On January 11, 2018, the Centers for Medicare and Medicaid Services (CMS) announced a historic shift in its policy to promote work or “community engagement” requirements for Medicaid beneficiaries as a condition of eligibility (
1). Under this policy, individuals can be required to work, go to school, or volunteer for at least 20 hours per week to qualify for Medicaid coverage (
2). Since January 2018, CMS has granted Section 1115 waivers to nine states— Arizona, Arkansas, Indiana, Kentucky, Maine, Michigan, New Hampshire, Ohio, and Wisconsin— to allow state-specific modifications to implement work requirements in their Medicaid programs (
3). As many as 22 million of the 28 million adults (nearly 80%) with Medicaid coverage nationally could be affected by work requirements (
4,
5).
Individuals with behavioral health conditions, such as mental or substance use disorders, constitute a large proportion of the Medicaid population and may be particularly vulnerable to health risks should they experience coverage loss associated with a work requirement (
6,
7). Such individuals have greater baseline health and employment challenges (
8,
9). For example, people with depression and anxiety have a greater likelihood of unemployment, absences from work, and lower productivity at work, compared with individuals without these conditions (
8–
11). Furthermore, similar to those with other chronic conditions, such as asthma and rheumatoid arthritis, people with mental health or substance use disorders (behavioral health disorders) often have fluctuating impairments in their functioning; some days they may be able to work a full-time job, and other days they may not be able to leave the house (
12).
Obtaining Medicaid coverage may allow individuals to improve their health and, thus, their ability to work and maintain employment (
13). However, health and job-related outcomes may be more difficult to achieve for enrollees with behavioral health diagnoses, compared with enrollees with other chronic conditions. It is uncertain whether coverage can facilitate improvements in health and job-related outcomes for individuals with behavioral health conditions, who often have greater baseline challenges in health and employment.
We focused our study on examining the current effects of expanded Medicaid coverage on health and job-related outcomes of enrollees with behavioral health diagnoses. In 2014, Michigan expanded its Medicaid program under a Section 1115 waiver program, which is providing coverage to approximately 670,000 low-income adults as of February 2019. In our prior work, we found that, among the general Medicaid expansion population in Michigan, enrollees who reported improvements in physical and mental health were more likely to report improvements in ability to work and to seek employment (
13). In this study, we sought to assess the impact of Michigan’s Medicaid expansion program (“Healthy Michigan Plan”) on health and employment outcomes of enrollees with and without behavioral health diagnoses.
Methods
Study Design
From January to October 2016, we conducted a telephone survey of 4,090 Healthy Michigan Plan enrollees. The survey was conducted approximately 2 years after implementation of the plan in April 2014. The survey was part of an evaluation of the Healthy Michigan Plan under contract with the Michigan Department of Health and Human Services (MDHHS). The study was deemed exempt and informed consent was waived by the University of Michigan and MDHHS institutional review boards, because the study was a federally mandated evaluation of a public program. We described the survey methods elsewhere (
13,
14). Briefly, we included enrollees ages 19 to 64, with Healthy Michigan Plan enrollment for at least 12 months prior to sampling and at least 9 months in a Healthy Michigan Plan managed care plan; preferred language of English, Spanish, or Arabic; and a complete Michigan address and phone number in the MDHHS Medicaid claims data warehouse used in the study. We used sampling stratified by income and geographic region and conducted telephone interviews with enrollees in English, Arabic, and Spanish. The interviews lasted approximately 1 hour.
The study sample included 4,108 Healthy Michigan Plan enrollees (weighted N=379,627) who completed the survey. We excluded from analysis 18 surveys in which more than 20% of the data were missing, leaving 4,090 respondents with fully completed surveys (weighted response rate=53.7% [using the American Association for Public Opinion Research’s response rate formula]) (
15). Compared with respondents, nonrespondents were more likely to be younger, male, or have a Detroit residence. We applied nonresponse adjustment to sampling weights, controlling for age, gender, race and ethnicity, enrollment month, sampling strata, sampling month, and the interaction between sampling strata and sampling month (
16). Further, we controlled for any discrepancy between the sample and the population through an iterative proportional fitting method on age, gender, race or ethnicity, enrollment month, and sampling strata (
17).
Identification of the Behavioral Health Diagnosis Group
We identified enrollee respondents with potential diagnoses of behavioral health disorders by searching Medicaid administrative claims in the 24-month period prior to survey sampling. The behavioral health diagnosis group was defined as having at least one claims diagnosis from the Mental and Behavioral Disorders Value Set from the 2016 Healthcare Effectiveness Data and Information Set; we excluded tobacco use disorder from the eligible list. We used this method of identifying any individual with one or more behavioral health claims diagnoses to identify enrollee respondents with potential behavioral health diagnoses. This method yielded 2,040 enrollee respondents (48.3% of all survey respondents) as having a potential behavioral health diagnosis. We note that behavioral health diagnoses that appear in claims (treated prevalence) differ from actual prevalence of behavioral health conditions.
Measures
We examined both health and job-related outcomes in the study.
Health outcomes.
We asked about perceived changes in health status through the following items: “Overall, since you enrolled in the Healthy Michigan Plan, would you say your physical health has gotten better, stayed the same, or gotten worse?”; and “Overall, since you enrolled in the Healthy Michigan Plan, would you say your mental and emotional health has gotten better, stayed the same, or gotten worse?” Responses were dichotomized as “better” versus all other responses.
Job-related outcomes.
We first assessed current employment status (employed or self-employed, out of work, homemaker, student, retired, unable to work). For respondents who were working, we asked about perceived changes in ability to work through the following items: “In the past 12 months, about how many days did you miss work because of illness or injury (do not include maternity leave)?”; “Compared to the 12 months before this time, was this more, less, or about the same?”; and “Has getting health insurance through the Healthy Michigan Plan helped you do a better job at work?” Response options were “yes” or “no.”
We also asked about perceived changes in job seeking by assessing agreement with the following statements: “Having health insurance through the Healthy Michigan Plan has made me better able to look for a job” (among those not working); and “Having health insurance through the Healthy Michigan Plan helped me get a better job” (among those with a recent job change but currently working). Responses were dichotomized as “strongly agree or agree” versus “neutral, disagree, or strongly disagree.”
The survey also included standard measures of demographic characteristics, health status, insurance status, health care access, and health care utilization from established national surveys (
18–
22). Covariates included age, gender, race, income, self-reported health status, presence of any chronic health condition, and functional limitation.
Statistical Analysis
We used descriptive statistics to report individual survey responses. Differences between enrollees with and without behavioral health diagnoses were assessed in bivariate analyses that used chi-square tests for categorical variables and t tests for continuous variables. For each group of enrollees, we used multivariable logistic regression analysis to assess the association between reported physical or mental health improvements and job-related outcomes, adjusting for the covariates noted above. To assess whether enrollees with and without behavioral health diagnoses differed in the association between health improvement and job-related outcomes in multivariable analyses, we conducted additional analyses, including an interaction term between an indicator variable for the behavioral health diagnosis group and health improvement. We weighted all analyses to account for sampling and nonresponse using Stata version 14.2. We considered two-sided alpha values of less than .05 to be statistically significant.
Results
Demographic, Health, and Employment Characteristics
Nearly one-half of survey respondents (48.3%) had at least one Medicaid claim with a behavioral health diagnosis (
Table 1). Enrollees with behavioral health diagnoses were more likely than those without to have lower incomes (57.4% versus 46.6%, with incomes 0% to 35% of the federal poverty level, p<0.001). The behavioral health diagnosis group also had greater prevalence of having one or more chronic health conditions (84.3% versus 55.1%, p<0.001) and fair or poor health status (39.8% versus 20.3%, p<0.001) than those without behavioral health diagnoses. Enrollees with behavioral health diagnoses had greater numbers of poor physical health days and mental health days than enrollees without behavioral health diagnoses (p<0.001).
With regard to employment, 43.3% of enrollees with behavioral health diagnoses were employed or self-employed, compared with 54.0% of enrollees without behavioral health diagnoses (p<0.001) (
Table 2). Nearly one in five (18.3%) enrollees with behavioral health diagnoses reported they were unable to work, compared with 4.6% of enrollees without behavioral health diagnoses (p<0.001) (
Table 2).
Changes in Health Since Enrollment
Although both groups reported health improvements since Healthy Michigan Plan enrollment, enrollees with behavioral health diagnoses were more likely than enrollees without behavioral health diagnoses to report improvements in their physical health (51.2% versus 44.6%, p<0.001) and mental or emotional health (45.0% versus 31.8%, p<0.001).
Reported Ability to Work
Among employed enrollees, those with behavioral health diagnoses missed more workdays in the past 12 months than those without behavioral health diagnoses (10.7 versus 5.6 days, p<0.001), but those with behavioral health diagnoses were also more likely than enrollees without behavioral health diagnoses to report that enrollment in the Healthy Michigan Plan helped them to do a better job at work (76.4% versus 64.1%, p<0.001).
Changes in Job Seeking Since Enrollment
Of the respondents who were out of work, those with and without behavioral health diagnoses were equally likely to strongly agree or agree that enrollment in the Healthy Michigan Plan made them better able to look for a job (55.8% of enrollees with behavioral health diagnoses and 53.2% of enrollees without behavioral health diagnoses). Similarly, of the respondents who had a recent job change, those with and without behavioral health diagnoses were equally likely to agree or strongly agree that the Healthy Michigan Plan helped them get a better job (40.0% versus 34.3%).
Associations Between Health Improvements and Job-Related Outcomes
In adjusted multivariable analyses, for both enrollees with a behavioral health diagnosis (adjusted odds ratio [AOR]=1.11, 95% confidence interval [CI]=.83–1.48) and those without a behavioral health diagnosis (AOR=1.03, 95% CI=.80–1.34), physical or mental health improvement since Healthy Michigan Plan enrollment was not associated with current employment (
Table 3). However, both enrollees with behavioral health diagnoses and enrollees without behavioral health diagnoses who reported improved health were more likely than enrollees without improved health to report that Medicaid expansion coverage helped them do a better job at work (AOR for enrollees with behavioral health diagnoses=5.62, 95% CI=3.68–8.59; AOR for enrollees without behavioral health diagnoses=3.27, 95% CI=2.33–4.60), made them better able to look for a job (AOR for enrollees with behavioral health diagnoses=2.71, 95% CI=1.61–4.59; AOR for enrollees without behavioral health diagnoses=3.16, 95% CI=1.78–5.61), and, for those with a recent job change, helped them get a better job (AOR for enrollees with behavioral health diagnoses=5.38, 95% CI=2.24–12.94; AOR for enrollees without behavioral health diagnoses=2.65, 95% CI=1.23–5.69).
When comparing changes in job-related outcomes between enrollees with and without behavioral health diagnoses, an association was found between improvements in health and better job-related outcomes. Among enrollees with improved health, those with a behavioral health diagnosis were as likely as enrollees without a behavioral health diagnosis to report improved ability to work, job seeking, and current employment.
Discussion
In this survey of Medicaid expansion enrollees with and without claims-based behavioral health diagnoses in Michigan, we found that enrollees with a behavioral health diagnosis were more likely than enrollees without a behavioral health diagnosis to have chronic conditions, poor health, and lower incomes. However, enrollees with behavioral health diagnoses were also more likely to report improvements in physical and mental health and to report improvements in their ability to work. Compared with those without behavioral health diagnoses, enrollees with behavioral health diagnoses were as likely to demonstrate an association between improved health and ability to work, as well as improved job seeking. Overall, among enrollees with behavioral health diagnoses, Medicaid expansion appears to be more effective at improving health and at least equally as effective at improving job-related outcomes as it is among enrollees without behavioral health diagnoses.
Nationally, the Affordable Care Act’s (ACA) Medicaid expansion was projected to increase access to and receipt of behavioral health treatment (
23). People with behavioral health disorders reported gains in insurance coverage and access to care after the ACA (
24). For persons with substance use disorders in particular, those with Medicaid were twice as likely as those with private insurance or no insurance to have received treatment services, including outpatient and inpatient services and medication-assisted treatment, in 2016 (
25). Although prior studies have found that unmet behavioral health care needs were associated with lower likelihood of working status (
26), other studies of treatment interventions that reduce behavioral health symptom burden found modest associated improvements in work productivity and labor supply (
27–
29).
Nationally, enrollees with behavioral health conditions are disproportionately represented in Medicaid expansion populations and may have stood to gain more from Medicaid expansion because they started with greater challenges in accessing health care (
30), in experiencing good health (
30,
31), and in maintaining employment at baseline (
8–
10). In addition, Olesen and colleagues (
11) have suggested a reciprocal relationship between mental health and employment, given that poor mental health was identified as both a consequence and a risk factor for unemployment in their longitudinal, population-level study of working-age adults in Australia. Improved access to treatment associated with Medicaid coverage may set a positive cascade in motion, in which good mental health, recovery from a substance use disorder, or improvements in overall health facilitate employment, which further improves mental health and sobriety.
This study should be interpreted within the context of its potential limitations. First, claims-based identification of behavioral health diagnoses may differ from self-report or medical records and from the actual prevalence of these conditions. However, administrative data have been shown to have satisfactory concordance with medical record diagnoses of behavioral health conditions (
32,
33). Second, our method of selecting enrollees with one or more claims with a behavioral health diagnosis in a 24-month period was intended to identify those with potential behavioral health diagnoses, but not to confirm diagnosis. This is a more sensitive than specific method for identifying behavioral health diagnoses and could bias our findings toward the null when comparing enrollees with behavioral health diagnoses and enrollees without behavioral health diagnoses.
Our selection method also selected for a heterogeneous group, with no differentiation between individuals with serious or persistent mental illness and individuals with mild or moderate behavioral health conditions, which could overestimate the number of people with behavioral health needs. We also did not distinguish between enrollees with mental and substance use disorder diagnoses who may have different health care needs and different treatment resources from one another; we also were not able to distinguish which enrollees in our study may have co-occurring mental health and substance use disorders. Third, self-reported outcomes may be limited by recall bias and social desirability bias. However, unless such bias differs between those with and without behavioral health diagnoses, our conclusions about differences between the two groups should hold.
Our self-reported outcomes are also limited to specific groups. For example, we asked respondents about their ability to work only if they were employed. Fourth, we do not have survey data from prior to Medicaid expansion implementation in 2014. Job-related outcomes may have differed between enrollees with and without behavioral health diagnoses at baseline. In addition, because individuals with behavioral health diagnoses are more likely to have worse health and employment at baseline, they may be more likely to report improvements in these outcomes, compared with individuals without behavioral health diagnoses. Fifth, this was a cross-sectional study among Medicaid enrollees conducted after Medicaid expansion, which limits inferences about causality. Last, the study was conducted in one Medicaid expansion state, and experiences of enrollees may vary in states with different program features.
Conclusions
We found that enrollees with behavioral health diagnoses reported significant improvements in health and job-related outcomes associated with Medicaid expansion coverage and that this coverage appeared equally effective for improving job-related outcomes among enrollees with and without behavioral health diagnoses. For low-income people with behavioral health conditions, treatment and recovery services may be accessible only through Medicaid coverage. However, this key group may be at particular risk of coverage loss under Medicaid work requirements because of baseline difficulty in navigating the job market and potential challenges with administrative documentation requirements. Our findings suggest that Medicaid coverage itself may improve employment outcomes and that coverage interruptions for enrollees with behavioral health diagnoses should be minimized to maintain favorable health and employment outcomes.