University of Southern California Dissertations and Theses

Essays on the U.S. market for substance use treatment and the impact of Medicaid policy reform

(USC Thesis Other)

Essays on the U.S. market for substance use treatment and the impact of Medicaid policy reform

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON THE U.S. MARKET FOR SUBSTANCE USE TREATMENT AND THE IMPACT
OF MEDICAID POLICY REFORM
by
Yimin Ge
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PUBLIC POLICY AND MANAGEMENT)
August 2023
Copyright 2023 Yimin Ge
Acknowledgements
I would like to express my sincerest gratitude to my dissertation committee chair, Prof. Ros-
alie Pacula, who guided me through this journey step by step with her persistent generosity, kind-
ness, and passion. It was her guidance, support, and encouragement that made this work possible,
and it was her patience and expertise that drove my academic growth.
I am grateful to Prof. John Romley, my advisor for the first two years and my qualification
exam chair. His continuous advice on all my work has been an indispensable source of inspiration
for thinking and improving. I am also grateful to Prof. Jason Doctor and Prof. Seth Seabury, who
gave me critical feedback for my dissertation papers. Their expertise and scholarly input have been
invaluable.
I have also received countless help from other faculties on different projects, including Prof.
Bryan Tysinger, Prof. Julie Zissimopoulos, Prof. Jakub Hlavka, Prof. Alex Capron, and Prof.
Michelle Keller. I will forever remember and appreciate their advice and help. I will also remember
and appreciate all the help I have received from Julie Kim and Anna Parks, who have made my
every transition and every progress in school so much easier, for being patient and caring.
I would also like to thank my colleagues, friends, and fellow graduate students for their ca-
maraderie and moral support. Their presence enriches this path to completion. In particular, I
would like to thank Seema Pessar, who was a strong support during my transition to new work.
Her understanding and generous help were instrumental in overcoming my challenges.
ii
I am indebted to my parents and other family members for their unwavering love and un-
derstanding. Their constant support and belief in my abilities have been the cornerstone of my
academic journey. I extend my heartfelt gratitude to my partner, for his love and company, for
being the only one I can talk to during the most difficult time of mine.
I acknowledge the financial support provided by USC Graduate School, USC Sol Price School
of Public Policy, and RAND-USC Schaeffer Opioid Policy Tools and Information Center (OPTIC),
without which this research would not have been possible. Their support enabled me to pursue my
passion for knowledge and contribute to my field.
The past five years were truly a challenging period in my life, impacted by the COVID-19
pandemic and my personal health problems. But all the support and help I received from everyone
around me made the time also unforgettable and valuable.
Thank you all for being part of this incredible journey.
iii
Table of Contents
Acknowledgements ii
List of Tables vii
List of Figures x
Abstract xi
Chapter 1: Introduction 1
1.1 Substance Abuse and Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Public Policies on Expanding Substance Abuse Treatment . . . . . . . . . . . . . 2
Chapter 2: The Impact of Medicaid 1115 Substance Use Disorder Demonstration Waivers
on the Availability of Substance Abuse Treatment Facilities and Services 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Theoretical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4.1.1 Outcome Measures . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4.1.2 Policy Measures . . . . . . . . . . . . . . . . . . . . . . . . . . 17
iv
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
2.4.1.3 Additional Control Variables . . . . . . . . . . . . . . . . . . . 18
2.4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.2.1 Instrumental Variable Approach . . . . . . . . . . . . . . . . . . 19
2.4.2.2 Empirical Models . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5.1 First Stage Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5.2 Impact of Waivers on the Availability of Treatment Facilities and Services . 25
2.5.3 Supplementary Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5.3.1 Heterogeneity in Policy Effects Due to Facility Ownership . . . 27
2.5.3.2 Heterogeneity Based on Primarily Mental Health Versus Primar-
ily Substance Abuse Facilities . . . . . . . . . . . . . . . . . . . 28
2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Chapter 3: Does Medicaid Institutions for Mental Diseases Exclusion Waiver Crowd
Out the Access to Treatment Among Commercially Insured Substance Use Disorder
Patients? 46
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.1 Substance Abuse Treatment and Access to It . . . . . . . . . . . . . . . . 49
3.2.2 Medicaid IMD Waivers for SUD and their Impact on the Commercially
Insured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.6 Conclusion and Limitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
v
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 4: Examining the Spillover Effect of Institutions for Mental Diseases Exclusion
Waiver and Identifying Predictors on the Quality of Substance Abuse Treatment Ser-
vices Received in Commercial Insurance 73
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.1 Quality Measures for SUD Treatment and the Change of Treatment Qual-
ity Overtime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2.2 Adopting Medicaid Payment Waivers on SUD . . . . . . . . . . . . . . . 78
4.3 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.1.1 Outcome measures . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.3.1.2 Outcome trends during 2010-2021 . . . . . . . . . . . . . . . . 82
4.3.1.3 Individual- and state-level covariates . . . . . . . . . . . . . . . 83
4.3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4.1 Initiation and engagement of treatment . . . . . . . . . . . . . . . . . . . 86
4.4.2 Followup care after residential services . . . . . . . . . . . . . . . . . . . 87
4.4.3 Average length of stay in treatment episodes . . . . . . . . . . . . . . . . . 88
4.5 Conclusion and Limitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 5: Conclusion and ongoing work 108
5.1 Main findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.3
References
Appendices
Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
vi
113
121
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
List of Tables
2.1 2015-2020 States with Effective Section 1115 Demonstration Waiver in SUD . . . 33
2.2 Summary Statistics of Key Variables 2014-2020 . . . . . . . . . . . . . . . . . . . 34
2.3 Data Sources of Key Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Potential Instrumental Variable Selection . . . . . . . . . . . . . . . . . . . . . . . 36
2.5 F-statistics of Potential Instrumental Variables . . . . . . . . . . . . . . . . . . . . 36
2.6 First Stage Result Estimating the Likelihood of a State Adopting a Section 1115
SUD Waiver, with a Linear Probability Model . . . . . . . . . . . . . . . . . . . . 37
2.7 Examining the Waiver Effect on the Number of Facilities, with a Mixed-effect
Poisson Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.8 Examining the Waiver Effect on the Number of Facilities without Addressing En-
dogeneity, with a Mixed-Effect Poisson Model . . . . . . . . . . . . . . . . . . . . 39
2.9 Examining the Waiver Effect on the Facility-level Likelihood of Offering Co-
occurring Treatment, with a Logit Regression Model (with Imputation) . . . . . . 40
2.10 Examining the Waiver Effect on the Facility-level Likelihood of Offering Co-
occurring Treatment, with a Logit Regression Model (with No Imputation) . . . . 41
2.11 Examining the Waiver Effect on the Percentage of SA Facilities Offering Co-
occurring Disorder Treatment, with a Linear Probability Model . . . . . . . . . . . 42
vii
2.12 Examining the Waiver Effect on the Percentage of SA Facilities Offering Co-
occurring Disorder Treatment by Facility Ownership Type, with a Linear Prob-
ability Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.13 Examining the Waiver Effect on the Percentage of Facilities Offering Co-occurring
Disorder Treatment, with a Linear Probability Model (on Licensed MH Treatment
Facilities) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.14 Comparing Percentage of Facilities Offering Treatment in Mental Health and Sub-
stance Abuse among Primarily Substance Abuse Facilities and Primarily Mental
Health Facilities, with Paired-t Test . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 2015-2020 IMD Waiver in SUD (effective in a full year) . . . . . . . . . . . . . . 67
3.2 Data Sources of Key Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3 Sample Characteristics for Evaluating IMD Exclusion Waiver . . . . . . . . . . . . 69
3.4 Summary Statistics for State-level Co-variates . . . . . . . . . . . . . . . . . . . . 70
3.5 Synthetic Difference-in-Difference Estimation Results . . . . . . . . . . . . . . . 71
3.6 Examining the Waiver Effect on the Access Rate to Different Levels of Treatment,
with a Two-way FE Linear Probability Model . . . . . . . . . . . . . . . . . . . . 72
4.1 2015-2021 IMD Waiver in Substance Use Disorder (SUD) . . . . . . . . . . . . . 95
4.2 Demographic and Clinical Characteristics for SUD Patients (Patient-level) for the
Period 2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.3 Demographic and Clinical Characteristics for SUD Patients Starting New Episodes
with OP (Episode-level) in the Optum Commercial Insurance Data for the Period
2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4 Demographic and Clinical Characteristics for SUD Patients Receiving Residen-
tial Care (Episode-level) in the Optum Commercial Insurance Data for the Period
2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5 Data Sources of Key Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
viii
4.6 Summary Statistics for State-level Co-variates for the Period 2010-2021 . . . . . . 100
4.7 Initiation and Engagement Rate (Starting with OP) by Group in the Optum Com-
mercial Insurance Data for the Period 2010-2021 . . . . . . . . . . . . . . . . . . 101
4.8 Continuation Rate after Residential Services by Group in the Optum Commercial
Insurance Data for the Period 2010-2021 . . . . . . . . . . . . . . . . . . . . . . . 102
4.9 Average Length of Stay by Group in the Optum Commercial Insurance Data for
the Period 2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.10 GEE with Logit Link Function Estimating the Rate of Initiation for the Period
2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.11 GEE with Logit Link Function Estimating the Rate of Engagement for the Period
2010-2021 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.12 GEE with Logit Link Function Estimating the Rate of Treatment within 14 Days
after Residential Services for the Period 2010-2021 . . . . . . . . . . . . . . . . . 106
4.13 GEE with Poisson Link Function Estimating the Average Length of Stay in Treat-
ment Episodes for the Period 2010-2021 . . . . . . . . . . . . . . . . . . . . . . . 107
A.1 Optum File Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.2 ICD Codes for Substance Use Disorders . . . . . . . . . . . . . . . . . . . . . . . 123
A.3 ICD Codes for Mental Health Problems . . . . . . . . . . . . . . . . . . . . . . . 124
A.4 HCPCS/CPT Codes for Substance Use Disorders Treatment at Outpatient Level . . 125
A.5 HCPCS/CPT Codes for Substance Use Disorders Treatment at Inpatient Level . . . 126
A.6 HCPCS/CPT Codes for Substance Use Disorders Treatment at Intensive Outpa-
tient or Partial Hospitalization Level . . . . . . . . . . . . . . . . . . . . . . . . . 126
A.7 HCPCS/CPT Codes for Medication Assisted Treatment . . . . . . . . . . . . . . . 127
ix
List of Figures
2.1 Number of States with Effective Section 1115 Demonstration Waiver in SUD
2015-2020 by Year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1 Trends of SUD Patients Receiving Inpatient Care . . . . . . . . . . . . . . . . . . 63
3.2 Trends of SUD Patients Receiving Outpatient Care . . . . . . . . . . . . . . . . . 64
3.3 Trends of SUD Patients Receiving Intensive Outpatient Care . . . . . . . . . . . . 65
3.4 Trends of SUD Patients Receiving MAT Care . . . . . . . . . . . . . . . . . . . . 66
4.1 Trends of Initiation Rate among SUD Patients in Optum Commercial Insurance Data 91
4.2 Trends of Engagement Rate among SUD Patients in Optum Commercial Insurance
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3 Trends of Continuation Rate after Residential Services among SUD Patients in
Optum Commercial Insurance Data . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.4 Trends of Length of Stay per Episode among SUD Patients in Optum Commercial
Insurance Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
x
Abstract
State and federal governments have enacted numerous policies to increase access to evidence-
based substance abuse and mental health treatment in the United States, including Medicaid pay-
ment exclusion waivers as part of Section 1115 Substance Use Disorder (SUD) demonstrations,
which are also called IMD waivers. As of the end of 2020, 28 states obtained such waivers. Using
data from the annual versions of the National Directory of Substance Abuse Treatment Facilities,
the second chapter examines changes in the number of substance abuse treatment facilities and the
likelihood of those substance abuse treatment facilities offering treatment for co-occurring disor-
ders in mental health using an instrumental variable, two-stage residual inclusion method. It shows
that states with Section 1115 SUD waivers have a 16% in rate of having substance abuse treatment
facilities (extensive margin) and have a 28% increase in the odds of a substance abuse treatment
facility offering co-occurring treatment (intensive margin). The waiver effect is more substantial
among residential facilities than non-residential facilities and larger among private, for-profit facil-
ities than public or governmental facilities. The findings support those found in other studies that
payment reform is an important lever for improving access to evidence-based treatment for those
suffering from substance abuse and mental health comorbidity and adding new evidence on the
policy effects of expanding the treatment market and transforming the service provision.
The third chapter examines whether adopting Medicaid IMD exclusion waivers crowds out the
access to treatment settings of those commercially insured SUD patients. The analysis is repeated
for SUD patients with mental health comorbidities and compares the differences between the two
groups of patients. It finds that commercial plan enrollees use more outpatient SUD treatment
xi
(41.5%) compared to inpatient (14.9%) and intensive outpatient/partial hospitalization (1.4%). The
study uses synthetic control difference-in-difference design to examine the waiver effect. Instead
of finding a crowd-out impact on access, the findings suggest adopting an IMD exclusion waiver
contributes to a 0.2 percentage point increase in access to intensive outpatient or partial hospital-
ization treatment services among those commercially insured SUD patients. No impact is found
on access to inpatient, outpatient, and medication-assisted treatment. This study offers insights
into access to treatment among commercially insured SUD patients. It provides evidence of the
importance of Medicaid policies in changing the whole treatment market besides the Medicaid
population.
Using quality measures developed by the Washington Circle Group, the fourth chapter ex-
amines the spillover effect of Medicaid IMD waivers on the quality of SUD treatment received
by commercial insurance between 2010-2021 and identifies important predictors of quality mea-
sures. Descriptive trends show that the initiation, engagement, continuation rates after residential
care, and average length of stay in treatment episodes have not improved during the 12-year ob-
servation period. Using generalized estimating equations with logit and Poisson functions at the
patient-episode level, I find that the adoption of Medicaid payment waivers does not negatively
impact the treatment quality among commercially insured but is shown to have positively affected
the engagement rate. Some critical factors that are associated with higher initiation, engagement,
continuation after residential care rates, and a longer average length of stay in treatment include
gender (being female) and ethnicity (being white). Mental health comorbidity is associated with
a lower initiation and engagement rate and shorter length of stay in treatment, but it is associated
with a higher likelihood of receiving follow-up care after residential services.
xii
Chapter 1
Introduction
1.1 Substance Abuse and Treatment
Substance abuse refers to a pattern of excessive or harmful use of psychoactive substances,
including drugs or alcohol (World Health Organization Regional Office for Africa, n.d.). It brings
severely negative mental and physical health consequences to the abuser, and criminal or anti-social
behavior may also occur when a person is under the influence of a drug (Hart & Ksir, 2022).
Substance use is a severe public health problem in the United States. More than 932,000
people have died from a drug overdose over the past two decades (CDC, 2021). The rising trend
of overdose death, from 37,000 in 2010 to more than 100,000 in 2020 (CDC, 2021), draws wide
public attention. Though the number of people suffering from substance abuse problems increased
drastically, the rate of getting treatments has remained low. Only 9.6% percent of people aged 12 or
older who needed substance abuse treatment in 2020 received some form of treatment (SAMHSA,
2022b). Some common reasons for people not receiving substance abuse treatment include not
being ready to stop using (36.7 %), having no health coverage or not being able to afford treatment
(24.9 %), not knowing where to go for treatment (17.9 %), and not finding a program that offered
the type of treatment wanted (15.8 %) (SAMHSA, 2022b), which can be summarized into self-
motivation, financial barriers and shortage of supply.
1
Substance abuse problems frequently co-occur with other behavioral health conditions, and
patients sometimes can hardly make treatment decisions themselves. While the relapsing nature
of the disease often requires multiple episodes of treatment and step-down care (McLellan, 2002).
The uniqueness of behavioral health treatments makes their coverage unpopular in the insurance
market. Since many behavioral health problems are more persistent and have less clear clinical
measures than other illnesses, health plans are likely chosen by persons with those conditions and
overused by patients. Therefore, the high cost of treatments can hardly be managed by commercial
insurance due to adverse selection and moral hazard (Frank et al., 1997). As a result, the public
sector, including Medicare and Medicaid, has been the primary payer for the healthcare expenses
of substance use disorders(SUD) services (Mark et al., 2011; Mark et al., 2016). Even for people
who can get treatments, the limited capacity of treatment facilities has often led to a waiting time
before enrollment. For example, according to the Treatment Episode Data Set (TEDS) in 2020,
21% of all admissions have to wait up to a week (between 1-7 days) to enter treatment, and in the
case of opioid addiction, it is 26% that have to wait up to a week (SAMHSA, 2022c). 10% must
wait eight or more days (i.e., at least a week and as long as a month) to enter treatment (SAMHSA,
2022c).
1.2 Public Policies on Expanding Substance Abuse Treatment
Public policies can help address organizational or systematic problems that prevent people
from receiving substance abuse treatment, particularly regarding access and cost. Policies have
done this by increasing funding to support supply expansion and implementing laws that require
more insurance coverage, thereby reducing costs to the patient.
As early as the 1970s and 1980s, several states enacted laws requiring private health plans op-
erating in the state to offer a specific minimum set of mental health benefits (18 states), alcohol ben-
efits (38 states), and drug abuse benefits (25 states) (Barry et al., 2010). The Domenici-Wellstone
2
amendment enacted in September 1996, commonly known as the Mental Health Parity Act, is con-
sidered a significant step forward in calling for parity in behavioral health coverage, which aims
to eliminate certain limits on coverage for behavioral health care under private insurance (Frank
et al., 1997). Addiction treatment, however, was excluded from this legislation.
The Paul Wellstone and Pete Domenici Mental Health Parity and Addiction Equity Act (MH-
PAEA) of 2008, which became effective in 2010, requires most insurance plans to provide the same
mental health and addiction care benefits as other medical care. MHPAEA goes beyond the 1996
Mental Health Parity Act by extending parity to addiction services and applying these requirements
to quantitative and non-quantitative treatment limits. The Affordable Care Act (ACA) in 2010
expanded the reach of the MHPAEA by specifying substance use disorder (SUD) treatment as an
essential health benefit to be covered by all health insurance plans competing in state exchanges. It
also prevented all insurance plans from discriminating against people based on their conditions. In
2014, the ACA expanded Medicaid to cover adults below 138 percent of the federal poverty level,
further expanding the reach of the MHPAEA, which Medicaid was subject to, by increasing the
number of insured people suffering from SUDs.
As drug abuse and overdose deaths have continued in the United States, more policy efforts
have been undertaken since 2010. For example, the Tribal Law and Order Act of 2010 was passed,
promoting interagency coordination and collaboration in providing alcohol and substance abuse
prevention and treatment for those in Indian country (SAMHSA, 2022a). The Comprehensive Ad-
diction and Recovery Act (CARA) of 2016 authorized over 181 million dollars each year for almost
seven years to respond to the epidemic of opioid abuse, created multiple evidence-based treatment
programs for different groups of people suffering from SUDs and expanded the availability of
naloxone (SAMHSA, 2022a). The 21st Century Cures Act codified the Center for Behavioral
Health Statistics and Quality (CBHSQ), which acts as the primary federal agency responsible for
collecting and analyzing behavioral health data, conducting national surveys such as the National
Survey on Drug Use and Health (SAMHSA, 2022a). The Act also created the National Mental
3
Health and Substance Use Policy Laboratory, which evaluates and promotes evidence-based prac-
tices and services for substance use disorders, mainly focusing on opioids (SAMHSA, 2022a).
Finally, the SUPPORT Act of 2018 provides State Targeted Response Grants (STR) and creates a
first responder training program and comprehensive opioid recovery centers (SAMHSA, 2022a).
As the largest single-payer of substance abuse treatment, Medicaid has made additional efforts
to expand access and lower SUD treatment costs since the general expansion of coverage under
ACA. An important strategy has been providing Institutions for Mental Diseases (IMDs) Waivers
under the Section 1115 demonstration program. These waivers allow states to obtain federal Med-
icaid funds for services received by nonelderly adults in an IMD, which is defined as a “hospital,
nursing facility, or other institution of more than 16 beds, that is primarily engaged in providing
diagnosis, treatment, or care of persons with mental diseases, including medical attention, nursing
care, and related services (Musumeci et al., 2019).” Under the IMD exclusion rule created under
the 1965 Medicaid law, the IMD exclusion prohibits Medicaid benefits from paying any nonelderly
adults for services in IMDs. These waivers allow states to circumvent that exclusion. In 2015, the
Centers for Medicare and Medicaid (CMS) issued the first formal guidance for states to apply for
IMD exclusion waivers under the Section 1115 demonstration projects. Later in 2017, under the
Trump administration, the CMS issued enhanced guidance with a more streamlined application
procedure for states to apply for such waivers.
Some studies have shown that the Medicaid IMD exclusion waiver has generated spillover
effects on the treatment market. For example, the acceptance of Medicaid increased in residential
and intensive outpatient facilities, and the delivery of medications increased at outpatient facili-
ties with the adoption of the IMD exclusion waivers (Maclean et al., 2021). Those waivers for
SUDs are also found to be positively associated with the likelihood of providing opioid use dis-
order(OUD) treatment programs in hospitals (Chang et al., 2023). Giving particular attention to
the IMD waivers in SUD, the following three chapters of my dissertation each take a unique angle
of the US market for substance abuse treatments from both the facilities and patients who need
4
treatment. The first paper (second chapter here) looks at the treatment facilities nationwide, ex-
amining the impact of IMD SUD waivers on the total number of available treatment facilities and
the provision of co-occurring mental health and substance abuse treatments. The next chapter ana-
lyzes whether this expansion influenced access to different levels of care among the commercially
insured, who may have been crowded out by growth that only occurred among Medicaid enrollees,
thereby considering the spillover effects of IMD SUD waivers on the commercially insured. The
fourth chapter takes a closer look at the SUD treatment, focusing on the quality provided to those
with commercial insurance using widely accepted process measures that indicate the provision of
higher-quality SUD treatment. It, therefore, examines whether expanding coverage of inpatient
services for Medicaid enrollees may generate a spillover effect on the quality of treatment received
by those with commercial insurance.
5
Chapter 2
The Impact of Medicaid 1115 Substance Use Disorder Demonstration Waivers on the
Availability of Substance Abuse Treatment Facilities and Services
2.1 Introduction
Substance use and mental health problems have risen substantially in the United States over
the past three decades. In 2020, 46.3 million people aged 12 or older, which is approximately 16.5
percent of this population, had a substance use disorder (SUD)(SAMHSA, 2022b). Among adults
aged 18 or older in 2020, 22.8 percent (or 57.8 million people) had some form of mental illness,
and 5.5 percent (or 14.1 million people) had serious mental illnesses (SAMHSA, 2022b). Among
this adult population, 7.6 percent (or 19.4 million) had co-occurring mental illness and SUD in
2020 (SAMHSA, 2022b). Mental health, SUDs, and chronic physical health illnesses commonly
co-occur (Naren et al., 2022; Owens et al., 2018), and patients with serious mental illness (SMI),
such as schizophrenia, have higher rates of drug use than the general population (NIDA, 2020).
Untreated SUDs are associated with increased risks for various mental and physical conditions
(CMS, 2015), as well as suicide and overdose (Forray & Yonkers, 2021). The co-occurring disorder
is defined as the coexistence of a mental illness and a substance use disorder (SAMHSA, 2023).
Research has found advantages in providing simultaneous treatments to patients with co-
occurring disorders, as substance use and mental health disorders usually interact with each other
(Butler et al., 2008; T. M. Kelly & Daley, 2013; Mangrum et al., 2006; Orford, 2008; Ziedonis,
6
2004). However, since moral hazard and adverse selection are even more severe in the mental
health care marketplace than in the general medical care marketplace, private insurance is less
generous covering and paying for these services (Frank et al., 1997; Frank & McGuire, 2000).
As a result, the public sector has been the primary payer for the healthcare expenses of mental
health and addiction services, including Medicare and Medicaid. Federal behavioral health spend-
ing provides a safety net for people with behavioral health treatment needs (Levit et al., 2013).
Despite substantial governmental expenditures, access to treatment has persistently been limited
(SAMHSA, 2021).
Rising substance abuse (SA) and mental health (MH) problems have led to a call for public
health policies that can reduce the financial burden of treatment, remove socioeconomic barriers
to getting adequate care, and potentially increase access to more effective care. Among various
health policies, Medicaid programs have allowed states to demonstrate potentially cost-saving,
beneficial care or delivery practices through the Medicaid Section 1115 demonstration program.
Through this program, states can test new Medicaid approaches that differ from federal program
rules and still get paid for them. In 2015, states began seeking approvals for demonstration projects
that could expand access to inpatient substance use services for nonelderly adults in Institutions for
Mental Diseases (or IMDs) under formal guidance issued by the Center for Medicare and Medicaid
Services (CMS) (KFF, 2021), so they are also sometimes referred as “IMD exclusion waivers”.
States must abide by evaluation goals set by CMS to receive approval. For example, combining
SA, MH, and primary care aligns with one of the CMS’s performance goals in promoting care
integration by issuing Section 1115 demonstration waivers in SA and MH services.
In this paper, I test whether adopting Medicaid waivers specifically targeting IMDs can en-
courage facilities to offer co-occurring treatment among SA treatment facilities. I also examine
whether presumed cost savings created by integrating SA and MH care stimulate growth in the
number of SA treatment facilities. While initial work looking at the impact of those 1115 waivers
suggests they have been effective at expanding treatment, they are limited in consideration of which
7
margin treatment expands (Maclean et al., 2021; O’Brien et al., 2022). I extend this research by
considering impacts on both the extensive and intensive margins of treatment facilities and address
the problem of potential policy endogeneity - states’ Medicaid programs may self-select to apply
for waivers based on their unique conditions, which are unable to be identified by state-level trends
comparisons in difference-in-difference design.
2.2 Background
In July 2015, the Obama Administration issued a letter to all state Medicaid directors inviting
them to apply for Section 1115 demonstration projects and receive Medicaid funding for SUD
services (Cohen et al., 2021). This guidance became the first from CMS in applying for Section
1115 SUD waivers. Later in November 2017, under the Trump Administration, CMS issued new
guidance with a more streamlined approach for states to apply for these waivers (Cohen et al.,
2021). By the end of 2020, 28 states had IMD payment exclusion waivers for SUD treatment
(KFF, 2021). See Figure 2.1 for the number of states that adopt waivers each year, and Table 2.1
for a list of those states that obtained a Section 1115 SUD waiver during the time period (2015-
2020) I will be evaluating.
One of the most important components of the 1115 SUD waivers is the inpatient service
reimbursement to nonelderly patients receiving care within an IMD, which has been explicitly
forbidden under Medicaid since 1965. The so-called “IMD exclusion” prohibits Medicaid from
providing federal funds to states for any services received by individuals aged 21 to 64 within an
IMD, which are defined as treatment facilities with more than 16 beds primarily treating patients
for mental health conditions (Mitchell, 2019). The intention of the IMD exclusion in the origi-
nal Medicaid Act was to make states responsible for paying for mental health patients who need
long-term inpatient care from psychiatric settings and discourage the institutionalization of mental
health patients (Geller, 2000). However, it has reduced the availability of inpatient care for people
8
with chronic mental illnesses. The United States has closed almost 97% of its state hospital beds
since the mid-1950s (TreatmentAdvocacyCenter, 2016). Providers of psychiatric services have un-
dergone a massive transformation from inpatient to outpatient settings and from public to private
providers, while the development of community care remains inadequate (Geller, 2000).
A recent study compared the acceptance of Medicaid, and other types of health coverage,
self-pay arrangements, and provision of charity care in residential and outpatient SUD treatment
facilities after states adopted Section 1115 IMD SUD waivers (Maclean et al., 2021). Acceptance
of patients with Medicaid insurance was estimated to have increased by 34 percent at residential
treatment facilities and 9 percent at intensive outpatient facilities after two years of the waivers
being implemented (Maclean et al., 2021). The study, while significant, suffers from some poten-
tially serious limitations. These include the narrow scope of study outcomes, a short follow-up
period post-waiver adoption, and the potential endogeneity of adoption of the 1115 SUD waivers.
I extend the scope of evaluation to the number of facilities and the share of SA facilities that
offer co-occurring treatment in this paper, allowing me to consider both the extensive and intensive
margins on which these waivers might impact markets. I also use an instrumental variable approach
to address the endogeneity problem caused by the correlation between the pursuit of SUD waivers
and unmeasured state-level factors that might influence the availability of SA-MH services in the
marketplace. Finally, I make an additional contribution by improving the quality of the data used
to consider this question. Specifically, I merge data previously available through RAND that links
the National Survey of Substance Abuse Treatment Services (N-SSATS) data longitudinally across
different years (Cantor et al., 2022) with licensing data available through Data Axle. Doing so
allows me to capture facilities that are not consistently recorded in N-SSATS in some years and
those that change names due to being bought out by other facilities.
9
2.3 Theoretical Framework
In the treatment market for SA and MH, assuming homogeneous firms with identical cost
functions and perfectly substitutable inputs to production, the average profit ( π) for each facility
when N facilities are in the market offering a type of service (s), can be given as:
π
s
= V
s
∗ Z
N
∗ d
s
− F
s
(2.1)
where V
s
is the variable profit when N (N≥ 0) facilities offer a type of service (s), which
is defined as price minus average variable cost (A VC) (Abraham et al., 2007). Z (Z≥ 0) is the
demand in the market, defined as the number of patients needing each type of service s. The
variable s includes SA care services, MH care services, or a combination of the two services
produced in either residential or outpatient settings, d
s
is the individual demand for the service of
type s, and F
s
(F
s
≥ 0) (assumed to be constant by state and year, and the differences of costs by
facility are included in A VC) is the fixed cost of offering the service s. A facility chooses to enter
the market whenπ
s
is greater than 0. Suppose each facility in the market can only offer SA or MH
treatment. In this situation, the profit function for the facility is either
⇒π
sa
=(Price
sa
− A VC
sa
)∗ U
sa
N
sa
− F
sa
, (2.2)
or
⇒π
mh
=(Price
mh
− A VC
mh
)∗ U
mh
N
mh
− F
mh
, (2.3)
where U is Z∗ d
s
, the number of total units served. A firm will offer either SA or MH services
depending on the higher profit (assuming positive profits can be achieved quickly before firm
entry).
10
I hypothesize that several factors affect profits change if SA facilities offer integrated care in
MH. First, there is a reduced fixed cost by condensing buildings or sites. Space for operation can be
reduced and packed into one instead of two locations, saving fixed costs in rent, equipment, and site
construction. Second, A VC can be reduced by combining management teams. It is hypothesized
that in a market with facilities that can only offer either SA or MH care, patients with co-occurring
disorders must go to two separate facilities to treat both conditions. The average variable costs of
treating a patient could be double because the patient sees two sets of medical and administrative
staff, two sets of management services to oversee the treatment, and so on. Some of these costs
could be reduced if a patient could receive both treatments at a single facility. Third, reduced A VC
can also be achieved by combining expert teams. Treatment productivity for treating both SA and
MH may be higher in one facility. Suppose a facility chooses to combine SA and MH treatment
services, co-treatment is likely more efficient since some experts, such as psychotherapists in both
areas, can treat both conditions. Fourth, there is an increase in market demand Z. The high rate of
co-morbidity also makes receiving treatments at two facilities less efficient. For example, a patient
suffering from mental health issues can use an intoxicating substance to self-medicate if substance
use is not addressed during MH treatment. This can lead to treatment cycles. On the contrary,
if facilities enhance co-treatment, more treatment slots are freed as patients receive care at one
facility instead of two. Facilities could treat more patients, assuming more patients are waiting to
be treated than the available slots. Finally, individual demand d
s
also increases. As more patients
with co-occurring disorders can be identified and receive co-occurring treatment, their demand for
services increases, and the length of treatment is likely to increase for more complicated problems.
Assuming in this simplified market, the number of SA and MH facilities is evenly distributed,
and the number does not change initially when facilities start to combine SA and MH services.
Then, let
N= N
mh
= N
sa
= N
mhsa
. (2.4)
11
The add-up profit for an SA facility and an MH facility would be:
π
sa
+π
mh
=(Price
sa
− A VC
sa
)∗ U
sa
N
sa
+(Price
mh
− A VC
mh
)∗ U
mh
N
mh
− F
sa
− F
mh
(2.5)
=
(Price
sa
− A VC
sa
)∗ U
sa
+(Price
mh
− A VC
mh
)∗ U
mh
N
− (F
sa
+ F
mh
).
The profit of owning two facilities offering integrated SA and MH care would be:
2∗ π
mhsa
=
(Price
mhsa
− A VC
mhsa
)∗ U
mhsa
+(Price
mhsa
− A VC
mhsa
)∗ U
mhsa
N
− 2∗ F
mhsa
(2.6)
If the policy allows facilities to achieve cost savings by integrating MH and SA services, then
facilities will do so if and only if,
A VC
mhsa
< A VC
mh
; (2.7)
A VC
mhsa
< A VC
sa
; (2.8)
U
mhsa
= Z
mhsa
∗ d
mhsa
> Z
mh
∗ d
mh
= U
mh
; (2.9)
U
mhsa
= Z
mhsa
∗ d
mhsa
> Z
sa
∗ d
sa
= U
sa
; (2.10)
12
2∗ F
mhsa
< F
sa
+ F
mh
. (2.11)
Therefore, the profit of owning two facilities offering integrated SA and MH care may be greater
than owning two facilities offering SA and MH care separately, or:
2∗ π
mhsa
>π
sa
+π
mh
(2.12)
The Medicaid 1115 SUD waivers expand the reimbursement ranges of services in the treat-
ment facilities, especially the residential facilities, and promote integrated care as set by evaluation
goals. Thus, the intervention of those waivers could potentially lead to more integrated care and
reduced operation costs under my hypothesized conditions. In the short run, the rising profit would
incentivize more facilities to enter the market until the profits are lowered to an equilibrium, or
zero profit in a “perfectly competitive” market, assuming it exists in the treatment market and all
types of facilities – for-profit or nonprofit, behave in the same manner.
In this initial framework, when there is a “perfectly competitive” market without the interven-
tion of this policy,
N=
F
s
(Price
s
− A VC
s
)∗ U
s
. (2.13)
E[N]= E[
F
s
(Price
s
− A VC
s
)∗ U
s
]. (2.14)
After the introduction of waivers into the market, N changes with the changes of other pa-
rameters in the above equation, making it (N) a function of factors that influence price, cost, and
market size of producing MH and SA jointly versus the production of these separately. Those
13
factors include policies P, the demand for services D, socioeconomic variables X, market compet-
itiveness measures M, and within-industry operational costs W. Drawn from the specification in
Abraham et al. (2007), functions for F
s
, V
s
(Price
s
-A VC
s
), and U
s
can be written as:
F
s
= exp(α
1
∗ P+α
2
∗ X+ε
F
), (2.15)
V
s
= exp(β
1
∗ P+β
2
∗ X+β
3
∗ M+β
4
∗ W+ε
V
), (2.16)
U
s
= exp(γ
1
∗ P+γ
3
∗ X+γ
2
∗ D+ε
U
). (2.17)
Then the number of facilities can be modeled as:
E[N]= exp(δ
1
∗ P+δ
2
∗ X+δ
3
∗ M+δ
4
∗ D+δ
5
∗ W+ε). (2.18)
I will choose proxy variables from each of those categories for the empirical analysis.
My empirical work tests two hypotheses directly from the theoretical framework described
above. First, motivated by the cost savings that could potentially be achieved by integrating SA
and MH services, facilities are more likely to offer co-occurring treatment after adopting Section
1115 SUD waivers. As a result, a larger share of SA facilities will provide co-occurring treatment
after adopting SUD waivers within a state. Second, driven by the extra profits created by integrating
SA and MH care, more facilities may choose to enter or stay in the market. Then adopting Section
1115 SUD waivers is likely associated with growth in the total number of SA treatment facilities.
2.4 Data and Methods
I use a two-stage residual inclusion method with an instrumental variable to estimate the effect
of Section 1115 SUD waivers on services offered in facilities licensed to provide SUD services.
14
The primary data used for estimation is reconstructed panel data based on the public version of the
National Survey of Substance Abuse Treatment Services (N-SSATS) data.
2.4.1 Data
My primary data source comes originally from various annual versions of the National Sur-
vey of Substance Abuse Treatment Services (N-SSATS), a voluntary national survey of SUD treat-
ment providers conducted by the US Substance Abuse and Mental Health Services Administration
(SAMHSA), which uses state directories of specialty treatment providers registered in each state as
its sampling frame. The sampling frame compiled by SAMHSA from state directories of licensed
SUD treatment facilities includes the facility’s address. SAMHSA, through its voluntary survey
N-SSATS, obtains from these providers information on the facility’s types of care, operation of the
facility(public or private), service programs, and forms of payment accepted. I use two versions
of the data from this primary data source in the main and sensitivity analysis separately. One is
the panel version linking facilities over time, produced by the RAND Corporation (Cantor et al.,
2022), and enhanced by my work linking with Data Axle licensed business data. The other is the
publicly available version of N-SSATS data.
The public cross-sectional version of the N-SSATS data made available by SAMHSA is not
ideal for answering questions about the opening or closing of treatment facilities over time (a
change in treatment availability within a single location over time) because treatment facilities do
not consistently report data to state directors every year for reasons that are not well understood.
Therefore, I use a panel dataset built up on the original facilities in the directory to overcome
the issue of inconsistent reporting that affects the interpretation of treatment facility availability.
The original data in the directory is matched by locations across years using an algorithm adopted
by researchers at RAND and then cross-checked using additional facility information available
in Data Axle’s business database. Both the algorithm and check with Data Axle cannot identify
whether the facilities with the same address refer to one or multiple facilities. So I keep one
15
facility corresponding to one address, relying on available information about the facilities’ service
type, e.g., Standard Industrial Classification(SIC) code. By keeping one facility at each address,
I can also see whether there is a growth of facilities measured by the addresses (i.e., whether
patients have more places to go for treatment), at the same time eliminating the potential problem
of repetitive measures.
Data Axle is a data service company offering various industries data-related solutions, includ-
ing data processing and database management. They provide a nationwide, real-time licensed busi-
ness database with licensed facilities providing mental health and substance use-related services.
Without supplemental information indicating that some treatment facilities are still operating, one
might incorrectly presume that the licensed facility is closed. The database from Data Axle is used
to fill in gaps that remained in RAND’s enhanced data set, which only relies on information in state
directories over time. However, it does not contain all the same descriptors of facilities as reported
in state directories. Missing data on facility characteristics available in N-SSATS but not reported
in Data Axle are imputed. That enables me to construct a longitudinal data set of all facilities
operating in specific addresses for 2011-2020. I narrow the range of businesses using the North
American Industry Classification System (NAICS) code to companies and organizations offering
mental health and substance use services, such as hospitals, treatment centers, social advocacy
groups, public health programs, etc. I further filter the facilities based on each facility’s Stan-
dard Industrial Classification (SIC) descriptions. There are 162,051 observations in the panel data
for 2011-2020, consisting of about 16,205 facilities each year on average. Of those observations,
17.6% (28,545) come from Data Axle, and 10.2% (16,501) are imputed based on the available
information for filling the gap years between observed years.
2.4.1.1 Outcome Measures
From the facility-level data, I focus on two primary measures: the number of SA facilities
within a state in a given year and whether a facility in my sample offers co-occurring disorder
16
treatment each year. The annual number of facilities is calculated by taking the total number of
recorded facilities in the facility-level panel data every year. And I directly use the binary value of
a facility offering co-occurring disorder treatment each year in the dataset.
2.4.1.2 Policy Measures
The key policy variable is the adoption of a Medicaid 1115 SUD waiver in a given state during
a given year. The data comes from RAND Opioid Policy Tools and Information Center (OPTIC)
policy database. OPTIC identifies and extracts existing legal information on states’ waiver status
from the Prescription Drug Abuse Policy System (PDAPS). I also double-checked the data against
information provided on Medicaid’s website (CMS, 2023b), as well as National Health Law Pro-
gram’s report (Cohen et al., 2021). Twenty-eight states adopted a Section 1115 demonstration
waiver in SUD during 2015-2020 (KFF, 2021) (Table 2.1). Using binary values and fractions, I
construct states’ 1115 SUD waiver status in a given year. Fractions are calculated based on the
number of months out of 12 that the waiver is effective. I consider the waiver effective for a given
month if it became effective by the 7th of January or by the 3rd of any subsequent month. Since
laws are generally set to start on the first business day of a year or month, and because of weekends
(and in the case of the New Year holiday), I impose this rule to reflect variations in state laws in
the treatment of the beginning of the calendar year or month from a business perspective.
From 1993 to 2009, nine states approved Section 1115 waivers permitting them to access
federal Medicaid funds for providing behavioral health services in IMDs. All except one of those
waivers were phased out by 2009. Due to the fact that those past waivers are not specifically on
SUD, their lack of emphasis on the quality of SUD care based on the continuum of care models,
and their reduced impact of being phased out several years ago, I do not consider the possible
remaining effect of them.
17
The federal and state governments have also implemented some other policies during my
study’s observation period, including the Comprehensive Addiction and Recovery Act (CARA) of
2016, the 21st Century Cures Act of 2016, and the SUPPORT Act of 2018. My analysis does not
consider their impact because they are national laws, and all states are subject to their effects. And
more importantly, CARA emphasizes access to medication for treatment in primary care, which
is less likely to influence facilities’ behavior in expansion or service provision. The 21st Century
Cures Act mainly addresses the leadership and accountability for behavioral health disorders at
the federal level. The SUPPORT Act provides grant funds for states to tackle the opioid crisis.
Though states may receive different levels of grant funds, the starting time of those funds is after
2018, which limits the impact because of the short time in my observation period. The levels also
depend on state-level factors such as the severity of the substance use problems, states’ demand for
treatment, other available resources, socioeconomic conditions, etc., which are already accounted
for in my model.
2.4.1.3 Additional Control Variables
State-level time-varying covariates are selected based on equation 13, which contains major
components of factors that may influence the number of facilities providing care. Since the two
primary outcomes of interest are supply indicators, I examine each outcome using the same set of
state-level covariates that potentially affect treatment supply. Those components include concur-
rent policies P in each state that might influence the need or access to SUD treatment (i.e., ACA
Medicaid expansion, recreational marijuana laws, parity laws requiring covered benefits for SUD),
the demand for services D (i.e., the share of state population enrolled in Medicaid, share serving
in armed forces, opioid prescribing rate, share with serious mental illness, share using cocaine,
and population size), socioeconomic variables X (i.e., unemployment rate, poverty rate, education
level, age 18-64 without health insurance), market competitiveness measures M (i.e., availability
of psychiatrist as a percentage of the population), and within-industry operational costs W (i.e.,
18
nursing assistant wage). Relevant literature has provided some evidence supporting the variable
selection. For example, higher wage level is associated with physician consultation length and
the number of doctor consultations (Kim et al., 2022; Lenhart, 2017; Qin et al., 2013), likely in-
fluencing the number of available treatment facilities and services. ACA Medicaid expansion is
found to be associated with increased alcohol and opioid treatment admissions in the US (Mulia
et al., 2022), which is also an important factor affecting the treatment supply. Table 2.2 provides a
descriptive summary of the variables in the study. Table 2.3 lists the data source of each variable.
2.4.2 Methods
2.4.2.1 Instrumental Variable Approach
Adopting a Section 1115 SUD demonstration waiver in a state is not random. Unobserved
state factors will likely influence a state’s decision to pursue a Section 1115 SUD waiver. The
unmeasured state factors can be correlated with the policy variable (adoption of 1115 SUD waiver)
and thus lead to a biased estimate of the effect of policy adoption on my treatment outcomes. A
political cloud usually surrounds a policy’s adoption. The interactions of policymakers, political
parties, governmental agencies, and other interest groups make it difficult to unravel this process.
For example, factors such as donations from nonprofit organizations to political parties and specific
interest groups’ connections with elected officials can make a difference in promoting a policy’s
adoption, at the same time, can also affect the market expansion of treatment and services. To test
whether there is such a concern, I use an instrumental variables approach to examine the effect of
1115 SUD waivers.
My selection of potential instrumental variables is based on three potential drivers of endo-
geneity. The first source of endogeneity comes from states’ involvement with Medicaid policies.
The application of those waivers is an interactive and negotiate process between state policymak-
ers and CMS, so the past attempts of a state that has applied for a Section 1115 waiver, regardless
19
of the outcomes, including approved, disapproved, expired, pending, terminated, or withdrawn,
could be an essential indicator of how experienced a state is in waiver application and a predictor
of success in being granted a waiver. With a similar idea, the number of past attempts to apply
for Section 1915 waivers may influence the resource allocated for 1115 waivers. Waiver authority
in section 1915 of the Social Security Act is another vehicle states can use to test new or existing
ways to deliver and pay for health care services in Medicaid and the Children’s Health Insurance
Program (CHIP). States’ applications for waivers under Section 1915 may affect their intent to
apply for Section 1115 waivers. The second source of endogeneity comes from political ideology
related to the need or desire to provide these benefits. For example, a state’s dominating political
ideology is likely to influence the decision-making process of policymakers. I construct a variable
to indicate whether a state’s controlling party is Democratic, Republican, or divided. I also calcu-
late the percentage of seats in the Senate and House of Representatives that the Democratic party
takes as a potential instrument candidate. Moreover, I include an instrument variable to indicate
whether the state’s controlling party is affiliated with the incumbent president. The third source of
endogeneity may come from other state policies that can influence the need to expand treatment
- such as how aggressively the state is tackling the opioid epidemic. I consider policies that are
likely to influence the waiver adoption decision but not likely to influence facilities’ decisions on
expansion or service provision. Specifically, those include 1) whether a state has laws requiring
prescribers to check the Prescription Drug Monitoring Program (PDMP) before prescribing con-
trolled substances; 2) whether a state has a Good Samaritan overdose prevention law; 3) whether a
state has an involuntary commitment law on substance users; 4) whether a state has any naloxone
access laws to provide civil or criminal immunity for lay responders administering naloxone or
allow for third-party prescribing to provide naloxone. As I am not certain whether any of these
potential sources of endogeneity is genuinely a concern, I go to the data and empirically assess
each. A summary of the instrumental variable candidates is listed in Table 2.4. The table also
gives the estimated association between the instrument and the 1115 SUD waiver. Then I select
the four variables significantly associated with adopting the 1115 SUD waiver to test their validity
20
as potential instruments. The test result is provided in Table 2.5, according to which I can rule
out the number of past attempts in applying for 1915 waivers, and the controlling party is affili-
ated with the incumbent president. The Cragg-Donald Wald F statistics of Good Samaritan law
and naloxone law are close, but the Kleibergen-Paap rk Wald F statistic of naloxone law is greater
than that of Good Samaritan law. Theoretically, the naloxone law is also more likely to affect the
adoption of 1115 waivers than the Good Samaritan law on overdose since expanding the naloxone
law can lead to more resource input in dispensing naloxone. I also consider the combination of
naloxone law and Good Samaritan law on overdose as the instruments. However, the F statistics
of the combination are much lower (Table 2.5). So I choose the presence of a naloxone access law
as the primary instrumental variable.
The naloxone access law is one of several policies states have adopted to mitigate the harms of
opioid misuse in recent years. These laws can expand the distribution and use of naloxone beyond
first responders and providers, thereby increasing the chances that someone who may accidentally
overdose on opioids can be resuscitated. These laws have been found to be significantly associated
with increased naloxone dispensing overall (Xu et al., 2018) and expanded coverage on naloxone
through Medicaid (Gangal et al., 2020; Gertner et al., 2018). Theoretically, if a state’s Medicaid
spends more money on naloxone dispensing, developing new programs could be more difficult
due to the budget-neutral requirement under Section 1115 demonstrations.
1
Increased Medicaid
spending in some areas reduces a state’s ability to develop new programs under Section 1115
demonstration, especially those with high demands on financial input such as IMD coverage. As
a result, if a state has a naloxone law that potentially increases Medicaid coverage on naloxone,
its Medicaid may be financially less likely to adopt a Medicaid 1115 SUD waiver if it does not
take money away from other priorities. Moreover, facilities’ operations or expansion usually are
not directly influenced by the eligibility expansion of a particular medication. In the model, I
account for factors associated with the demand for opioids, so the remaining variation left for the
1
Medicaid spending under Section 1115 demonstrations must be budget neutral, meaning that federal spending
under the demonstration cannot exceed projected costs in the absence of the demonstration.
21
naloxone access law to cover is the policy interactions described above. I also test the endogeneity
of naloxone law in predicting the adoption of an 1115 SUD waiver, and the endogeneity concern
is ruled out. So the naloxone law status satisfies the requirements of being a good instrument.
2.4.2.2 Empirical Models
I use a two-stage residual inclusion estimation method for evaluating the impact of Section
1115 SUD waivers on my two outcomes of interest (Terza et al., 2008). My desire to consider a
non-linear version of the second equation motivates using the two-stage residual inclusion model
instead of a traditional two-stage least square approach. Considering California is the only state
that adopted its waiver in 2015 and Virginia, the next state to adopt a waiver had the waiver at
the end of 2016(effective in 2017), and most of the other waivers became effective after 2017,
I choose the period for analysis to be 2014 -2020 to reduce the potential unaccountable impact
with an extended period. The first stage for the two parts of analysis is the same. I estimate
the likelihood that a state adopts a Section 1115 SUD waiver using a linear probability model
(ordinary least squares (OLS) regression) that includes measures of demand, policy indicators,
other socioeconomic and market measures, and my selected instrumental variable, the presence of
a naloxone law. From this first stage model, I calculate the predicted residual. To estimate the
waiver effect on the facility number, I use a mixed-effect Poisson model in the second stage, which
includes both the endogenous waiver variable and the predicted residual obtained from the first
stage and the other controls in the first stage. The mixed-effect Poisson model adds a variance of
the error term on top of the state-fixed effect. I chose the mixed-effect Poisson model for estimating
the change of facility numbers over time because it allows me to also account for correlation in the
error terms among states within the same region, which might influence the number of facilities
in those states. Such factors may include policy spill-overs, inter-state collaborations, and so on.
For example, if an entity owns multiple facilities across different states, the decision to open a new
22
subsidiary facility is interrelated, and the state-fixed effect does not capture that. The model can
be described as follows:
Waiver
it
=β
0
+β
1
∗ InstrumentalVariable
it
+β
2
∗ D
it
+β
3
∗ P
it
+β
4
∗ X
it
+β
5
∗ M
it
+β
6
∗ W
it
(2.19)
+α
i
+τ
t
+ε
it
FacilityCount
it
=γ
0
+γ
1
∗ Waiver
it
+γ
2
∗ D
it
+γ
3
∗ P
it
+γ
4
∗ X
it
+γ
5
∗ M
it
+γ
6
∗ W
it
(2.20)
+γ
7
∗ PredictedResidual+α
i
+ρ
i
+τ
t
+ε
it
P includes ACA Medicaid expansion status, recreational marijuana law, and Good Samaritan
law on SUD. D consists of the state’s population who are Medicaid beneficiaries, the percentage of
the state population in the armed forces or veterans, the opioid prescribing rate, the state prevalence
of cocaine use, the state prevalence of serious mental illness, and the population size. X includes
the unemployment rate, poverty rate, percentage of the population over age 25 with college or
more education, and those aged 18 to 64 without health insurance. M refers to the availability of
psychiatrists. W refers to the average hourly wage for nursing assistants. I also include the state
fixed effect α
i
, state random effect ρ
i
, and year fixed effect τ
t
in the first stage to account for the
unobserved time-invariant state heterogeneity and the shared national shocks each year that affects
SA treatment.
I use binary outcomes of whether a facility offers co-occurring treatment with facility-level
panel data. The first stage estimation is the same as the previous part. In the second stage, I
use logit regression for the first group and OLS for the second group. Since facilities may have
different operating and management methods, which can systematically affect their decisions over
time, I add a facility-level fixed effect ( λ
i
) for the facility-level panel data analysis. This facility-
level fixed effect allows me to address those facility-level decisions that are more closely related to
providing specific care or not, including co-occurring care in SA and MH. The logit model suffers
23
from the perfect separation problem here if adding state fixed effect. Alternatively, according to
the Census Bureau, I substitute the state-fixed effect with a regional fixed effect by dividing the
country into nine regions.
Data without imputed values are also included for a sensitivity analysis.
The second stage for estimating whether the facility offers SA and MH co-occurring treatment
services is estimated with the following equations:
FacilityO f f erCoTreatment
it
=δ
0
+δ
1
∗ Waiver
it
+δ
2
∗ D
it
+δ
3
∗ P
it
(2.21)
+δ
4
∗ X
it
+δ
5
∗ M
it
+δ
6
∗ W
it
+δ
7
∗ PredictedResidual+λ
i
+τ
t
+ε
it
I also conduct a sensitivity analysis using the percentage of SA treatment facilities in a state
offering co-occurring treatment each year with the public version of the N-SSATS data. The first
stage estimation is the same as the previous part, and the second stage is estimated with the equation
below, with the state fixed effect α
i
again:
ShareO fCoTreatmentFacilities
it
=γ
0
+γ
1
∗ Waiver
it
+γ
2
∗ D
it
+γ
3
∗ P
it
(2.22)
+γ
4
∗ X
it
+γ
5
∗ M
it
+γ
6
∗ W
it
+γ
7
∗ PredictedResidual+α
i
+τ
t
+ε
it
2.5 Results
2.5.1 First Stage Results
Results from the first stage model, estimated using ordinary linear regression, are presented
in Table 2.6. A state with any naloxone access law is 31% less likely to adopt 1115 SUD waivers
than states without. In the first stage diagnostic test, the Cragg-Donald Wald F statistic gives a
24
value of 20.56, and the Kleibergen-Paap rk Wald F statistic is 35.58. According to these F statistic
values, the selected instrumental variable has a promising significance level in predicting waiver
adoption. The Cragg-Donald Wald F statistic of 20.56 is greater than the critical value of 16.38
given by Stock (2005). I also consider the “tF” test procedure explained by Lee et al. (2022) to
assess the validity of my instrument and obtain the valid t-ratio inference for IV . According to Lee
and McCary’s adjusted critical value function, I obtain a standard error inflated by 1.12, which
would deflate my original t-statistic (-5.96) by 1.06. The adjusted test result is still significant,
indicating my instrument is valid.
2.5.2 Impact of Waivers on the Availability of Treatment Facilities and Services
I find that the adoption of a Section 1115 SUD waiver is significantly associated with the
total number of treatment facilities (Table 2.7). The log difference of expected facility counts after
adopting a waiver changed by a positive 0.146, which means that after adopting 1115 SUD waiver,
it is estimated, on average, a state has 1.16(i.e., e
0.146
) times more SA treatment facilities than
states that do not have a waiver. The significant association between the predicted residual from
the 1st stage and the number of facilities also confirm the existence of policy endogeneity and the
need to adjust for it to reduce bias when estimating its effects. The endogeneity problem tends to
underestimate the coefficient (Table 2.8). By accounting for the endogeneity of waiver adoption,
the results suggest a more substantial effect than ignoring the endogeneity. Since the estimated
variance of the constant per state (the state-level random effect) is 0.358, which is non-zero and
not close to zero, I have evidence that the random effect is beneficial.
Using the facility-level panel data with imputations, I find that adopting a waiver causes an
increase of around 28% (i.e., e
0.25
− 1) in a facility’s odds of providing co-occurring treatment and
an increase of around 89% (i.e., e
0.636
− 1) in a residential facility’s odds of providing co-occurring
treatment (Table 2.9).
25
Further, when repeating the steps for sensitivity analysis with no imputed values, I obtain a
result that all facilities are more likely to offer co-occurring treatment after a state adopts a Section
1115 SUD waiver (Table 2.10). The estimates from the logit regression indicate that adopting
Section 1115 SUD waivers leads to a 0.278 increase in the log-odds ratio of offering co-occurring
treatment. In other words, there is a 32% (i.e., e
0.278
− 1) increase in the odds that a SA treatment
facility in my sample offers co-occurring treatment when a state adopts a Section 115 SUD waiver.
Again, this effect is larger among residential facilities in the sample, dropping imputed values. In
these models, I find that the adoption of a Section 1115 SUD waiver results in a 0.543 increase
in the log-odds ratio of offering co-occurring treatment among residential SA treatment facilities,
which implies a 72% (i.e., e
0.543
− 1) increase in the odds of providing co-occurring SA and MH
treatment.
What is worth being noted here is that the predicted residual from the first stage is only
significantly associated with all facilities in general. If only a residential facility sample is used
for analysis, the predicted residual’s association from the first stage is no longer significant. It
tells that the endogeneity of the waiver does not specifically affect the residential facilities but
the treatment facilities in general. The uncaptured influential force (e.g., lobbying on resource
allocation/treatment input) is more likely to promote a broad market expansion and integration
rather than targeting residential benefits.
I also use a linear probability model, with the public N-SSATS cross-sectional data, to es-
timate the waiver effect with the percentages of facilities that provide co-occurring SA and MH
treatment in each state for each year as the outcome variable. I use the linear probability model here
because the logit and probit model suffers from perfect separation problems if adding state fixed
effect, an important control in this state-level analysis with aggregated cross-sectional data. The
adoption of the SUD waiver is associated with approximately a 5.5 percentage point increase in
the percentage of SA treatment facilities that provide SA and MH co-occurring treatment. Again,
the effect is greater among residential SA treatment facilities. Adopting a waiver results in an 11.5
26
percentage point increase in the share of SA residential facilities offering co-occurring treatment
in SA and MH. Estimates are exhibited in Table 2.11.
2.5.3 Supplementary Analyses
I conduct separate analyses among all facilities and residential facilities separately. Facilities
with different ownership types are also compared, including private for-profit, private, not-for-
profit, and public or government-owned facilities.
I have only run analyses based on state directories of SA treatment facilities in the main anal-
ysis. Though the distinction between SA and MH treatment facilities is not always clear, facilities
that were licensed as MH facilities may respond differently to the Section 1115 waivers than SA-
licensed treatment facilities. To supplement my existing analysis, I also compare the results of SA
treatment facilities to MH treatment facilities and see if SA treatment facilities respond more to
SUD waiver. To test this, I conduct similar analyses using the National Mental Health Services
Survey (N-MHSS) data, which captures MH-licensed treatment facilities that complement facil-
ities in N-SSATS. It collects information on the location, organization, structure, services, and
utilization of MH treatment facilities. N-MHSS data may contain some facilities that overlap with
those in N-SSATS data, which are licensed as both mental health and substance use treatment facil-
ities. Still, a significant amount of specialized substance treatment facilities, such as detoxification
facilities, transitional housing, halfway house, or sober home, are not included in N-MHSS data.
2.5.3.1 Heterogeneity in Policy Effects Due to Facility Ownership
Facilities with different ownership types may vary in response to this state policy change,
particularly if some private treatment facilities are less likely to accept Medicaid insurance. Non-
profit or public treatment facilities may be more willing to accept Medicaid payment but less
27
willing to make operational changes. So the differences among these different ownership types are
worth exploring.
During the period 2014-2020, on average, there were 5,377 private for-profit facilities, 7,618
private non-profit facilities, and 1,567 public or government-owned facilities each year, so private
for-profit facilities represent a good share of all facilities in the marketplace.
To test whether ownership type influences the responsiveness to this Medicaid waiver adop-
tion, I conduct separate analyses by ownership type. The results are in Table 2.12. Interestingly,
private facilities drive the general increase in the share of SA treatment facilities offering co-
occurring treatment in response to this policy. While the adoption of Section 1115 SUD waivers
is associated with a rise in the percentage of SA treatment facilities providing co-occurring SA
and MH treatment in both private for-profit and non-profit facilities, I find a more prominent effect
among private for-profit facilities (12.9 percentage point increase) compared to private non-profit
facilities (7.06 percentage point increase). I see no impact on public or governmental agencies.
2.5.3.2 Heterogeneity Based on Primarily Mental Health Versus Primarily Substance Abuse
Facilities
It is possible that facilities licensed as MH treatment facilities respond differently to the Sec-
tion 1115 waivers compared to facilities licensed as SA treatment facilities. To test this, I conduct
similar analyses using the National Mental Health Services Survey (N-MHSS) data, another survey
conducted by SAMHSA but includes facilities that are licensed as MH treatment facilities. Again,
SAMHSA reports information on the location, organization, structure, services, and utilization of
MH treatment facilities. N-MHSS data contains about 12,220 facilities annually nationwide from
2014-2020. The results are displayed in Table 2.13. Interestingly, I do not find a significant associ-
ation between waiver adoption and the percentage of mental health facilities offering co-occurring
28
treatment. One explanation is probably due to the eligibility restriction of receiving SUD pay-
ment from Medicaid if licensed as an MH facility. Another possible reason could be that many
MH facilities already offer co-occurring treatment, making mental health facilities less sensitive
to the 1115 SUD waiver. I run a paired t-test between the percentages calculated using NSSATS
and NMHSS separately. The results are presented in Table 2.14. The percentage of co-occurring
treatment among MH facilities is significantly larger than that of SA treatment facilities.
2.6 Conclusions
Medicaid Section 1115 SUD waivers are one of several policies to improve the substance
abuse crisis in the United States. This study advances the understanding of the impact of Section
1115 SUD waivers on SA treatment facilities. According to the results, a state implementing a
waiver causes a 1.16 increase in the incidence rate of having SA treatment facilities on average
by contrasting the average value of states with waiver with states without waiver, suggesting a
sizeable impact on the extensive margin. The impact of Section 1115 SUD waivers on the intensive
margin is even greater. Facilities are more likely to provide co-occurring treatment services in
SA and MH. A facility’s odds of providing co-occurring treatment increase by 28% after the state
adopts the waiver. A residential facility’s odds of offering co-occurring treatment increased by 89%
with the adoption of a waiver. Similarly, the adoption of the SUD waiver causes approximately a
5.5 percentage point increase in the share of SA facilities that provide SA and MH co-occurring
treatment. Among residential facilities, adopting a waiver results in an 11.5 percentage point
increase in the percentage offering co-occurring treatment in SA and MH.
My study takes a longer observation period than previous studies, with a range of states adopt-
ing Medicaid Section 1115 SUD waivers during 2015-2020, and considers a broader margin on
which the waivers generate impact (Maclean et al., 2021; O’Brien et al., 2022). I also address
the endogeneity problem of implementing waivers that other studies do not address adequately if
29
using a difference-in-difference design only. The endogeneity problem tends to underestimate the
coefficient (Table 2.12). By accounting for the endogeneity of waiver adoption, the results suggest
a larger effect compared to ignoring the endogeneity.
Supplementary analyses suggest that the change in the availability of treatment comes pri-
marily from private, for-profit facilities. With the implementation of Section 1115 SUD waivers,
there is a 12.9 percentage point increase among private for-profit facilities and a 7.06 percent-
age point increase in private non-profit facilities. In comparison, no such effect is found among
public or governmental agencies. This result further indicates that the theoretical framework for
motivating these hypotheses, while overly simplified, is still helpful in describing the underlying
behavior. Private facilities are more likely to be motivated by extra profit and make operational
adjustments. While on the contrary, even though public or governmental facilities should be more
likely to accept Medicaid payments, they are less likely to make service changes driven by extra
profit.
Importantly, I also find in sensitivity analyses that waiver adoption has no significant impact
on the share of mental health facilities offering co-occurring treatment. These facilities are much
more likely to provide co-occurring treatment in the first place than SA treatment facilities.
However, my study has several limitations. First, to convert the original N-SSATS data into
a panel dataset, I have to impute the values that are not reported in the survey. While the data
also faces a missing value problem because of those unreported issues, if I remove all the imputed
observations, I conduct two sets of analyses with imputed values and without, each of which is
not perfect regarding the data availability. To compensate for that, I have done further analyses
taking percentages as my outcome of interest, assuming reported facilities are randomly distributed
across years. Second, due to the limited data and information availability, it is difficult to track the
specific provisions of the nine states that had approved waivers providing federal Medicaid funds
for behavioral health services in IMDs and analyze their potential effect on adopting new waivers.
Since those past waivers did not target SUD benefits, I assume the new waivers are completely
30
independent. Third, though my model gives some explanations of the possible mechanisms behind
my hypothesis and results, it relies on many assumptions and fails to consider more factors such
as costs, facility types, etc. Finally, the concept of “integrated care” also includes primary physical
care and different ways of collaboration. My study only addresses a part of it. Later research may
look into other service expansions and inter-facility cooperation in providing care.
Nonetheless, despite these limitations, the analysis shows that adopting the Section 1115 SUD
waivers can be helpful for promoting SUD treatment facilities, particularly better-integrated care
for SA and MH treatment. The effects are not simply a shift among existing SA treatment providers
(the intensive margin) but also encouraging new facilities to pop up (the extensive margin). It
provides further evidence of the importance of payment reform (in this case, the coverage of in-
tegrative care through Medicaid) on access to and the delivery of evidence-based treatment. By
addressing the endogeneity with policy adoption, the findings suggest a more substantial effect
than the condition without addressing the issue.
31
Figure 2.1: Number of States with Effective Section 1115 Demonstration Waiver in SUD 2015-
2020 by Year
Notes: Years refer to when the waiver becomes effective. SUD is Substance Use Disorder. Data comes from RAND
Opioid Policy Tools and Information Center (OPTIC) policy database
32
Table 2.1: 2015-2020 States with Effective Section 1115 Demonstration Waiver in SUD
Effective Year N States
2015 1 CA
2017 5 V A, MD, MA, NJ, UT
2018 10 IL, IN, KY , LA, NH, PA, VT, WA, WV , WI
2019 10 AK, DE, KS, MI, MN, NE, NM, NC, OH, RI
2020 2 DC, ID
Total 28
Notes:Data comes from RAND Opioid Policy Tools and Information Center (OPTIC) policy database. SUD
is Substance Use Disorder. The effective year is the year when the waiver is adopted regardless of the specific
point in time.
33
Table 2.2: Summary Statistics of Key Variables 2014-2020
Count Mean Std. Dev Min Max
Outcome Variables
Number of SA treatment facilities 357 318 293 35 1888
Whether a SA treatment facility
provides co-occurring disorders treatment 110,795 0.466 0.499 0 1
Percentage of SA treatment providing co-occurring
disorders treatment in residential facilities 357 0.553 0.135 0.182 1
Percentage of SA treatment providing
co-occurring disorders treatment 357 0.484 0.107 0.164 0.781
Independent Variables
Section 1115 SUD waivers 357 0.182 0.372 0 1
ACA Medicaid Expansion Status 357 0.619 0.483 0 1
Parity law in SUD 357 0.444 0.490 0 1
Recreational marijuana law 357 0.145 0.349 0 1
Share of Medicaid enrollees 357 0.217 0.066 0.090 0.428
Unemployment rate 357 0.049 0.017 0.024 0.128
Poverty rate 357 0.121 0.033 0.037 0.231
Higher education rate 357 0.614 0.054 0.441 0.778
Armed force rate 357 0.007 0.010 9.93e-06 0.062
Age 18 to 64 without health insurance rate 357 0.127 0.051 0.038 0.285
Availability of psychiatrist 357 0.000 0.000 0.000 0.001
Opioid prescribing rate 357 0.628 0.214 0.250 1.352
Serious mental illness rate 357 0.007 0.002 0.004 0.014
Using cocaine rate 357 0.016 0.005 0.007 0.040
Nursing assistant wage 357 13.850 1.990 9.900 20.430
Population 357 6,364,066 7,203,519 579,054 3.94e+07
Notes: SA is substance abuse. A co-occurring disorder is defined as the coexistence of both a mental
illness and a substance use disorder. The first and second outcome variables are based on reconstructed
and imputed panel version N-SSATS data. The third and fourth outcome variables are based on the public
version of N-SSATS data. Please see the data section for more details.
34
Table 2.3: Data Sources of Key Variables
Variables Data sources
Number of facilities National Survey of Substance Abuse Treatment Services (N-SSATS)
Providing co-occurring treatment National Survey of Substance Abuse Treatment Services (N-SSATS)
Percentage of co-occurring treatment National Survey of Substance Abuse Treatment Services (N-SSATS)
ACA Medicaid expansion Kaiser Family Foundation
Parity law in SUD Paritytrack.org, NCSL
Recreational marijuana law Kaiser Family Foundation
Share of Medicaid enrollees Kaiser Family Foundation
Unemployment rate University of Kentucky Center for Poverty Research
Poverty rate University of Kentucky Center for Poverty Research
Higher education rate American Community Survey (ACS)
Armed force rate American Community Survey (ACS)
Age 18 to 64 without health insurance rate American Community Survey (ACS)
Availability of psychiatrists Area Health Resource File
Opioid prescribing rate Center for Disease Control and Prevention (CDC)
Serious mental illness rate National Survey of Drug Use and Health (NSDUH)
Using cocaine rate National Survey of Drug Use and Health (NSDUH)
Nursing assistant wage US Bureau of Labor Statistics
Population University of Kentucky Center for Poverty Research
Note: N-SSATS data is reconstructed into a panel version data for the main analysis.
35
Table 2.4: Potential Instrumental Variable Selection
1115 SUD waiver
Number of past attempts in 1115 waiver applications 0.042
Number of past attempts in 1915 waiver applications -0.061*
Percentage of Democratic seats 0.327
State controlling party (Democratic) -0.068
Controlling party affiliated with president -0.072*
Must access PDMP -0.108
Good Samaritan law -0.299***
Involuntary commitment law -0.008
Naloxone law -0.312***
Notes: The first stage model is used to predict the likelihood of adopting 1115 SUD waiver, with different
instrument candidates. PDMP is Prescription Drug Monitoring Program. Must access PDMP means that
a state has laws requiring prescribers to check PDMP before prescribing controlled substances. Statistical
significance is indicated as follows: * indicates significance at the 10% level, ** indicates significance at the
5% level, and *** indicates significance at the 1% level.
Table 2.5: F-statistics of Potential Instrumental Variables
Cragg-Donald Wald Kleibergen-Paap rk Wald
F statistic F statistic
Number of past attempts 4.834 3.630
in 1915 waiver application
Controlling party 2.924 3.474
affiliated with president
Good Samaritan law 22.989 30.279
Naloxone law 20.558 35.576
Good Samaritan + naloxone law 16.418 26.217
Notes: Potential instrumental variables are selected based on the results in Table 2.4. The validity of the
instrumental variables is tested and F-statistics are reported.
36
Table 2.6: First Stage Result Estimating the Likelihood of a State Adopting a Section 1115 SUD
Waiver, with a Linear Probability Model
V ARIABLES Section 1115 SUD waivers
Any naloxone law -0.313***
(0.052)
ACA Medicaid expansion status 0.098
(0.110)
Parity law in SUD -0.201***
(0.072)
Recreational marijuana law 0.099
(0.085)
Share of Medicaid enrollees 1.093
(1.249)
Unemployment rate 1.611
(2.656)
Poverty rate -1.472
(1.1563)
Higher education rate 0.169
(2.5468)
Armed force rate 5.047
(5.1138)
Age 18 to 64 without health insurance rate -1.989
(2.2175)
Availability of psychiatrists -319.200
(3,965.96)
Opioid prescribing rate 0.216
(0.3592)
Serious mental illness rate -41.420*
(24.270)
Using cocaine rate -8.490
(6.880)
Nursing assistant wage 0.002
(0.063)
Population -1.80e-07**
(8.41e-08)
State FE Yes
Year FE Yes
Constant 1.010
(1.670)
Observations 357
Diagnostic statistics
Cragg-Donald Wald F statistic 20.560
Kleibergen-Paap rk Wald F statistic 35.580
R-squared 0.644
Notes: The first variable is the instrumental variable. Robust standard errors are included in parentheses.
Statistical significance is indicated as follows: * indicates significance at the 10% level, ** indicates signif-
icance at the 5% level, and *** indicates significance at the 1% level.
37
Table 2.7: Examining the Waiver Effect on the Number of Facilities, with a Mixed-effect Poisson
Model
V ARIABLES Number of facilities Number of facilities
(coefficients converted to IRR)
Section 1115 SUD waivers 0.146*** 1.160***
(0.047) (0.054)
Predicted Residual from the 1st stage -0.100** 0.905**
(0.049) (0.044)
Parity law in SUD -0.009 0.991
(0.021) (0.021)
ACA Medicaid expansion status 0.034 1.030
(0.025) (0.025)
Recreational marijuana law -0.044** 0.957**
(0.019) (0.018)
Share of Medicaid enrollees -0.561** 0.570**
(0.235) (0.134)
Unemployment rate 1.060* 2.900*
(0.569) (1.650)
Poverty rate 0.432 1.540
(0.308) (0.474)
Higher education rate -0.729 0.483
(0.654) (0.316)
Armed force rate 0.607 1.840
(1.180) (2.160)
Age 18 to 64 without health insurance rate -0.073 0.930
(0.516) (0.479)
Availability of psychiatrists -1,090.11 0.000
(856.24) 0.000
Opioid prescribing rate -0.234*** 0.792***
(0.080) (0.063)
Serious mental illness rate 9.430 12,518
(6.540) (81,850.460 )
Using cocaine rate -6.80*** 0.001***
(1.600) (0.002)
Nursing assistant wage -0.002 0.998
(0.012) (0.012)
Population 6.94e-08*** 1.000
(1.05e-08) (1.05e-08)
Constant 5.708*** 301.370
(0.477) (143.610)
State-level random effect variance 0.338 0.338
(0.071) (0.071)
State FE Yes Yes
Year FE Yes Yes
Observations 357 357
Number of groups 51 51
Notes: IRR is Incidence Rate Ratio. Standard errors are included in parentheses. In this model, I use
enhanced panel data aggregated to the state level. Statistical significance is indicated as follows: * indicates
significance at the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the
1% level.
38
Table 2.8: Examining the Waiver Effect on the Number of Facilities without Addressing Endo-
geneity, with a Mixed-Effect Poisson Model
(1) (2)
V ARIABLES Number of facilities Number of facilities
(coefficients converted to IRR)
Section 1115 SUD waivers 0.054*** 1.050***
(0.013) (0.014)
Parity law in SUD -0.031* 0.970*
(0.018) (0.018)
ACA Medicaid expansion status 0.034 1.030
(0.024) (0.025)
Recreational marijuana law -0.032* 0.968*
(0.018) (0.018)
Share of Medicaid enrollees -0.450** 0.638**
(0.229) (0.146)
Unemployment rate 1.162** 3.200**
(0.567) (1.810)
Poverty rate 0.329 1.390
(0.304) (0.422)
Higher education rate -0.621 (0.537)
(0.655) (0.352)
Armed force rate 1.226 3.410
(1.136) 3.870
Age 18 to 64 without health insurance -0.260 0.771
(0.509) (0.392)
Availability of psychiatrists -1,460* 0.000
(848.8) 0.000
Opioid prescribing rate -0.216*** 0.8060***
(0.0796) (0.0642)
Serious mental illness 5.075 159.996
(6.198) (991.72)
Using cocaine rate -7.189*** 0.001***
(1.595) (0.001)
Nursing assistant wage 0.003 1.003
(0.011) (0.011)
Population 6.16e-08*** 1
(1.04e-08) (1.04e-08)
State-level random effect (variance of error term) 0.358 0.358
(0.078) (0.078)
Constant 5.708*** 301.310***
(0.480) (144.710)
State FE Yes Yes
Year FE Yes Yes
Observations 357 357
Number of groups 51 51
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Notes: IRR is Incidence Rate Ratio. Standard errors are included in parentheses. Statistical significance is
indicated as follows: * indicates significance at the 10% level, ** indicates significance at the 5% level, and
*** indicates significance at the 1% level.
39
Table 2.9: Examining the Waiver Effect on the Facility-level Likelihood of Offering Co-occurring
Treatment, with a Logit Regression Model (with Imputation)
(1) (2) 3) (4)
V ARIABLES Co-occurring Co-occurring Co-occurring treatment Co-occurring
treatment treatment treatment treatment
in Odds Ratio (residential facilities) (residential facilities)
in Odds Ratio
Section 1115 SUD waivers 0.250* 1.280* 0.636** 1.890**
(0.141) (0.181) (0.308 ) (0.582)
Predicted Residual -0.278* 0.758* -0.527 0.591
from the 1st stage (0.151) (0.115) (0.328 ) (0.194)
ACA Medicaid 0.012 1.010 -0.152 0.859
expansion status (0.088) (0.089) (0.203 ) (0.175)
Parity law in SUD 0.253*** 1.290*** 0.216 1.240
(0.067) (0.086) (0.135 ) (0.167)
Recreational marijuana law -0.059 0.942 -0.164 0.849
(0.066) (0.062) (0.132 ) (0.112)
Share of Medicaid enrollees -1.750** 0.174** -0.564 0.569
(0.839) (0.146 ) (1.680) (0.957)
Unemployment rate -1.400 0.248 -2.010 0.134
(1.200) (0.297) (2.510) (0.337)
Poverty rate 1.47 4.350 3.340 28.210
(1.100) (4.780) (2.360) (66.580)
Higher education rate 4.950** 140.980** 1.590 4.890
(2.390) (336.380) (5.120) (25.040)
Armed force rate -7.000 0.001 -6.900 0.001
(4.360) (0.004) (9.710) (0.010)
Age 18 to 64 without -0.471 0.624 0.128 1.140
health insurance rate 1.220 (0.759) (2.540) (2.890)
Availability of psychiatrists 10,134.750** 0.000** 26,320.87*** 0.000***
(4715.20) (0.000) (10,033.82 ) (0.000)
Opioid prescribing rate -0.214 0.808 0.060 1.060
(0.260) (0.210) (0.592 ) (0.628)
Serious mental illness rate 61.620*** 5.76e+26*** 95.890** 4.43e+41**
(19.390) (1.12e+28) ( 43.480 ) (1.93e+43 )
Using cocaine rate -6.100 0.002 -14.700 4.14e-07
(5.880) (0.013) (13.860) ( 5.74e-06 )
Nursing assistant wage 0.112*** 1.120*** 0.192** 1.210**
(0.039) (0.044) (0.080 ) (0.097)
Population 9.45e-08 1 -1.93e-07 0.999
(7.40e-08) (7.40e-08) (1.46e-07 ) (1.46e-07)
Facility FE Yes Yes Yes Yes
Region FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Observations 42,011 42,011 9,773 9,773
Number of facilities 6,912 6,912 1,630 1,630
Notes: Standard errors are included in parentheses. In this model, I use the enhanced panel data at the
facility level, with a facility-level fixed effect. Statistical significance is indicated as follows: * indicates
significance at the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the
1% level.
40
Table 2.10: Examining the Waiver Effect on the Facility-level Likelihood of Offering Co-occurring
Treatment, with a Logit Regression Model (with No Imputation)
(1) (2) 3) (4)
V ARIABLES Co-occurring Co-occurring Co-occurring treatment Co-occurring
treatment treatment treatment treatment
in Odds Ratio (residential facilities) (residential facilities)
in Odds Ratio
Section 1115 SUD waivers 0.278* 1.320* 0.543* 1.720**
(0.146) (0.192) (0.315 ) (0.542)
Predicted Residual -0.321** 0.726** -0.440 0.644
from the 1st stage (0.156) (0.113) (0.335 ) (0.216)
ACA Medicaid 0.007 1.007 -0.080 0.923
expansion status (0.091) (0.092) (0.207 ) (0.191)
Parity law in SUD 0.276*** 1.310*** 0.233* 1.260*
(0.0691) (0.091) (0.138 ) (0.174)
Recreational marijuana law -0.062 0.940 -0.180 0.835
(0.068) (0.064) (0.133 ) (0.111)
Share of Medicaid enrollees -1.830** 0.161** -0.535 0.586
(0.863) (0.146 ) ( 1.710) (0.999)
Unemployment rate -1.480 0.228 -1.390 0.248
(1.250) (0.297) (2.600) (0.645)
Poverty rate 1.340 3.810 3.040 20.970
(1.130) (4.780) (2.400) (50.290)
Higher education rate 5.520** 250.620** 3.150 23.240
(2.470) (336.380) (5.210) (121.160)
Armed force rate -7.960* 0.0003* -5.290 0.005
(4.580) (0.004) (10.000) (0.050)
Age 18 to 64 without -0.491 0.612 0.608 1.840
health insurance rate 1.250 (0.759) (2.580) (4.730)
Availability of psychiatrists 9,994.850** 0.000** 26,304.360*** 0.000***
(4883.100) (0.000) (10,271.030 ) (0.000)
Opioid prescribing rate -0.237 0.789 0.086 1.090
(0.271) (0.210) (0.604) (0.658)
Serious mental illness rate 63.470*** 3.66e+27*** 99.620** 1.84e+43**
(20.260) (1.12e+28) (44.640) (8.19e+44 )
Using cocaine rate -6.420 0.002 -19.390 3.79e-09
(6.150) (0.013) (14.250) (5.40e-08 )
Nursing assistant wage 0.114*** 1.120*** 0.196** 1.220**
(0.041) (0.044) (0.082) (0.099)
Population 1.05e-07 1 -1.49e-07 0.999
(7.70e-08) (7.70e-08) (1.50e-07 ) (1.50e-07 )
Facility FE Yes Yes Yes Yes
Region FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Observations 39,835 39,835 9,469 9,469
Number of facilities 6,910 6,910 1,632 1,632
Notes: Standard errors are included in parentheses. In this model, I use the enhanced panel data at the
facility level, with a facility-level fixed effect. Statistical significance is indicated as follows: * indicates
significance at the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the
1% level.
41
Table 2.11: Examining the Waiver Effect on the Percentage of SA Facilities Offering Co-occurring
Disorder Treatment, with a Linear Probability Model
(1) (2)
V ARIABLES Percentage for all SA facilities Percentage for residential SA facilities
Section 1115 SUD waivers 0.055* 0.115*
(0.032) (0.066)
ACA Medicaid expansion status 0.015 0.058**
(0.015) (0.023)
Parity law in SUD 0.014 0.019
(0.014) (0.027)
Recreational marijuana law -0.011 -0.031
(0.014) (0.025)
Share of Medicaid enrollees -0.427*** -0.176
(0.147) (0.287)
Unemployment rate 0.055 -0.402
(0.316) (0.506)
Poverty rate -0.224 -0.503
(0.189) (0.345)
Higher education rate 0.196 -1.467*
(0.356) (0.756)
Armed force rate 0.409 -0.061
(0.553) (0.738)
Age 18 to 64 without health insurance -0.349 -0.214
(0.274) (0.498)
Availability of psychiatrists 449.600 2,957**
(594.000) (1,274)
Opioid prescribing rate 0.021 -0.104
(0.049) (0.105)
Serious mental illness rate 1.702 -1.862
(4.469) (8.114)
Using cocaine rate 1.993* 4.123**
(1.019) (1.805)
Nursing assistant wage -0.007 0.024*
(0.007) (0.013)
Population 3.45e-08*** 1.09e-08
(1.10e-08) (2.36e-08)
Constant 0.224 0.845*
(0.240) (0.501)
State FE Yes Yes
Year FE Yes Yes
Observations 357 357
R-squared 0.898 0.763
Notes: SA stands for Substance Abuse. Standard errors are included in parentheses. In this model, I use
the public N-SSATS data aggregated to the state level. I use OLS with the ivreg2 command in Stata to
estimate and only list the 2nd stage output here. Statistical significance is indicated as follows: * indicates
significance at the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the
1% level.
42
Table 2.12: Examining the Waiver Effect on the Percentage of SA Facilities Offering Co-occurring
Disorder Treatment by Facility Ownership Type, with a Linear Probability Model
(1) (2) (3)
V ARIABLES Private For-profit Private Not-for-profit Public Agencies
Section 1115 SUD waivers 0.129* 0.071* -0.122
(0.075) (0.037) (0.104)
ACA Medicaid expansion status -0.018 0.073*** 0.083*
(0.030) (0.023) (0.045)
Parity law in SUD 0.027 0.007 0.080**
(0.030) (0.017) (0.037)
Recreational marijuana law -0.035 -0.019 -0.001
(0.035) (0.016) (0.036)
Share of Medicaid enrollees 0.285 -0.350* -0.556
(0.448) (0.208) (0.472)
Unemployment rate -1.234* 0.005 0.638
(0.719) (0.413) (0.881)
Poverty rate 0.014 -0.427* -0.731
(0.468) (0.224) (0.665)
Higher education rate 1.470 -0.057 -0.134
(0.973) (0.547) (1.576)
Armed force rate 1.077 0.518 0.745
(1.478) (0.712) (1.394)
Age 18 to 64 without health insurance rate 0.847 0.311 -0.862
(0.826) (0.373) (0.869)
Availability of psychiatrists 621.900 -372.500 -1,643
(1,545) (827.600) (2,012)
Opioid prescribing rate 0.181 -0.015 -0.337
(0.117) (0.061) (0.208)
Serious mental illness rate 9.227 -5.333 -10.700
(10.490) (5.490) (18.730)
Using cocaine rate 5.364*** 0.886 2.248
(2.078) (1.202) (4.197)
Nursing assistant wage -0.028 0.003 -0.039*
(0.017) (0.008) (0.020)
Population 7.57e-08*** 2.44e-08 1.27e-08
(2.38e-08) (1.48e-08) (4.01e-08)
Constant -1.051 0.355 1.668
(0.646) (0.350) (1.190)
State FE Yes Yes Yes
Year FE Yes Yes Yes
Observations 357 357 356
R-squared 0.679 0.874 0.599
Notes: SA stands for Substance Abuse. Standard errors are included in parentheses. In this model, I use
the public N-SSATS data aggregated to the state level. I use OLS with the ivreg2 command in Stata to
estimate and only list the 2nd stage output here. Statistical significance is indicated as follows: * indicates
significance at the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the
1% level.
43
Table 2.13: Examining the Waiver Effect on the Percentage of Facilities Offering Co-occurring
Disorder Treatment, with a Linear Probability Model (on Licensed MH Treatment Facilities)
(1)
V ARIABLES Percentage offering co-occurring treatment
Section 1115 SUD waivers -0.018
(0.036)
ACA Medicaid expansion status -0.023*
(0.013)
Parity law in SUD -0.011
(0.015)
Recreational marijuana law -0.025**
(0.012)
Share of Medicaid enrollees 0.145
(0.162)
Unemployment rate -0.211
(0.432)
Poverty rate -0.301
(0.203)
Higher education rate 0.024
(0.444)
Armed force rate 0.265
(0.931)
Age 18 to 64 without health insurance -0.402
(0.316)
Availability of psychiatrists 498.800
(709.200)
Opioid prescribing rate -0.133**
(0.058)
Serious mental illness rate 2.215
(4.670)
Using cocaine rate -0.136
(1.196)
Nursing assistant wage 0.010
(0.008)
Population -8.55e-09
(1.40e-08)
Constant 0.757**
(0.298)
State FE Yes
Year FE Yes
Observations 357
R-squared 0.901
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Notes: MH is Mental Health. Standard errors are included in parentheses. In this model, I use the public
N-NMHSS data aggregated to the state level. I use OLS with the ivreg2 command in Stata to estimate and
only list the 2nd stage output here. Statistical significance is indicated as follows: * indicates significance at
the 10% level, ** indicates significance at the 5% level, and *** indicates significance at the 1% level.
44
Table 2.14: Comparing Percentage of Facilities Offering Treatment in Mental Health and Sub-
stance Abuse among Primarily Substance Abuse Facilities and Primarily Mental Health Facilities,
with Paired-t Test
Variable Mean Std. Dev. N
Percentage among Licensed Mental Health Facilities 0.707 0.119 357
Percentage among Licensed Substance Abuse Facilities 0.484 0.107 357
Difference 0.223 0.137
t statistic 30.900
Degrees of freedom 356
H0: mean(difference) = 0
Ha: mean(difference)> 0
Pr(T> t) = 0.000
Notes: I use N-SSATS data and N-MHSS data to compare. Percentages refer to the share of facilities
offering MH and SA care in each state for each year. Percentages are calculated at the state-year level. N
refers to the total observations of 51 states in 7 years 2014-2020.
45
Chapter 3
Does Medicaid Institutions for Mental Diseases Exclusion Waiver Crowd Out the Access to
Treatment Among Commercially Insured Substance Use Disorder Patients?
3.1 Introduction
Substance use disorders are a leading public health problem in the United States. More than
932,000 people have died from a drug overdose over the past two decades (CDC, 2021). The
rising trend of overdose death has been extremely concerning during the last decade, from 37,000
in 2010 to more than 100,000 in 2020 (CDC, 2021; Saloner et al., 2022). However, access to
quality treatment has been limited over past decades due to high costs, lack of insurance coverage,
and low availability of treatment providers (Saloner et al., 2022). The National Survey on Drug Use
and Health (NSDUH) reports that among people aged 12 or older in 2021 who needed substance
use treatment in the past year, only 9.6% received any substance use treatment
1
, and fewer (6.8%)
received substance use treatment at a specialty facility
2
in the past year (SAMHSA, 2021). People
suffering from substance use disorder have a high risk of developing comorbid physical and mental
health conditions, which result in higher spending on treatment (CMS, 2015).
1
According to SAMHSA, receiving any substance use treatment includes treatment received at any location, such
as a hospital (inpatient), rehabilitation facility (inpatient or outpatient), mental health center, emergency room, private
doctor’s office, self-help group, prison/jail, and virtual services.
2
Specialty facilities for substance use treatment include drug and alcohol rehabilitation facility (inpatient or out-
patient), hospital (inpatient only), or mental health center
46
Public policies have been seeking ways of improving access to treatment services for SUD
patients through insurance coverage. The Paul Wellstone and Pete Domenici Mental Health Parity
and Addiction Equity Act (MHPAEA) of 2008, which became effective in 2010, requires most in-
surance plans to provide the same mental health and addiction care benefits as other medical care.
MHPAEA goes beyond the 1996 Mental Health Parity Act by extending parity to addiction ser-
vices in addition to specific mental health conditions and relaxing quantitative and non-quantitative
treatment limits. The Affordable Care Act (ACA) 2010 specified SUD treatment as an essential
health benefit and prohibited insurance plans from discriminating against people based on their
conditions. In 2014, the ACA further expanded Medicaid to cover adults below 138 percent of the
federal poverty level and improved the insurance rate among people suffering from SUDs. Since
low-income people are at high risk of having substance use problems (Olfson et al., 2018), evi-
dence has found that Medicaid expansion contributes to an increase in insurance coverage among
SUD patients and allows them to receive more treatment (Olfson et al., 2018; Saloner & Maclean,
2020).
Medicaid Institutions for Mental Diseases (IMDs) Waivers under the Section 1115 demon-
stration program are another policy reform aiming to expand access through payment coverage for
SUD patients. These waivers allow states to obtain federal Medicaid funds for services received
by nonelderly adults in an IMD, which is defined as a “hospital, nursing facility, or other insti-
tution of more than 16 beds, that is primarily engaged in providing diagnosis, treatment, or care
of persons with mental diseases, including medical attention, nursing care, and related services
(Musumeci et al., 2019)”. Under the IMD exclusion rule created under the 1965 Medicaid law,
the IMD exclusion prohibits Medicaid benefits from paying any nonelderly adults for services in
IMDs.
The public sector, including Medicare and Medicaid, has been the primary payer for the
healthcare expenses of substance use disorders(SUD) services (Mark et al., 2011; Mark et al.,
47
2016), since commercial insurance has been limited in paying for substance abuse treatment ser-
vices due to the severe moral hazard and adverse selection problems of behavioral health treatment
in the private insurance market (Frank & McGuire, 2000). In 2014, only 18% of SUD expenditures
were paid by private health insurance (Mark et al., 2016). However, requirements for coverage
of SUD treatment services in commercial insurance have also grown, especially after the imple-
mentation of MHPAEA and ACA. So understanding the evolution of access through commercial
insurance is a crucial aspect of improving access to treatment for SUD patients in general.
Expanded SUD coverage and benefits have grown among those with Medicaid, Medicare, and
among private commercial insurance, while the availability of treatment has remained somewhat
limited. Even for people who can receive treatment, wait time before treatment is also a barrier
preventing patients from engaging or continuing their treatments. According to the Treatment
Episode Data Set (TEDS) in 2020, 21% of all admissions have to wait up to a week (between 1-7
days) to enter treatment, and in the case of opioid addiction, it is 26% that have to wait up to a
week (SAMHSA, 2022c). 10% must wait eight or more days (i.e., at least a week and as long as a
month) to enter treatment (SAMHSA, 2022c). SUD usually requires a significant amount of time
involved in treatment. The median length of stay (LOS) among discharges that completed treat-
ment is 91 days for intensive outpatient treatment
3
, 78 days for long-term residential treatment, 59
days for outpatient treatment, 26 days for short-term residential treatment, and 14 days for hospital
residential treatment (SAMHSA, 2022c). The waiting time and typical LOS usually required for
treatment raise questions about crowd-out and spillover effects of these policies targeting the pub-
licly insured. Exploring the interactive effects of public policy changes, public insurance programs,
and commercial insurance can offer more insights into future policymaking.
Medicaid policy improving access for those with Medicaid insurance may generate unin-
tended spillover effects on the commercially insured in light of supply shortages of some SUD
3
According to American Addiction Centers, “intensive outpatient programs (IOPs) are treatment programs used to
address addictions, depression, eating disorders, or other dependencies that do not require detoxification or round-the-
clock supervision. They enable patients to continue their normal, day-to-day lives in a way that residential treatment
programs do not.” (https://americanaddictioncenters.org/intensive-outpatient-programs)
48
treatment providers. For example, Knudsen et al. (2022) finds Medicaid expansion is associ-
ated with a significant increase in buprenorphine utilization paid by Medicaid and a decrease in
buprenorphine utilization paid by commercial insurance. Another study by Meinhofer and Wit-
man (2018), however, finds Medicaid expansion contributes to an 18% increase in aggregate opi-
oid admissions, which is driven by admissions from Medicaid beneficiaries, without crowding out
admissions from individuals enrolled in other health insurance.
This paper examines the spillover effect of Medicaid IMD exclusion waivers on those com-
mercially insured SUD patients’ access to different treatment settings. I also repeat the analysis
for SUD patients with mental health comorbidities and compare the differences between the two
groups of patients.
3.2 Background
3.2.1 Substance Abuse Treatment and Access to It
Three primary SUD treatment forms are detoxification, cognitive and behavioral therapies,
and medication-assisted therapies (ClevelandClinic, 2022). In general, the rate of receiving treat-
ment for substance abuse has remained low for patients who need treatment (SAMHSA, 2021).
Like many illnesses, different levels of care are required for patients suffering from SUD, de-
pending on the severity of the conditions. The American Society of Addiction Medicine (ASAM)
has established five main levels of care based on the severity of illness, which they describe as a
continuum of care for substance abuse treatment. The five levels of treatment include early inter-
vention services, outpatient services, intensive outpatient or partial hospitalization services, resi-
dential or inpatient services, and medically managed intensive inpatient services (CenterforSub-
stanceAbuseTreatment, 2006). Medication-Assisted-Treatment, or “MAT”, has been integrated
into all the levels of services except early intervention services for those patients who can benefit
49
from it, as ASAM recognizes it as an important part of evidence-based treatment services for SUD.
As each of these types of treatment can occur in a wide range of settings (e.g. community settings
4
, doctor’s offices, health centers, hospitals, and specialty facilities), and as no data source provides
a comprehensive look across all of them, it can be difficult to identify how many people receive
care at each level. When most researchers look at treatment, instead of examining the different lev-
els of care, they look at location or type in terms of residential, outpatient, and medication-assisted
treatment. The National Survey of Substance Abuse Treatment Services (N-SSATS) and the Na-
tional Survey on Drug Use and Health (NSDUH) are the most commonly used data sources. They
do not provide information about the level of care received in a given location, only the location of
service and use of MAT.
For example, Saloner et al. (2022) explored the trend in self-reported receipt of substance use
treatment with data from the NSDUH. The NSDUH is a nationally representative survey of the US
household population that includes questions about substance use and past year receipt of treat-
ment. The Saloner et al. study, which examined trends from these data, covered the use patterns
by setting types, including outpatient programs, inpatient programs, hospitals or emergency de-
partments, mental health centers, doctor’s offices, self-help groups, and jails or prisons. They also
covered the change of use by payers such as Medicaid, Medicare, and private insurance. Though
the survey data capture those different types of payers and settings, the self-reported measures are
subject to recall and social desirability bias (Saloner et al., 2022). Also, the measures do not pro-
vide information relevant to understanding the quality or level of care received in each setting and
by whom.
TEDS is another potential data source providing both type and level of treatment services.
TEDS compiles client-level data for substance abuse treatment admissions from state agency data
systems (SAMHSA, 2020). It comprises admissions and discharge data from individuals 12 years
or older. State data systems collect publicly-funded admissions data or privately funded admissions
4
treatment programs funded by the government (e.g. through block grants) can happen in churches, schools, and
community health centers.
50
from facilities that receive public funding (SAMHSA, 2020). The records come specifically for
admission or discharge with demographic information and substance abuse characteristics, but no
person-level ID for tracking over time. Thus, even though TEDS data can capture multiple episodes
of treatment from the same patient with the level of care information, they do not have a unique
patient identifier across different treatment facilities that can allow a researcher to identify the same
patient passing through various facilities as part of one episode of treatment (e.g., stepping up their
treatment or stepping it down).
Therefore, claims data are the only source of information where the setting of service can be
coupled with measures of the intensity of service, the types of providers engaged in treatment at a
given location, as well as whether transitions across settings are part of the same treatment episode
for a given individual or separate episodes of treatment. With claims data, I can tell if a small
number of high-intensity users are repeatedly gaining access to treatment or if an actual expansion
in the number of people receiving appropriate evidenced-based therapy. In addition, I can address
the inconsistency and inaccuracy of self-reporting data with commercial insurance claims data,
identifying levels of treatment using standard clinical procedure codes. The analysis can provide
a more accurate and comprehensive understanding of the use of substance abuse treatment among
patients with private insurance payers.
3.2.2 Medicaid IMD Waivers for SUD and their Impact on the Commercially Insured
In 2015, the Centers for Medicare and Medicaid (CMS) issued the first formal guidance for
states to apply for IMD exclusion waivers under the Section 1115 demonstration projects. Later in
2017, the CMS issued another guidance with a more streamlined application procedure for states to
apply for such waivers. The primary goal of CMS encouraging states to apply for IMD exclusion
waivers is to improve access to critical levels of care for SUD patients across a “continuum of
care” (CMS, 2017), where the continuum of care is defined as a treatment system in which clients
51
receive treatment services at an appropriate level based on their needs and then switch to a higher
or lower level of care according to their conditions (CenterforSubstanceAbuseTreatment, 2006).
CMS issued the new guidance for the Section 1115 demonstration waiver in SUD in 2015.
California was the first state to obtain a waiver in 2015, but by the end of 2020, there were 28 states
with these SUD IMD Exclusion waivers (KFF, 2021).
The IMD SUD exclusion waivers allow more Medicaid enrollees to receive SUD services
in residential treatment facilities. Those services might be delivered by the residential treatment
facility itself or by an outside agency (a health care agency or outpatient treatment provider) with
which the residential facility coordinates care. Thus the IMD SUD exclusion waiver could impact
access to and receipt of SUD care for patients in both residential and non-residential treatment
facilities. A recent study showed that the acceptance of Medicaid increased at residential and
intensive outpatient facilities, and the delivery of medications increased with the adoption of the
IMD exclusion waivers (Maclean et al., 2021). It is also found that hospitals in states that have
adopted Medicaid 1115 waivers for SUDs are more likely to provide opioid use disorder(OUD)
treatment programs (Chang et al., 2023).
Thus far, no study has carefully considered the potential spillover effects of these policies on
the commercially insured. If the supply of treatment facilities or outpatient treatment slots does
not change much in the short term or the increase does not reach the rise of overall patients in need,
then expanded use of SUD treatment services by Medicaid patients in residential facilities enabled
by these waivers potentially crowds out SUD patients with other types of insurance. But if these
policies and others being adopted with federal funding expand the total number of facilities or
total treatment slots available (as shown in the first chapter) or the general availability to patients
in need, then it is possible these policies do not crowd out the privately insured. Furthermore,
efficiency gains that might occur through care coordination and integration offered by residential
facilities now able to treat both mental health and SUD conditions simultaneously could generate
better outcomes from treatment, thereby reducing the need for subsequent treatment by patients
52
who might otherwise have to cycle through the system multiple times as care for only mental
health or SUD is addressed. Those impacts might vary by service type since the waivers’ primary
focus is on residential care.
In this chapter, I will specifically analyze the impact of IMD waivers on each level of care,
including residential, outpatient, intensive outpatient, and MAT, among commercially insured in-
dividuals so that I can understand if these waivers lead to a crowd-out of particular types of care
(shifting the commercially insured out of residential and into outpatient and/or MAT services only,
for example) or if the market can expand to meet the greater need without any deleterious impacts
on the commercially insured.
Expanding these Medicaid IMD SUD exclusion waivers did not happen in a vacuum. Many
other state and federal policy changes have occurred during this period, including the Comprehen-
sive Addiction and Recovery Act (CARA) of 2016, the 21st Century Cures Act of 2016, and the
SUPPORT Act of 2018. Similar to the considerations in the previous chapter, my analysis does
not consider their impact because most of them happen at the national level. Thus all states are
subject to their effects. The SUPPORT Act provides grant funds for states to tackle the opioid
crisis. Though states may receive different levels of grant funds, the starting time of those funds
is after 2018, which limits the impact because of the short time in my observation period. The
levels also depend on state-level factors such as the severity of the substance use problems, states’
demand for treatment, other available resources, socioeconomic conditions, etc., which are already
accounted for in my model. More importantly, the sample for analysis in this chapter focuses only
on one provider in the commercial insurance market, and the indirect effect of the grant fund is
much smaller.
53
3.3 Data and Methods
3.3.1 Data
The primary data source for the outcomes in this paper comes from the Clinformatics® Data
Mart, a database covering administrative health claims for members of a national managed care
company affiliated with Optum. Those administrative claims are submitted by providers and phar-
macies for payment and comprise patients with commercial health plans and Medicare Advantage.
The database includes about 15-20 million annual covered lives, and I use the nine years from 2012
to 2020. I exclude Medicare Advantage claims in my sample and focus only on commercial health
plan claims of individuals between the ages of 21 and 64, as this is the population targeted by the
IMD exclusion rule (nonelderly adults).
Data on state IMD SUD exclusion waivers comes from RAND Opioid Policy Tools and In-
formation Center (OPTIC) policy database. Legal information on states’ waiver status is provided
by the Prescription Drug Abuse Policy System (PDAPS) and is verified against policy documents
published on Medicaid’s website (CMS, 2023b), as well as National Health Law Program’s report
(Cohen et al., 2021). As CMS can approve requests for these waivers throughout the calendar year,
and the immediate effectiveness of a waiver does not mean changes take place immediately, in the
analysis, I assume a waiver is fully effective for the following year. This results in 27 states that
adopted waivers between 2015 to 2020, as shown in Table 1. Since medication coverage is not
included in Massachusetts’ waiver, I exclude Massachusetts when examining the IMD waiver’s
effect on medication access.
I use the RAND Opioid Policy Tools and Information Center (OPTIC) policy database to con-
struct my other state policy variables. OPTIC identifies existing data from sources on state opioid
policies, including (but not limited to) the National Alliance for Model State Drug Laws, the Na-
tional Conference of State Legislatures, the Prescription Drug Abuse Policy Surveillance System
54
(PDAPs) and the PDMP Training and Technical Assistance Center. It reviews inconsistencies in
definitions and legal dates across these databases and assesses whether they are true inconsisten-
cies or a function of differences in policy definitions. OPTIC then extracts a consistently defined
policy of relevance to researchers from these data and its original legal research, facilitating repli-
cation in analyses. The policy variables I include in my analyses here are the Affordable Care Act
(ACA) Medicaid expansion status, the presence of naloxone prescribing law, parity law in SUD,
and recreational marijuana law.
Other state-level demographic and substance use characteristics variables are obtained from
sources including Kaiser Family Foundation, University of Kentucky Center for Poverty Research,
American Community Survey (ACS), Area Health Resource File, Center for Disease Control and
Prevention (CDC), and National Survey of Drug Use and Health (NSDUH). A detailed table of
data sources is provided in Table 3.2.
I select patients from 2012 to 2020, suffering from SUDs in general and SUD patients with
mental health comorbidities, using ICD-10 codes for 2015-2020 and ICD-9 codes before 2015.
The codes used for selection are listed in Appendix Table A.2-A.3. I use CPT/HCPCS procedure
codes to tell the treatment a SUD patient receives at a specific care level, and I group them into
outpatient services, intensive outpatient or partial hospitalization services, inpatient services, and
medications for treatment. The codes for identifying the critical levels of care are provided in
Appendix Table A.4-A.7. Specialty facilities for substance use treatment usually include hospitals,
drug or alcohol rehabilitation facilities, and mental health centers (SAMHSA, 2021). Though I
cannot make clear distinctions on each type of facility based on my data, I exclude some places of
service based on place of service (POS) codes, including assisted living facilities, skilled nursing
facilities, nursing facilities, custodial care facilities, hospices, ambulances, mass immunization
centers, end-stage renal disease treatment facilities, and independent laboratories.
Three sub-tables in Optum’s database are used for analysis: the medical claims table with
procedure codes, the medical diagnosis table, and the enrollment table. Patients diagnosed with
55
substance use and mental health problems are identified with diagnosis codes from the medical
diagnosis table. The medical claims table with procedure codes contains information on the proce-
dure codes indicating what procedures a patient receives and the place of services indicating what
types of facilities the services are received. The patient enrollment table contains the state, birth
year, gender, and race to identify patients’ demographic background information. The structure of
the data files with available variables is listed in Appendix Table A.1.
To combine the diagnosis file with SUD patients and the medical record file with procedure
codes, I match the two parts of information by the patient ID number and the year receiving the
relevant procedure and diagnosis of the relevant problem. One record is kept at the patient-year
level to indicate that the patient diagnosed with SUD gets treatment at a critical level of care of
my interest. The demographic information of those patients is matched from the enrollment file.
Since some patients may seek treatment from a specialty facility out of state, I use the first service
date of their first SUD procedure to determine the state they are in at the time of the procedure.
Specifically, the matched year is the most recent year before their first service date.
The SUD patient sample size for estimation for 2012-2020 is 4,321,292. And the size of SUD
patients with MH comorbidity sample for 2012-2020 is 1,127,475. A summary of the sample’s
characteristics is provided in Table 3.3. Overall, summary statistics show that outpatient care ac-
cess is greater than other types of SUD treatment for those with Optum commercial insurance.
Over the period 2012-2020, around 37.3% of patients diagnosed with a SUD problem get some
form of outpatient care each year, and around 40.0% of patients diagnosed with a SUD and MH
comorbidity receive outpatient care annually. The use of inpatient care is lower than outpatient
care. Specifically, 15.4% of diagnosed SUD patients receive inpatient care annually during the
study period, and 26.0% of diagnosed SUD patients with MH comorbidity receive inpatient care
annually. The lowest utilization is intensive outpatient and partial hospitalization care. On aver-
age, about 1.3% of SUD patients receive intensive outpatient or partial hospitalization treatment
annually, and about 3.6% of SUD patients with MH comorbidity use intensive outpatient or partial
56
hospitalization care each year. The use of medication for SUD patients is also low. On average,
about 0.5% of SUD patients use medications for treatment each year. Of those with MH comor-
bidity, 1.1% of them use medications for treating SUD.
Access to the inpatient level of care shows a decreasing trend over 2012-2020 for both SUD
patients in general and the group of SUD patients with MH comorbidity (Figure 3.1). A cutoff
point in 2016 for the inpatient trend could be caused by the ICD9 to ICD10 code change. The
outpatient level utilization rates for SUD care also show a decreasing trend over the study period,
of which the comorbid group has a similar trend and slope with the general SUD group (Figure
3.2). The decreasing use of inpatient and outpatient services also raises questions on whether IMD
waivers have any crowd-out effect contributed to it. The trend of receiving intensive outpatient
or partial hospitalization increases throughout the study period (Figure 3.3). MH comorbid group
has a higher increasing rate than the general sample and shows a slightly decreasing trend after
2016. According to Figure 3.4, access to MAT for SUD patients is very low but increasing over the
observation period. The general sample group and the MH comorbid group have similar trends. In
contrast, the general sample group has a higher increasing rate during the first half of the period,
and the MH comorbid group has a higher increasing rate during the second half (Figure 3.4).
3.3.2 Methods
The standard approach would be two-way fixed effects (a transformed difference-in-difference
method) for examining the effects of the IMD exclusion waivers. However, as shown in Table 1,
the states experience an incremental adoption process, with only a few states adopted in the initial
periods. I only have one effective state for 2016, three effective states for 2017, and 7 effective
states for 2018. Though the treated units are increasing, the size is still relatively small compared to
those untreated states that serve as the control group. Despite the high possibility that the parallel
trend assumption for the difference-in-difference method may fail due to the small treated group
size at each period, the synthetic control method is often considered a better way to use when the
57
treated group is very small. Therefore, I consider a synthetic difference-in-difference method here
to examine the effect of SUD waivers. The synthetic difference-in-difference method combines
features of the synthetic control and difference-in-difference methods. It re-weights and matches
pre-exposure trends, weakening the reliance on parallel trend assumption, at the same time being
invariant to additive unit-level shifts like the regular difference in difference method (Arkhangel-
sky et al., 2021). Since most states adopted the waivers after 2017, the post-treatment period for
observing the treatment effect is limited for most states. The method also allows staggered treat-
ment effect comparisons. The estimates give the impact of IMD waivers on states that adopt them
across the study period and aggregate the effect through different treated years. Specifically, the
empirical model is:
Y
it
=β
it
∗ W
it
+λ
it
∗ X
it
+α
i
+γ
t
+ε; (3.1)
where Y
it
is the outcomes of interest, which are the rates of having access to outpatient SUD
care, inpatient SUD care, and intensive outpatient or partial hospitalization care, W
it
is the waiver
treatment, X
it
is other time-varying co-variates, α
i
is the state fixed effect, and γ
t
is the year fixed
effect. The synthetic difference-in-difference method tries to estimate the coefficients by adding
both weights ˆ ω
sdid
to align pre-exposure trends in the outcome of unexposed units with those
for the exposed units and time weights
ˆ
δ
sdid
to balance pre-exposure periods with post-exposure
periods, shown in the equation below (Arkhangelsky et al., 2021):
(
ˆ
β
sdid
,
ˆ
λ, ˆ α, ˆ γ)= arg min
β,λ,α,γ
n N
∑
i=1
T
∑
t=1
(Y
it
− β
it
∗ W
it
− λ
it
∗ X
it
− α
i
− γ
t
)
2
ˆ ω
sdid
ˆ
δ
sdid
o
(3.2)
State-level co-variates include policy measures (ACA Medicaid expansion status, parity law in
SUD, naloxone prescribing law, recreational marijuana law), the prevalence of SUD and MH need
(opioid prescribing rate, the share of the population with any mental illnesses, with serious mental
illnesses, and using cocaine), treatment resources (number of hospitals, and number of community
mental health center) and socio-economic measures (poverty rate, unemployment rate, population
education level, the percentage of Medicaid enrollees as a share of the population, the population
58
share of people in the armed force, and the percentage without health insurance). Descriptions and
summary statistics of all the co-variates are provided in Table 3.4.
3.4 Results
The results of examining the spillover effect of IMD exclusion waivers on access to treatment
among those commercially insured SUD patients are summarized in Table 3.5. No effect is found
on the access rate to inpatient care, outpatient care, and medication-assisted treatment for both
SUD patients in general and patients with MH comorbidities with the adoption of IMD waivers.
However, I find the adoption of an IMD exclusion waiver leads to an average treatment effect
of a 0.23 percentage point increase in the use of an intensive outpatient or partial hospitalization
care for patients with an SUD diagnosis and a 0.4 percentage increase in the use of intensive
outpatient or partial hospitalizations for patients diagnosed with SUD and MH comorbidities.
Given that I find no effect on inpatient treatment services by either group, these policies do not
crowd out the inpatient care use among commercially insured patients. Also, no evidence supports
any impact of IMD exclusion waivers on outpatient care and medication use among commercially
insured SUD patients in general and those with MH comorbidities. This finding also rules out the
concerns that expanded Medicaid benefits may crowd out the use among other insurance enrollees
on different levels of care. On the contrary, the IMD waivers are shown to increase access to
intensive outpatient or partial hospitalization for those with commercial insurance for both groups
of patients.
3.5 Sensitivity Analysis
Since the synthetic difference-in-difference method does not allow fractional numbers as a
policy intervention indicator, I conduct a sensitivity analysis using the traditional two-way fixed
59
effect model to compare the results. Though the two-way fixed effect method is not ideal in the
situation here, as explained in the method section, it has an advantage in allowing fractional policy
indicators so that waivers that become effective at different times of the year can be differentiated.
The results are shown in Table 3.6. Adoption of a Medicaid 1115 SUD waiver contributes to
a 0.18 percentage point increase in the use of an intensive outpatient or partial hospitalization at a
10% significance level, which is slightly lower than but close to the estimation from the synthetic
difference in difference. The result also suggests that adopting a waiver contributes to an increase
in access rate to inpatient care at a 10% significance level, while the synthetic control method does
not indicate that. The access rates to outpatient and MAT are not affected by the adoption of the
waiver, as indicated by the results.
3.6 Conclusion and Limitation
Commercial insurance plays an important role in paying for SUD treatment, among which
outpatient care takes the largest share, followed by inpatient care and intensive outpatient care or
partial hospitalization. The findings of my study demonstrate the changing role of commercial
insurance. The study results show commercial insurance use in paying for inpatient and outpatient
SUD care has decreased over the past decade, while the trends in paying for intensive outpatient
or partial hospitalization and medicated-assisted treatments for SUD patients have generally in-
creased. SUD patients suffering from MH comorbidities have higher rates of receiving care at
each critical level compared to SUD patients in general. This result may be because patients with
mental health comorbidities generally have more treatment needs. The trends in using these lev-
els of care for SUD patients with MH comorbidities are similar to SUD patients in general. The
very limited sample size of patients that receive MAT could be due to the lack of coverage in
commercial insurance on medications such as buprenorphine.
60
Another potential mechanism affecting the trend of outpatient and inpatient care may be a
selection effect of patients with substance use and mental health problems. Specifically, those
patients are generally less likely to be employed and have more barriers to getting insured in
commercial insurance over time. The selection issue may be worse for those with SUD and MH
comorbidity. As a result, more those patients with SUD and MH problems may end up switching
from commercial insurance to Medicaid. The rate of getting access to treatment through commer-
cial plans may decrease over time. This provides a possible explanation for the decreasing trend in
the outpatient and inpatient access rate reflected in my data.
Medicaid IMD exclusion waivers aim to expand access to quality treatment for SUD patients.
Medicaid enrollees can directly benefit from expanded coverage in IMDs, which raises concerns
about crowding out access to care among other types of insurance enrollees because of the ex-
pansion in access to treatment among Medicaid enrollees. The results of my study negate those
concerns, from which I do not find any effect of adopting IMD waivers on access to outpatient,
inpatient, and medication-assisted treatments. More interestingly, the results suggest that adopting
IMD exclusion waivers positively affects the use of intensive outpatient or partial hospitalization
treatment for SUD patients in general and SUD patients with MH comorbidities. The reason be-
hind my finding may be that Medicaid input in expanding access to treatment for SUD patients
improves care coordination and stimulates the development of some intermediate care, such as
intensive outpatient or partial hospitalization, which has been underused before. The lower cost of
commercial insurance in paying for inpatient care freed by the public funds enables insurers to pay
more for this intermediate level of care.
My study has several limitations. First, due to the need for more data on some critical co-
variate measures, my observation period after the waiver intervention can only extend to 2020,
which is still short. More effects may be observed over a more extended period. Second, patients
may be enrolled in different types of insurance (i.e., commercial, Medicare, Medicaid) simultane-
ously. I cannot tell whether a patient is also enrolled in Medicaid or regular Medicare outside my
61
data sample. A patient who is also enrolled in Medicaid can affect his utilization of commercial
insurance. Patients with only commercial insurance may behave differently from those who are
also enrolled in other types of insurance. Third, since there is no standard code procedure code
guide on differentiating those codes by levels of care, I have to make some inferences based on
code descriptions and places of service. As a result, the rate of using different levels of care may
be inaccurate, affecting the estimation and comparison. Moreover, some outpatient and inpatient
procedural codes do not identify whether the main reason for treatment is SUD or not. I only keep
observations that contain information such as specific SUD-related codes or receiving treatments
at a substance use treatment facility. The rate of access to care may be underestimated. And as the
SUD treatments get more integrated into primary care, this underestimation may become larger
over time.
62
Figure 3.1: Trends of SUD Patients Receiving Inpatient Care
Notes: SUD is Substance Use Disorders. MH is Mental Health. The blue line sample excludes patients that
do not have MH comorbidity from the orange line - SUD patients sample. 95% confidence intervals are noted in the
figure. Data is from the Clinformatics® Data Mart, a database covering administrative health claims for members of a
national managed care company affiliated with Optum.
63
Figure 3.2: Trends of SUD Patients Receiving Outpatient Care
Notes: SUD is Substance Use Disorders. MH is Mental Health. The blue line sample excludes patients that
do not have MH comorbidity from the orange line - SUD patients sample. 95% confidence intervals are noted in the
figure. Data is from the Clinformatics® Data Mart, a database covering administrative health claims for members of a
national managed care company affiliated with Optum.
64
Figure 3.3: Trends of SUD Patients Receiving Intensive Outpatient Care
Notes: SUD is Substance Use Disorders. MH is Mental Health. The blue line sample excludes patients that
do not have MH comorbidity from the orange line - SUD patients sample. 95% confidence intervals are noted in
the figure. Intensive outpatient care refers to treatment programs that are often used to address addictions or other
behavioral health problems that enable patients to continue with their normal lives in a way that residential treatment
programs do not, but involve more intensive treatment commitment than typical outpatient services do. Data is from
the Clinformatics® Data Mart, a database covering administrative health claims for members of a national managed
care company affiliated with Optum.
65
Figure 3.4: Trends of SUD Patients Receiving MAT Care
Notes: SUD is Substance Use Disorders. MAT is medication-assisted treatment. MH is Mental Health. The
blue line sample excludes patients that do not have MH comorbidity from the orange line - SUD patients sample.
95% confidence intervals are noted in the figure. Data is from the Clinformatics® Data Mart, a database covering
administrative health claims for members of a national managed care company affiliated with Optum.
66
Table 3.1: 2015-2020 IMD Waiver in SUD (effective in a full year)
Effective Year (as a full year) N States
2016 1 CA
2017 2 MD, V A
2018 4 MA, NJ, UT, WV
2019 13 AK, IL, IN, KY , LA, NH, NM,
OR, PA, RI, VT, WA, WI
2020 7 DC, DE, MI, MN,
NE, NC, OH
Total 27
Notes: SUD is Substance Use Disorder. I assume the effective year is the following year if a waiver is
adopted in the middle of a year. Data is from RAND Opioid Policy Tools and Information Center (OPTIC)
policy database.
67
Table 3.2: Data Sources of Key Variables
Variables Data sources
ACA Medicaid expansion RAND Opioid Policy Tools and Information Center (OPTIC)
Parity law in SUD RAND Opioid Policy Tools and Information Center (OPTIC)
Recreational marijuana law RAND Opioid Policy Tools and Information Center (OPTIC)
Naloxone prescribing law RAND Opioid Policy Tools and Information Center (OPTIC)
Opioid prescribing rate Center for Disease Control and Prevention (CDC)
Serious mental illness rate National Survey of Drug Use and Health (NSDUH)
Using cocaine rate National Survey of Drug Use and Health (NSDUH)
Number of community mental health centers Area Health Resource File
Number of hospitals Area Health Resource File
Availability of emergency department Area Health Resource File
Availability of psychiatrists Area Health Resource File
Poverty rate University of Kentucky Center for Poverty Research
Unemployment rate University of Kentucky Center for Poverty Research
Higher education rate American Community Survey (ACS)
Share of Medicaid enrollees Kaiser Family Foundation (KFF)
Armed force rate American Community Survey (ACS)
Age 18 to 64 without health insurance rate American Community Survey (ACS)
Notes: ACA is Affordable Care Act. SUD is Substance Use Disorder.
68
Table 3.3: Sample Characteristics for Evaluating IMD Exclusion Waiver
Variable Definition Mean Std. Dev.
SUD Sample: N = 4,321,292
Inpatient access rate Percentage of patients using inpatient care 0.154 0.361
Outpatient access rate Percentage of patients using outpatient care 0.373 0.484
Intensive outpatient access rate Percentage of patients using intensive outpatient care 0.013 0.1160
MAT rate Percentage of using MAT 0.006 0.076
Age Average age 43.8 12.2
Female Percentage of female 0.446 0.497
White Percentage of white 0.764 0.424
SUD with MH comorbidities Sample: N = 1,127,475
Inpatient access rate Percentage of patients using inpatient care 0.260 0.439
Outpatient access rate Percentage of patients using outpatient care 0.380 0.485
Intensive outpatient access rate Percentage of patients using intensive outpatient care 0.036 0.186
MAT rate Percentage of using MAT 0.011 0.106
Age Average age 42.6 12.2
Female Percentage of female 0.537 0.499
White Percentage of white 0.813 0.390
Notes: SUD is Substance Use Disorder. MH is Mental Health. MAT is Medication-Assisted Treatment.
69
Table 3.4: Summary Statistics for State-level Co-variates
count mean sd min max
ACA Medicaid expansion status 459 0.481 0.498 0 1
Parity law in SUD 459 0.424 0.489 0 1
Naloxone prescribing law 459 0.694 0.437 0 1
Recreational marijuana law 459 0.117 0.318 0 1
Opioid prescribing rate 459 0.677 0.236 0.250 1.438
Serious mental illness 459 0.007 0.002 0.004 0.014
Using cocaine 459 0.015 0.005 0.007 0.041
Poverty rate 459 0.125 0.035 0.037 0.231
Unemployment rate 459 0.054 0.019 0.024 0.128
Higher education rate 459 0.609 0.055 0.439 0.778
Armed force rate 459 0.007 0.010 0.000 0.063
Share of Medicaid enrollees 459 0.209 0.065 0.090 0.429
Share of age 18 to 64 without health insurance 459 0.146 0.065 0.038 0.343
Number of community mental health center 459 4.9 10.4 0 123
Number of hospital 459 120.9 102 12 606
Number of total SUD diagnoses 459 9,414.6 10,746.6 63 70,599
Notes: ACA is Affordable Care Act. SUD is Substance Use Disorder. All the numbers are the original value
according to descriptions. The opioid prescribing rate can exceed 1.
70
Table 3.5: Synthetic Difference-in-Difference Estimation Results
Panel A. SUD Patients in general
Rate of using Rate of using Rate of using Rate of using
inpatient care outpatient care intensive outpatient care medications
Treated -0.0008 0.0022 0.0023** -0.0028
( 0.0033) (0.0028 ) (0.0011) (0.0022)
N 459 459 459 459
Panel B. SUD Patients with MH comorbidities
Rate of using Rate of using Rate of using Rate of using
inpatient care outpatient care intensive outpatient care medications
Treated 0.0018 -0.0001 0.0041* -0.0032
(0.0067) ( 0.002) (0.0024) (0.0043)
N 459 459 459 459
Notes: SUD is Substance Use Disorder. MH is Mental Health. Standard errors are included in parenthe-
ses. Statistical significance is indicated as follows: * indicates significance at the 10% level, ** indicates
significance at the 5% level, and *** indicates significance at the 1% level.
71
Table 3.6: Examining the Waiver Effect on the Access Rate to Different Levels of Treatment, with
a Two-way FE Linear Probability Model
(1) (2) 3) (4)
V ARIABLES Outpatient Inpatient Intensive Outpatient MAT
Section 1115 SUD waivers -0.0042 0.0044* 0.0017* 0.0010
(0.0059) (0.0026) (0.0010 ) (0.0007)
ACA Medicaid expansion status 0.0040 -0.0037 -0.0001 0.0014*
(0.0067) (0.0029) (0.0011) (0.0008)
Parity law in SUD 0.0147* 0.0095*** 0.0047*** -0.0041***
(0.0081) (0.0035) (0.0014) (0.0009)
Naloxone prescribing law 0.0157*** .0015 .0011 -0.0013**
(0.0060) (0.0026) (0.0010) (0.0007)
Recreational marijuana law 0.0007 0.0027 0.0029** 0.0030***
(0.0073) (0.0032) (0.0012) (0.0008)
Opioid prescribing rate -0.0062 0.0277*** 0.0061 0.0009
(0.0251) (0.0109) (0.0043) (0.0028)
Serious mental illness 2.5957 2.8843*** 0.2519 -0.2436
(2.3692) (1.0320) (0.4031) (0.2662)
Using cocaine 0.9137 -0.0452 0.1168 -0.1990***
(0.6590) (0.2871) (0.1121) (0.0740)
Poverty rate 0.2407** 0.0617 0.0041 0.0216*
(0.1095) (0.0477) (0.0186) (0.0123)
Unemployment rate 0.5313*** -0.1922** -0.0411 -0.0074
(0.1925) (0.0838) (0.0327) (0.0216)
Higher education rate -0.0843 0.0819 -0.0634* 0.0185
(0.2158) (0.0940) (0.0327) (0.0242)
Armed force rate -0.1372 0.2667 0.1255* 0.0466
(0.4113) (0.1792) (0.0670) (0.0462)
Share of Medicaid enrollees -0.0936 0.0277 -0.0097 0.0236**
(0.0902) (0.0393) ( 0.0153) (0.0101)
Age 18 to 64 without health insurance -0.2889*** 0.0539 -0.0100 0.0504***
(0.1099) (0.0749) (0.0187) (0.0123)
Number of community mental health center 0.00001 0.0001 0.0000 4.74e-06
(0.0002) (0.0001) (0.0000) (0.00002)
Number of hospital -0.0006 0.0002 0.0002* -9.02e-06
(0.0005) (0.0002) (0.0000) (0.0001)
Number of total SUD diagnoses -2.68e-06*** -3.04e-07 2.58e-07** -3.31e-09
(6.88e-07) (3.00e-07) (1.17e-07) (7.73e-08)
State FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Notes: ACA is Affordable Care Act. SUD is Substance Use Disorder. Standard errors are included in
parentheses. Statistical significance is indicated as follows: * indicates significance at the 10% level, **
indicates significance at the 5% level, and *** indicates significance at the 1% level.
72
Chapter 4
Examining the Spillover Effect of Institutions for Mental Diseases Exclusion Waiver and
Identifying Predictors on the Quality of Substance Abuse Treatment Services Received in
Commercial Insurance
4.1 Introduction
Substance abuse is one of the most severe public health challenges in the United States. The U.S. government
has been dedicating significant attention and investing a tremendous amount of money to improve access to substance
use treatment, especially evidence-based medication-assisted therapy (MAT), in the past twenty years, under laws such
as the Mental Health Parity and Addiction Equity Act of 2008, the Affordable Care Act of 2010, the Tribal Law and
Order Act of 2010, the Comprehensive Addiction and Recovery Act (CARA) of 2016, the 21st Century Cures Act
of 2016, the SUPPORT Act of 2018, COVID policies expanding access to treatment (SAMHSA, 2022a), as well as
policy changes made by the Centers for Medicare and Medicaid Services (CMS) (CMS, 2015, 2017). Evaluations
of these efforts, including my work in the previous chapter, show that access to treatment has expanded for those
with Medicaid and some with commercial insurance (Knudsen et al., 2022; Maclean et al., 2021). The uninsured rate
among substance use patients has also declined (Olfson et al., 2018; Saloner & Maclean, 2020). Evidence on access
expansion to quality treatment services, such as MAT, is promising (Knudsen et al., 2022; Maclean et al., 2021).
However, it is less clear if the quality of treatment services, beyond the general access to different levels of treatment,
has improved over the past two decades, especially under some recent policy changes.
Substance use disorder (SUD) treatments usually involve psychosocial treatments and medication treatments,
or both. Non-pharmacological specialty treatment services can be provided in inpatient hospitals, outpatient and
inpatient psychiatric hospitals, substance abuse treatment facilities, and community mental health centers (SAMHSA,
2021). Cost can be prohibitive for gaining access to high-quality addiction services in some of these settings for many
73
patients (Ali et al., 2017). Moral hazard and adverse selection problems in the insurance market (Frank et al., 1997)
have led many insurance companies to fail to provide enough coverage for SUD treatment services despite recent
mandates of some coverage required by the ACA and the MHPAEA (Hayes, 2023; Kennedy, 2019). But research has
identified other barriers in addition to cost and general access. For example, since SUDs co-occur with other mental
health conditions at a high rate (CMS, 2017), it is sometimes difficult for patients to make treatment decisions for
themselves and continue engaging in their treatment (Dixon et al., 2016). Moreover, the chronic, relapsing nature of
the disease (McLellan, 2002) means that multiple episodes of treatment and step-down care are necessary before the
patient experiences a proper recovery (Davidson & Schmutte, 2020), and few facilities or insurance carriers support
this for these patients.
Not unlike other areas of health care, patients often have difficulty identifying “high-quality” addiction treatment
services. Measurement of the quality of SUD treatment has developed gradually over the past two decades, even among
the health research community, starting with work conducted by the Washington Circle Group (WCG), which was ini-
tiated in March 1998 (McCorry et al., 2000). Developed from the classic framework by Donabedian (Donabedian,
1980), the SUD quality measures generally include five categories: structure, process, outcomes, access, and patient
experience (Garnick et al., 2006). The Center for Medicaid and Medicare Services (CMS) today focuses on three
quality metrics developed under contract with Mathematica Policy Research, Brandeis University, and the National
Committee for Quality Assurance (NCQA), which similarly focus on continuity of care after medically-managed with-
drawal from alcohol and drugs, use of pharmacotherapy for opioid use disorder, and continuity of care after inpatient
or residential treatment for SUD (CMS, 2023a). Because of the chronic, relapsing nature of the disease (McLellan,
2002), measurement of treatment success (and therefore the quality of treatment services) has moved farther away
from a focus on abstinence. Instead, it focuses more on process and access measures across treatment settings (e.g.
“warm handoffs,” which are captured through the continuity of care measures) rather than on days abstinent. As there
is no consensus on a universal quality measure for SUD treatment, increased identification, initiation, engagement in
treatment, and improved adherence to treatment are frequently used as indications of improved quality of SUD care
(Garnick et al., 2009; SAMHSA, 2021).
Public policies in recent years have focused on expanding access to care as well as improving the quality of
treatment. Medicaid plays an essential role in addressing the treatment needs of many patients with SUD as it is
one of the top payers of those services (Miles, 2019). States can implement experimental programs through Section
1115 demonstration projects, including financing services that Medicaid has not previously covered. The Medicaid
Institutions for Mental Diseases (“IMD”) waiver under the Section 1115 demonstration programs is one of the inno-
vative reforms, allowing states to obtain federal Medicaid funds for services received in IMDs by nonelderly adults
(Musumeci et al., 2019). Those waivers intend to expand the residential SUD treatment among Medicaid enrollees
74
and increase access to the full continuum of SUD care, which the American Society of Addiction Medicine (ASAM),
as well as other health service providers, believed would improve the treatment quality of SUD treatment.
In previous chapters of this dissertation, I sought to evaluate the impact of IMD waivers on the availability of
and access to treatment among the commercially insured. In this chapter, I assess whether IMD waivers are associated
with the quality of treatment received by commercially insured patients. Using nationwide medical claims data on
commercially insured patients receiving care in all 50 states over the period 2010-2021, I evaluate the spillover effect
of IMD waiver adoption on four measures of treatment quality that regulatory agencies, including NCQA and CMS,
have adopted. This set of commercial claims data also allows me to identify some critical predictors associated with
quality improvement in behavioral health treatment among commercially enrolled patients, which can offer some vital
insights to compare them with other public insurance enrollees.
4.2 Background
4.2.1 Quality Measures for SUD Treatment and the Change of Treatment Quality Overtime
Treatment quality measures are essential; with them, it is easier for policies and programs to identify and target
improvement areas or to compare individual providers’ performance against standards (Watkins et al., 2015). It is
believed that access to treatment for SUD should be accompanied by improved quality to ensure better treatment
outcomes (Watkins et al., 2015).
While there is no unique standard evaluation system for measuring the quality of substance use treatment, federal
agencies have developed multiple measures for tracking the quality of substance use disorder treatment, using provider
and patient outcomes. Provider-based measures have been introduced by the National Institute on Drug Abuse (NIDA),
National Institute on Alcohol Abuse and Alcoholism (NIAAA), and SAMHSA through research products conducted
over decades (Mark et al., 2020). These agencies describe some common signs of higher-quality programs, which in-
clude using evidence-based behavioral health therapies, offering medications for addiction, being accredited, attending
to mental health and physical health needs, offering recovery support services, being readily available when needed,
personalizing the treatment plan for each patient, and patients staying in treatment long enough and receiving contin-
uous monitoring with adjustments to the treatment plan as needed (Mark et al., 2020). Some of those measures are
captured by indicators of patients’ access and adherence to treatment, such as the continuation of receiving treatments
and length of treatment episodes.
75
On the patient outcome side, for example, the CMS has contracted with Mathematica Policy Research, Brandeis
University, and the National Committee for Quality Assurance and developed three measures, including continuity of
care after medically-managed withdrawal from alcohol and drugs, use of pharmacotherapy for opioid use disorder,
and continuity of care after inpatient or residential treatment for substance use disorder (CMS, 2023a). The WCG, a
multidisciplinary group convened by SAMHSA, consisting of providers, researchers, health plan representatives, and
public policymakers, aimed to develop and disseminate performance measures for SUD treatment services (Garnick
et al., 2009). The measures they developed and proposed to capture treatment quality include measures of initia-
tion and engagement in treatment, which have been endorsed by the National Quality Forum and adopted by NCQA
as Healthcare Effectiveness Data and Information Set (HEDIS) measures (Garnick et al., 2009; NationalCommit-
teeforQualityAssurance, 2023; NationalQualityForum(NQF), 2012), as well as other process measures indicating the
continuity of treatment. Since WCG gives unambiguous definitions of those measures, they are easy to construct using
my claims data. I choose to use those measures as my primary outcomes of interest and will discuss each of these
next.
WCG emphasized performance measures in four categories: identification, initiation, engagement, and continu-
ity. The first three measures were designed early in 2000 for private insurance (Garnick et al., 2009), and continuity
measures were developed nearly ten years later. Identification is defined as the percentage of adult enrollees with a
substance abuse claim, which contains a diagnosis of substance abuse or dependence or a specific substance abuse-
related service (Garnick et al., 2009). Initiation is defined as the percent of individuals who have an index outpatient
(OP) or intensive outpatient (IOP) service with no other substance abuse services in the previous 60 days and who
have received a second SUD service (other than detoxification or crisis care) within 14 days after the index service
(Garnick et al., 2009). Engagement measures the percentage of individuals who initiated OP or IOP substance abuse
treatment and received two additional services within 30 days of initiation (Garnick et al., 2009). In addition to these
basic measures, WCG identifies several continuity of care measures, with no service-free period required as with the
initiation and engagement measures. First, there is the continuity of care after assessment, which is a measure defined
as the percent of individuals who have a positive assessment for substance abuse and received another substance abuse-
related service (other than detoxification or crisis care) within 14 days (Garnick et al., 2009), and the assessment can
be in any form in all settings. They also construct a continuity of care after detoxification measure, which is defined
as the percent of individuals receiving a detoxification service who get another substance abuse-related service (other
than detoxification or crisis care) within 14 days of discharge from detoxification (Garnick et al., 2009). The continu-
ity of care after residential services measure is defined as the percent of individuals who have a stay that is followed
by another substance abuse-related service (other than detoxification or crisis care) within 14 days after discharge
(Garnick et al., 2009).
76
Researchers and regulatory agencies have used these WCG quality metrics to assess the quality of treatment
using data from different types of insurance plans over the past two decades since the measures were tested and widely
accepted. For example, NCQA reported the national estimates of the initiation and engagement rate of alcohol or
other drug (AOD) dependence treatment, varying by different types of insurance plans, including commercial HMO,
commercial PPO, Medicaid HMO, Medicare HMO, and Medicare PPO (NationalCommitteeforQualityAssurance,
2023). The average initiation rate of AOD dependence treatment for commercial HMO and PPO ranges from 34% to
47% during 2004-2021, with the highest rate in 2004-2006 and the lowest rate in 2015-2016. The average engagement
rate of AOD dependence treatment for commercial HMO and PPO ranges from 12% to 16%, with the highest rate
in 2008 and the lowest rate in 2016. The initiation rate is slightly higher in Medicaid for the whole period, and the
engagement rate is slightly higher in Medicaid after 2015. The numbers indicate that the initiation of AOD treatment
rate on average does not improve over the past two decades, even showing a decreasing trend for most types of
insurance covered, and that is the same for the engagement rate (NationalCommitteeforQualityAssurance, 2023). The
limited improvement in these quality metrics over time and notable declines in quality measures for the commercially
insured, while there is a modest improvement for those with Medicaid insurance, over 2011 - 2021 in particular, raises
questions as to whether policies requiring improved SUD treatment in Medicaid may have had a negative association
with the care received by the commercially insured. To understand if there are associations or potential spillover
effects of state Medicaid policies on the commercially insured, I will use individual-level commercial claims data
from insurers covering beneficiaries from all states nationwide.
Several studies have evaluated patient, health care system, and policy factors that are positively and negatively
associated with the quality of treatment received, using the measures discussed above (Chavez et al., 2022; Mark et al.,
2003; B. Stein et al., 2000; B. D. Stein et al., 2009). For example, using Medicaid data on adolescents, Chavez et al.
(2022) find 49.5% of patients in their sample initiated treatment and 48.5% engaged in treatment. They also find that
higher levels of clinical need, including mental health comorbidity and multiple SUD diagnoses, are associated with
higher odds of initiating treatment but lower odds of engaging in treatment. Other factors have also been found to be
positively associated with receiving follow-up care after detoxification or residential services, such as female gender,
mental health comorbidity, and outpatient copayments (Mark et al., 2003; B. Stein et al., 2000; B. D. Stein et al.,
2009).
State and federal policies may have also impacted access to higher quality treatment, and indeed numerous
studies have examined whether these policies have impacted initiation, engagement, and follow-up treatment after
detoxification or residential services, using data on the commercially insured. For example, Azzone et al. (2011) find
that the Federal Employees Health Benefits (FEHB) program, which offered comprehensive parity for both mental
77
disorders and substance use disorders, is associated with a higher probability of identification while having no signif-
icant association with initiation and engagement rates. Similarly, when examining the impact of the Paul Wellstone
and Pete Domenici Mental Health Parity and Addiction Equity Act (MHPAEA) on SUD treatment, Busch et al. (2014)
do not find a significant change in the identification, treatment initiation, and treatment engagement using Aetna Inc
health plans in 10 states.
Many studies that examine predicting factors of quality measures were very early studies using old datasets,
and few of them use a large nationwide database. On the policy side, since Medicaid IMD waivers happened fairly
recently, few studies have looked at the spillover effects of those waivers, and no studies have considered the impacts
on quality measures for those with commercial insurance. This research fills that gap by examining the effect of this
recent policy change while, at the same time, identifying significant correlates of quality treatment using a more recent
national commercial database.
4.2.2 Adopting Medicaid Payment Waivers on SUD
Medicaid has enacted several policies to improve the well-being of enrollees suffering from SUD by increasing
coverage and improving the quality of care. The Affordable Care Act (ACA) required states that expand Medicaid
eligibility to include behavioral health under “alternative benefit plans.” It expanded the reach of the Paul Wellstone,
and Pete Domenici Mental Health Parity and Addiction Equity Act of 2008 to alternative benefit plans and Medicaid
managed care plans (Andrews et al., 2018). Surveys found that the number of states providing residential addiction
treatment and medications for opioid use disorders under Medicaid increased substantially, comparing data from
2017 to 2014 (Andrews et al., 2018). Expanding Medicaid eligibility under Affordable Care Act increased insurance
coverage among SUD patients and gave them access to more treatment (Olfson et al., 2018; Saloner & Maclean, 2020).
Evidence also suggested that admissions to substance use treatment increased four years after Medicaid expansion,
and more people entered treatment in expansion states compared to non-expansion states (Saloner & Maclean, 2020).
Besides the benefit and access expansions on addiction treatment, quality improvement is also an important
goal of CMS to address SUD treatment problems. For example, they developed measures for the Medicaid Innova-
tion Accelerator Program’s (IAP) Reducing SUD area (CMS, 2023a), which are used to assess treatment program
performance and achieve better outcomes.
Medicaid also allows states to apply for payment waivers under Section 1115 programs to expand payment cov-
erage for SUD services. Many states use such waivers to cover IMD payments not allowed under Medicaid rules. CMS
78
issued the initial formal guidance for states to apply for Medicaid waivers addressing SUD payments in 2015. Later
in 2017, CMS issued another guidance with a more streamlined application process under the Trump administration.
California was the first state to obtain the waiver in 2015. By the end of 2021, 32 states had implemented a Medicaid
1115 waiver in SUD (KFF, 2021). Specifically, a list of the waiver implementations by year and state is provided in
Table 4.1.
IMD waivers aim to promote a full continuum of care for SUD patients at critical levels of care. They also
focus on improving the quality, accessibility, and outcomes of SUD treatments. Though the primary objective of
these waivers is to improve care for Medicaid beneficiaries, their effects can spill over to those with other forms of
insurance. Spillover effects on access to the SUD treatment market of many Medicaid policies have been demonstrated
in the literature (Knudsen et al., 2022; Meinhofer & Witman, 2018). In the case of these IMD waivers in particular,
a study by Maclean et al. (2021) shows that adopting IMD waivers is associated with expanded Medicaid payment
acceptance and provision of medications at multiple locations and Chang et al. (2023) shows an association between
the adoption of an IMD waiver and hospitals’ likelihood of providing opioid treatment programs.
Patients enrolled in other types of insurance, such as Medicare and commercial insurance, may also experience
a change in the quality of the SUD services they are receiving in light of policies adopted in Medicaid. In terms of
the quality of care received by the non-Medicaid insured patients, these spillovers may go one of two ways, however.
First, expanded access to higher-quality treatment created by Medicaid policies may crowd out access to those without
Medicaid insurance because of limited treatment capacity. In other words, Medicaid enrollees may get priority when
initiating, engaging, and continuing treatment services, given explicit funding for these services. However, it may
also be the case that the work facilities and health care systems undertake to improve access and coordination of care
across all levels of care may benefit any patient receiving services from those facilities, not just Medicaid enrollees.
If policies expand the number of facilities or available slots where higher quality care is occurring, then there may be
positive spillover effects on the commercially insured. Moreover, the availability of these services in the general area
may also improve the information provided to patients and their caregivers about the availability of the full continuum
of care in these areas, as was shown in the previous chapter. So, systematic improvements in initiation, engagement,
or continuation rates targeting one group of enrollees may have positive spillovers to those with other insurance.
79
4.3 Data and Methods
4.3.1 Data
As my primary data source, I use insurance claims level data from the Clinformatics® Data Mart, a database
covering administrative health claims for members of a national managed care company affiliated with Optum. Ad-
ministrative claims for patients with Optum health insurance are submitted by providers and pharmacies for payment.
The database includes about 15-20 million annual covered lives, including those on Medicare Advantage plans. I
exclude Medicare Advantage claims from the sample and focus only on commercial health plan claims of individuals
between the ages of 21 and 64, as this is the population targeted by the IMD exclusion rule (nonelderly adults).
Three sub-tables in Optum’s database are used to generate the analytic data file for this analysis: the medical
diagnosis table, the medical claims table, and the enrollment table. The medical diagnosis table contains all diagnosis
records for each patient, from which I identify patients with a substance use disorder and those with any mental
health comorbidity. The medical claims table provides the specific medical procedures a patient received and the
place of services, indicating what types of facilities the services are offered. The patient enrollment table contains
enrollment date, state, year of birth, gender, and race to identify patients’ demographic background information. I first
identify patients suffering from substance abuse and mental health disorders using ICD-9 and ICD-10 codes (due to
the transition in ICD versions in 2015) in the medical diagnosis table. Then I link these patients’ medical procedures
and enrollment data once they are identified using the unique patient ID number in each file. Patients are included in
my sample if they have a SUD diagnosis up to a year before receiving any SUD procedure. The specific codes used
for selecting patients suffering from SUD or SUD and MH conditions are listed in Appendix Tables A.2, and A.3.
HCPCS/CPT codes are used to identify the different procedures received by patients with a SUD diagnosis. I exclude
patients receiving services from places of service that cannot be considered treatment facilities for SUD clinically
based on place of service (POS) codes in the dataset, including assisted living facilities, skilled nursing facilities,
nursing facilities, custodial care facilities, hospices, ambulances, mass immunization centers, end-stage renal disease
treatment facilities, and independent laboratories.
After identifying patients and matching their patient ID numbers, I construct the analytic data file so that the
patient episode of treatment is the unit of observation. Prior treatment diagnoses and histories are constructed and
matched to each patient episode based on the date of that episode and past records up to it. The state of residency at
the time of enrollment in a given year is used to identify the geographic location of each treatment episode after that
enrollment date and before the next enrollment date.
80
4.3.1.1 Outcome measures
Following the Washington Circle group’s definitions, I construct the following outcome measures for the study:
SUD treatment initiation rate, SUD treatment engagement rate, and continuation rate after residential services for a
SUD. As is standard for the initiation and engagement measures, I consider patients who have an index outpatient
(OP) or intensive outpatient (IOP) service with a 60-day service-free period before the outpatient (OP) or intensive
outpatient (IOP) service (Garnick et al., 2009). No service-free period is required for the continuation measure after
residential services (Garnick et al., 2009). I also calculate the length of stay (LOS) per episode on average across all
treatment modalities as another outcome that may be more directly comparable to measures from other studies.
Since the claims data do not provide information on whether a patient’s assessment received for SUD is positive
or negative, I do not include the continuation measure after a positive assessment of a SUD as a measure. I also
exclude the continuity of care after detoxification measure, which is also a part of the WC group’s measures. I exclude
it because the total annual records of detoxification are under 100 before 2016, which is too low for comparison with
other measures in the study.
The SUD treatment initiation rate is defined as the percent of individuals who have an index outpatient (OP)
or intensive outpatient (IOP) service with no other substance abuse services in the previous 60 days and received a
second substance abuse service (other than detoxification or crisis care) within 14 days after the index service (Garnick
et al., 2009). Based on observations at the patient-service date level, I can identify 7,553,130 new episodes starting
with outpatient and intensive outpatient substance abuse treatment service 60 days before no other substance abuse
service was received across 6,341,971 patients receiving a SUD diagnosis. Twenty-three percent of these identified
new treatment episodes received follow-up services in the next 14 days, deemed “successful initiations.”
For counting engagement, I similarly identify the new episodes starting with OP or IOP substance abuse treat-
ment that involved two additional services within 30 days after initiation for those with a SUD diagnosis within one
year. Over my entire sample period, I find that 44% of all successfully initiated treatment episodes (but only 10% of all
identified treatment episodes) meet the criteria for a successful engagement. I construct the same measure of success-
ful engagement for SUD patients with mental health comorbidity. Among this group of episodes of co-morbid patients
(49% of all those new episodes), I find that 20% had successfully initiated their treatments, and 8% had successful
engagements.
Based on WC’s definition, the continuity of care for residential treatment is defined as the percentage of indi-
viduals who have a residential stay that is followed by another service (other than detoxification or crisis care) within
81
14 days after discharge, with no service-free period required before the residential stay (Garnick et al., 2009). To
construct the continuation measure after residential care, I can identify 5,336,367 episodes of residential treatment
among patients with a prior SUD diagnosis in one year for the entire study period. Among those episodes of residen-
tial treatment, 12% of them were followed by another service within 14 days after discharge. Those follow-ups are
counted as continuity of care after residential services. Among those episodes, 37% are with MH comorbidity, and
12% meet the definition of continuity of care after residential services.
In my study, the length of stay (LOS) refers to the average number of days of a single treatment episode across
locations of services for a patient. This measure differs from some other duration measures (e.g., the treatment duration
in B. D. Stein et al. (2009)) in SUD studies, which include the gap times between any two services, but this measure
more accurately describes the care that involves step-down or step-up care required by a patient and is a better reflection
of the actual time in treatment, regardless of the location of treatment. Since the WC group uses a 60-day service-free
period to determine the start of a new episode, I consider two treatments within 60 days of each other as belonging
to one treatment episode. Specifically, “one episode” here to estimate LOS refers to services of one patient that has
a 60-day service-free period before the first day of any service, whatever service they receive, and followed by any
services that are within 60 days since the last day of service of the previous one. I remove those repetitive days for
treatments that overlap each other. Patients may have some “stand-alone” treatments that occur only once with no
further treatment 60 days before or 60 days after. Those are also counted as distinct episodes, so the average length of
stay calculated in my study tends to be short, which is around 5.43 days.
4.3.1.2 Outcome trends during 2010-2021
Trends in each of these measures over time, adjusted for patient and state-level factors, are shown in Figures
4.1-4.4. According to the NCQA statistics, over the period 2010-2021, the average initiation rate of commercial HMO
and commercial PPO decreased, with a clear downward trend during the first half and a moderate upward trend during
the second half, while the Medicaid HMO increased overall, at an increasing rate during the second half of the period
(NationalCommitteeforQualityAssurance, 2023). The initiation rate in my data, seen in Figure 4.1, shows a relatively
stable trend during the 12 years 2010-2021, with some minor ups and downs. Patients with MH comorbidity tend
to have a lower likelihood of initiating treatment. The NCQA statistics show that the average engagement rate for
commercial HMO and commercial PPO decreased over the first half of the observation period and slightly increased
during the second half. That of the Medicaid HMO also decreased during the first half but increased more during the
second half. In my data, Figure 4.2 shows that from 2010-2021, the engagement rate in my commercial data had an
upward trend. The trendline can be divided into three periods, 2010-2013, 2013-2016, and after 2016, reflecting a
82
steady periodical rise, though some small decreases occurred during each period. Patients with MH comorbidity again
tend to have a lower likelihood of engaging in their treatment. Beyond those two measures also adopted by NCQA,
in Figure 4.3, the continuation rate after residential services shows a decreasing trend after 2014 for SUD patients in
general, and the rates do not show much change for patients with MH comorbidity. The continuation rate for patients
with MH comorbidity is close to that of patients in general. The length of stay among SUD patients decreases if
comparing 2021 to 2010, but overall trend lines do not show a clear changing pattern over 2010-2021. On average,
patients with MH comorbidity tend to have shorter lengths of stay in the treatment episodes (Figure 4.4).
4.3.1.3 Individual- and state-level covariates
The demographic information and prior diagnoses and services of patients receiving any treatment are shown in
Table 4.2. Males are more highly represented in the patient sample (55.3%) than females (44.7%), although the gap
decreases when examined based on the initiation of new treatment episodes. Whites take up the majority of the patient
sample (71.0%), followed by black (10.8%) and Hispanic patients (9.0%). Asians are the smallest minority group in
the sample (2.0%). Nearly half (49.0%) of the patients have ever received a mental health diagnosis. The number of
SUD diagnoses and past SUD services received are calculated as the average across the observation period for each
patient. Most patients have received 1-2 SUD diagnoses (74.2%) and 0-4 SUD services before getting a new treatment
service (65.5%). The number of months since the first enrollment in the plan is distributed relatively evenly, with a
larger share in the≥ 50 months group (28.6%), which is more than four years. The number of newly enrolled patients
(0-4 months) is less than other groups (9.1%) but is also a sizeable amount.
Table 4.3 - 4.4 give the same demographic information, prior diagnoses, and services statistics based on episodes.
Namely, the numbers in the tables are counting the episodes. Table 4.3 is for the episodes initiating treatment from
outpatient (OP) services, and Table 4.4 is for the episodes receiving residential services. Similarly, Males (52.9%
and 57.4%) are more than females (47.1% and 42.6%) in both samples. Whites take up the majority in both samples
(73.7% and 71.8%). Patients tend to be older in the residential sample (51.5% for age group 50-64) compared to the
starting new treatment episode sample (42.4% for age group 50-64). A significant part of those episodes are patients
with mental health co-morbidity in both samples, and the share in the initiating new episodes with the OP sample
(48.6%) is larger than that in the receiving residential services sample (36.0%).
As with the previous dissertation chapters, the data on state IMD exclusion SUD waivers comes from the RAND
Opioid Policy Tools and Information Center (OPTIC) policy database and is verified against policy documents pub-
lished on CMS’ website (CMS, 2023b) and the National Health Law Program’s report (Cohen et al., 2021). I compare
83
the effective date of those SUD waivers to the first treatment date of an episode of patients who reside in that state and
create a dummy variable to indicate the presence of an effective waiver.
I use the RAND OPTIC policy database to construct other state policy variables. OPTIC identifies existing data
from sources on state opioid policies, including (but not limited to) the National Alliance for Model State Drug Laws,
the National Conference of State Legislatures, the Prescription Drug Abuse Policy Surveillance System (PDAPs),
and the PDMP Training and Technical Assistance Center. It reviews inconsistencies in definitions and legal dates
across these databases and assesses whether they are actual inconsistencies or a function of differences in policy
definitions. OPTIC then extracts a consistently defined policy relevant to researchers from these data and its original
legal research, facilitating analysis replication. Specifically, I include in the analyses here the presence of the following
laws: Affordable Care Act (ACA), Medicaid expansion status, parity law in SUD, naloxone prescribing law, and
recreational marijuana law.
Some state-level variables are also controlled and compared. Those state-level demographic and substance use
characteristics variables are obtained from sources including Kaiser Family Foundation, the University of Kentucky
Center for Poverty Research, American Community Survey (ACS), Area Health Resource File, Center for Disease
Control and Prevention (CDC), and National Survey of Drug Use and Health (NSDUH). Table 4.5 provides a detailed
source of each variable.
Other policies that are adopted around the same period are taken into the model. ACA Medicaid expansion
status in each state affects the level of treatments the Medicaid population can get for SUD care. The mental health
parity law status may indicate how a state regulates its behavioral health treatment plans. The specific SUD parity law
directly benefits patients in treatment need. Naloxone prescribing law is an important indicator of how severe opioid
overdoses are in the state, as naloxone can reverse the effects of an opioid overdose and save lives. And recreational
marijuana law may serve as a proxy of how liberal a state’s drug market is being regulated.
Other state-level controlling variables include the opioid prescribing rate, the prevalence rate of having any
mental illnesses, the prevalence rate of using cocaine, the unemployment rate, the share of the population receiving
higher education, the percentage of Medicaid enrollees, and the share of private health insurance). Descriptions and
summary statistics are provided in Table 4.6.
84
4.3.2 Methods
Around 85% of patients in the sample data have more than two treatment episodes. Considering the longitudinal
nature of the data, in which repeated observations are recorded for each patient, I use generalized estimating equa-
tions (GEE) which also allow for within-group correlations. Logit link functions are used to examine the association
between the IMD waivers and the rates of initiation, engagement, and continuation after residential care, adjusting for
both individual-level and state-level characteristics. The empirical model is:
ln
Y
i j
1− Y
i j
=β
0
+β
1
∗ Waiver+β
2
∗ X
i j
(4.1)
+β
3
∗ T
i j
+β
4
∗ Z
i j
+β
5
∗ Year+β
6
∗ State,
where Y
i j
are the likelihood that patients successfully initiate, engage, and continue their treatment. Waiver is the
IMD waiver. X
i j
is the individual demographic characteristics, including age, gender, and race. T
i j
refers to clinical
and enrollment characteristics, including the number of past SUD diagnoses, past SUD treatments, the length of time
since first enrollment in the plan, and MH comorbidity. Z
i j
are state-level factors that may also be associated with
treatment, such as demographics and other state policies related to treatment. The subscripts i and j denote individual
and episode. State and year-fixed effects are also included in the model.
For examining the association of the IMD waiver and the average length of stay in treatment episodes, I use GEE
with Poisson regression as the link function and use the same set of controlling variables as above:
lnE(lengtho f stay)
i j
=γ
0
+γ
1
∗ Waiver+γ
2
∗ X
i j
(4.2)
+γ
3
∗ T
i j
+γ
4
∗ Z
i j
+γ
5
∗ Year+γ
6
∗ State.
Odds ratios in the first three models and incidence ratios in the fourth model of all variables are generated by
exponentiating the parameter estimates of these models.
4.4 Results
Table 4.7-4.9 show the initiation, engagement, and continuation after residential services and average LOS statis-
tics by demographic and clinical characteristic groups. Females and whites have higher initiation rates, engagement
rates, and continuity rates in this group of commercially insured individuals, and they also tend to have more extended
85
stays in treatment episodes. The youngest age group (21-29) has higher initiation and engagement rates than older
age groups, while the difference is unclear in the continuation rate after residential services. The young age group
(21-29) also tends to have a longer average length of stay in episodes. The mental health comorbidity group shows
lower initiation and engagement rates and shorter length of stay in episodes but a higher continuation rate after res-
idential services. Initiation and engagement rates are also higher in the greater number of diagnoses (≥ 20) and past
SUD services group (≥ 50) but lower among newly enrolled patients’ episodes. The continuation rate after residential
services is higher in the greater number of past SUD service group (≥ 50) and among patients that are enrolled earlier.
Average LOS is longer in the greater number of diagnoses (≥ 20) and the greater number of past SUD services (≥ 50)
groups and is also longer among newly enrolled patients’ episodes.
4.4.1 Initiation and engagement of treatment
Calculated odds ratios from estimating model (1) for initiation are provided in Table 4.10. Having an IMD
waiver does not affect the likelihood of initiating treatment. The results suggest that females are more likely to initiate
treatment (OR 1.09; 95% CI 1.085 to 1.095). The white group is also more likely to initiate treatment (OR 1.012; 95%
CI 1.007 to 1.018) than other ethnic groups. Being older contributes negatively to the likelihood of initiating treatment
but on a very small magnitude (OR 0.999; 95% CI 0.998 to 0.999). Patients having mental health comorbidity are
less likely to initiate treatment (OR 0.686; 95% CI 0.683 to 0.689) compared to those who do not have mental health
comorbidity. Moreover, having a higher number of previous treatments makes a patient slightly more likely to initiate
treatment (OR 1.004; 95% CI 1.004 to 1.004), while a higher number of past SUD diagnoses is negatively associated
with initiating treatment (OR 0.998; 95% CI 0.998 - 0.999). At the same time, being enrolled in the plan earlier (longer
length since first enrollment) is also shown to reduce the likelihood of initiating SUD treatment (OR 0.999; 95% CI
0.999 to 0.999).
Regarding other policy effects, I find that ACA Medicaid expansion is significantly associated with a higher
initiation rate (OR 1.021; 95% CI 1.003 to 1.039). Having a parity law in SUD (OR 1.041; 95% CI 1.022 to 1.059)
and having a recreational marijuana law (OR 1.067; 95% CI 1.043 to 1.091) are also positively associated with a
higher likelihood of initiation. On the contrary, a naloxone prescribing law is negatively associated with the likelihood
of initiation (OR 0.986; 95% CI 0.976 to 0.997). Some other state-level factors that are associated with a higher
initiation rate include opioid prescribing rate (OR 1.1; 95% CI 1.049 to 1.154), any mental illness prevalence rate
(OR 39.746; 95% CI 6.794 to 232.479), unemployment rate (OR 3.347; 95% CI 2.252 to 4.974), as well as cocaine
prevalence rate (OR 3.196; 95% CI 0.959 to 10.655) but at a less significant level (5% - 10%).
86
Table 4.11 provides the calculated odds ratios associated with the engagement rate for SUD treatment. IMD
waivers are found to have a small but significant effect on leading to a higher likelihood of engaging in treatment
(OR 1.026; 95% CI 1.014 to 1.039). Similar to initiation, females are found to be more likely to engage in treatment
than males (OR 1.077; 95% CI 1.07 to 1.087). Whites are more likely to engage in treatment compared to other
ethnic groups (OR 1.052; 95% CI 1.044 to 1.06). Being older is associated with being slightly less likely to engage
in treatment (OR 0.995; 95% CI 0.995 to 0.996). Patients with mental health comorbidity are less likely to engage in
treatment (OR 0.638; 95% CI 0.634 to 0.642). Prior SUD treatment contributes to a higher likelihood of engaging in
treatment (OR 1.004; 95% CI 1.004 to 1.004), while the earlier a patient enrolled in the plan, the patient is a little bit
less likely to engage in treatment (OR 0.999; 95% CI 0.999 to 0.999). Unlike initiation, more past SUD diagnoses are
associated with a larger likelihood of treatment engagement (OR 1.003; 95% CI 1.002 to 1.004).
In terms of state-level factors, patients living in states that have parity laws (OR 1.041; 95% CI 1.022 to 1.059),
expanded Medicaid under the ACA (OR 1.021; 95% CI 1.003 to 1.039), or recreational marijuana laws (OR 1.067;
95% CI 1.043 to 1.091) are more likely to engage in treatment. However, mental health parity is associated with a
smaller likelihood of treatment engagement (OR 0.984; 95% CI 0.97 to 0.998).
4.4.2 Followup care after residential services
The estimates of the continuation rate after residential services are presented in Table 4.12. IMD waivers are not
found significantly impact the follow-up rate after residential care. Again, females (OR 1.116; 95% CI 1.106 to 1.127)
and whites (OR 1.108; 95% CI 1.006 to 1.096) are more likely to receive follow-up care after receiving residential
services than males and other racial groups. Age is negatively associated with follow-up treatment after residential
care (OR 0.995; 95% CI 0.995 to 0.995). Different from initiation and engagement, patients having mental health
comorbidity are more likely to receive follow-up treatment after residential care compared to those who do not have
(OR 1.033; 95% CI 1.024 to 1.042). A higher number of past SUD diagnoses (OR 1.004; 95% CI 1.004 to 1.005),
as well as being earlier getting enrolled in the plan (OR 1.0006; 95% CI 1.0004 to 1.0007) are both significantly
associated with a higher likelihood of receiving follow-up care after residential services. In contrast, more previous
SUD treatments are associated with a lower likelihood of receiving follow-up care after residential services (OR 0.996;
95% CI 0.996 to 0.997).
At the state level, Medicaid expansion under ACA slightly contributes to a lower follow-up rate after residential
services (OR 0.977; 95% CI 0.952 to 1.004). And the presence of a general mental health parity also contributes to
a lower follow-up rate after residential services (OR 0.974; 95% CI 0.953 to 0.997). Different from initiation and
87
engagement, a higher cocaine prevalence rate (OR 0.087; 95% CI 0.006 to 1.251) and a higher unemployment rate
(OR 0.346; 95% CI 0.147 to 0.816) are associated with a lower likelihood of receiving follow-up care after residential
services. A higher private health insurance rate is also associated with a lower likelihood of receiving follow-up care
after residential services (OR 0.29; 95% CI 0.138 to 0.61).
4.4.3 Average length of stay in treatment episodes
The incidence rate ratios (IRRs) obtained from estimating GEE with a Poisson link function on the average
length of stay in treatment episodes are provided in Table 4.13. I failed to find a significant effect of adopting an IMD
waiver on the average length of stay in episodes. Similar to what was found previously for initiation, engagement,
and continuity of care, females are more likely to stay longer in treatment than males (IRR 1.088; 95% CI 1.082 to
2.46), whites are more likely to stay longer in treatment than other racial groups (IRR 1.101; 95% CI 1.093 to 1.107),
and being older is associated with a shorter length of stay in treatment episodes on average (IRR 0.995; 95% CI 0.994
to 0.995). On average, patients with mental health comorbidity have a shorter stay in treatment than patients without
mental health comorbidity (IRR 0.445; 95% CI 0.444 to 0.448). More past SUD diagnoses (IRR 1.002; 95% CI 1.002
to 1.002) and treatments (IRR 1.009; 95% CI 1.008 to 1.01) are associated with a longer stay in treatment episodes
on average. Earlier enrollment in the plan is associated with a shorter stay in treatment IRR (0.998; 95% CI 0.998 to
0.998).
At the state level, come contributing factors to a longer average stay in treatment include parity law in SUD
(IRR 1.026; 95% CI 1.011 to 1.041), recreational marijuana law (IRR 1.063; 95% CI 1.044 to 1.083), higher mental
illness prevalence rate (IRR 6.835; 95% CI 0.879 to 53.144), higher cocaine prevalence rate (IRR 4.988; 95% CI 1.203
to 20.697), higher unemployment rate (IRR 1.872; 95% CI 1.195 to 2.933), and higher rate of having private health
insurance (IRR 1.853; 95% CI 1.235 to 2.784). On the contrary, naloxone prescribing law is associated with a shorter
average length of stay in treatment (IRR 0.969; 95% CI 0.957 to 0.98).
4.5 Conclusion and Limitation
In the previous chapter, I examined access to different levels of SUD treatments by commercial insurance en-
rollees. I counted receiving any treatment as “access to treatment” as long as the procedure codes fell in any one
category of outpatient, intensive outpatient, inpatient, or medication-assisted treatment. In this chapter, I take advan-
tage of the longitudinal claims data at the patient’s level and track the continuation of care, allowing me to consider
88
the impact of this policy on the quality of SUD treatment received. I found access to outpatient care is about 40%
on average in the previous chapter, while in this chapter, I find that the actual initiation rate falls slightly above 20%
on average. This contrast shows that fewer SUD patients actually initiate treatment even if they have access to it.
Furthermore, the analysis of trends of the four measures indicates that the quality of receiving SUD care based on
process measures has not improved much over the past decade.
My analysis provides evidence that females are more likely to initiate and engage in SUD treatment than males.
Females are also more likely to receive follow-up care after residential services and tend to have longer lengths of stay
in treatment episodes on average than males. Similarly, whites are more likely to initiate and engage in SUD treatment
than other racial groups. They are also more likely to receive follow-up care after residential services and tend to
have longer stays in treatment episodes on average than other racial groups. Among nonelderly adults, older people
are less likely to initiate, engage, and continue their treatments after residential care and tend to have shorter stays
in treatment. People with mental health comorbidity are significantly less likely to initiate and engage in treatment
but are more likely to continue treatment after residential care than patients without mental health comorbidity. On
average, patients with mental health comorbidity have shorter stays in treatment.
Medicaid IMD waivers expand coverage eligibility for Medicaid enrollees for their inpatient treatment services
in SUD. Due to the limitation of treatment capacity and continuity of care required for SUD patients, this expansion
among Medicaid enrollees raises questions about having a crowd-out effect on commercial insurance enrollees receiv-
ing continuing SUD care. However, my study does not find any crowd-out effect of IMD waivers on quality measures,
including initiation, continuation after residential services, and the average length of stay in treatment episodes among
commercially insured SUD patients. Interestingly, the engagement rate among commercially insured SUD patients is
higher under the impact of IMD waivers. This finding is consistent with the hypothesis that facilities and healthcare
systems that undertake to improve access and coordination of care across all levels of care may benefit any patient
receiving services from those facilities, not just Medicaid enrollees.
In the previous chapters, I argued that IMD waivers opened up opportunities for facilities to integrate services,
increase care supply, and for patients to seek treatment at some intermediate level, such as intensive outpatient or partial
hospitalization, which was underused as reflected by insurance claims. In this chapter, I provide further evidence that
IMD waivers generate stimulating effects on patients’ utilization of treatment services, reflected by their engagement
rate in treatment. As my data comes from commercial insurance enrollees, it tells that the systematic improvements
in getting quality SUD treatment targeting one group of patients have positive spillovers to other insurance enrollees.
This result may also be because Medicaid is the largest payer of SUD treatment, and Medicaid enrollees comprise the
majority of the patient group. And thus, Medicaid policy reforms are making a significant impact on the whole market.
89
In the meantime, the trends of quality changes displayed in this chapter using quality indicators tell us that
quality is hard to improve. The reason may be that quality treatment services are determined by both the provider and
the patient. Policies should recognize the factors at the individual level that affect treatment quality, such as mental
health comorbidity, other health complications, and demographic and socioeconomic barriers, to help groups that are
under more challenges in receiving care.
My study also has several limitations. First, due to the complexity of measuring quality and the data availability,
the Washington Circle group measures can only reflect the quality change from the treatment process aspect. More
spillover changes may be found if evaluations can be done using other types of measures, such as patient outcomes and
satisfaction. Second, since my data do not provide information on SUD assessment results of patients and contain a
very small amount of detoxification cases, I am not able to provide a comparative evaluation based on continuation after
positive assessment and detoxification, which are important quality measures that frequently occur in other studies.
Third, due to the limitation of my data, I do not have many details that can account for more variations of each patient
and each episode. Given the limitations, the actual role of the IMD waiver plays in affecting the quality of treatment
has to be evaluated carefully and needs more rigorous data.
90
Figure 4.1: Trends of Initiation Rate among SUD Patients in Optum Commercial Insurance Data
Notes: Initiation is defined as the percent of individuals who have an index outpatient (OP) or intensive outpatient
(IOP) service with no other substance abuse services in the previous 60 days and who have received a second SUD
service (other than detoxification or crisis care) within 14 days after the index service. SUD is Substance Use Disorder.
MH is Mental Health. The blue line sample excludes patients that do not have MH comorbidity from the orange line
- SUD patients sample. 95% confidence intervals are noted in the figure. Data is from the Clinformatics® Data Mart,
a database covering administrative health claims for members of a national managed care company affiliated with
Optum.
91
Figure 4.2: Trends of Engagement Rate among SUD Patients in Optum Commercial Insurance
Data
Notes: Engagement is defined as the percentage of individuals who initiated OP or IOP substance abuse treatment and
received two additional services within 30 days of initiation. SUD is Substance Use Disorder. MH is Mental Health.
The blue line sample excludes patients that do not have MH comorbidity from the orange line - SUD patients sample.
95% confidence intervals are noted in the figure. Data is from the Clinformatics® Data Mart, a database covering
administrative health claims for members of a national managed care company affiliated with Optum.
92
Figure 4.3: Trends of Continuation Rate after Residential Services among SUD Patients in Optum
Commercial Insurance Data
Notes: The continuity of care with respect to residential treatment is defined as the percentage of individuals
who have a residential stay that is followed by another service (other than detoxification or crisis care) within 14 days
after discharge, with no service-free period required before the residential stay. SUD is Substance Use Disorder. MH
is Mental Health. The blue line sample excludes patients that do not have MH comorbidity from the orange line -
SUD patients sample. 95% confidence intervals are noted in the figure. Data is from the Clinformatics® Data Mart,
a database covering administrative health claims for members of a national managed care company affiliated with
Optum.
93
Figure 4.4: Trends of Length of Stay per Episode among SUD Patients in Optum Commercial
Insurance Data
Notes: The length of stay (LOS) is defined as the average number of days of a single treatment episode across
locations of services for a patient, and I consider two treatments that occur within 60 days of each other as belonging
to one treatment episode. SUD is Substance Use Disorder. MH is Mental Health. The blue line sample excludes
patients that do not have MH comorbidity from the orange line - SUD patients sample. 95% confidence intervals are
noted in the figure. Data is from the Clinformatics® Data Mart, a database covering administrative health claims for
members of a national managed care company affiliated with Optum.
94
Table 4.1: 2015-2021 IMD Waiver in Substance Use Disorder (SUD)
Effective Year N States
2015 1 CA
2017 5 MD, MA, NJ, V A, UT
2018 10 IL, IN, KY , LA, NH, PA, VT, WA, WV , WI
2019 10 AK, DE, KS, MI, MN, NE, NM, NC, OH, RI
2020 2 DC, ID
2021 4 CO, ME, OK, OR
Total 32
Notes: The effective year is the year when the waiver is adopted regardless of the specific point in time.
Data is from RAND Opioid Policy Tools and Information Center (OPTIC). policy database
95
Table 4.2: Demographic and Clinical Characteristics for SUD Patients (Patient-level) for the Period
2010-2021
SUD Patients: N=3,190,840
N %
Gender
Female 1,426,431 44.7
Male 1,763,908 55.3
Unknown 501 0.0
Race
White 2,264,303 71.0
Black 345,580 10.8
Hispanic 286,126 9.0
Asian 65,856 2.0
Unknown 228,975 7.2
Age
Age 21-29 575,882 18.0
Age 30-39 722,450 22.6
Age 40-49 770,783 24.2
Age 50-64 1,121,725 35.2
MH comorbidity
Yes 1,561,939 49.0
No 1,628,901 51.0
Number of diagnosis
1 - 2 2,367,863 74.2
3 - 4 486,658 15.2
5 - 9 215,987 6.8
10 - 19 83,117 2.6
≥ 20 37,215 1.2
Number of past SUD services
0 - 4 2,090,884 65.5
5 - 9 569,178 17.8
10 - 19 330,839 10.4
20 - 49 166,698 5.2
≥ 50 33,241 1.1
Number of months since first enrollment
0 - 4 290,396 9.1
5 - 9 486,329 15.2
10 - 19 650,630 20.4
20 - 49 852,226 26.7
≥ 50 911,259 28.6
Notes: SUD is Substance Use Disorders. MH is mental health. The sample is at patient level, which means
only one record is kept for each patient. MH comorbidity is considered if the patient has an MH diagnosis up
to one year prior to their first-time SUD treatment. Since the data is longitudinal, the numbers of diagnoses,
past treatments, and months since first enrollment are the average across the whole period.
96
Table 4.3: Demographic and Clinical Characteristics for SUD Patients Starting New Episodes with
OP (Episode-level) in the Optum Commercial Insurance Data for the Period 2010-2021
Starting new treatment episodes with OP: N=7,553,130
N %
Gender
Female 3,558,713 47.1
Male 3,993,256 52.9
Unknown 1,161 0.0
Race
White 5,568,867 73.7
Black 801,329 10.6
Hispanic 629,384 8.3
Asian 145,998 1.9
Unknown 407,552 5.4
Age
Age 21-29 1,000,796 13.3
Age 30-39 1,505,162 19.9
Age 40-49 1,840,627 24.4
Age 50-64 3,206,545 42.4
MH comorbidity
Yes 3,672,703 48.6
No 3,880,427 51.4
Initiation
Yes 1,738,533 23.0
No 5,814,597 77.0
Engagement
Yes 763,330 10.1
No 6,789,800 89.9
Number of diagnosis
1 - 2 6,293,535 83.3
3 - 4 772,663 10.2
5 - 9 342,381 4.5
10 - 19 96,937 1.3
≥ 20 47,614 0.6
Number of past SUD services
0 - 4 4,809,493 63.7
5 - 9 1,173,868 15.5
10 - 19 882,082 11.7
20 - 49 553,810 7.3
≥ 50 133,877 1.8
Number of months since first enrollment
0 - 4 596,985 7.9
5 - 9 725,486 9.6
10 - 19 1,235,852 16.4
20 - 49 2,016,086 26.7
≥ 50 2,978,721 39.4
Notes: OP refers to outpatient. SUD is Substance Use Disorder. MH is mental health. The two samples
are at episode level, which means every episode is kept for each patient. Since the data is longitudinal, the
numbers of diagnoses, past treatments, and months since first enrollment are the average across the whole
period.
97
Table 4.4: Demographic and Clinical Characteristics for SUD Patients Receiving Residential Care
(Episode-level) in the Optum Commercial Insurance Data for the Period 2010-2021
Receiving residential care: N=5,486,362
N %
Gender
Female 2,337,524 42.6
Male 3,146,214 57.4
Unknown 2,601 0.0
Race
White 3,939,822 71.8
Black 656,570 12.0
Hispanic 473,125 8.6
Asian 93,950 1.7
Unknown 322,895 5.9
Age
Age 21-29 747,666 13.6
Age 30-39 781,775 14.3
Age 40-49 1,133,598 20.7
Age 50-64 2,823,323 51.5
MH comorbidity
Yes 1,974,388 36.0
No 3,511,974 64.0
Continuation of care
Yes 620,840 11.3
No 4,865,522 88.7
Number of diagnosis
1 - 2 2,453,571 44.7
3 - 4 1,020,432 18.6
5 - 9 933,502 17.0
10 - 19 528,458 9.6
≥ 20 550,399 10.0
Number of past SUD services
0 - 4 1,494,893 27.3
5 - 9 780,132 14.2
10 - 19 939,340 17.1
20 - 49 1,195,095 21.8
≥ 50 1,076,902 19.6
Number of months since first enrollment
0 - 4 499,376 9.1
5 - 9 587,952 10.7
10 - 19 861,827 15.7
20 - 49 1,435,557 26.2
≥ 50 2,101,650 38.3
Notes: SUD is Substance Use Disorders. MH is mental health. The two samples are at episode-level, which
means every episode is kept for each patient. Since the data is longitudinal, the numbers of diagnoses, past
treatments, and months since first enrollment are the average across the whole period.
98
Table 4.5: Data Sources of Key Variables
Variables Data sources
Policies
ACA Medicaid expansion RAND Opioid Policy Tools and Information Center (OPTIC)
Parity law in SUD RAND Opioid Policy Tools and Information Center (OPTIC)
Naloxone prescribing law RAND Opioid Policy Tools and Information Center (OPTIC)
Recreational marijuana law RAND Opioid Policy Tools and Information Center (OPTIC)
State Characteristics
Opioid prescribing rate Center for Disease Control and Prevention (CDC)
Any mental illness rate National Survey of Drug Use and Health (NSDUH)
Using cocaine rate National Survey of Drug Use and Health (NSDUH)
Unemployment rate University of Kentucky Center for Poverty Research
Share of receiving higher education American Community Survey (ACS)
Share of private health insurance American Community Survey (ACS)
Share of Medicaid enrollees Kaiser Family Foundation (KFF)
Notes: ACA is Affordable Care Act. SUD is a substance use disorder.
99
Table 4.6: Summary Statistics for State-level Co-variates for the Period 2010-2021
mean sd min max
Policies
ACA Medicaid Expansion Status 0.481 0.498 0 1
Parity law in SUD 0.694 0.437 0 1
Mental health parity 0.161 0.344 0 1
Naloxone prescribing law 0.676 0.448 0 1
Recreational marijuana law 0.076 0.264 0 1
State Characteristics
Opioid prescribing rate 0.677 0.236 0.25 1.438
Any mental illness rate 0.025 0.005 0.017 0.046
Using cocaine rate 0.015 0.004 0.007 0.041
Unemployment rate 0.054 0.019 0.024 0.128
Higher education rate 0.609 0.055 0.439 0.778
Share of private health insurance 0.684 0.057 0.532 0.806
Share of Medicaid enrollees 0.209 0.065 .09 0.429
Notes: ACA is Affordable Care Act. SUD is a substance use disorder. All the numbers are the original value
according to descriptions. The opioid prescribing rate can exceed 1.
100
Table 4.7: Initiation and Engagement Rate (Starting with OP) by Group in the Optum Commercial
Insurance Data for the Period 2010-2021
N % Initiation % Engagement
Gender
Female 3,558,713 23.9 10.5
Male 3,993,256 22.2 9.7
Race
White 5,568,867 23.0 10.1
Black 801,329 22.4 9.5
Hispanic 629,384 23.8 10.4
Asia 145,998 22.6 9.7
Age
21 - 29 1,000,796 23.5 11.2
30 - 39 1,505,162 22.8 10.2
40 - 49 1,840,627 22.8 9.8
50 - 64 3,206,545 23.1 9.9
MH comorbidity
Yes 3,672,703 19.7 8.2
No 3,880,427 26.2 11.9
Number of diagnosis
1 - 2 6,293,535 23.2 10.0
3 - 4 772,663 21.2 10.2
5 - 9 342,381 22.9 13.3
10 - 19 96,937 25.8 16.1
≥ 20 47,614 29.2 22.8
Number of past SUD services
0 - 4 4,809,493 23.1 10.3
5 - 9 1,173,868 20.2 7.8
10 - 19 882,082 22.8 9.4
20 - 49 553,810 26.5 12.2
≥ 50 133,877 32.8 18.2
Number of months since first enrollment
< 5 596,985 27.3 13.3
≥ 5 and< 10 725,486 23.7 10.5
≥ 10 and< 20 1,235,852 22.1 9.5
≥ 20 and< 50 2,016,086 22.3 9.7
≥ 50 2,978,721 22.7 9.7
Notes: OP refers to outpatient. MH is mental health. The number N and rates are based on episode level,
which means each episode is taken into calculation.
101
Table 4.8: Continuation Rate after Residential Services by Group in the Optum Commercial Insur-
ance Data for the Period 2010-2021
N % Continuation after
residential services
Gender
Female 2,284,222 12.3
Male 3,049,544 11.3
Race
White 3,823,842 12.1
Black 645,730 10.3
Hispanic 462,461 11.3
Asia 91,904 11.2
Age
21 - 29 700,799 12.0
30 - 39 746,828 12.8
40 - 49 1,100,398 12.6
50 - 64 2,788,342 11.0
MH comorbidity
Yes 1,959,829 11.9
No 3,376,538 11.6
Number of diagnosis
1 - 2 2,453,571 11.8
3 - 4 1,020,432 12.2
5 - 9 933,502 11.8
10 - 19 528,458 10.6
≥ 20 550,399 7.5
Number of past SUD services
0 - 4 1,494,893 13.3
5 - 9 780,132 13.3
10 - 19 939,340 11.8
20 - 49 1,195,095 10.5
≥ 50 1,076,902 7.6
Number of months since first enrollment
0 - 4 499,376 11.0
5 - 9 587,952 11.2
10 - 19 861,827 11.2
20 - 49 1,435,557 11.2
≥ 50 2,101,650 11.5
Notes: MH is mental health. The number N and rates are based on episode level, which means each episode
is taken into calculation.
102
Table 4.9: Average Length of Stay by Group in the Optum Commercial Insurance Data for the
Period 2010-2021
N Average length
of stay (LOS)
Gender
Female 3,891,365 5.3
Male 4,458,354 4.8
Unknown 1,362
Race
White 6,141,672 5.2
Black 892,422 4.7
Hispanic 699,116 4.7
Asia 160,419 4.4
Unknown 457,452
Age
21 - 29 1,141,550 5.6
30 - 39 1,656,567 4.9
40 - 49 2,025,450 4.9
50 - 64 3,527,514 5.0
MH comorbidity
Yes 3,954,8196 3.0
No 4,396,262 6.9
Number of diagnosis
1 - 2 6,968,185 4.9
3 - 4 833,875 4.6
5 - 9 376,381 6.0
10 - 19 113,361 8.9
≥ 20 59,279 13.9
Number of past SUD services
0 - 4 5,422,147 5.1
5 - 9 1,239,501 3.6
10 - 19 939,384 4.5
20 - 49 598,731 6.6
≥ 50 151,318 12.3
Number of months since first enrollment
0 - 4 858,695 7.1
5 - 9 813,699 4.9
10 - 19 1,307,054 4.5
20 - 49 2,151,740 4.7
≥ 50 3,219,893 5.1
Notes: MH is mental health. The number N and length of stay are based on episode level, which means
each episode is taken into calculation. LOS here counts only stays actually in treatment, differing from the
duration that usually covers gaps within one episode.
103
Table 4.10: GEE with Logit Link Function Estimating the Rate of Initiation for the Period 2010-
2021
Odds Ratio 95% Confidence Interval P-value
Waiver
IMD waiver 1.007 0.997 - 1.016 NS
Individual Characteristics
Female 1.09 1.085 - 1.095 <0.0001
White 1.012 1.007 - 1.018 <0.0001
Age 0.999 0.998 - 0.999 <0.0001
Mental health comorbidity 0.686 0.683 - 0.689 <0.0001
Number of past SUD diagnosis 0.998 0.998 - 0.999 <0.0001
Number of previous SUD treatments 1.004 1.004 - 1.004 <0.0001
Number of months since first enrollment 0.999 0.999 - 0.999 <0.0001
Policy Effects
ACA Medicaid expansion 1.018 1.006 - 1.031 0.005
Parity in SUD 1.02 1.007 - 1.034 0.002
Mental health parity 0.993 0.983 - 1.004 NS
Naloxone prescribing law 0.986 0.976 - 0.997 0.01
Recreational marijuana law 1.037 1.02 - 1.054 <.0001
State Characteristics
Opioid prescribing rate 1.10 1.049 - 1.154 <.0001
Any mental illness rate 39.746 6.794 - 232.479 0.001
Using cocaine rate 3.196 0.959 - 10.655 0.059
Unemployment rate 3.347 2.252 - 4.974 <0.0001
Higher education rate 1.062 0.652 - 1.729 NS
Private health insurance rate 0.998 0.702 - 1.419 NS
Medicaid enrollment rate 1.014 0.867 - 1.185 NS
Notes: NS is Not Significant. IMD waiver is the Medicaid Institute for Mental Disease exclusion waiver, the
policy of interest. SUD is a substance use disorder. ACA is Affordable Care Act. The model is estimated
using generalized estimating equations (GEE), using logit link function, clustering at the patient level, with
state and year fixed effects.
104
Table 4.11: GEE with Logit Link Function Estimating the Rate of Engagement for the Period
2010-2021
Odds Ratio 95% Confidence Interval P-value
Waiver
IMD waiver 1.026 1.014 - 1.039 <0.0001
Individual Characteristics
Female 1.077 1.07 - 1.087 <0.0001
White 1.052 1.044 - 1.06 <0.0001
Age 0.995 0.995 - 0.996 <0.0001
Mental health comorbidity 0.638 0.634 - 0.642 <0.0001
Number of past SUD diagnosis 1.003 1.002 - 1.004 <0.0001
Number of previous SUD treatments 1.004 1.004 - 1.004 <0.0001
Number of months since first enrollment 0.999 0.999 - 0.999 <0.0001
Policy Effects
ACA Medicaid expansion 1.021 1.003 - 1.039 0.022
Parity in SUD 1.041 1.022 - 1.059 <0.0001
Mental health parity 0.984 0.97 - 0.998 0.03
Naloxone prescribing law 0.991 0.977 - 1.006 NS
Recreational marijuana law 1.067 1.043 - 1.091 <.0001
State Characteristics
Opioid prescribing rate 1.125 1.052 - 1.204 0.0006
Any mental illness rate 533.575 46.82 - 6,080.243 <.0001
Using cocaine rate 11.719 2.259 - 60.782 0.0034
Unemployment rate 4.624 2.664 - 8.026 <0.0001
Higher education rate 0.679 0.344 - 1.339 NS
Private health insurance rate 1.385 0.848 - 2.261 NS
Medicaid enrollment rate 0.949 0.762 - 1.182 NS
Notes: NS is Not Significant. IMD waiver is the Medicaid Institute for Mental Disease exclusion waiver, the
policy of interest. SUD is a substance use disorder. ACA is Affordable Care Act. The model is estimated
using generalized estimating equations (GEE), using logit link function, clustering at the patient level, with
state and year fixed effects.
105
Table 4.12: GEE with Logit Link Function Estimating the Rate of Treatment within 14 Days after
Residential Services for the Period 2010-2021
Odds Ratio 95% Confidence Interval P-value
Waiver
IMD waiver 0.984 0.964 - 1.004 NS
Individual Characteristics
Female 1.116 1.106 - 1.127 <0.0001
White 1.108 1.006 - 1.096 <0.0001
Age 0.995 0.995 - 0.995 <0.0001
Mental health comorbidity 1.033 1.024- 1.042 <0.0001
Number of past SUD diagnosis 1.004 1.004 - 1.005 <0.0001
Number of previous SUD treatments 0.996 0.996 - 0.997 <0.0001
Number of months since first enrollment 1.0006 1.0004 - 1.0007 <0.0001
Policy Effects
ACA Medicaid expansion 0.977 0.952 - 1.004 0.091
Parity in SUD 1.009 0.981 - 1.037 NS
Mental health parity 0.974 0.953 - 0.997 0.023
Naloxone prescribing law 0.984 0.962 - 1.006 NS
Recreational marijuana law 0.982 0.947 - 1.019 NS
State Characteristics
Opioid prescribing rate 1.062 0.959 - 1.176 NS
Any mental illness rate 21.846 0.511 - 933.835 NS
Using cocaine rate 0.087 0.006 - 1.251 0.073
Unemployment rate 0.346 0.147 - 0.816 0.015
Higher education rate 2.18 0.757 - 6.275 NS
Private health insurance rate 0.29 0.138 - 0.61 0.001
Medicaid enrollment rate 0.927 0.664 - 1.293 NS
Notes: NS is Not Significant. ACA is Affordable Care Act. SUD is a substance use disorder. The model is
estimated using generalized estimating equations (GEE), using logit link function, clustering at the patient
level, with state and year fixed effects.
106
Table 4.13: GEE with Poisson Link Function Estimating the Average Length of Stay in Treatment
Episodes for the Period 2010-2021
IRR 95% Confidence Interval P-value
Waiver
IMD waiver 0.995 0.984 - 1.006 NS
Individual Characteristics
Female 1.088 1.082 - 2.46 <0.0001
White 1.101 1.093 - 1.107 <0.0001
Age 0.995 0.994 - 0.995 <0.0001
Mental health comorbidity 0.445 0.444 - 0.448 <0.0001
Number of past SUD diagnosis 1.009 1.008 - 1.01 <0.0001
Number of previous SUD treatments 1.002 1.002 - 1.002 <0.0001
Number of months since first enrollment 0.998 0.998- 0.998 <0.0001
Policy Effects
ACA Medicaid expansion 0.993 0.993 - 1.023 NS
Parity in SUD 1.026 1.011 - 1.041 0.0006
Mental health parity 0.996 0.984 - 1.008 NS
Naloxone prescribing law 0.969 0.957 - 0.98 <0.0001
Recreational marijuana law 1.063 1.044 - 1.083 <0.0001
State Characteristics
Opioid prescribing rate 1.04 0.983 - 1.101 NS
Any mental illness rate 6.835 0.879 - 53.144 0.066
Using cocaine rate 4.988 1.203 - 20.697 0.027
Unemployment rate 1.872 1.195 - 2.933 0.006
Higher education rate 0.79 0.441 - 1.413 NS
Share of private health insurance 1.853 1.235 - 2.784 0.003
Share of Medicaid enrollment 1.055 0.884 - 1.259 NS
Notes: NS is Not Significant. ACA is Affordable Care Act. SUD is a substance use disorder. The model
is estimated using generalized estimating equations (GEE), using Poisson distribution, and clustering at the
patient level, with state and year-fixed effects.
107
Chapter 5
Conclusion and ongoing work
5.1 Main findings
Federal and state governments in the US have made tons of efforts to tackle the severe public health problem of
substance abuse, mainly by increasing the availability of treatment, lowering the cost of treatment, and improving the
quality of treatment. Every policy has its target areas, but the effect generated may be more comprehensive.
With special attention to a recent Medicaid policy change on SUD payments under the Section 1115 demonstra-
tion program, my dissertation provides an overview of the US treatment market for SUD from both the facility and
patient sides. Medicaid Section 1115 SUD waiver (IMD waiver) is one of the several policies focusing on payment
reform and coverage expansion so that more services on SUD can be paid through Medicaid. The IMD waiver di-
rectly impacts non-elderly Medicaid enrollees that receive inpatient SUD services and emphasizes the improvement
of quality care.
My work analyzes how policy changes influence facilities’ characteristics in terms of the general availability
and services offered, and how they vary by ownership and service types. I also examine treatment access patterns
and quality changes among commercially insured SUD patients. Under the policy context created by IMD waivers
in SUD, the analysis advances the understanding of the direct impact of those SUD waivers. It explores potential
spillover effects on patients insured under commercial plans.
In the second chapter, I provide a simplified analytical framework on how a facility might respond to a policy
change on payment expansion. Then my empirical examination finds that a state implementing a waiver is associated
with a 1.16 times the incidence rate of having SA treatment facilities on average by contrasting the average value of
states with waiver with states without waiver, suggesting a sizeable impact on the extensive margin. On the intensive
108
margin, facilities will likely provide co-occurring treatment services in SA and MH with a waiver. A facility’s odds
of delivering co-occurring treatment increase by 28% after the state adopts the waiver. A residential facility’s odds
of offering co-occurring treatment increased by 89% with the adoption of a waiver. I also find that the change in
the availability of treatment comes primarily from private, for-profit facilities. With the adoption of Section 1115
SUD waivers, there is a 12.9 percentage point increase among private for-profit facilities and a 7.06 percentage point
increase in private non-profit facilities. While on the other hand, no such effect is found among public or governmental
agencies.
In the third chapter, I find that commercial insurance enrollees use outpatient care the most in paying for SUD
treatment, followed by inpatient care, intensive outpatient care or partial hospitalization, and medication-assisted
therapies. The study shows a decreasing trend in paying for inpatient and outpatient SUD care over the past decade
in commercial insurance, while the trends in paying for intensive outpatient or partial hospitalization and medication-
assisted treatments for SUD patients have increased. SUD patients who are also suffering from MH comorbidity have
higher rates of receiving care at each critical level of care, compared to SUD patients in general, with similar trajectory
trends. Moreover, my findings suggest that the adoption of IMD exclusion waivers contributes to a 0.2 percentage point
increase in the rate of using intensive outpatient or partial hospitalization treatment for SUD patients in general. It
also contributes to a 0.4 percentage point increase in access to intensive outpatient or partial hospitalization treatment
among SUD patients with MH comorbidity.
In the fourth chapter, by taking advantage of the longitudinal commercial claims data at the patient’s level, I
track the continuation of care, advancing the third chapter from care access to receiving quality care and incorporating
individual socioeconomic and clinical characteristics. The analysis shows that being female and being in the white
racial group are associated with higher initiation, engagement, continuation after residential care rates, and longer
days of stay in treatment episodes. Patients with mental health comorbidity are associated with a lower initiation and
engagement rate and shorter length of stay in treatment but with a higher likelihood of receiving follow-up care after
residential services. The presence of an IMD SUD waiver is found to have an increasing effect on the likelihood of
engaging in treatment (OR 1.026; 95% CI 1.014 to 1.039) but not on other quality indicators.
5.2 Implications
To sum up, my analysis provides evidence of a series of spillover effects generated by the Medicaid IMD waivers
in SUD on both facilities’ operation and patients’ utilization of SUD care, beyond the direct impact of those waivers
that enables more non-elderly Medicaid enrolled SUD patients to receive more inpatient care.
109
On the facility side, the expansion on extensive and intensive margins of facilities reflects an underlying driving
force on service expansion generated by the policy change on coverage increase. My analysis shows that private
facilities are more likely to be motivated by extra profit and make operational adjustments. On the contrary, even
though public or governmental facilities should be more likely to accept Medicaid payments, they are less likely to
make service changes driven by extra profit. This result corroborates my analytical framework that extra profit margin
is an essential driving force on service change and expansion, and private facilities respond more naturally to this
force.
On the patient side, due to the limitation of treatment capacity and continuity of care required for SUD patients,
the Medicaid coverage expansion through IMD waivers raises questions about having a crowd-out effect on commer-
cial insurance enrollees in receiving continuing SUD care. However, my analysis suggests that commercial insurance
responds to this policy change by showing a slight increase in access to intensive outpatient or partial hospitalization
care, even though the average utilization of this level of care is very low. The mechanism could be that since Medicaid
allows more coverage on inpatient SUD care, some commercial plans’ premiums are freed up to create more oppor-
tunities to cover some intermediate levels of care. More Medicaid input in expanding access to treatment for SUD
patients also improves care coordination. Coordination and integration can simultaneously stimulate the development
of these intermediate care, such as intensive outpatient or partial hospitalization, which is generally underused. The
policy change improves the engagement rate among commercial insurance enrollees in the sample, which further sup-
ports the hypothesis on the overall efficiency gain in coordinating and integration. In addition, the fact that Medicaid
enrollees take the largest share of the patient population may also determine the importance of any Medicaid policy
changes.
One thing worth noting is that, in Chapter Three, my outcomes come from all kinds of procedures related to
SUD at the four different treatment levels, which result in a rate of access to outpatient care of around 40 percent. At
the same time, the actual initiation rate of SUD in Chapter Four falls to slightly above twenty percent on average. The
engagement and continuation rates are even lower. This contrast shows that the rates of SUD patients getting quality
treatment among commercially insured are meager, even though it may be easy for patients to get some simple form
of treatment.
Based on the analysis, it is hopeful to see that those IMD waivers, to some extent, help with increasing the
availability of treatment services, getting access to some intermediate levels of care even with commercial insurance,
and even improving the quality of treatment, though the rates of utilization and receiving quality care remain low.
Prioritizing resource allocation to some critical levels of care, such as inpatient, could generate efficiency gains for
the whole treatment connection and transitions. More policy efforts should target the quality of care, including care
110
integration and aftercare support resources. And since there is always a selection problem in the private health insur-
ance market, it may be more efficient to encourage ancillary or intermediate supportive care in commercial insurance
instead of mandating all the same benefits.
Quality care requires cooperation from providers and patients, and factors at the individual level are much more
uncontrollable. Health complications, living situations, cultural backgrounds, and ultimately a willingness to receive
treatment are all something that can hardly be solved at the aggregate level, but that does not mean public policies
have reasons to stop making new efforts.
5.3 Future work
In the past three years, the COVID pandemic has tremendously impacted the world in all sorts of ways. People
also faced unprecedented challenges with substance abuse problems. Deteriorating financial circumstances, disrup-
tions to work or job loss, reduced social support due to lockdown, and increased psychological distress during the
pandemic (Czeisler et al., 2020; Robinson & Daly, 2021) were all contributing factors to more substance use (Linas
et al., 2021; Stephenson, 2021). Governments made a range of state and federal policies to conquer the public emer-
gency, some of which affected access to and utilization of SUD treatment during the pandemic (Pessar et al., 2021).
Some policies temporarily relieved stress during a public emergency, but some were made permanently to solve more
problems effectively. My next step of the research plan is to closely examine those policy changes made during the
pandemic and assess their impacts compared with some other concurrent policies.
Many studies have identified that socioeconomic factors such as housing, family structure, and relationships are
associated with substance use and treatment engagement (Brown et al., 2011; S. M. Kelly et al., 2010; Mak et al.,
2010; Soto-Nevarez et al., 2021). Another direction I plan to take is exploring other types of policies, such as housing
policy, that may generate an impact on people with behavioral health problems. Previous studies have analyzed the
effects of interventions on housing programs and drop-in and shelter services but produced mixed results (Wang et al.,
2019). I hope to add more research output in this area and identify what policies can be effective and to what extent.
Real lives are behind all those numbers above. People have different reasons that make them addicted to sub-
stances. More research is needed to understand behavioral health problems and assess and adjust policies that always
start with good intentions.
“Why do you call it the wild side?”
“There seems to be an idea here that ... it might be dangerous down here ... or might be ... mainstream society is a
111
little bit intimidated by it or afraid of it ... I said it in jest ... because it’s not the wild side at all, it’s just a neighborhood
of people that are experiencing a lot of pain and dealing with a lot of trauma. ”
“What kind of pain do people typically go through?”
“Well ... high poverty area, there’s all kinds of ... people survive all kinds of trauma ... from early childhood abuse to
... you know ... relationship pain ... yeah ... just ...... people have gone through a lot down here. ”
“What does fentanyl feel like when you take it?”
“Just a big hug. ”
—— from ENDEVR Documentary: OpioidTragedy: InsidetheFentanylCrisis|TenDollarDeathTrip
112
References
Abraham, J. M., Gaynor, M., & V ogt, W. B. (2007). Entry and competition in local hospital mar-
kets. Journal of Industrial Economics, 55(2), 265–288. https://doi.org/10.1111/j.1467-
6451.2007.00311.x
Ali, M. M., Teich, J. L., & Mutter, R. (2017). Reasons for not seeking substance use disorder
treatment: Variations by health insurance coverage. J Behav Health Serv Res, 44(1), 63–
74. https://doi.org/10.1007/s11414-016-9538-3
Andrews, C. M., Grogan, C. M., Smith, B. T., Abraham, A. J., Pollack, H. A., Humphreys, K.,
Westlake, M. A., & Friedmann, P. D. (2018). Medicaid benefits for addiction treatment
expanded after implementation of the affordable care act. Health Affairs, 37(8), 1216–
1222. https://doi.org/10.1377/hlthaff.2018.0272
Arkhangelsky, D., Athey, S., Hirshberg, D. A., Imbens, G. W., & Wager, S. (2021). Synthetic
difference-in-differences. American Economic Review, 111(12), 4088–4118. https://doi.
org/10.1257/aer.20190159
Azzone, V ., Frank, R. G., Normand, S.-L. T., & Burnam, M. A. (2011). Effect of insurance parity
on substance abuse treatment. Psychiatric Services, 62(2), 129–134. https://doi.org/10.
1176/ps.62.2.pss6202 0129
Barry, C. L., Huskamp, H. A., & Goldman, H. H. (2010). A political history of federal mental
health and addiction insurance parity. Milbank Q, 88(3), 404–33. https://doi.org/10.1111/j.
1468-0009.2010.00605.x
Brown, C. H., Bennett, M. E., Li, L., & Bellack, A. S. (2011). Predictors of initiation and engage-
ment in substance abuse treatment among individuals with co-occurring serious mental
illness and substance use disorders. Addict Behav, 36(5), 439–47. https://doi.org/10.1016/
j.addbeh.2010.12.001
Busch, S. H., Epstein, A. J., Harhay, M. O., Fiellin, D. A., Un, H., Leader, J., D., & Barry, C. L.
(2014). The effects of federal parity on substance use disorder treatment. Am J Manag
Care, 20(1), 76–82.
Butler, M., Kane, R. L., McAlpine, D., Kathol, R. G., Fu, S. S., Hagedorn, H., & Wilt, T. J. (2008).
Integration of mental health/substance abuse and primary care. Evid Rep Technol Assess
(Full Rep), (173), 1–362.
113
Cantor, J. H., Deyoreo, M., Hanson, R., Kofner, A., Kravitz, D., Salas, A., Stein, B. D., & Kapinos,
K. A. (2022). Patterns in geographic distribution of substance use disorder treatment facil-
ities in the us and accepted forms of payment from 2010 to 2021. JAMA Network Open,
5(11), e2241128. https://doi.org/10.1001/jamanetworkopen.2022.41128
CDC. (2021). Wide-ranging online data for epidemiologic research (wonder) (Report). CDC, Na-
tional Center for Health Statistics. http://wonder.cdc.gov
CenterforSubstanceAbuseTreatment. (2006). Substance abuse: Clinical issues in intensive out-
patient treatment. (V ol. Chapter 3. Intensive Outpatient Treatment and the Continuum of
Care.). Substance Abuse; Mental Health Services Administration (US). https://www.ncbi.
nlm.nih.gov/books/NBK64088/
Chang, J. E., Cronin, C. E., Lindenfeld, Z., Pag´ an, J. A., & Franz, B. (2023). Association of medi-
caid expansion and 1115 waivers for substance use disorders with hospital provision of opi-
oid use disorder services: A cross sectional study. BMC Health Services Research, 23(1).
https://doi.org/10.1186/s12913-023-09035-0
Chavez, L. J., Steelesmith, D. L., Bridge, J. A., & Fontanella, C. A. (2022). Predictors of substance
use disorder treatment initiation and engagement among adolescents enrolled in medicaid.
Subst Abus, 43(1), 1260–1267. https://doi.org/10.1080/08897077.2022.2074603
ClevelandClinic. (2022). Substance use disorder (sud): What is the treatment for substance use
disorder?
CMS. (2015). Smd 15-003 re: New service delivery opportunities for individuals with a substance
use disorder. Centers for Medicare Medicaid Services. https://www.medicaid.gov/sites/
default/files/federal-policy-guidance/downloads/SMD15003.pdf
CMS. (2017). Smd 17-003 re: Strategies to address the opioid epidemic. Centers for Medicare
Medicaid Services. https : / / www. medicaid . gov / federal - policy - guidance / downloads /
smd17003.pdf
CMS. (2023a). Reducing substance use disorders: Quality measures (Report). Centers for Medi-
careMedicaid. https://www.medicaid.gov/resources- for- states/innovation- accelerator-
program / functional - areas / quality - measurement / reducing - substance - use - disorders -
quality-measures/index.html
CMS. (2023b). State waivers list (Report). Centers for Medicare Medicaid Services. https://www.
medicaid.gov/medicaid/section-1115-demo/demonstration-and-waiver-list/index.html
Cohen, C., Hern´ andez-Delgado, H., & Robles-Fradet, A. (2021). Medicaid section 1115 waivers
for substance use disorders: A review (Report). National Health Law Program. https : / /
healthlaw.org/wp-content/uploads/2021/06/FINAL Medicaid-1115-Waivers-for-SUD-
updated.pdf
114
Czeisler, M., Lane, R. I., Petrosky, E., Wiley, J. F., Christensen, A., Njai, R., Weaver, M. D., Rob-
bins, R., Facer-Childs, E. R., Barger, L. K., Czeisler, C. A., Howard, M. E., & Rajaratnam,
S. M. W. (2020). Mental health, substance use, and suicidal ideation during the covid-19
pandemic - united states, june 24-30, 2020. MMWR Morb Mortal Wkly Rep, 69(32), 1049–
1057. https://doi.org/10.15585/mmwr.mm6932a1
Davidson, L., & Schmutte, T. (2020). What is the meaning of recovery? (1st ed.). Palgrave Macmil-
lan Cham. https://doi.org/https://doi.org/10.1007/978-3-030-11908-9
Dixon, L. B., Holoshitz, Y ., & Nossel, I. (2016). Treatment engagement of individuals experiencing
mental illness: Review and update. World Psychiatry, 15(1), 13–20. https://doi.org/10.
1002/wps.20306
Donabedian, A. (1980). Explorations in quality assessment and monitoring: The definition of qual-
ity and approaches to its assessment. Ache Management.
Forray, A., & Yonkers, K. A. (2021). The collision of mental health, substance use disorder, and
suicide. Obstetrics Gynecology, 137(6). https://journals.lww.com/greenjournal/Fulltext/
2021/06000/The Collision of Mental Health, Substance Use.16.aspx
Frank, R. G., Koyanagi, C., & McGuire, T. G. (1997). The politics and economics of mental health
‘parity’ laws. Health Affairs, 16(4), 108–119. https://doi.org/10.1377/hlthaff.16.4.108
Frank, R. G., & McGuire, T. G. (2000). Chapter 16 economics and mental health. In Handbook of
health economics (pp. 893–954). Elsevier. https://doi.org/https://doi.org/10.1016/S1574-
0064(00)80029-3
Gangal, N. S., Hincapie, A. L., Jandarov, R., Frede, S. M., Boone, J. M., Mackinnon, N. J., Koech-
lin, K., Defiore-Hyrmer, J., Holthusen, A., & Heaton, P. C. (2020). Association between
a state law allowing pharmacists to dispense naloxone without a prescription and nalox-
one dispensing rates. JAMA Network Open, 3(1), e1920310. https : / / doi . org / 10 . 1001 /
jamanetworkopen.2019.20310
Garnick, D. W., Horgan, C. M., & Chalk, M. (2006). Performance measures for alcohol and other
drug services. Alcohol Res Health, 29(1), 19–26.
Garnick, D. W., Lee, M. T., Horgan, C. M., & Acevedo, A. (2009). Adapting washington cir-
cle performance measures for public sector substance abuse treatment systems. Journal of
Substance Abuse Treatment, 36(3), 265–277. https://doi.org/10.1016/j.jsat.2008.06.008
Geller, J. L. (2000). Excluding institutions for mental diseases from federal reimbursement for
services: Strategy or tragedy? Psychiatric Services, 51(11), 1397–1403. %3CGo%20to%
20ISI%3E://WOS:000090027700010
Gertner, A. K., Domino, M. E., & Davis, C. S. (2018). Do naloxone access laws increase outpatient
naloxone prescriptions? evidence from medicaid. Drug and Alcohol Dependence, 190, 37–
41. https://doi.org/https://doi.org/10.1016/j.drugalcdep.2018.05.014
115
Hart, C., & Ksir, C. (2022). Drugs, society, and human behavior (18th ed.) McGraw-Hill.
Hayes, N. (2023). For addiction treatment, longer is better. but insurance companies usually cut it
short. STAT news. https://www.statnews.com/2023/01/13/insurers-subvert-long-term-
addiction-treatment-success/
Kelly, S. M., O’Grady, K. E., Schwartz, R. P., Peterson, J. A., Wilson, M. E., & Brown, B. S.
(2010). The relationship of social support to treatment entry and engagement: The commu-
nity assessment inventory. Subst Abus, 31(1), 43–52. https://doi.org/10.1080/08897070903442640
Kelly, T. M., & Daley, D. C. (2013). Integrated treatment of substance use and psychiatric disor-
ders. Social Work in Public Health, 28(3-4), 388–406. https://doi.org/10.1080/19371918.
2013.774673
Kennedy, P. J. (2019). Mental health and addiction care are poorly covered by insurance networks.
STAT news. https://www.statnews.com/2019/12/10/mental-health-addiction-care-poorly-
covered-by-insurance-networks/
KFF. (2021). Medicaid waiver tracker: Approved and pending section 1115 waivers by state (Re-
port). KFF (Kaiser Family Foundation). https://www.kff.org/report-section/section-1115-
waiver-tracker-definitions/?preview id=368642
Kim, J.-R., Byun, S.-J., Son, D.-S., & Kim, H. A. (2022). Correlation between the ratio of physician
consultation fees to hourly minimum wage and consultation length: A cross-sectional study
of nine countries. BMJ Open, 12(12), e064369. https://doi.org/10.1136/bmjopen-2022-
064369
Knudsen, H. K., Hartman, J., & Walsh, S. L. (2022). The effect of medicaid expansion on state-
level utilization of buprenorphine for opioid use disorder in the united states. Drug Alcohol
Depend, 232, 109336. https://doi.org/10.1016/j.drugalcdep.2022.109336
Lee, D. S., McCrary, J., Moreira, M. J., & Porter, J. (2022). Valid ¡i¿t¡/i¿-ratio inference for iv.
American Economic Review, 112(10), 3260–3290. https://doi.org/10.1257/aer.20211063
Lenhart, O. (2017). The impact of minimum wages on population health: Evidence from 24 oecd
countries. The European Journal of Health Economics, 18(8), 1031–1039. https://doi.org/
10.1007/s10198-016-0847-5
Levit, K. R., Mark, T. L., Coffey, R. M., Frankel, S., Santora, P., Vandivort-Warren, R., & Malone,
K. (2013). Federal spending on behavioral health accelerated during recession as individu-
als lost employer insurance. Health Aff (Millwood), 32(5), 952–62. https://doi.org/10.1377/
hlthaff.2012.1065
Linas, B. P., Savinkina, A., Barbosa, C., Mueller, P. P., Cerd´ a, M., Keyes, K., & Chhatwal, J. (2021).
A clash of epidemics: Impact of the covid-19 pandemic response on opioid overdose. J
Subst Abuse Treat, 120, 108158. https://doi.org/10.1016/j.jsat.2020.108158
116
Maclean, J. C., Wen, H. F., Simon, K. I., & Saloner, B. (2021). Institutions for mental diseases
medicaid waivers: Impact on payments for substance use treatment facilities. Health Af-
fairs, 40(2), 326–333. %3CGo%20to%20ISI%3E://WOS:000614609300019
Mak, K. K., Ho, S. Y ., Thomas, G. N., Schooling, C. M., McGhee, S. M., & Lam, T. H. (2010).
Family structure, parent-child conversation time and substance use among chinese adoles-
cents. BMC Public Health, 10, 503. https://doi.org/10.1186/1471-2458-10-503
Mangrum, L. F., Spence, R. T., & Lopez, M. (2006). Integrated versus parallel treatment of co-
occurring psychiatric and substance use disorders. Journal of Substance Abuse Treatment,
30(1), 79–84. https://doi.org/10.1016/j.jsat.2005.10.004
Mark, T. L., Dilonardo, J. D., Chalk, M., & Coffey, R. (2003). Factors associated with the receipt
of treatment following detoxification. J Subst Abuse Treat, 24(4), 299–304. https://doi.org/
10.1016/s0740-5472(03)00039-4
Mark, T. L., Dowd, W. N., & Council, C. L. (2020). In Tracking the quality of addiction treatment
over time and across states: Using the federal government’s ”signs” of higher quality.
https://doi.org/10.3768/rtipress.2020.rr.0040.2007
Mark, T. L., Levit, K. R., Vandivort-Warren, R., Buck, J. A., & Coffey, R. M. (2011). Changes in us
spending on mental health and substance abuse treatment, 1986-2005, and implications for
policy. Health Aff (Millwood), 30(2), 284–92. https://doi.org/10.1377/hlthaff.2010.0765
Mark, T. L., Yee, T., Levit, K. R., Camacho-Cook, J., Cutler, E., & Carroll, C. D. (2016). Insurance
financing increased for mental health conditions but not for substance use disorders, 1986-
2014. Health Aff (Millwood), 35(6), 958–65. https://doi.org/10.1377/hlthaff.2016.0002
McCorry, F., Garnick, D. W., Bartlett, J., Cotter, F., Chalk, M., Babor, T., Falcon, S., Gastfriend,
D. R., Gelber, S., Harrison, P. A., Horgan, C., McLellan, A. T., Merrill, J., Posada, H.,
Robinson-Beale, R., & Weisner, C. (2000). Developing performance measures for alcohol
and other drug services in managed care plans. The Joint Commission Journal on Quality
Improvement, 26(11), 633–643. https://doi.org/https://doi.org/10.1016/S1070-3241(00)
26054-9
McLellan, A. T. (2002). Have we evaluated addiction treatment correctly? implications from a
chronic care perspective. Addiction, 97(3), 249–252. https://doi.org/https://doi.org/10.
1046/j.1360-0443.2002.00127.x
Meinhofer, A., & Witman, A. E. (2018). The role of health insurance on treatment for opioid use
disorders: Evidence from the affordable care act medicaid expansion. J Health Econ, 60,
177–197. https://doi.org/10.1016/j.jhealeco.2018.06.004
Miles, J. (2019). Changes in residential substance use treatment service access resulting from re-
cent medicaid section 1115 waivers. Psychiatr Serv, 70(5), 428–431. https://doi.org/10.
1176/appi.ps.201800543
117
Mitchell, A. (2019). Medicaid’s institutions for mental disease (imd) exclusion (Report). Congres-
sional Research Service. https://sgp.fas.org/crs/misc/IF10222.pdf
Mulia, N., Lui, C. K., Bensley, K. M. K., & Subbaraman, M. S. (2022). Effects of medicaid ex-
pansion on alcohol and opioid treatment admissions in u.s. racial/ethnic groups. Drug and
Alcohol Dependence, 231, 109242. https://doi.org/https://doi.org/10.1016/j.drugalcdep.
2021.109242
Musumeci, M., Chidambaram, P., & Orgera, K. (2019). State options for medicaid coverage of
inpatient behavioral health services (Report). https://www.kff.org/medicaid/report/state-
options-for-medicaid-coverage-of-inpatient-behavioral-health-services/
Naren, T., Cook, J., Armstrong, F., & Manger, S. (2022). Lifestyle interventions in the management
of substance use disorder. Australian Journal of General Practice, 51(8), 560–564. https:
//doi.org/10.31128/ajgp-02-22-6331
NationalCommitteeforQualityAssurance. (2023). Initiation and engagement of alcohol and other
drug abuse or dependence treatment (iet) (Report). National Committee for Quality Assur-
ance. https://www.ncqa.org/hedis/measures/initiation-and-engagement-of-alcohol-and-
other-drug-abuse-or-dependence-treatment/
NationalQualityForum(NQF). (2012). Behavioral health phase 1, 2012 final report (Report). Na-
tional Quality Forum. https://www.qualityforum.org/Publications/2012/12/Behavioral
Health Phase 1, 2012.aspx
NIDA. (2020). Common comorbidities with substance use disorders research report (Report).
National Institute on Drug Abuse. https://nida.nih.gov/publications/research- reports/
common-comorbidities-substance-use-disorders/introduction
O’Brien, P. L., Stewart, M. T., Shields, M. C., White, M., Dubenitz, J., Dey, J., & Mulvaney-Day,
N. (2022). Residential treatment and medication treatment for opioid use disorder: The
role of state medicaid innovations in advancing the field. Drug Alcohol Depend Rep, 4.
https://doi.org/10.1016/j.dadr.2022.100087
Olfson, M., Wall, M., Barry, C. L., Mauro, C., & Mojtabai, R. (2018). Impact of medicaid expan-
sion on coverage and treatment of low-income adults with substance use disorders. Health
Aff (Millwood), 37(8), 1208–1215. https://doi.org/10.1377/hlthaff.2018.0124
Orford, J. (2008). [commentary] joining the queue of dissenters. Addiction, 103(5), 706–707. https:
//doi.org/10.1111/j.1360-0443.2007.02128.x
Owens, P., Heslin, K., Fingar, K., & Weiss, A. (2018). Co-occurrence of physical health conditions
and mental health and substance use conditions among adult inpatient stays, 2010 versus
2014. (Report). Agency for Healthcare Research and Quality (US). https://www.ncbi.nlm.
nih.gov/books/NBK534105/
118
Pessar, S. C., Boustead, A., Ge, Y ., Smart, R., & Pacula, R. L. (2021). Assessment of state and
federal health policies for opioid use disorder treatment during the covid-19 pandemic
and beyond. JAMA Health Forum, 2(11), e213833–e213833. https : / / doi . org / 10 . 1001 /
jamahealthforum.2021.3833
Qin, X., Li, L., & Hsieh, C.-R. (2013). Too few doctors or too low wages? labor supply of health
care professionals in china. China Economic Review, 24, 150–164. https://doi.org/10.1016/
j.chieco.2012.12.002
Robinson, E., & Daly, M. (2021). Explaining the rise and fall of psychological distress during the
covid-19 crisis in the united states: Longitudinal evidence from the understanding america
study. Br J Health Psychol, 26(2), 570–587. https://doi.org/10.1111/bjhp.12493
Saloner, B., Li, W., Bandara, S. N., McGinty, E. E., & Barry, C. L. (2022). Trends in the use of
treatment for substance use disorders, 2010-19. Health Aff (Millwood), 41(5), 696–702.
https://doi.org/10.1377/hlthaff.2021.01767
Saloner, B., & Maclean, J. C. (2020). Specialty substance use disorder treatment admissions steadily
increased in the four years after medicaid expansion. Health Aff (Millwood), 39(3), 453–
461. https://doi.org/10.1377/hlthaff.2019.01428
SAMHSA. (2020). Treatment episode data set (teds). https://www.samhsa.gov/data/data- we-
collect/teds-treatment-episode-data-set
SAMHSA. (2021). Key substance use and mental health indicatorsin the united states: Results
from the 2020 national survey on drug use and health (hhs publication no.pep21-07-01-
003, nsduh series h-56) (Report). Center for Behavioral Health Statistics, Quality, Sub-
stance Abuse, and Mental Health Services Administration. https://www.samhsa.gov/data/
report/2020-nsduh-annual-national-report
SAMHSA. (2022a). Federal laws related to samhsa (Report). Substance Abuse and Mental Health
Services Administration. https://www.samhsa.gov/about-us/who-we-are/laws-regulations
SAMHSA. (2022b). Key substance use and mental health indicators in the united states: Results
from the 2021 national survey on drug use and health (Report). Center for Behavioral
Health Statistics, Quality, Substance Abuse, and Mental Health Services Administration.
https://www.samhsa.gov/data/report/2021-nsduh-annual-national-report
SAMHSA. (2022c). Treatment episode data set (teds): 2020. admissions to and discharges from
publicly funded substance use treatment facilities. (Report). Center for Behavioral Health
Statistics, Quality, Substance Abuse, and Mental Health Services Administration.
SAMHSA. (2023). Co-occurring disorders and other health conditions (Report). Substance Abuse
and Mental Health Services Administration. https : / / www . samhsa . gov / medications -
substance - use - disorders / medications - counseling - related - conditions / co - occurring -
119
disorders# :
∼ : text = The % 20coexistence % 20of % 20both % 20a , affected % 20by % 20a %
20mental%20illness.
Soto-Nevarez, A., Stevens, E., & Jason, L. A. (2021). Housing, relationships, and substance use
among formerly incarcerated females. Alcohol Treat Q, 39(2), 238–250. https://doi.org/10.
1080/07347324.2020.1862007
Stein, B., Orlando, M., & Sturm, R. (2000). The effect of copayments on drug and alcohol treat-
ment following inpatient detoxification under managed care. Psychiatr Serv, 51(2), 195–8.
https://doi.org/10.1176/appi.ps.51.2.195
Stein, B. D., Kogan, J. N., & Sorbero, M. (2009). Substance abuse detoxification and residential
treatment among medicaid-enrolled adults: Rates and duration of subsequent treatment.
Drug and Alcohol Dependence, 104(1-2), 100–106. https://doi.org/10.1016/j.drugalcdep.
2009.04.008
Stephenson, J. (2021). Cdc warns of surge in drug overdose deaths during covid-19. JAMA Health
Forum, 2(1), e210001–e210001. https://doi.org/10.1001/jamahealthforum.2021.0001
Stock, M., James H.; Yogo. (2005). Testing for weak instruments in linear iv regression. In D. W.
Andrews & J. H. Stock (Eds.), Identification and inference in econometric models: Essays
in honor of thomas j. rothenberg (pp. 80–108). Cambridge University Press.
Terza, J. V ., Basu, A., & Rathouz, P. J. (2008). Two-stage residual inclusion estimation: Addressing
endogeneity in health econometric modeling. Journal of Health Economics, 27(3), 531–
543. https://doi.org/10.1016/j.jhealeco.2007.09.009
TreatmentAdvocacyCenter. (2016). The medicaid imd exclusion and mental illness discrimination
(Report). Treatment Advocacy Center. https://www.treatmentadvocacycenter.org/storage/
documents/backgrounders/imd-exclusion-and-discrimination.pdf
Wang, J. Z., Mott, S., Magwood, O., Mathew, C., McLellan, A., Kpade, V ., Gaba, P., Kozloff, N.,
Pottie, K., & Andermann, A. (2019). The impact of interventions for youth experiencing
homelessness on housing, mental health, substance use, and family cohesion: A systematic
review. BMC Public Health, 19(1), 1528. https://doi.org/10.1186/s12889-019-7856-0
Watkins, K. E., Farmer, C. M., De Vries, D., & Hepner, K. A. (2015). Without quality measures,
increasing access to substance use treatment may not improve patient outcomes. RAND
Corporation. https://www.rand.org/pubs/research briefs/RB9816.html
World Health Organization Regional Office for Africa. (n.d.). Substance abuse [Accessed: [07/13/2023]].
Xu, J., Davis, C. S., Cruz, M., & Lurie, P. (2018). State naloxone access laws are associated with
an increase in the number of naloxone prescriptions dispensed in retail pharmacies. Drug
and Alcohol Dependence, 189, 37–41. https://doi.org/https://doi.org/10.1016/j.drugalcdep.
2018.04.020
120
Ziedonis, D. M. (2004). Integrated treatment of co-occurring mental illness and addiction: Clini-
cal intervention, program, and system perspectives. CNS Spectrums, 9(12), 892–904, 925.
https://doi.org/10.1017/s1092852900009718
121
Appendices
Table A.1: Optum File Tables
Table Types Diagnosis Table Medical Claims Table Enrollment Table
patid (patient ID number) patid (patient ID number) patid (patient ID number)
fst dt (first service date) fst dt (first service date) state
Variables diag (diagnosis code) pos (place of service) yrdob (year of birth)
proc cd (procedure codes: CPT/HCPCS) gdr cd (gender)
race
122
Table A.2: ICD Codes for Substance Use Disorders
ICD-9 ICD-10
Code Description Code Description
291 Alcohol-induced mental disorders F10 Alcohol related disorders
292 Drug-induced psychotic disorders F11 Opioid related disorders
303 Alcohol dependence syndrome F12 Cannabis related disorders
304 Drug dependence F13 Sedative, hypnotic, or anxiolytic related disorders
305 Nondependent abuse of drugs F14 Cocaine related disorders
967 Poisoning by sedatives and hypnotics F15 Other stimulant related disorders
968 Poisoning by other central nervous system
depressants and anesthetics F16 Hallucinogen related disorders
969 Poisoning by psychotropic agents F17 Nicotine dependence
970 Poisoning by central nervous system stimulants F18 Inhalant related disorders
E850 Accidental poisoning by
analgesics antipyretics and antirheumatics F19 Other psychoactive substance related disorders
E935 Analgesics antipyretics and antirheumatics T39 Poisoning by, adverse effect of and underdosing
causing adverse effects in therapeutic use of nonopioid analgesics, antipyretics and antirheumatics
E940 Central nervous system stimulants T40 Poisoning by, adverse effect of and underdosing
causing adverse effects in therapeutic use of narcotics and psychodysleptics [hallucinogens]
T42 Poisoning by, adverse effect of and underdosing
of antiepileptic, sedative- hypnotic and antiparkinsonism drugs
T43 Poisoning by, adverse effect of and underdosing
of psychotropic drugs, not elsewhere classified
123
Table A.3: ICD Codes for Mental Health Problems
ICD-9 ICD-10
Code Description Code Description
293 Transient mental disorders due to conditions classified elsewhere F20 Schizophrenia
294 Persistent mental disorders due to conditions classified elsewhere F21 Schizotypal disorder
295 Schizophrenic disorders F22 Delusional disorders
296 Episodic mood disorders F23 Brief psychotic disorder
297 Delusional disorders F24 Shared psychotic disorder
298 Other nonorganic psychoses F25 Schizoaffective disorders
300 Anxiety, dissociative and somatoform disorders F28 Other psychotic disorder not due to
a substance or known physiological condition
301 Personality disorders F29 Unspecified psychosis not due to
a substance or known physiological condition
302 Sexual and gender identity disorders F30 Manic episode
308 Acute reaction to stress F31 Bipolar disorder
309 Adjustment reaction F32 Depressive episode
311 Depressive disorder, not elsewhere classified F33 Major depressive disorder, recurrent
F34 Persistent mood [affective] disorders
F39 Unspecified mood [affective] disorder
F40 Phobic anxiety disorders
F41 Other anxiety disorders
F42 Obsessive-compulsive disorder
F43 Reaction to severe stress, and adjustment disorders
F44 Dissociative and conversion disorders
F45 Somatoform disorders
F48 Other nonpsychotic mental disorders
F50 Eating disorders
F51 Sleep disorders not due to
a substance or known physiological condition
F52 Sexual dysfunction not due to
a substance or known physiological condition
F53 Mental and behavioral disorders associated with
the puerperium, not elsewhere classified
F54 Psychological and behavioral factors associated with
disorders or diseases classified elsewhere
F55 Abuse of non-psychoactive substances
F59 Unspecified behavioral syndromes associated with
physiological disturbances and physical factors
F60 Specific personality disorders
F63 Impulse disorders
F64 Gender identity disorders
F65 Paraphilias
F66 Other sexual disorders
F68 Other disorders of adult personality and behavior
F69 Unspecified disorder of adult personality and behavior
124
Table A.4: HCPCS/CPT Codes for Substance Use Disorders Treatment at Outpatient Level
Code Description
HCPCS Codes
G2086 Office-based treatment for opioid use disorder, including development of the treatment plan, care coordination,
individual therapy and group therapy and counseling; at least 70 minutes in the first calendar month
G2087 Office-based treatment for opioid use disorder, including care coordination,
individual therapy and group therapy and counseling; at least 60 minutes in a subsequent calendar month
G2088 Office-based treatment for opioid use disorder, including care coordination,
individual therapy and group therapy and counseling; each additional 30 minutes beyond the first 120 minutes
H0005 Alcohol and/or drug services; group counseling by a clinician
H0022 Alcohol and/or drug intervention service (planned facilitation)
H0047 Alcohol and/or other drug abuse services, not otherwise specified
T0016 Alcohol and/or substance abuse services, family/couple counseling
CPT Codes
90791 Psychiatric diagnostic evaluation
90792 Psychiatric diagnostic evaluation
90832 Psychotherapy, 30 minutes with patient and/or family member
90834 Psychotherapy, 45 minutes with patient and/or family member
90837 Psychotherapy, 60 minutes with patient and/or family member
125
Table A.5: HCPCS/CPT Codes for Substance Use Disorders Treatment at Inpatient Level
Code Description
HCPCS Codes
H0017 OBehavioral health; residential (hospital residential treatment program),
without room and board, per diem
H0018 Behavioral health; short-term residential (non-hospital residential
treatment program), without room and board, per diem
H0019 Behavioral health; long-term residential (non-medical, non-acute care in a residential treatment program
where stay is typically longer than 30 days), without room and board, per diem
T2048 Behavioral health; long-term care residential (non-acute care in a residential treatment program
where stay is typically longer than 30 days), with room and board, per diem
Table A.6: HCPCS/CPT Codes for Substance Use Disorders Treatment at Intensive Outpatient or
Partial Hospitalization Level
Code Description
HCPCS Codes
G0410 Group psychotherapy other than of a multiple-family group,
in a partial hospitalization setting, approximately 45 to 50 minutes
G0411 Interactive group psychotherapy, in a partial hospitalization setting,
approximately 45 to 50 minutes
H0015 Alcohol and/or drug services; intensive outpatient (treatment program that operates
at least 3 hours/day and at least 3 days/week and is based on an individualized treatment plan),
including assessment, counseling; crisis intervention, and activity therapies or education
H0035 Mental health partial hospitalization, treatment, less than 24 hours
H2012 Behavioral health day treatment, per hour
S9480 Intensive outpatient psychiatric services, per diem
126
Table A.7: HCPCS/CPT Codes for Medication Assisted Treatment
Code Description
HCPCS Codes
G2067 Medication assisted treatment, methadone; weekly bundle including dispensing
and/or administration, substance use counseling, individual and group therapy, and
toxicology testing, if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2068 Medication assisted treatment, buprenorphine (oral); weekly bundle including
dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2069 Medication assisted treatment, buprenorphine (injectable); weekly bundle including
dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2070 Medication assisted treatment, buprenorphine (implant insertion); weekly bundle
including dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2071 Medication assisted treatment, buprenorphine (implant removal); weekly bundle
including dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2072 Medication assisted treatment, buprenorphine (implant insertion and removal); weekly bundle
including dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2073 Medication assisted treatment, naltrexone; weekly bundle including dispensing
and/or administration, substance use counseling, individual and group therapy, and
toxicology testing if performed (provision of the services by a medicare-enrolled opioid treatment program)
G2074 Medication assisted treatment, weekly bundle not including the drug, including
substance use counseling, individual and group therapy, and toxicology testing if
performed (provision of the services by a medicare-enrolled opioid treatment program)
G2075 Medication assisted treatment, medication not otherwise specified; weekly bundle
including dispensing and/or administration, substance use counseling, individual and group therapy,
and toxicology testing, if performed (provision of the services by a medicare-enrolled opioid treatment program)
H0020 Alcohol and/or drug services; methadone administration and/or service (provision of the drug by a licensed program)
J0570 Buprenorphine implant (Probuphine), 74.2 mg buprenorphine
J0571 Buprenorphine, oral, 1 mg
J0572 Buprenorphine/naloxone, oral, less than or equal to 3 mg buprenorphine
J0573 Buprenorphine/naloxone, oral, greater than 3 mg, but less than or equal to 6 mg buprenorphine
J0574 Buprenorphine/naloxone, oral, greater than 6 mg, but less than or equal to 10 mg buprenorphine
J0575 Buprenorphine/naloxone, oral, greater than 10 mg buprenorphine
J0592 Injection, buprenorphine hydrochloride, 0.1 mg
S0109 Methadone, oral, 5 mg
J2310 Naloxone hydrochloride, injection, per 1 mg
J2315 Injection, naltrexone, depot from, 1mg
Q9991 Injection, buprenorphine extended-release (Sublocade), less than or equal to 100 mg
Q9992 Injection, buprenorphine extended-release (Sublocade), greater than 100 mg
127

Abstract (if available)

Abstract State and federal governments have enacted numerous policies to increase access to evidence-based substance abuse and mental health treatment in the United States, including Medicaid payment exclusion waivers as part of Section 1115 Substance Use Disorder (SUD) demonstrations. Using data from the annual versions of the National Directory of Substance Abuse Treatment Facilities, the second chapter examines changes in the number of specialty substance abuse treatment facilities and the likelihood of those substance abuse treatment facilities offering treatment for co-occurring disorders in mental health. The third chapter examines the spillover effect of Medicaid IMD exclusion waivers on the access to different types of treatment settings of those commercially insured SUD patients. The fourth chapter identifies important predictors of quality measures and examines the association of those payment waivers with the quality of SUD treatment received by commercial insurance between 2010-2021.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Homelessness and substance use treatment: using multiple methods to understand risks, consequences, and unmet treatment needs among young adults

PDF

Crohn’s disease: health outcomes and resource utilization in the biologic era

PDF

Trauma-related treatment gains among women with histories of interpersonal violence and co-occurring mental health and substance abuse disorders

PDF

The impact of childhood trauma on substance use and mental health during the SARS-CoV-2 pandemic among young adults

PDF

The interplay between social connection and substance use

PDF

Topics in U.S. healthcare

PDF

Integrative care strategies for older adults experiencing co-occurring substance use and mental health disorders (I-CARE)…

PDF

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

PDF

Essays on development economics and adolescent behavior

PDF

Essays on treatment effect and policy learning

PDF

Essays in opioid use and abuse

PDF

Three essays on health, aging, and retirement

PDF

Essays examining nutrition behavior and policy in California

PDF

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

PDF

Interagency collaboration: cultivating resources and strengthening substance use service delivery in child welfare

PDF

Essays on work, retirement, and fostering longer working lives

PDF

Global health diplomacy: a new era of health in U.S. foreign policy

PDF

Climate change, air pollution and health: how policy can induce adaptation

PDF

Health care utilization and spending of the U.S. aging population

PDF

Three essays on the evaluation of long-term care insurance policies

Asset Metadata

Creator Ge, Yimin (author)

Core Title Essays on the U.S. market for substance use treatment and the impact of Medicaid policy reform

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Degree Conferral Date 2023-08

Publication Date 08/04/2023

Defense Date 07/17/2023

Publisher University of Southern California. Libraries (digital)

Tag health insurance,Health policy,Medicaid,OAI-PMH Harvest,substance abuse treatment

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Pacula, Rosalie Liccardo (committee chair), Doctor, Jason (committee member), Seabury, Seth A. (committee member), Romley, John A. (committee member)

Creator Email amyyiminge@gmail.com,yiminge@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113296397

Unique identifier UC113296397

Identifier etd-GeYimin-12195.pdf (filename)

Legacy Identifier etd-GeYimin-12195

Document Type Dissertation

Rights Ge, Yimin

Internet Media Type application/pdf

Type texts

Source 20230807-usctheses-batch-1079 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA

Repository Email cisadmin@lib.usc.edu