057.13 Telemedicine; coverage of health care 
services; definitions 
 
A. An evidence of coverage issued, delivered or renewed by 
a health care services organization on or after January 1, 
2015 must provide coverage for health care services that are 
provided through telemedicine if the health care service 
would be covered were it provided through in-person 
consultation between the enrollee and a health care provider 
and provided to an enrollee receiving the service in a rural 
region of this state. The evidence of coverage may limit the 
coverage to those health care providers who are members of 
the health care services organization's provider network. 
2013 Arizona Revised Statutes 
Title 20 - Insurance 
§ 20-841.09  
 
A. All contracts issued, delivered or renewed on or after 
January 1, 2015 must provide coverage for health care 
services that are provided through telemedicine if the health 
care service would be covered were it provided through in-
person consultation between the subscriber and a health 
care provider and provided to a subscriber receiving the 
service in a rural region of this state. The contract may limit 
the coverage to those health care providers who are 
members of the corporation's provider network. 
 
2015 Arizona Revised Statutes 
Title 20 - Insurance 
§ 20-1057.13 Telemedicine; coverage of health care 
services; definitions 
 
A. An evidence of coverage issued, delivered or renewed by 
a health care services organization on or after January 1, 
2015 must provide coverage for health care services that are 
provided through telemedicine if the health care service 
would be covered were it provided through in-person 
consultation between the enrollee and a health care provider 
and provided to an enrollee receiving the service in a rural 
region of this state. The evidence of coverage may limit the 
coverage to those health care providers who are members of 
the health care services organization's provider network. 
Arkansas 
2016 Arkansas Code 
Title 23 - Public Utilities and Regulated Industries 
Subtitle 3 - Insurance 
Chapter 79 - Insurance Policies Generally 
Subchapter 16 - -- Coverage for Services Provided Through 
Telemedicine 
§ 23-79-1602. Coverage for telemedicine 
 
(2) Subject to subdivision (d)(1) of this section, a health 
benefit plan shall reimburse a physician licensed by the 
board for healthcare services provided through telemedicine 
on the same basis as the health benefit plan reimburses a 
physician for the same healthcare services provided in 
person. 
2016 Arkansas Code 
Title 23 - Public Utilities and Regulated Industries 
Subtitle 3 - Insurance 
Chapter 79 - Insurance Policies Generally 
Subchapter 16 - -- Coverage for Services Provided Through 
Telemedicine 
§ 23-79-1602. Coverage for telemedicine 
 
(c) (1) A health benefit plan shall cover the services of a 
physician who is licensed by the Arkansas State Medical 
Board for healthcare services through telemedicine on the 
same basis as the health benefit plan provides coverage for 
the same healthcare services provided by the physician in 
person. 
California 
Telehealth Advancement Act of 2011 (AB 415) 
Section 1374.13 
(d) No health care service plan shall limit the type of setting 
CA Health & Safety Code § 1374.13 
(b) It is the intent of the Legislature to recognize the 
practice of telehealth as a legitimate means by which an 
128 
where services are provided for the patient or by the health 
care provider before payment is made for the covered 
services 
appropriately provided through telehealth, subject to the 
terms and conditions of the contract entered into between 
the enrollee or subscriber and the health care service plan, 
and between the health care service plan and its 
participating providers or provider groups. 
 
individual may receive health care services from a health 
care provider without in-person contact with the health care 
provider. 
 
(c) No health care service plan shall require that in-person 
contact occur between a health care provider and a patient 
before payment is made for the covered services 
appropriately provided through telehealth, subject to the 
terms and conditions of the contract entered into between 
the enrollee or subscriber and the health care service plan, 
and between the health care service plan and its 
participating providers or provider groups. 
 
(d) No health care service plan shall limit the type of setting 
where services are provided for the patient or by the health 
care provider before payment is made for the covered 
services appropriately provided through telehealth, subject 
to the terms and conditions of the contract entered into 
between the enrollee or subscriber and the health care 
service plan, and between the health care service plan and 
its participating providers or provider groups. 
 
Colorado 
2016 Colorado Revised Statutes 
Title 10 - Insurance 
Health Care Coverage 
Article 16 - Health Care Coverage 
Part 1 - General Provisions 
§ 10-16-123. Telehealth - definitions 
 
(b) Subject to all terms and conditions of the health benefit 
plan, a carrier shall reimburse the treating participating 
provider or the consulting participating provider for the 
diagnosis, consultation, or treatment of the covered person 
delivered through telehealth on the same basis that the 
carrier is responsible for reimbursing that provider for the 
provision of the same service through in-person 
consultation or contact by that provider. A carrier shall not 
deny coverage of a health care service that is a covered 
benefit because the service is provided through telehealth 
rather than in-person consultation or contact between the 
participating provider or, subject to section 10-16-704, the 
nonparticipating provider and the covered person where the 
health care service is appropriately provided through 
telehealth. Section 10-16-704 applies to this paragraph (b). 
2016 Colorado Revised Statutes 
Title 10 - Insurance 
Health Care Coverage 
Article 16 - Health Care Coverage 
Part 1 - General Provisions 
§ 10-16-123. Telehealth – definitions 
 
(2) On or after January 1, 2002, no health benefit plan that 
is issued, amended, or renewed for a person residing in a 
county with one hundred fifty thousand or fewer residents 
may require face-to-face contact between a provider and a 
covered person for services appropriately provided through 
telemedicine, pursuant to section 12-36-106 (1) (g), C.R.S., 
subject to all terms and conditions of the health benefit 
plan, if such county has the technology necessary for the 
provision of telemedicine. Any health benefits provided 
through telemedicine shall meet the same standard of care 
as for in-person care. Nothing in this section shall require 
the use of telemedicine when in-person care by a 
participating provider is available to a covered person 
within the carrier's network and within the member's 
geographic area. 
 
Delaware 
2015 Delaware Code 
Title 18 - Insurance Code 
CHAPTER 33. HEALTH INSURANCE CONTRACTS 
§ 3370 Telemedicine 
2015 Delaware Code 
Title 18 - Insurance Code 
CHAPTER 33. HEALTH INSURANCE CONTRACTS 
§ 3370 Telemedicine 
129 
(e) An insurer, health service corporation, or health 
maintenance organization shall reimburse the treating 
provider or the consulting provider for the diagnosis, 
consultation, or treatment of the insured delivered through 
telemedicine services on the same basis and at least at the 
rate that the insurer, health service corporation, or health 
maintenance organization is responsible for coverage for 
the provision of the same service through in-person 
consultation or contact. Payment for telemedicine 
interactions shall include reasonable compensation to the 
originating or distant site for the transmission cost incurred 
during the delivery of health-care services. 
(d) An insurer, health service corporation, or health 
maintenance organization shall not exclude a service for 
coverage solely because the service is provided through 
telemedicine services and is not provided through in-person 
consultation or contact between a health-care provider and a 
patient for services appropriately provided through 
telemedicine services. 
Florida No legal statute No legal statute 
Georgia 
O.C.G.A. 33-24-56.4 (2010) 
33-24-56.4. Payment for telemedicine services 
(c) It is the intent of the General Assembly to mitigate 
geographic discrimination in the delivery of health care by 
recognizing the application of and payment for covered 
medical care provided by means of telemedicine, provided 
that such services are provided by a physician or by another 
health care practitioner or professional acting within the 
scope of practice of such health care practitioner or 
professional and in accordance with the provisions of Code 
Section 43-34-31. 
O.C.G.A. 33-24-56.4 (2010) 
33-24-56.4. Payment for telemedicine services 
(d) On and after July 1, 2005, every health benefit policy 
that is issued, amended, or renewed shall include payment 
for services that are covered under such health benefit 
policy and are appropriately provided through telemedicine 
in accordance with Code Section 43-34-31 and generally 
accepted health care practices and standards prevailing in 
the applicable professional community at the time the 
services were provided. The coverage required in this Code 
section may be subject to all terms and conditions of the 
applicable health benefit plan. 
Hawaii 
2013 Hawaii Revised Statutes 
TITLE 24. INSURANCE 
431. Insurance Code 
431:10A-116.3 Coverage for telehealth. 
 
2016 Hawaii Revised Statutes 
TITLE 24. INSURANCE 
432D. Health Maintenance Organization Act 
432D-23.5 Coverage for telehealth. 
2013 Hawaii Revised Statutes 
TITLE 24. INSURANCE 
431. Insurance Code 
431:10A-116.3 Coverage for telehealth. 
 
Iowa No legal statute No legal statute 
Idaho No legal statute No legal statute 
Massachusetts No legal statute No legal statute 
North Carolina No legal statute No legal statute 
Ohio No legal statute No legal statute 
Pennsylvania No legal statute No legal statute 
South Carolina No legal statute No legal statute 
Virginia 
2016 Code of Virginia 
Title 38.2 - Insurance 
Chapter 34 - Provisions Relating to Accident and Sickness 
Insurance 
§ 38.2-3418.16. Coverage for telemedicine services 
 
D. An insurer, corporation, or health maintenance 
organization shall not be required to reimburse the treating 
provider or the consulting provider for technical fees or 
2016 Code of Virginia 
Title 38.2 - Insurance 
Chapter 34 - Provisions Relating to Accident and Sickness 
Insurance 
§ 38.2-3418.16. Coverage for telemedicine services 
 
C. An insurer, corporation, or health maintenance 
organization shall not exclude a service for coverage solely 
because the service is provided through telemedicine 
130 
costs for the provision of telemedicine services; however, 
such insurer, corporation, or health maintenance 
organization shall reimburse the treating provider or the 
consulting provider for the diagnosis, consultation, or 
treatment of the insured delivered through telemedicine 
services on the same basis that the insurer, corporation, or 
health maintenance organization is responsible for coverage 
for the provision of the same service through face-to-face 
consultation or contact. 
services and is not provided through face-to-face 
consultation or contact between a health care provider and a 
patient for services appropriately provided through 
telemedicine services. 
 
Wisconsin No legal statute No legal statute 
Wyoming No legal statute No legal statute 
131 
Appendix C2. CPT Codes 
Categorization Procedure Codes 
Asynchronous Any Modifier = GQ OR 
 99091, 99453, 99454, 99457, 99458, 99473, 99474 
  
Synchronous Any Modifier in (G0, GT, or 95) OR 
Place of Service = 02 OR 
G2010, G2012, G2061, G2062, G2063, Q3014, T1014 
 
  
132 
Appendix C3. Data Description 
Variable Description Data Source(s) 
Dependent Variable 
Total Telehealth Visits All patient to provider telehealth visits originating from a 3-digit 
ZCTA for a given year 
FairHealth 
Asynchronous Telehealth Visits All patient to provider asynchronous telehealth visits originating from 
a 3-digit ZCTA for a given year 
FairHealth 
Synchronous Telehealth Visits All patient to provider asynchronous telehealth visits originating from 
a 3-digit ZCTA for a given year 
FairHealth 
   
Independent Variables   
Consumer Broadband % The number of Census tabulation blocks within a 3-digit ZCTA with 
at least one fixed broadband provider offering fiber to the end user 
for a given year 
FCC Form 477, 
US Census ACS, 
ArcGIS Pro 2.7 
Providers in Service Area Average value of the number of primary care providers within a 30-
mile driving radius of 5-digit ZCTAs within a 3-digit ZCTA for a 
given year 
InfoGroup 
Historical Business 
Data, NPPES NPI, 
US Census ACS, 
ArcGIS Pro 2.7 
   
Rural Healthcare Program Controls 
RHP Applicants: Individual % Number of applicants to the Rural Healthcare Program that were 
completed by individual providers for a given year 
USAC 
Total RHP Funding Total dollar value provided by the Rural Healthcare Program to a 
given 3-digit ZCTA for a given year 
USAC 
RHP Clinics % Number of primary care clinics within a 3-digit ZCTA receiving 
fundings from the Rural Healthcare Program for a given year 
USAC, InfoGroup 
Historical Business 
Data, NPPES NPI, 
US Census ACS, 
ArcGIS Pro 2.7 
   
Demographic Controls 
Average Household Income Average household income for households within a 3-digit ZCTA for 
a given year 
US Census ACS, 
ArcGIS Pro 2.7 
Asian Asian population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Black Black population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Native Native population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Other Other population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Age: Under 18 Population under the age of 18 within a 3-digit ZCTA for a given 
year 
US Census ACS, 
ArcGIS Pro 2.7 
Age: 18 to 24 Population between 18 and 24 within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Age: 25 to 34 Population between 25 and 34 within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Age: 35 to 44 Population between 35 and 44 within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Age: 45 to 54 Population between 45 and 54 within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Age: 55 to 64 Population between 55 and 64 within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Sex Ratio Ratio of males to females in a 3-digit ZCTA in a given year. US Census ACS, 
ArcGIS Pro 2.7 
Married Married population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Separated/Divorced Separated/Divorced population within a 3-digit ZCTA for a given 
year 
US Census ACS, 
ArcGIS Pro 2.7 
133 
Widowed Widowed population within a 3-digit ZCTA for a given year US Census ACS, 
ArcGIS Pro 2.7 
Education: Greater Than High 
School 
Population that has at least a high school education/high school 
education equivalent or greater within a 3-digit ZCTA for a given 
year 
US Census ACS, 
ArcGIS Pro 2.7 
Establishments with 1,000 
Employees or More % 
Percent of firms employing 1,000 employees or more within a given 
3-digit ZCTA for a given year 
US Census 
Economic Census, 
ArcGIS Pro 2.7 
Employer Based Insurance Number of individuals with employer based private insurance within 
a given 3-digit ZCTA for a given year 
US Census ACS, 
ArcGIS Pro 2.7 
Direct Purchase Insurance Number of individuals with direct purchase private insurance within a 
given 3-digit ZCTA for a given year 
US Census ACS, 
ArcGIS Pro 2.7 
Inward Migration Number of individuals migrating into a given 3-digit ZCTA for a 
given year that are privately insured 
US Census ACS, 
ArcGIS Pro 2.7 
Employed: Part Time Number of individuals with private insurance that employed part time 
in a given 3-digit ZCTA for a given year 
US Census ACS, 
ArcGIS Pro 2.7 
Employed: Full Time Number of individuals with private insurance that employed full time 
in a given 3-digit ZCTA for a given year 
US Census ACS, 
ArcGIS Pro 2.7 
 
  
134 
Appendix C4. ZCTA3 Variable Summary Statistics 
     
 Mean SD Min Max 
Dependent and Independent Variables     
All Usage 709456.826 741880.044 47118.000 7641408.00
0 
Telehealth Usage % 0.002 0.004 0.000 0.059 
Synchronous Visits 1584.125 6719.135 0.000 91512.000 
Asynchronous Visits 79.660 348.115 0.000 6540.000 
Telehealth Visits Total 1663.785 6842.909 0.000 91639.000 
Consumer Broadband % 0.191 0.207 0.000 1.000 
Providers in Service Area 935.834 1293.525 5.019 6533.588 
     
Rural Healthcare Program Controls     
RHP Applicants: Individual % 0.254 0.337 0.000 1.000 
Total RHP Funding 363109.764 525453.728 0.000 3991760.36
0 
RHP Clinics % 0.101 0.180 0.000 1.000 
     
Demographic Controls     
Average Household Income 82362.314 23780.810 47331.817 196391.355 
Asian 35217.382 73168.997 161.000 629601.000 
Black 67781.739 103393.825 86.000 808397.000 
Native 3345.709 7310.283 33.000 70095.000 
Other 30149.561 67139.786 32.000 674245.000 
Age: Under 18 118832.131 102473.175 7228.000 628430.000 
Age: 18 to 24 50673.220 43142.194 2387.000 289152.000 
Age: 25 to 34 73549.140 70001.984 3465.000 485308.000 
Age: 35 to 44 68367.811 61136.426 3447.000 355923.000 
Age: 45 to 54 71648.346 61439.613 3653.000 362992.000 
Age: 55 to 64 67405.407 53061.439 4617.000 328810.000 
Sex Ratio 97.280 3.870 88.265 110.779 
Married 194372.392 153364.338 13716.000 1007980.00
0 
Separated/Divorced 55453.906 46622.856 3336.000 279535.000 
Widowed 25153.159 19942.396 2164.000 106376.000 
Education: Greater Than High School 185077.990 155300.603 10975.000 1014669.00
0 
Establishments with 1,000 Employees or 
More % 
0.000 0.001 0.000 0.004 
Direct Purchase Insurance 72234.075 59549.363 7426.000 346121.000 
Employer Insurance 284345.458 232083.985 15368.000 1601331.00
0 
Inward Migration 76157.866 64679.206 4214.000 356402.000 
Employed: Part Time 60774.000 51003.480 3267.000 321545.000 
Employed: Full Time 171053.633 147821.106 10936.000 873006.000 
Observations 904 
 
  
135 
Appendix C5. Leads and Lags Analysis 
 Negative Binomial 
 (1) (2) 
 β(SE) β(SE) 
Parity X Pre-RHP Expansion -0.01725 (0.079)   
2 Years Before   0.06078 (0.113) 
1 Year Before   -0.06742 (0.069) 
   
Parity X Post-RHP Expansion -0.18058 (0.067)***   
1 Year After   -0.18642 (0.067)*** 
   
RHP Controls   
RHP Applicants: Individual % 0.34404 (0.219) 0.33971 (0.218) 
Total RHP Funding -0.00000 (0.000) -0.00000 (0.000) 
RHP Clinics % -0.43839 (0.484) -0.44463 (0.483) 
   
ln Alpha [Dispersion Parameter] -1.94505 (0.086)*** -1.94754 (0.086)*** 
Demographic Controls Yes Yes 
Year FE Yes Yes 
ZCTA FE Yes Yes 
State FE Yes Yes 
Number of Observations 904 904 
* p<0.10, ** p<0.05, *** p<0.01