Medicaid Expansion Boosted Specialty Treatment Episodes For Substance Use Disorder In Expansion States, 2010–22

Abstract

The Affordable Care Act (ACA) enabled states to expand Medicaid to low-income adults and required expansion programs to cover substance use disorder (SUD) treatment. Extending prior research, we analyzed more recent effects of ACA Medicaid expansions on specialty SUD treatment, using 2010–22 all-payer data on treatment episodes. This period coincides with the worsening national drug overdose epidemic, as well as changes to Medicaid policy through program redesign and under the COVID-19 public health emergency. Using difference-in-differences methods, we found that after expansion, episodes to specialty treatment increased by 28 percent in expansion states compared with nonexpansion states. Financial protection through Medicaid as a source of insurance and payment for services also increased significantly in expansion states compared with nonexpansion states. Medicaid expansion is an important program for increasing access to SUD care for a population with high levels of need.

The Medicaid expansion under the Affordable Care Act (ACA) provided a historic investment in federally funded insurance coverage for low-income adults.¹ Medicaid expansion has been especially important for people with substance use disorder (SUD).² Before the ACA, people with SUD experienced high rates of uninsurance and often relied on state or local safety-net grants to access treatment.³ StateMedicaid programs varied in coverage of SUD treatment, with some not covering any services.⁴ Beginning in 2014, the ACA provided enhanced federal matching funds to states to enroll adults with incomes below 138 percent of the federal poverty level in Medicaid. The ACA included SUD treatment as one of the essential health benefits and also extended provision of the federal mental health and SUD parity law to Medicaid.⁵ Medicaid expansion was adopted by twenty-six states (including Washington, D.C.) in 2014; by mid-2024, this figure had increased to forty-one.⁶

Prior analyses of Medicaid expansion, mostly focused on the first few years of implementation, have found meaningful increases in coverage and financing, but less clear results for access to and use of care. Specifically, studies using survey data found that expansion substantially decreased the uninsurance rate among low-income adults, with most of the impact concentrated in expansion states.^7–9 Another analysis, using administrative data, likewise found that the policy resulted in a major shift away from uninsured to Medicaid-enrolled people in treatment. Correspondingly, there was a large shift in expansion states toward treatment covered by Medicaid and a meaningful offsetting decrease in state and local funding.¹⁰

Medicaid expansion occurred during a worsening opioid crisis that coincided with declining mental health status¹¹ and increased the need for treatment nationally.¹² However, several analyses found that in the early years of Medicaid expansion, there was no discernible increase in the total number of people receiving SUD treatment in expansion states compared with nonexpansion states,^7,10 potentially because of people with SUD initially delaying seeking care as a result of factors such as limited availability of services and stigma.^13,14 An analysis using administrative data through 2017 found that total episodes to treatment began increasing significantly in expansion compared with nonexpansion states after 2015.¹⁴ Medicaid expansion had an even stronger impact for episodes that included medications for opioid use disorder (MOUD), potentially showing the increased importance of the program in covering highly effective treatments.¹⁵

The current study extends the analysis of Medicaid expansion, using national data on specialty SUD treatment episodes through 2022. Examining Medicaid expansion’s evolving dynamics is important because the overdose crisis continued to accelerate during this period, largely as a result of the spread of fentanyl in the illicit drug supply.¹⁶ Further analysis is also important because of Medicaid’s continued evolution, including the broader coverage of residential services through Section 1115 waivers and increased provision of MOUD—for example, several states added methadone maintenance to their programs after 2017.^17,18 Finally, our more recent study period included the first three years of the COVID-19 pandemic, when federal public health emergency provisions were implemented to allow for less in-person contact with SUD treatment programs¹⁹ and during a time when Medicaid programs were incentivized by the federal government from disenrolling members.²⁰

Study Data And Methods

DATA

The 2010–22 Treatment Episode Data Set (TEDS) was used to measure episodes in specialty treatment programs. States are required to report data to the Substance Abuse and Mental Health Services Administration (SAMHSA) on specialty SUD treatment episodes delivered by providers who receive public funding or are tracked for other state-specific reasons for people ages twelve and older.²¹ Data from SAMHSA suggest that in 2020, 51 percent of specialty SUD treatment centers received public funding.²² TEDS included approximately two million episodes per year and contained information on individual demographics, substance use (up to three substances), treatment setting, and services. There were 23,115,435 episodes in the 2010–22 TEDS; we aggregated the data to the state-year level (n = 640 state-years).We excluded detoxification-only episodes because such episodes were not considered treatment under consensus guidelines.²³ Despite mandates to do so, some states did not report to TEDS in all years. Supplementary exhibit 1 in the online appendix presents states’ reporting in each year of our study period.²⁴

For each state-year, we examined overall episodes and stratified the number of episodes by treatment setting: residential (hospital inpatient unit or nonacute rehabilitation facility), intensive outpatient (ambulatory treatment that lasts a minimum of two to three hours per day for three or more days per week), and nonintensive outpatient (less than two hours per day or occurs on fewer than three days per week).

We used data from the Current Population Survey and the Census Bureau^25,26 to convert episode counts to rates per 100,000 nonelderly adult state residents. We examined heterogeneity in total episodes by primary substance, MOUD treatment for opioid-involved episodes, and clinical characteristics (for example, co-occurring mental health disorder).

We investigated the source of insurance and expected payer for each episode. States were not required to report insurance and payment information, and data were missing for many states. We included state-year observations with no more than 25 percent missing in these variables, excluding 234 (37 percent) state-year pairs in the insurance sample and 269 (42 percent) state-year pairs in the payment sample.¹⁵ We constructed the share of episodes in which people had each primary insurance coverage source (n = 406): Medicaid, private, other, and uninsured. The analysis of expected payers (n = 371) included Medicaid, private insurance, state and local government grants and vouchers, and self or other. Supplementary exhibits 2 and 3 list state-year pairs in the insurance and payment samples.²⁴

METHODS

To study the effect of ACA Medicaid expansion on specialty SUD treatment episodes, we used a two-stage difference-in-differences procedure robust to bias that can arise when estimating time-varying treatments.^27,28 In the first stage, using untreated observations, the relationships between the treatment variable (ACA Medicaid expansion) and the time-varying covariates and fixed effects were estimated. The outcome was then residualized using these parameter estimates. In the second stage, the residualized outcome was regressed on the treatment variable for all observations. Standard errors accounted for within-state correlations and the first-stage prediction, allowing for the possibilities that outcomes for the same state may have been correlated over time²⁹ and that predicted values had less variation than unpredicted values.³⁰

We used data from KFF for states’ ACA Medicaid expansion status.⁶ We constructed a binary indicator for state Medicaid expansion, with states that expanded Medicaid by 2022 coded 0 before expansion and 1 in the year of expansion and thereafter. For states that did not expand Medicaid in January of the expansion year, we coded the first partial year of expansion as the effective year. States that did not expand Medicaid by the end of 2022 (that is, those states that expanded Medicaid in later years or did not expand by the time of writing) were coded 0 in all years of the study period. Supplementary exhibit 4 reports the exact coding of each state by year.²⁴

Regressions were adjusted for state- and region-by-year fixed effects and time-varying state-level characteristics: demographics,²⁵ unemployment rate,²⁶ and poverty rate.²⁶ Demo-graphics included age, sex, education, race, ethnicity, and country of birth. We included state-level characteristics to account for differences across states that might have predicted SUD outcomes and the propensity to expand Medicaid, allowing us to better isolate the impact of ACA Medicaid expansion from other concurrent policies or shocks.

SENSITIVITY ANALYSES

We conducted sensitivity analyses to assess the robustness of findings (reported in supplementary exhibits 5–21).²⁴ First, difference-in-differences methods assume that in the absence of the ACA Medicaid expansion, treated and untreated states would have followed the same trend in outcomes (“parallel trends”). This assumption is untestable, as counterfactual outcomes were not observed for treatment states. We followed the literature and estimated an “event study.”^15,31–35

We used alternative specifications and samples. First, we used alternative regression specifications: unweighted, without time-varying state-level controls, including additional time-varying state-level controls (that is, number of SUD treatment providers, governor’s political party, unemployment rate, social insurance policies, COVID-19 deaths, and SUD-related Medic- aid policies), and replacing region-by-year fixed effects with year fixed effects.^{25,26,33,36–46} Second, we changed the estimation approach (two-way fixed-effects regression), explored the impact of alternative TEDS sample inclusion criteria (for example, retaining the balanced sample of states that reported data to TEDS in each year, 2010–22; dropping states that implemented sub-stantial Medicaid expansion for low-income, nondisabled adults before 2014;⁴⁷ dropping states that expanded Medicaid 2020–22; and dropping 2020), and lagged the ACA Medicaid expansion variable by one year. Finally, we explored the sensitivity of our insurance and payment status findings to the use of alternative approaches to dealing with missing data.

LIMITATIONS

This study had limitations. First, results can be interpreted causally only if the parallel trends assumption held; although we estimated an event study, this assumption was untestable. Second, despite states being mandated to report data to SAMHSA, not all states reported to TEDS in all years. Third, estimation of Medicaid expansion effects in the longer term were identified according to the experience of early-adopting states.

Study Results

Supplementary exhibit 5 provides summary statistics for expansion states (before expansion) and nonexpansion states.²⁴ Supplementary exhibits 6–8 report trends in specialty SUD treatment episodes, insurance, and expected payer in 2010–22, respectively.²⁴ Episodes increased through approximately 2018 and then declined, with the decline being particularly stark during the COVID-19 pandemic.

Our difference-in-differences specification examined the average effect across all postexpansion years (exhibit 1). We present estimates as rates per 100,000 nonelderly adult state residents each year in exhibit 1, dividing these rates by the baseline to provide a relative change measure. We found that total episode rates increased by 28 percent in expansion states relative to the preexpansion mean in expansion states. This increase occurred across most treatment settings: Rates of residential episodes increased by 26 percent in expansion states compared with nonexpansion states after policy adoption, whereas rates of intensive outpatient episodes increased by 35 percent. The coefficient estimate in the nonintensive outpatient episode specification suggested an increase of 27 percent, but this estimate was not statistically different from zero.

EXHIBIT 1.

Effect of Affordable Care Act Medicaid expansion on substance use disorder treatment episodes per 100,000 nonelderly adult state residents, 2010–22

	Medicaid expansion estimated effect (coefficient)	Years after Medicaid expansion				Preexpansion mean in expansion states
Outcomes (episode types)		0–1	2–3	4–5	6 or more
All episodes
No. per 100,000	2,149**	1,018***	2,989****	3,017**	1,826	7,600
Change (%)	28	13	39	40	24	—a
Residential
No. per 100,000	425****	61	454***	656***	544**	1,635
Change (%)	26	4	28	40	33	—a
Intensive outpatient
No. per 100,000	411**	201*	696***	502*	300	1,167
Change (%)	35	17	60	43	26	—a
Nonintensive outpatient
No. per 100,000	1,313	756**	1,839***	1,858	983	4,798
Change (%)	27	16	38	39	20	—a

SOURCE Authors’ analysis of information for 2010–22 from the Treatment Episodes Data Set (TEDS), linked to data on years of adoption of the Medicaid expansion by state.

NOTES N = 640 state-year observations. Episode types are defined in the text. Detoxification-only episodes were excluded. Results based on two-stage difference-in-differences procedure. The regression included state-level demographics (age, sex, education, race, ethnicity, and country of birth), unemployment rate, and poverty rate; state nonelderly population; state fixed effects; and region-by-year fixed effects. The unit of observation is a state in a year. Data were weighted by the state nonelderly population. Percent change is relative to the preexpansion mean in expansion states and was calculated as the quotient of the coefficient estimate and preexpansion mean multiplied by 100.

^aNot applicable.

^*p < 0.10

^**p < 0.05

^***p < 0.01

^****p < 0.001

Exhibit 1 also disaggregates results by period: 0–1, 2–3, 4–5, and 6 or more years after Medicaid expansion. Total episodes increased through 4–5 years postexpansion, but the coefficient estimate on the 6 or more years postexpansion indicator was not statistically different from zero. More specifically, total episodes increased by 13 percent, 39 percent, and 40 percent 0–1, 2–3, and 4–5 years postexpansion, respectively. Residential episodes did not increase immediately after expansion, but the 2–3, 4–5, and 6 or more years postexpansion episodes in this modality increased 28 percent, 40 percent, and 33 percent, respectively. Intensive outpatient episodes increased by 17 percent, 60 percent, and 43 percent 0–1, 2–3, and 4–5 years postexpansion, respectively, but for the 6 or more years indicator, the coefficient estimate was not statistically significant. Nonintensive outpatient episodes increased by 16 percent and 38 percent 0–1 and 2–3 years postexpansion, respectively, but coefficient estimates on the more distal policy lag variables were not statistically significant.

We also examined differences by insurance status (exhibit 2). The share of treatment episodes in which the patient had Medicaid surged by 26 percentage points (“points”) or 118 percent compared with nonexpansion states when averaged across all postexpansion years. This increase in Medicaid was primarily offset by relatively fewer people who were uninsured (−19 points, or −33 percent), and to a lesser extent by declines in private coverage (−2 points, or −20 percent) and other coverage (−5 points, or −45 percent), although the latter estimates were not statistically significant. Exhibit 2 also shows changes in insurance status during the post period, by years after expansion. Medicaid coverage increased steadily over time, with increases of 59 percent, 109 percent, 141 percent, and 155 percent 0–1, 2–3, 4–5, and 6 or more years postexpansion, respectively. None of the coefficient estimates for private and other coverage were statistically significant. Uninsurance among people in treatment decreased by 25 percent, 39 percent, and 37 percent 0–1, 2–3, and 4–5 years postexpansion, respectively, but the coefficient estimate for 6 or more years was not statistically significant.

EXHIBIT 2.

Effect of Affordable Care Act Medicaid expansion on the insurance source status of substance use disorder treatment clients, 2010–22

	Medicaid expansion estimated effect (coefficient)	Years after Medicaid expansion				Preexpansion mean in expansion states
Insurance status		0–1	2–3	4–5	6 or more
Medicaid
Share of episodes covered	0.26****	0.13****	0.24****	0.31****	0.34****	0.22
Change (%)	118	59	109	141	155	—a
Private
Share of episodes covered	−0.02	−0.01	−0.04	−0.04	−0.01	0.1
Change (%)	−20	−10	−40	−40	−10	—a
Other
Share of episodes covered	−0.05	0.01	0.01	−0.06	−0.13	0.11
Change (%)	−45	9	9	−55	−118	—a
Uninsured
Share of episodes covered	−0.19**	−0.14****	−0.22****	−0.21**	−0.19	0.57
Change (%)	−33	−25	−39	−37	−33	—a

SOURCE Authors’ analysis of information for 2010–22 from the Treatment Episode Data Set (TEDS), linked to data on years of adoption of the Medicaid expansion.

NOTES N = 406 state-year observations (234 state-year pairs had more than 25 percent missing data and were excluded in the calculations). Detoxification-only episodes were excluded. Results based on two-stage difference-in-differences procedure, with regression details in the exhibit 1 notes. Data were weighted by the state nonelderly population. Percent change is relative to the preexpansion mean in expansion states and was calculated as the quotient of the coefficient estimate and preexpansion mean multiplied by 100.

^aNot applicable.

^**p < 0.05

^****p < 0.001

ACA Medicaid expansion also had meaningful impacts on the expected episode payment source among people receiving SUD treatment (exhibit 3). The probability of Medicaid as the expected payer increased by 23 points (144 percent) in expansion states relative to nonexpansion states, but coefficient estimates for other expected payer sources were not statistically significant. Results also show that Medicaid as an expected payer increased monotonically through 4–5 years postexpansion, by 81 percent, 156 percent, and 200 percent 0–1, 2–3, and 4–5 years, respectively, after the policy change. The coefficient estimate for 6 or more years postexpansion was not statistically significant. Similarly, coefficient estimates for other expected payer sources were generally not statistically different from zero.

Exhibit 3.

Effect of Affordable Care Act Medicaid expansion on the payment source status of substance use disorder treatment clients, 2010–22

	Medicaid expansion estimated effect (coefficient)	Years after Medicaid expansion				Preexpansion mean in expansion states
Expected payment source		0–1	2–3	4–5	6 or more
Medicaid
Expected payer probability	0.23**	0.13**	0.25***	0.32***	0.24	0.16
Change (%)	144	81	156	200	150	—a
Private
Expected payer probability	0.02	0.01	0.01	0.01	0.03	0.05
Change (%)	40	20	20	20	60	—a
State and local government
Expected payer probability	−0.14	−0.08	−0.18	−0.17	−0.14	0.45
Change (%)	−31	−18	−40	−38	−31	—a
Self or other
Expected payer probability	−0.11	−0.05	−0.09	−0.16*	−0.13	0.34
Change (%)	−32	−15	−26	−47	−38	—a

SOURCE Authors’ analysis of information for 2010–22 from the Treatment Episode Data Set (TEDS), linked to data on years of adoption of the Medicaid expansion.

NOTES N = 371 state-year observations (269 state-year pairs had more than 25 percent missing data and were excluded in the calculations). Detoxification-only episodes were excluded. Results based on two-stage difference-in-differences procedure, with regression details in the exhibit 1 notes. Data were weighted by the state nonelderly population. Percent change is relative to the preexpansion mean in expansion states and was calculated as the quotient of the coefficient estimate and preexpansion mean multiplied by 100.

^aNot applicable.

^**p < 0.05

^***p < 0.01

Exhibit 4 illustrates heterogeneity in total episode rates to SUD treatment by primary substance. Our overall change in episodes appears to have been driven by increases in treatment where the primary substance was alcohol (23 percent), opioids (39 percent), benzodiazepines (38 percent), and other (98 percent).

Exhibit 4.

Effect of Affordable Care Act Medicaid expansion on substance use disorder treatment episodes per 100,000 nonelderly adult state residents, by primary substance type, 2010–22

Substance type	Medicaid expansion estimated effect (coefficient)	Preexpansion mean in expansion states	Change (%)
None	−23	140	−16
Alcohol	643**	2,773	23
Cocaine	−55	621	−9
Marijuana	5	1,131	0
Opioids	758**	1,962	39
Stimulants	9	812	1
Benzodiazepines	26***	69	38
Other	48*	49	98

SOURCE Authors’ analysis of information for 2010–22, from the Treatment Episode Data Set (TEDS), linked to data on years of adoption of the Medicaid expansion.

NOTES N = 640 state-year observations. Detoxification-only episodes were excluded. Results based on two-stage difference-in-differences procedure, with regression details in the exhibit 1 notes. Data were weighted by the state nonelderly population. Percent change is relative to the preexpansion mean in expansion states and was calculated as the quotient of the coefficient estimate and preexpansion mean multiplied by 100. Substance type was classified by the primary substance endorsed on the treatment record.

^*p < 0.10

^**p < 0.05

^***p < 0.01

All analyses examining the robustness of the regression specification upheld the validity of the study design. Further results are in the appendix.²⁴

Discussion

The ACA Medicaid expansion gave many low-income adults with SUD access to comprehensive drug and alcohol treatment at a time when national overdose death rates were rising rapidly and the need for care was acute. Our study found that Medicaid expansion resulted in a significant boost in specialty SUD treatment episodes: on average, 28 percent in expansion states compared with nonexpansion states. The effect strengthened after the first two years of expansion, and it occurred broadly across a variety of treatment settings, substances, and individual characteristics. However, gains in episodes and Medicaid as a source of payment may have stalled over longer time horizons (that is, six or more years postexpansion).

These findings provide further confirmation of prior research suggesting that Medicaid expansion increased treatment episodes but that the effect occurred with a delay and might not be permanent.¹⁵ That study, using TEDS data through 2017, found that episodes rates had increased to 36 percent over baseline in expansion versus nonexpansion states by four years after the expansion. The delayed effect might explain why other studies on Medicaid expansion that primarily focused on early expansion did not find significant changes in episodes in expansion states.⁶ The current study, incorporating 2018–22 data, found qualitatively similar effect sizes two to five years after expansion, with somewhat diminished effects for some outcomes six or more years postexpansion (the results are not identical for several reasons—for example, the inclusion of newer expansion states).

The results also show that episode changes were largest for people receiving treatment for alcohol, benzodiazepine, opioid, and other disorders. Opioid-related episodes, in particular, increased by 39 percent more in expansion than nonexpansion states (exhibit 4).

Increasing access to MOUD has been a national public health priority,⁴⁸ and state Medicaid programs have undertaken a variety of policy reforms to support MOUD access, including expanded coverage of methadone maintenance, elimination of prior authorization for buprenorphine,⁴³ and systemic interventions to improve access (such as enhanced payments for Medicaid health homes).⁴⁹ Although these efforts are not limited to Medicaid expansion states, there may be advantages to simultaneously improving MOUD coverage in the context of Medicaid expansion. For example, treatment programs may be more sensitive to Medicaid policy changes in states where they depend more heavily on Medicaid for revenue.

Our study extends prior work demonstrating how Medicaid expansion brought new revenue to specialty programs that had largely operated outside of health insurance before the ACA.^7–10 Adding 2018–22 data further increases the estimated impact of Medicaid as a source of coverage and, to a lesser extent, payment in expansion states compared with nonexpansion states. We found, for example, that Medicaid coverage increased by 59 percent immediately after the policy change and by 155 percent six or more years postpolicy (exhibit 2). These changes correspond to the influx of new patients with Medicaid, but they likely also reflect greater sophistication in how specialty programs in expansion states capitalized on potential Medicaid revenue through upgraded billing and claims processing capabilities.^10,50,51 These findings may also reflect changes in Medicaid continuum-of-care provisions, such as the availability of Section 1115 waivers to cover services in residential facilities.³²

Finally, our study findings should be considered against the backdrop of major upheaval occurring during the COVID-19 pandemic. The pandemic posed enduring challenges for low-income people with SUD, increasing solitary drug use, overdose, and residential instability.⁵² In general, Medicaid expansion might be a source of stability during periods of turmoil. Other research suggests that provisions to freeze Medicaid disenrollment were particularly important in Medicaid expansion states.⁵³ Changes in federal policy under the public health emergency permitted more people with SUD to access telehealth treatment and reduced barriers to methadone services.⁵⁴ Despite these changes, we observed an overall decline in specialty care episodes during the pandemic years, although it was relatively lower in expansion states. Thus, even in expansion states, there remains considerable room to engage in and sustain specialty SUD treatment and to regain lost ground.

Conclusion

Medicaid expansion remains a critical policy to encourage the use of SUD treatment, and the policy has an especially important role in providing effective treatment with medication in the current opioid crisis. Amid ongoing changes to Medicaid emerging from the COVID-19 pandemic, ensuring that the program remains a stable source of funding for specialty treatment can help people in crisis to access comprehensive, lifesaving care. ■