Impact of the ACA Medicaid Expansion on Utilization of Mental Health Care

The ACA’s Medicaid expansions increased insurance coverage for low-income Americans, among whom unmet need for mental health care is high. Empirical evidence regarding the impact of expanding insurance coverage on use of mental health services among low income and minority populations is lacking.

Methods:

Results:

Annual outpatient visits for mental health conditions increased by 0.513 (.053-.974) visits per person, from a baseline rate in ME states of 0.894 visits per person. However, no significant changes were observed in number of mental health related hospital stays, emergency department visits or prescription fills. The increase outpatient visits was limited to Hispanics and Non-Hispanic Whites, with no increase in service use observed among Non-Hispanic Blacks. There was no apparent increase in the number of users of outpatient mental health care (AOR=0.992, p=.942) and a marginally significant (p=.096) increase of 3.144 visits per user.

Discussion:

Medicaid expansion had a limited but positive impact on use of mental health services by low income Americans, though it may also have increased racial/ethnic disparities.

Evidence suggests that the expansion of Medicaid under the Affordable Care Act increased access to health care for low-income Americans(1, 2). In states that expanded Medicaid (henceforth ME states) at the beginning of 2014 Medicaid coverage increased by 10.5 percentage points and total insurance coverage by 7.4 percentage points by the second half of the year, relative to the states that did not adopt the expansion (Non-ME states)(3). Medicaid expansion also led to increases in use of some medical services, including visits to general practice physicians, new diagnoses of diabetes, and use of preventive services (1, 3).

Positive impacts of Medicaid expansion were also anticipated for mental health care(4, 5) since Medicaid is the single largest payer for mental health services in the US(6). Insurance coverage is a major barrier to mental health care(7), particularly among low-income individuals newly eligible under the ACA expansions. However, other barriers to care, such as provider supply(8), and low perceptions of need for treatment(9) may constrain the impact of Medicaid expansion on mental health care generally or in particular racial/ethnic groups. For instance, the number of psychiatrists who accept Medicaid as a form of payment has decreased since 2014(10). If the treatment system does not have adequate capacity, then increasing insurance coverage may not impact utilization.

There is evidence that use of mental health care increased in the years following implementation of the ACA(11–13), but this trend had been observed for many years prior to the ACA(14) and studies have yet to examine impacts attributable specifically to the Medicaid expansion as opposed to other components of the ACA. In addition, it is important to examine impacts of Medicaid expansion on different types of mental health care and across racial/ethnic groups. Increases in use of mental health care use in the years preceding the ACA were limited to use of psychiatric medication, raising concerns about the overall quality of that care due to the lack of concurrent increases in outpatient treatment (14). There is also concern that reducing financial barriers to care may increase the importance of remaining barriers that differentially impact minorities, leading to negative impacts on disparities in care. One study which modeled the impact of insurance expansion on disparities in care suggested that increased insurance coverage would reduce disparities between Hispanics and Whites, but not affect disparities between Blacks and Whites(15).

Understanding the impact of the Medicaid expansion on use of mental health services is important for documenting recent changes and for understanding how removing a financial barrier, which is one barrier to care among many, impacts use of services among low income individuals. In this paper we examine the impact of Medicaid expansions on use of mental health services by comparing trends in the ME states with concurrent trends in Non-ME states.

Methods

Data and Sample

We used data covering 2007 through 2015 from the Medical Expenditure Panel Survey (MEPS), a nationally representative survey that is conducted annually by the Agency for Healthcare Research and Quality to provide information about health status, health insurance, and health care utilization for the U.S. civilian noninstitutionalized population. We extracted information from the MEPS Household Full Year Consolidated Data File about patient level characteristics such as age, gender, marital status, race, education, poverty index, and K-6 scale of psychological distress. We obtained information about health care utilization from the MEPS Household Event Files, which consist of a series of files about prescription medicines, hospital inpatient stays, emergency room visits, outpatient department visits, and office-based medical provider visits.

Each event file has detailed clinical information, such as ICD-9-CM codes and Clinical Classification Software (CCS) codes, which are the ICD-9-CM codes regrouped in clinically homogenous categories (See details about the CCS at https://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp). Within each of the event files, mental disorder-related health care utilization was identified using a method developed by AHRQ(16), by which an event in the MEPS Household Event Files is considered as mental health-related if the event is associated with any conditions in CCS categories 650–663 and 670. For prescription medicines, we included visits with prescriptions of mental health drugs (stimulants, anxiolytics, antidepressants, antipsychotic, and mood stabilizers) regardless of the CCS category. Mental health drugs were identified from annual formularies by a psychiatrist.

As a study of the Medicaid expansion targeting low income groups, our analytic sample was restricted to adults, 18 years of age or older, with incomes at or below the eligibility threshold for Medicaid under the expansion rules, 138% of the federal poverty line. Four states that expanded Medicaid during 2015 were excluded from the analysis. The total sample included 52,739 respondents. Of these, 43,161 were interviewed prior to 2014 (24,301 in ME states and 18,860 in Non-ME states) and 12,578 were interviewed in 2014 or 2015 (7,157 in ME states and 5,421 in Non-ME states).

Outcome Measures

We used the AHRQ method of identifying mental disorder-related health care utilization in the MEPS to construct person-level count (# of events) and binary indicator (any visits vs no visits) measures for the following types of mental health care: emergency department visits; hospital stays; prescription fills; and outpatient visits. Events related to mental health treatment were identified by diagnosis of a psychiatric condition or use of a psychiatric medication. Information on hospital outpatient visits and office-based outpatient visits were combined to create measures of outpatient mental health visits. These two categories were combined because of the difficulty in distinguishing between them, given trends toward hospital acquisition of independent physician practices. (17) All measures are defined on a per-year basis.

Statistical Analysis

The above MEPS files were merged by state identifiers with a dataset of indicators of Medicaid expansion status by year at the AHRQ data center in Rockville, MD, where all analyses were performed.

We first applied the difference-in-differences (DID) approach to examine the impact of Medicaid expansion on the per-year quantity of service use by type of care, using survey design-adjusted linear regression models (model specifications detailed in the Supplemental Digital Appendix). The models included covariates for respondents’ age, sex, marital status (married, widowed/divorced/ separated, never married), race/ethnicity, K-6 score category (>13, 8–13, <8), and education level (<=High School/GED vs >High School/GED). Calendar years were modeled by dummy year indicators to allow for a flexible time trend. A treatment flag was used to indicate respondents in the ME states. The interaction between the treatment flag and an indicator for 2014 and later years represented the effect of Medicaid expansion.

Where the above model produced a significant DID estimate of the overall impact of the Medicaid expansion on the quantity of a particular type of service, we conducted additional investigations to distinguish impacts on the probability of using a service from impacts on the frequency of service use, and to estimate impacts within racial/ethnic groups. Impacts on the probability of service use were distinguished from impacts on the frequency of service use with a two-part modelling strategy, namely a binary part indicating whether a person used the particular type of care in a year, and a continuous part examining the frequency of service use during the year among the users. The binary part was modeled by a logistic DID model, and the continuous part was modeled by a linear DID regression on the subsample of service users. Impacts of the Medicaid expansion within racial/ethnic groups were examined by extending the original DID models to triple-difference models. The triple difference models include three-way interactions among race/ethnicity, time period, and expansion status, as well as all nested two-way interaction terms.

To test the assumption of parallel trends in the outcomes during the years prior to the Medicaid expansion, we conducted placebo tests by running DID models using the same specification as our analytic model but with the intervention set to an arbitrary date (18). A significant result would indicate a violation to a basic assumption of DID, i.e., the parallel trajectory assumption. The placebo tests found no significant violations. We also examined whether the results were altered by consideration of early Medicaid expansions in some states by including an additional time-varying indicator for respondents in the states that expanded Medicaid eligibility prior to January 2014 and the years following those early expansions. The time-varying state-level indicators for state-level expansion were not statistically significant in all analyses, and they had no impacts to the treatment effect estimates.

RESULTS

The sample is comprised of 29,827 respondents in ME states and 22,873 respondents in Non-ME states ( Table 1 ). Respondents in Non-ME and ME states are similar with respect to age, gender, educational attainment, and psychological distress. Respondents in ME states are slightly more likely to have been never married, more likely to be Hispanic or Non-Hispanic Other, and less likely to be Non-Hispanic Black than respondents in Non-ME states. All of the above individual characteristics are included as statistical controls in the adjusted models reported below.

Table 1.

Characteristics of the MEPS adult low-income 1 sample, 2007–2015, by state Medicaid expansion participation

MEPS=Medical Expenditures Panel Study; ME=Medicaid Expansion; K6=Kessler-6.

1 Low-income includes all adults from households with income up to 138% of the federal poverty level.

Results for the DID analysis for each of the mental health care event types are presented in Table 2 . The table shows the pre-post comparisons for the Non-ME and the ME states, the unadjusted comparison between these two differences, and the adjusted comparison between the two differences with statistical controls for respondent characteristics. With respect to inpatient stays, we find no pre-post differences in either the Non-ME or the ME states and a non-significant adjusted DID estimate. For ED visits, we find no pre-post difference in the Non-ME states and a significant increase of .008 events annually per person in the ME states. The DID estimate of an increase of .007 events annually per person is relatively large in magnitude compared to the baseline rates of .009 and .011 events annually per person in the Non-ME and ME states respectively, but the estimate does not reach statistical significance (p=.144).

Table 2.

Difference-in-Difference Analysis of Medicaid Expansion Impact on Mental Health Care Events

ME=Medicaid Expansion; P-P Dif.= Pre-Post Difference; DID=Difference in Differences; SE=Standard Error.

Sample (N=52,739) includes MEPS respondents with income less than 138% of poverty, 2007–2015.

However, neither the unadjusted or the adjusted DID estimates of the impact of the ME on ED visits reaches statistical significance. For prescription fills, the pre-post difference does not reach statistical significance in the Non-ME states, but it is positive and significant in the ME states. However, neither the unadjusted or the adjusted DID estimates of the impact of the ME on prescription fills reaches statistical significance.

For outpatient visits, the pre-post difference does not reach statistical significance in the Non-ME states, but it is positive and significant in the ME states. Both the unadjusted and the adjusted DID estimates of the impact of the ME on outpatient visits are statistically significant indicating an increase due to the ME of 0.513 visits per person per year.

We focus more detailed analyses on outpatient mental health care, the only event-type for which a significant impact of the Medicaid expansion was detected. DID estimates of the impact of ME on annual outpatient visits for Non-Hispanic Whites, Non-Hispanic Blacks and Hispanics are shown in Table 3 . In the Non-ME states, the pre-post differences in outpatient visits do not reach statistical significance for any of the three racial/ethnic groups. However, in the ME states, the pre-post differences are positive and significant for Non-Hispanic Whites and Hispanics. The adjusted DID estimate for Whites suggests an increase of 0.905 visits per person, and the adjusted DID estimate for Hispanics indicates an increase of 0.533 visits per person.

Table 3.

Difference-in-Differences Analysis of Medicaid Expansion Impact on Outpatient Mental Health Visits by Racial/Ethnic Group

ME=Medicaid Expansion; DID=Difference in Differences; P-P Dif.= Pre-Post Difference; OR=Odds Ratio; SE=Standard Error.

A two-part model was estimated to distinguish impacts on the likelihood of being treated from impacts on the number of outpatient visits per year among service users. Although the adjusted DID estimates for both parts of the two-part model fail to reach statistical significance at the conventional p=.05 level ( Table 4 ), the results suggest that the significant effect observed on outpatient visits is likely due to an increase in the number of visits among service users rather than an increase in the number of people receiving treatment. The odds of use of outpatient services increased to a similar degree in both the Non-ME states and the ME states, resulting in a DID estimate of the odds ratio very close to 1 (OR=0.992, p=0.942). In contrast, the average number of visits per user of care increased by over three visits per person in the ME states and decreased (non-significantly) by 0.452 visits per person in the Non-ME states. The DID estimate shows a relatively large increase of 3.140 visits annually per user of outpatient services in the ME relative to the Non-ME states. (p=0.096).

Table 4.

Two-Part Difference-in-Difference Analysis of Medication Expansion Impact on Outpatient Mental Health Care Use and Number of Visits among Users

ME=Medicaid Expansion; DID=Difference in Differences; P-P Dif.= Pre-Post Difference; OR=Odds Ratio; SE=Standard Error.

1 Results for ‘Any Visits’ estimated in logistic regression models with adjustment for age, sex, marital status, educational attainment and race/ethnicity. Pre- and Post- values are presented as odds of having one or more visits and P-P Dif. values are presented as odds ratios. DID values are ratios of odds ratios.

DISCUSSION

This study examined the impact of the Medicaid expansion under the ACA on use of four types of mental health services in nationally representative samples of low-income individuals during the first two years following implementation. The study’s strengths include consistent ascertainment of service use across the study period and a DID design that controls for the long-term trends toward increasing utilization of mental health services in the United States. The results suggest that the ME had an impact on utilization of outpatient mental health visits but not on use of other types of mental health services. Further investigation of the impact on outpatient mental health visits suggests that the change is more likely to have resulted from an increase in the number of visits among users than from an increase in the number of people using services. In addition, the change in use of outpatient mental health visits was significant among Non-Hispanic Whites and Hispanics, but not among Non-Hispanic Blacks.

The observed increase in the average number of annual mental health visits was not only statistically significant but meaningful in magnitude. At a population level, the increase of about .513 visits per person annually compares with a baseline (pre-2014) average of .438 in the Non-ME states and .894 visits in the ME states. Increasing the average number of outpatient mental health visits is a meaningful change which is likely to have positive public health effects. Discontinuation of treatment prior to completion of a full recommended course is common, with a large proportion of patients dropping out of treatment after the first visit(19). Moreover, lack of insurance is a strong predictor of dropout(19), suggesting that financial factors influence patients’ continuation of treatment even after they have decided to initiate treatment at their own cost. By lowering the cost of treatment to patients, the ME may have enabled patients to lengthen their course of treatment. Evidence shows that treatment is more effective when patients have more visits, even if they do not complete a full course of treatment(20), so even incremental increases in treatment duration can have positive impacts on treatment outcomes. For instance, a study of depression treatment found that a higher number of visits increased the value of an episode of treatment(21).

The results regarding impacts of ME across racial/ethnic groups are important given that the increase in access to care afforded by Medicaid coverage has the potential to reduce racial/ethnic disparities in care, due to the over-representation of racial/ethnic minorities in the low-income population that became newly eligible for Medicaid(22). In fact, after implementation of the ACA in 2014, rates of uninsurance declined more among Non-Latino Blacks and Latinos than among Whites(2, 23). However, financial cost of care is only one of multiple reasons that individuals with mental health problems do not seek care. Lack of perception of need for treatment, the most common reason that people with mental health conditions do not seek treatment(24), is more common among both Hispanics and Non-Hispanic Blacks than among Non-Hispanic Whites(9).

Our findings are consistent with those of a simulation of the impact of insurance expansion on racial/ethnic disparities in behavioral health service use, which was conducted prior to the ACA. That study found larger impacts of insurance expansion on service utilization among Non-Hispanic Whites and Hispanics than among Non-Hispanic Blacks(15). This pattern of effect across racial/ethnic groups differs from the pattern found for use of primary care services, where Non-Hispanic Blacks benefited more than Hispanics (25), suggesting that the impact of insurance on disparities may differ between behavioral health and other types of health care. The impact of insurance on disparities in behavioral health care between Non-Hispanic Whites and Non-Hispanic Blacks may be limited. Without more targeted efforts by health care institutions, these disparities are unlikely to change.

The negative findings with respect to mental health-related inpatient stays, emergency department visits, and prescription fills also deserve comment. Inpatient stays are likely to be highly constrained, without large capital expenditures to increase capacity. The lack of Medicaid reimbursement for inpatient hospitalization, i.e. the IMD exclusion, which remained in effect during the period under study, likely contributed to the lack of impact of ME on inpatient stays. Evidence regarding the impact of insurance coverage on emergency department use is mixed. The Oregon Medicaid Experiment, a rare randomized trial of Medicaid expansion, found that Medicaid coverage increased emergency department use, while quasi-experimental studies of the ME under the ACA have found evidence of increases(26), decreases(27) and no change(28). Our study found a significant increase in the ME States and no change in the Non-ME States in ED use specifically for mental health conditions, but the DID estimate, though positive, did not reach statistical significance. Statistical power for examining this relatively rare outcome may be an issue.

Finally, we expected to observe an impact of ME on mental health related prescription fills, since these medications are frequently prescribed by primary care providers and ME appears to have increased visits to primary care providers. In addition, the Oregon Medicaid Experiment found that Medicaid coverage increased use of mental health prescription medications(29). Our finding of no statistically significant change in mental health prescriptions suggests that cost may not be the sole determinative barrier to medication use, which is consistent with strong preferences in the general population for talk-therapy over medication for mental health conditions(30).

Limitations of the study should be noted. This study used repeat cross-sectional samples to investigate changes over time. Longitudinal data, which would show patterns of change within individuals over time would provide a more robust test of the impact of Medicaid expansion on mental health service use. In addition, our analyses of inpatient and ED visits have limited statistical power to definitively rule out meaningful effects of the Medicaid expansion. Low statistical power is particularly a concern regarding ED visits where the estimated magnitude of the effects of the Medicaid expansion is large relative to baseline rates of ED visits but does not reach statistical significance. Statistical power for analysis of prescription fills was more than adequate to detect meaningful effects giving us greater confidence in concluding that there was not an impact of Medicaid expansion on this outcome. Finally, this study examined only the first two years following expansion of Medicaid in the majority of states that participated. Additional impacts may be observed over a longer time frame.

Accumulating evidence regarding the impact of ME on use of behavioral health care suggests that insurance expansion can improve utilization patterns, particularly with respect to non-acute care, but that impacts are constrained by non-cost related factors. The impact of insurance may be limited to people who have already decided that they need care, increasing the duration, and likely the quality, of care that they receive. However, insurance expansion does not address other barriers to care, such as lack of perception of need for treatment. As a result, the ME may reinforce rather than reduce existing racial/ethnic disparities in care.