The Impacts of 1115 Medicaid Substance Use Disorder Waivers on Medicaid-Paid Use of Residential Treatment and Other Types of Services in 20 States

Abstract

Objective:

To assess the association between the implementation of 1115 Medicaid substance use disorder (SUD) waivers and changes in Medicaid-paid use of residential treatment and other types of services.

Study Setting and Design:

We compared 20 states with SUD waivers to 14 non-waiver states using a staggered difference-in-differences design. Primary outcomes were Medicaid-paid opioid-use disorder (OUD) related residential treatment stays and length of stay (LOS). Secondary outcomes included admissions and LOS for all-cause and OUD-related inpatient stays, psychiatric hospital admissions, emergency department (ED) visits, outpatient visits, and primary care visits.

Data Source and Analytic Sample:

We used the 2016–2021 Transformed Medicaid Statistical Information System (T-MSIS) Analytic Files (TAF). The analytic sample included Medicaid enrollees ages 18–64 with OUD.

Principal Findings:

On average, waiver implementation was associated with an increase in residential treatment stays (estimate: 0.4%; 95% CI: 0.1%–0.7%), OUD-related inpatient visits LOS (estimate: 0.3 days; 95% CI: 0.0%–0.5%), psychiatric hospital LOS (estimate: 1.0 days; 95% CI: 0.6 days–1.4 days), primary care visits (estimate: 3.0%; 95% CI: 1.2%–4.7%), and OUD-related primary care visits (estimate: 2.7%; 95% CI: 0.9%–4.4%); and a decline in all-cause inpatient visits (estimate: −0.9%; 95% CI: −1.9% to −0.0%) and OUD-related inpatient visits (estimate: −0.8%; 95% CI: −1.6% to −0.0%). Results for psychiatric hospital LOS and OUD-related primary care visits were sensitive to adjusting for pre-trends.

Among four early-adopting states (Indiana, Louisiana, New Jersey, Virginia), Medicaid-paid residential treatment increased 1–4 years following waiver implementation (e.g., 2-year estimate: 2.8%, 95% CI: 2.5%–3.0%), and inpatient visits declined 1–4 years following waiver implementation (e.g., 2-year estimate: −3.1%, 95% CI: −3.5% to −2.6%).

Conclusions:

SUD waivers were associated with a small increase in Medicaid-paid residential treatment and a decline in inpatient visits across states, with changes being concentrated among early-adopting states.

1 |. Introduction

Since the inception of the Medicaid program in 1965, federal policy has prohibited the use of federal funds for individuals ages 21–64 years who reside in so-called Institutions for Mental Disease (IMDs), defined as residential substance use disorder (SUD) and mental health treatment facilities with more than 16 beds. This “IMD exclusion” reflected states’ historical responsibility to finance psychiatric institutions and an emphasis on community-based care [1]. However, the IMD exclusion may be an important reason why many states have not covered SUD treatment in residential care settings [2]. More recently, such coverage gaps have been viewed as an impediment to an effective response to the opioid crisis [3].

In 2015, the Centers for Medicare & Medicaid Services (CMS) issued guidance that allowed states to apply for 1115 Medicaid SUD waivers and receive federal matching funds for SUD services provided in IMDs [4, 5]. A primary goal of the SUD waivers was to ensure access to the full continuum of SUD care in outpatient, residential, and inpatient care settings. Because coverage gaps primarily existed for residential care, waiver-related coverage changes were concentrated in this setting. A related goal was to improve access to lower levels of care and reduce care in emergency department (ED) and inpatient hospital settings. As of January 2025, 36 states and the District of Columbia had an approved SUD waiver [6].

Studies on SUD waivers and residential treatment thus far have mostly focused on changes in services provided by SUD treatment facilities, using data from the National Survey of Substance Abuse Treatment Services. For instance, a study on nine early SUD waiver states found a positive association between waiver implementation and the probability that residential treatment facilities accepted Medicaid [7]. Cunningham et al. focused on two early-waiver states, Maryland and Virginia, and reported similar findings [8]. SUD waiver adoption was also associated with an increase in the likelihood that residential treatment facilities offered co-occurring mental health and substance use treatment [9]. One study on SUD waivers used hospital data and found that waivers were associated with the existence of plans to implement opioid-related programs such as risk education in unadjusted analysis, but not when adjusting for hospital and geographic factors [10]. Taken together, these studies suggest a positive supply response to waivers by SUD treatment facilities; however, evidence on how such supply responses affected Medicaid members is currently lacking.

An earlier pilot demonstration may also provide evidence of plausible effects of SUD waivers. Specifically, CMS’s Medicaid Emergency Psychiatric Demonstration (2012–2015) provided federal Medicaid payments to 28 private IMDs in 11 states and the District of Columbia for the treatment of psychiatric emergencies among Medicaid enrollees ages 21–64. An evaluation of the demonstration using Medicaid and Medicare administrative records and a pre-post design did not find an association between the implementation of the demonstration and inpatient IMD admissions, IMD length of stay, emergency room visits, or emergency room boarding time [11].

The goal of this study was to examine the association between SUD waiver adoption in 20 states that implemented their SUD waiver between 2017 and 2021 and changes in the use of Medicaid-paid residential treatment and other services across the continuum of care. We focused on the opioid-use disorder (OUD) population as the primary target of the SUD waiver. To our knowledge, this is the first study to comprehensively assess changes in the use of residential treatment and other types of services among Medicaid members.

We expected that the requirement to cover all levels of SUD care and removal of the IMD exclusion would increase the use of Medicaid-paid residential treatment and length of stay. However, federal funds generally did not cover residential costs associated with room and board, which may dampen waiver effects [12]. Further, the waiver requires state-wide average lengths of stay for residential treatment to be 30 days, which could have reduced lengths of stay in this setting. Finally, we hypothesized that SUD waivers would decrease ED visits, inpatient stays, and lengths of inpatient stays while increasing outpatient and primary care visits, reflecting a shift from higher to lower levels of care.

2 |. Methods

2.1 |. Study Design and Setting

We used a difference-in-differences design to compare changes in outcomes among waiver states to states without waivers. The data source was the Transformed Medicaid Statistical Information System (T-MSIS) Analytic Files (TAF), a CMS dataset that starts with the year 2016. TAF data included enrollment and claims information of Medicaid beneficiaries enrolled in either fee-for-service or managed care plans from all states [13]. TAF is the successor of the Medicaid Analytic eXtract (MAX) file and is more comprehensive in terms of populations included (e.g., MAX did not include all Medicaid beneficiaries enrolled in managed care) and information provided (e.g., provider information).

We included 20 states (Alaska, Colorado, Delaware, Illinois, Indiana, Kansas, Kentucky, Louisiana, Michigan, Minnesota, Nebraska, New Hampshire, New Jersey, New Mexico, North Carolina, Oklahoma, Pennsylvania, Virginia, Washington, Wisconsin) that implemented their waivers between 2017 and 2021 (see Supporting Informations S1: Appendix section A1, Table A1 for details regarding implementation dates of these states). We used the demonstration start date as specified in CMS’s Special Terms and Conditions for a state’s waiver demonstration as the implementation date. Our comparison group included 14 states that did not adopt a waiver during the study period (Alabama, Arizona, Arkansas, Connecticut, Georgia, Hawaii, Iowa, Missouri, Nevada, New York, South Carolina, South Dakota, Texas, Wyoming).

We excluded five states (California, Maryland, Massachusetts, Oregon, Vermont) and the District of Columbia with waiver implementations prior to 2017 or in 2021 because of insufficient pre- or post-implementation quarters (which we specified as fewer than four quarters). We also excluded 11 states due to data quality concerns: six (Idaho, Maine, Ohio, Rhode Island, Utah, West Virginia) with waivers and five (Florida, Mississippi, Montana, North Dakota, Tennessee) without waivers. We used CMS’s Data Quality (DQ) Atlas [14] and a review of outcome trends for each state to assess data quality. We did not base our exclusion of states solely on the DQ Atlas because of evidence that its classification of data fields as high concern or unusable may not always be relevant for specific outcome measures [15]. Supporting Informations S1: Appendix section A-1, Table A-2 included details regarding exclusions of states.

2.2 |. Study Sample

The study sample consisted of adult Medicaid enrollees 18–64 years of age with OUD diagnoses, identified using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes F11.XXX. For each quarter of the study period, we included Medicaid enrollees with an OUD diagnosis during the current or previous quarter. We restricted our study sample to Medicaid beneficiaries who were fully enrolled during the current and previous quarters to ensure complete claims records for identifying OUD diagnoses. We also excluded Medicaid enrollees who were dually enrolled in Medicaid and Medicare (because we did not have Medicare claims records), resided in more than one state during a quarter, or had limited Medicaid benefits (see Supporting Informations S1: Appendix section A-1, Figure A-1.3 for a CONSORT diagram).

2.3 |. Variables

We examined two primary outcomes: (i) a binary indicator of OUD-related residential stay paid by Medicaid during the current calendar quarter and (ii) days of residential treatment paid by Medicaid among those in residential treatment. We used revenue, procedure, and place of service codes as described in prior studies to identify residential treatment in Medicaid claims [16–18]. We focused on OUD-related residential stays, defined as stays with an ICD-10 OUD code (F11.XXX) in any diagnostic field.

Secondary outcomes included all-cause and OUD-related ED visits; all-cause and OUD-related inpatient stays; lengths of all-cause and OUD-related inpatient stays; psychiatric hospital admissions and lengths of stays (LOS); outpatient specialty visits; and all-cause and OUD-related primary care visits. We used procedure, revenue, hospital type, and place of service codes in TAF Inpatient and Other Services files to identify ED visits, inpatient stays, psychiatric hospital admissions, and outpatient specialty visits. We defined psychiatric hospital admissions as admissions to an inpatient psychiatric hospital or IMD. These visits excluded admissions to general hospitals providing psychiatric care. Primary care visits were identified using procedure codes 99201–99205, 99211–99215, 99241–99245, 99271–99275, 99341–99345, 99347–99350, 99381–99387, 99391–99397, 99401–99404, 99411–99412, 99420, 99429, 99455–99456, 99499, G0402, G0438, G0439, G0463, and T1015 and provider taxonomy codes from servicing providers identified on claims. Procedure codes identify services typically provided in a primary care setting, such as office visits (e.g., code 99211) or consultation involving evaluation and management (e.g., code 99245) and were based on a previous study [17]. Outcomes were either expressed as binary indicators or in days. See Supporting Informations S1: Appendix section A-2 for details regarding definitions of outcomes.

Study covariates included age (18–24, 25–34, 35–44, 45–54, and 55–64; reference group: 18–24), sex (female, male; reference group: female), eligibility status (disabled adults, non-disabled adults, expansion adults, pregnant women, youth; reference group: disabled adults), presence of other substance use disorders (e.g., alcohol use disorder; reference group: no other substance use disorder), and presence of mental health conditions (reference group: no mental health conditions). Details regarding covariates are included in Supporting Informations S1: Appendix section A-3.

Approximately 0.7% of eligibility status information was missing; we used a combination of infilling from other quarters and imputation to replace missing records with eligibility categories. We used ICD-10 codes F10, F12-F16, F18, F19, F55, O355, O9931, and O9932 to identify other substance use disorders. ICD-10 codes for mental health conditions included F20-F29 (schizophrenia, schizotypal, delusional, and other nonmood psychotic disorders); F30-F34 and F39 (mood disorders); F40 (phobic anxiety disorders); F41 (other anxiety disorders); F42 (obsessive-compulsive disorder); F431 (post-traumatic stress disorder); F50 (eating disorders); F60 (specific personality disorders); and F9091 (attention-deficit hyperactivity disorder, unspecified type). We did not include race or ethnicity in regression models due to a high percentage of missing information in some states and years, as documented by the DQ Atlas.

As policy variables, we included an indicator for six study states that had participated in CMS’s earlier IMD demonstration (Alabama, Connecticut, Illinois, Missouri, North Carolina, and Washington). Eligibility status controlled for changes to the population through the Medicaid expansion, which occurred in five states during the study period (Louisiana, Virginia, Nebraska, Oklahoma, and Missouri). We also included indicators for time relative to methadone coverage changes in states that started covering methadone during the study period (reference group: last quarter prior to a methadone coverage change). Regression models with methadone coverage indicators did not converge for models with length of stay as outcome, and we therefore presented regression results with this covariate for binary outcomes only. Regarding buprenorphine prescribing, reliable information on the number of X-waivered providers did not exist, but prior research suggested similar changes in buprenorphine prescribing among waiver and non-waiver states [14].

2.4 |. Statistical Analysis

Our analysis was conducted at the enrollee-quarter level. All analyses started with the second quarter of 2016 (henceforth “baseline”) due to the one-quarter look-back period for identifying OUD. We used a staggered difference-in-differences design that accounted for arbitrary heterogeneity across waiver cohorts (states whose waivers began during the same quarter) and time relative to the waiver’s initiation [19]. We used linear regression models that included calendar quarter fixed effects (with the baseline quarter being excluded as reference quarter), indicators for each waiver cohort, and interaction terms between cohort indicators and time indicators (where we omitted the interaction term for the last quarter prior to waiver implementation, which thus constituted the reference period for difference-in-differences estimates). Models also included binary indicators, or indicators for each category, of covariates (except the respective reference group). We clustered standard errors at the state level.

We grouped waiver states into four cohorts based on the number of fully observed years following waiver adoption observed in the data: early waiver states (IN, LA, NJ, VA) had 4 years of observation; intermediate-early waiver states (AK, IL, KS, NH, NM, PA, WA, WI) had 3 years; intermediate-late waiver states (DE, KY, MI, MN, NC, NE) had 2 years; and late waiver states (OK, CO) had 1 year.

Following model estimation, we averaged estimates across cohorts to obtain difference-in-differences estimates by year following waiver implementation. We used enrollee-quarter weights to calculate population-weighted averages.

We tested whether outcome trends moved in parallel between waiver and non-waiver states prior to waiver implementation using the average difference-in-differences estimate for the pre-intervention period, which was zero under the null hypothesis of parallel trends, positive if outcomes declined more strongly or increased more slowly among waiver states compared to non-waiver states prior to waiver beginning, and negative otherwise. As a sensitivity analysis, we re-estimated the difference-in-differences models using trend-adjusted outcomes, where trend adjustment linearly increased or decreased the outcome slope in waiver states so that average outcomes between waiver and non-waiver states moved in parallel prior to waiver implementation. Supporting Informations S1: Appendix section A-4 includes details on regression specifications, calculation of average estimates, and trend adjustment. Analyses were completed using R version 4.3.3. We developed all code for estimating regression models. The Oregon Health & Science University Institutional Review Board (IRB) reviewed the study protocol and determined it was exempt.

3 |. Results

Our study included 1,234,324 Medicaid enrollees with OUD in waiver states and 561,571 enrollees with OUD in non-waiver states (Table 1). Enrollees in waiver states were more likely to be eligible due to the Medicaid expansion and to live in the Midwest or South, with a standardized difference exceeding 0.10 for these categories [20]. They were less likely to be 55 years or older, live in the Northeast or West, be eligible for Medicaid due to other categories, and have other substance use disorders or mental health conditions.

TABLE 1 |.

Characteristics of medicaid enrollees with OUD in states with and without SUD-IMD waiver, 2016–2021.

	Non-waiver states	Waiver states	Std diff
Number
Quarter observations	3,601,970	7,808,544
Individuals	561,571	1,234,324
Age (%)
18–24	212,585 (5.9%)	444,753 (6.0%)	0.00
25–34	1,173,220 (32.6%)	2,759,767 (37.0%)	0.09
35–44	946,667 (26.3%)	2,267,762 (30.4%)	0.09
45–54	733,025 (20.4%)	1,269,855 (17.0%)	0.09
55–65	536,473 (14.9%)	715,785 (9.6%)	0.16
Sex (%)
Male	1,947,227 (54.1%)	3,598,855 (48.3%)	0.12
Female	1,654,743 (45.9%)	3,859,067 (51.7%)	0.12
Eligibility group (%)
Disabled adults	738,865 (20.5%)	1,097,668 (14.7%)	0.15
Expansion adults	644,963 (17.9%)	4,377,540 (58.7%)	0.92
Other adults	2,135,689 (59.3%)	1,848,409 (24.8%)	0.75
Pregnant women	54,127 (1.5%)	89,364 (1.2%)	0.03
Youth	28,009 (0.8%)	36,937 (0.5%)	0.04
Unknown	317 (0.0%)	8004 (0.1%)	0.04
Region (%)
Midwest	178,343 (5.0%)	2,081,786 (27.9%)	0.65
Northeast	2,265,246 (62.9%)	2,119,265 (28.4%)	0.74
South	466,452 (12.9%)	2,175,223 (29.2%)	0.41
West	691,929 (19.2%)	1,081,648 (14.5%)	0.13
Comorbidities (%)
Other substance use disorders	1,167,497 (32.4%)	2,024,321 (27.1%)	0.12
Mental health conditions	1,392,776 (38.7%)	2,475,008 (33.2%)	0.11

Note: The table displays number and percentage values of patient characteristics residing in waiver and non-waiver states, and the standardized difference (absolute difference in percentage values relative to the standard deviation).

Abbreviations: IMD, Institution of Mental Disease; OUD, opioid use disorder; Std diff, standardized difference; SUD, substance use disorder.

Source: 2016–2021 transformed medicaid statistical information system analytic files.

At baseline, 3.9% of Medicaid enrollees with OUD residing in waiver states had an OUD-related residential treatment stay paid by Medicaid, compared to 0.7% of Medicaid enrollees residing in non-waiver states (Table 2). Waiver implementation was associated with a statistically significant increase in OUD-related residential treatment paid by Medicaid across all states (estimate: 0.4%; 95% CI: 0.1%–0.7%), but no significant change in residential treatment LOS (estimate: −1.0 days; 95% CI: −2.7 days to 0.8 days).

TABLE 2 |.

Association between SUD-IMD waivers implementation and changes in residential treatment, emergency department visits, inpatient visits, outpatient specialty care, and primary care visits.

	Non-waiver states, 2016 Q2	Waiver states, 2016 Q2	Difference in differences (covariate adjusted), estimate (95% CI)	Difference in differences (covariate and trend adjusted), estimate (95% CI)	Pre-waiver average
OUD-related residential treatment stays, N (%)	888 (0.7%)	8905 (3.9%)	0.4* (0.1, 0.7)	0.5** (0.2, 0.8)	0.1 (−0.3, 0.4)
OUD-related residential treatment LOS (mean [SD])	13.6 (9.3) days	9.8 (8.5) days	−1.0 (−2.7, 0.8)	−2.9** (−4.6, −1.1)	−1.2 (−2.8, 0.4)
ED visits, N (%)	47,465 (35.0%)	79,999 (35.3%)	−0.4 (−1.8, 0.9)	0.7 (−0.7, 2.0)	0.8 (0.0, 1.6)
OUD-related ED visits, N (%)	14,009 (10.3%)	18,904 (8.3%)	−0.4 (−1.3, 0.5)	0.2 (−0.7, 1.1)	0.4 (−0.1, 0.9)
Inpatient visits, N (%)	19,847 (14.6%)	25,415 (11.2%)	−0.9* (−1.9, −0.0)	−0.3 (−0.6, 1.2)	0.9 (−0.1, 1.8)
Inpatient visits LOS (mean [SD])	6.5 (1.3) days	6.7 (2.1) days	0.1 (0.0, 0.3)	0.1 (−0.1, 0.3)	0.0 (−0.2, 0.1)
OUD-related inpatient visits, N (%)	12,611 (9.3%)	16,038 (7.1%)	−0.8* (−1.6, 0.0)	0.3 (−0.5, 1.1)	0.8* (0.1, 1.5)
OUD-related inpatient visits LOS (mean [SD])	6.3 (1.3) days	6.4 (1.1) days	0.3* (0.0, 0.5)	0.0 (−0.2, 0.3)	−0.2 (−0.4, 0.1)
Psychiatric inpatient specialty hospital, N (%)	1378 (1.0%)	2587 (1.1%)	0.1 (−0.3, 0.5)	−0.6** (−1.0, −0.2)	−0.5 (−1.3, 0.3)
Psychiatric inpatient specialty LOS (mean [SD])	17.0 (15.7) days	8.8 (3.5) days	1.0*** (0.6, 1.4)	−0.6** (−1.0, −0.2)	−0.9** (−1.4, −0.4)
Outpatient specialty care visits, N (%)	24,585 (18.1%)	55,746 (24.6%)	0.4 (−1.6, 2.5)	−4.8*** (−6.9, −2.8)	−3.7** (−6.3, −1.1)
Primary care visits, N (%)	58,797 (43.3%)	95,421 (42.1%)	3.0** (1.2, 4.7)	0.3 (−1.4, 2.1)	−1.8 (−3.9, 0.3)
OUD-related primary care visits, N (%)	13,642 (10.1%)	23,646 (10.4%)	2.7** (0.9, 4.4)	−2.0* (−3.8, −0.3)	−3.3* (−6.1, −0.5)

Note: The table shows percentages (binary outcomes) or mean values (length of stay outcomes) at baseline (second quarter, 2016); difference-in-differences estimates; difference-in-differences estimates when adjusting for differential trends between waiver and non-waiver states prior to waiver implementation; and the average estimate of difference-in-differences estimates during the pre-waiver implementation period. Regression models included covariates as described in the manuscript. Methadone coverage status indicators were only included in regression models of binary outcomes, because regression models of length of stay outcomes did not converge when this covariate was added. Standard errors were clustered at the state level.

Abbreviations: 2016 Q2, 2016, second quarter; CI, confidence interval; ED, emergency department; IMD, Institution of Mental Disease; LOS, length of stay; N, number; OUD, opioid use disorder; SD, standard deviation; Std diff, standardized difference; SUD, substance use disorder.

Source: 2016–2021 transformed medicaid statistical information system analytic files.

^*p < 0.05.

^**p < 0.01.

^***p < 0.001.

Regarding secondary outcomes, waiver implementation was negatively associated with all-cause inpatient visits (estimate: −0.9%; 95% CI: −1.9% to −0.0%) and OUD-related inpatient visits (estimate: −0.8%; 95% CI: −1.6% to −0.0%). Several secondary outcomes were positively associated with waiver implementation: OUD-related inpatient visits LOS (estimate: 0.3 days; 95% CI: 0.0%–0.5%), psychiatric hospital LOS (estimate: 1.0 days; 95% CI: 0.6 days to 1.4 days), primary care visits (estimate: 3.0%; 95% CI: 1.2%–4.7%), and OUD-related primary care visits (estimate: 2.7%; 95% CI: 0.9%–4.4%). Other estimates were not statistically significant.

In assessment of parallel trends during the pre-intervention period, we observed that average difference-in-differences estimates for the pre-waiver period were statistically insignificant for most outcomes. Exceptions included OUD-related inpatient visits (estimate: 0.8%; 95% CI: 0.1%–1.5%), psychiatric hospital LOS (estimate: −0.9 days; 95% CI: −1.4 days to −0.4 days), outpatient visits (estimate: −3.7%; 95% CI: −6.3% to −1.1%), and OUD-related primary care visits (estimate: −3.3%; 95% CI: −6.1% to −0.5%). Difference-in-differences estimates for psychiatric hospital LOS and OUD-related primary care visits were sensitive to trend-adjustment (estimate, psychiatric hospital LOS: 0.6%; 95% CI: −1.0% to −0.2%; estimate, OUD-related primary care visits: −2.0%; 95% CI: −3.8% to −0.3%).

Moving to stratification analysis, we observed substantial variation in associations between the use of Medicaid-paid residential treatment and waiver implementation across waiver cohorts. Specifically, SUD waivers were associated with substantial increases in Medicaid-paid residential treatment among early waiver states during the first through fourth year following waiver implementation (e.g., estimate, 2 years after waiver start: 2.8%, 95% CI: 2.5%–3.0%; Figure 1). The first-year estimate for late waiver states was also positive and statistically significant but small (estimate: 0.3%; 95% CI: 0.1%–0.4%). By contrast, estimates for states in the intermediate-early and intermediate-late waiver cohorts were insignificant or negative and significant. We also found substantial heterogeneity in Medicaid-paid residential treatment rates across waiver cohorts at baseline, with levels below 1% among states in the early and late waiver cohorts but above 4% among states in the intermediate-early and intermediate-late waiver cohorts (see Supporting Informations S1: Appendix Table A-6).

FIGURE 1 |.

Association between SUD-IMD waivers and OUD-related residential stays, stratified by waiver cohorts and year since intervention, 2016–2021. See Supporting Informations S1: Appendix section A-4 for information regarding early, intermediate-early, intermediate-late, and late waiver groups. The right-hand side of the figure shows estimates (square) and 95% confidence intervals (horizontal lines). The column labeled “N” denotes the number of enrollee-quarter observations. Source: 2016–2021 transformed medicaid statistical information system analytic files.

With respect to stratification analyses for secondary outcomes, inpatient stays were negatively associated with SUD waiver implementation among early waiver states during all 4 years following waiver implementation. For instance, the relative decline for the second year following waiver implementation was −3.1% (95% CI: −3.5% to −2.6%; Figure 2).

FIGURE 2 |.

Association between SUD-IMD waivers and all-cause inpatient stays, stratified by waiver cohorts and year since intervention, 2016–2021. See Supporting Informations S1: Appendix section A-4 for information regarding early, intermediate-early, intermediate-late, and late waiver groups. The right-hand side of the figure shows estimates (square) and 95% confidence intervals (horizontal lines). The column labeled “N” denotes the number of enrollee-quarter observations. Source: 2016–2021 transformed medicaid statistical information system analytic files.

4 |. Discussion

In this study comparing 20 waiver states to 14 non-waiver states, waiver implementation was associated with a small increase in residential treatment stays among all waiver states. Waiver implementation was further associated with a decline in all-cause and OUD-related inpatient stays, and increases in OUD-related inpatient LOS, psychiatric hospital LOS, and all-cause and OUD-related primary care visits. Trends prior to waiver implementation between waiver and non-waiver states were not parallel for psychiatric hospital LOS and OUD-related primary care visits, and difference-in-differences estimates for these outcomes were sensitive to trend adjustment. Changes in residential treatment stays and inpatient visits were concentrated among four early waiver states (Indiana, Louisiana, New Jersey, and Virginia).

While SUD waivers were associated with a small increase in residential treatment stays overall, results varied substantially across states. Specifically, waiver implementation was not associated with increases in residential treatment among intermediate-early and intermediate-late waiver states. These states also had elevated levels of residential treatment at baseline indicating that they had a fairly well-developed residential treatment program in place prior to their waiver implementation. In the absence of a waiver, states may cover residential care in IMDs through state funds or, in states with a managed care systems, use the so-called “in lieu of” arrangement to receive federal funds for IMDs up to 15 days per calendar month. Thus, these waiver states may not have been required to make coverage changes that could have resulted in increases in residential treatment. Moreover, federal funds generally did not cover residential costs related to room and board, which may have reduced states’ financial incentives to expand residential treatment as part of the waiver.

By contrast, waiver implementation was associated with substantial increases in residential treatment among early waiver states (Virginia, New Jersey, Indiana, Louisiana). These states had low levels of Medicaid-paid residential treatment prior to waiver implementation that might reflect coverage limitations that were subsequently removed with the waiver. Because of low levels of Medicaid-paid residential treatment, they might also have made it a priority to improve this aspect of their OUD treatment system, for instance, by implementing outreach efforts to residential treatment providers or increasing reimbursement rates. Our ongoing mixed-methods analyses seek to identify factors associated with increases in residential treatment.

The finding that increases in Medicaid-paid residential treatment were concentrated among states with low levels of residential treatment at baseline further suggested that SUD waivers may eventually lead to more substantive increases in residential treatment. Non-waiver states in this study had low levels of Medicaid-paid residential treatment at baseline, and some of them implemented a waiver after the end of the study period. Future research using more recent data will be able to shed light on the longer-term effects of SUD waivers on residential treatment.

Our findings expand on prior research on SUD waivers and the IMD exclusion. The positive association between Medicaid-paid residential treatment stays among early waiver states was consistent with Maclean et al.’s finding of a positive association between waiver implementation and residential treatment facilities’ acceptance of Medicaid patients, suggesting that this supply response translated into an increase in the use of residential care among these states [7]. However, we also found that waiver-related increases in Medicaid-paid residential care were limited to some states, suggesting that the removal of the IMD exclusion by itself may not be sufficient to increase the use of residential care, as was suggested by the evaluation of the Medicaid Emergency Psychiatric Demonstration.

We found some evidence that SUD waivers shifted services from high-intensive care settings to lower levels of care among early-waiver states. This decline might be a result of waiver-related increases in residential stays among these states. However, it could also be related to increases in OUD medication treatment among these states [21].

Aside from findings related to residential care, we observed positive associations between waiver implementation and lengths of stay in psychiatric hospitals as well as OUD-related primary care visits. However, pre-waiver trends were nonparallel among waiver and non-waiver states for these outcomes, and corresponding difference-in-differences estimates were sensitive to trend adjustment. Thus, these results should be interpreted with caution.

5 |. Limitations

We note several limitations. First, TAF records had varying data quality. We excluded states based on the review of relevant TAF data elements, but there might be remaining data quality issues in states that were included in the study. Missing data also did not allow us to include race and ethnicity in the analyses. Second, we used an observational, quasi-experimental study design, and endogenous waiver applications may have introduced selection bias. Specifically, selection based on outcome trends may affect the validity of our analysis. To address this concern, we examined pre-waiver outcome trends for evidence of non-parallel trends between waiver and non-waiver states and estimated difference-in-differences models adjusted for pre-waiver implementation trends. Third, our study only included a subset of all states with SUD waivers, so findings may not generalize to all waiver states. Fourth, although we reported differences across waiver cohorts, the focus of our study was on estimating average changes. Future mixed methods work building on this analysis will examine states’ strategies to improve OUD treatment. Fifth, regression models with information about methadone coverage changes did not converge for length-of-stay outcomes, and we thus were only able to report results for these outcomes that did not include this covariate. Finally, the IMD exclusion may have resulted in residential treatment records not being included in TAF data among states with alternative funding mechanisms. To investigate this concern, we examined the prevalence of residential treatment by age for all waiver states and states stratified by residential coverage prior to waiver implementation (no coverage; coverage through in-lieu arrangement; coverage using state funds). We did not find sharp decreases in observed residential treatment at age 21 (the minimum age for which the IMD exclusion applies) for states covering residential treatment as one would expect if the IMD exclusion resulted in missing residential treatment records in TAF data (see Supporting Informations S1: Appendix Figures A2 and A3). However, we cannot fully rule out the possibility that these data did not include all residential treatment stays of Medicaid enrollees with OUD.

6 |. Conclusions

In this study, SUD waiver implementation was associated with increases in residential treatment and reductions in high-intensive care among 20 waiver states, with changes concentrated among early waiver states. While SUD waivers offer states a mechanism to expand residential treatment capacity, their impact on improving access to the broader continuum of care was limited and varied substantially across states.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section. Data S1: Supporting Information.

Summary.

• What is known on this topic

  • States applied for 1115 substance use disorder waivers to improve their substance use disorder (SUD) treatment system, with a focus on treatment for opioid use disorder (OUD).
  • Residential treatment has not been consistently covered by states’ Medicaid programs.
  • The waiver requires states to cover the full continuum of SUD and OUD care, including residential treatment.
• What this study adds

  • Waiver implementation was associated with an increase in Medicaid-paid residential treatment and a decline in inpatient stays among 20 states.
  • These changes were concentrated among four early-adopter states: Indiana, Louisiana, New Jersey, and Virginia.

Data Availability Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.