Medication-Assisted Treatment for Alcohol-Dependent Adults With Serious Mental Illness and Criminal Justice Involvement: Effects on Treatment Utilization and Outcomes

Eleven million people in the United States have serious psychiatric disorders, including schizophrenia, bipolar disorder, and major depression, and are at risk for a range of poor outcomes, including poor social functioning, frequent psychiatric hospitalizations, homelessness, and incarceration. For the nearly 25% who also have co-occurring substance use disorders, these risks are markedly exacerbated (1–6). Alcohol dependence, the predominating substance abuse problem in this population, undermines individuals’ stability and recovery and puts them at higher risk for costly hospitalizations and incarceration. Medication-assisted treatment (MAT) holds significant promise for ameliorating alcohol dependence among adults with serious mental illnesses.

Alcohol use disorders affect 20%−50% of adults with psychotic disorders, such as schizophrenia and bipolar disorder (6–8). Alcohol misuse can worsen psychiatric symptoms, reduce treatment adherence to psychotropic medications and psychosocial interventions, and complicate comorbid medical conditions (1–6). Furthermore, substance use disorders, including alcohol dependence, are strongly associated with risk for entering the criminal justice system. At least 75% of adults with serious mental illness who have had some involvement with the criminal justice system have co-occurring substance use disorders (7–10).

Many adults with serious mental illness do not receive adequate, well-integrated treatment for co-occurring disorders, and many receive no treatment at all (11–13). However, Medicaid expansion under the Affordable Care Act, the Mental Health Parity and Addiction Equity Act, and recent increases in federal funding for substance use disorder treatment, such as the 21st Century Cures Act, all hold promise for increasing the availability of and access to treatment for substance use disorders.

MAT—pharmacotherapies used for treating substance dependence in conjunction with psychosocial treatment—is an evidence-based practice for treating substance use disorders. The literature examining the use of MAT among people with co-occurring disorders demonstrates that MAT is generally well tolerated in people with serious mental illness, with little evidence of complicating interactions with psychotropic medications or of unique adverse side effects. A randomized controlled pilot study demonstrated significant reductions in alcohol craving and drinking days associated with the use of naltrexone in patients with bipolar disorder comorbid with alcohol dependence (14). Other studies have demonstrated promising results for use of naltrexone and disulfiram for treating alcohol dependence in patients with serious mental illness (15), specifically schizophrenia (16–18) or major depression (19), indicating good tolerability and significant reductions in drinking. Patient and provider attitudes regarding the use of monitored naltrexone for alcohol dependence in patients with schizophrenia have also been positive, with a majority of patients reporting improvements in mental health and reduced drinking (20).

Less is known about the extent to which MAT is adopted as part of routine care; however, it is clear that it is drastically underutilized (21–23). Moreover, the current focus lies heavily on the role of MAT in addressing the opioid epidemic, with far less attention given to its value in treating alcohol dependence, the most common substance use disorder among adults with serious mental illness. More effective treatment of alcohol dependence in this population could contribute to better mental health outcomes as well, with corresponding gains in quality of life, diminished burden of illness, and reductions in criminal justice involvement. Here, we estimated the effect of MAT on clinical and criminal justice-related outcomes in a sample of adults with serious mental illness, co-occurring alcohol dependence, and criminal justice involvement.

Method

Administrative data on treatment utilization and criminal justice events for 5,743 adults with serious mental illness, co-occurring alcohol dependence, and criminal justice involvement were merged from several public agencies in Connecticut. Our study population comprised adults ≥18 years old with a recorded diagnosis of schizophrenia spectrum disorder, bipolar disorder, or major depression, a recorded diagnosis of moderate to severe alcohol dependence, and some period of Medicaid enrollment during the study window (2002–2009). All persons had been in a Connecticut jail or prison for at least one night during 2002–2009 and began community-based treatment for alcohol dependence during 2003–2008, allowing for 12 months of observation before and after treatment initiation.

The Connecticut Department of Mental Health and Addiction Services provided data on demographic characteristics and clinical diagnoses for the study sample, along with outpatient treatment utilization and state psychiatric and substance abuse hospitalizations. The Connecticut Department of Social Services provided Medicaid service claims for MAT prescription fills, psychotropic medications for mental illness, outpatient service utilization, emergency department and crisis center visits, and psychiatric and substance abuse hospitalizations in community hospitals. The Department of Correction (in Connecticut, a unified system that produces administrative data on both jail and prison stays) provided data on days incarcerated, the Department of Public Safety provided data on arrest records, and the Connecticut Judicial Branch provided data on probation days. Data from these sources were matched, merged, and deidentified. The study was approved by the institutional review boards at Duke University School of Medicine and the Connecticut Department of Mental Health and Addiction Services.

MAT utilization was identified from Medicaid-covered pharmacy claims for acamprosate, disulfiram, and naltrexone (oral or extended-release [the latter was approved for treatment of alcohol dependence in 2006]). For the MAT group (N=896), the index treatment episode was defined as the first observed outpatient MAT maintenance episode. Treatment episodes were defined as periods of continuous treatment with no gaps longer than 14 days. For MAT pharmacy claims, the maintenance episode was defined as beginning on day 8, assuming that up to the first 7 days included detoxification or titration of the treatment regimen.

For the comparison group (N=4,847), the index treatment episode was defined as the first observed episode of outpatient substance abuse treatment or outpatient mental health treatment if accompanied by an alcohol dependence diagnosis. Comparison treatment episodes were also defined by sequential outpatient treatment visits not separated by more than 14 days. Except for MAT interventions, individuals in both study groups received a range of treatment services, including outpatient mental health treatment, outpatient substance abuse treatment, residential care, Affordable Care Act services, or case management. Although there was no imposed requirement that treatment service utilization be uniform across study groups, all outcome models were adjusted for use of each treatment type.

Measures

Outcomes.

Dependent variables were dichotomous indicators of any crisis-driven health care services and criminal reoffending during each month of observation. Crisis-driven health care was measured in three ways: any inpatient hospitalization for mental health, any inpatient hospitalization for substance use, and any receipt of crisis care at an emergency department or other crisis care provider. We also estimated the effect of MAT on adherence to psychotropic medications for mental illness, calculating the monthly medication possession ratio as the proportion of days in a month in which an individual had a supply of psychotropic medication appropriate for his or her primary psychiatric diagnosis, and we used a dichotomous indicator to determine whether the medication possession ratio was at least 80% within a given month. This approach is consistent with existing research using medication possession ratio as a validated proxy for medication adherence (24–28). Criminal offending was measured three ways: any days in jail or prison, any arrest convictions (e.g., trespassing, disturbing the peace, drug charges, violent crimes, motor vehicle moving violations, and property crimes), and only felony arrest convictions for offenses that typically carry prison sentences ≥2 years.

Explanatory variables.

To compare change over time for the two study groups (the MAT group compared with the non-MAT outpatient treatment group), we included dichotomous main-effects variables to indicate time period (the 12-month period following initiation of the index treatment episode [reference: pretreatment period]); study group membership (MAT treatment group [reference: comparison treatment group]); and an interaction term of time-by-study group.

Covariates.

Multivariate models adjusted for the effects of fixed and time-varying characteristics. Fixed characteristics were age at index treatment (data were missing for 28 persons, 0.49% of the study sample, imputed with the sample mean age); educational need, measured by the Connecticut Department of Correction on a scale of 1 (lowest) to 5 (highest) (data were missing for 150 persons, 2.61% of the study sample, imputed with the sample mean value); gender; primary psychiatric diagnosis (schizophrenia spectrum disorder, bipolar disorder, or major depressive disorder); and race/ethnicity (Caucasian, non-Hispanic; African American, non-Hispanic; Hispanic/Latino; or other).

The models included several important time-varying covariates during the observation period: any probation days; enrollment in Medicaid (≥15 days); any Supplemental Security Income; any outpatient service utilization (mental health, substance abuse, assertive community treatment, or residential care); a dichotomous indicator of the psychiatric medication possession ratio (included in all models except the medication possession ratio outcome model and lagged by 1 month to avoid temporal uncertainty between the medication possession ratio and outcomes of interest in a given month); and secular time (measured as 24 30-day periods numbered consecutively). We also adjusted for differential time in the community at risk (i.e., not institutionalized) by including as covariates the number of jail days (for inpatient outcome models), inpatient days (for the jail outcome model), and total community days (for arrests and emergency department/crisis outcome models).

An underlying assumption of medication possession ratio as a proxy measure for medication adherence is that the patient should continue treatment indefinitely with psychotropic medication for the reference psychiatric disorder. However, the duration of medication use for treating alcohol dependence is much more variable, and although evidence suggests that 6 months to 1 year should be the minimum, the optimal duration is not known (29, 30). Discontinuation is determined by the patient’s ability to maintain abstinence, readiness to discontinue, and engagement in other recovery activities, such as mutual-help groups (29). Thus, it was not appropriate to calculate a medication possession ratio for MAT medications, given that the treatment course and discontinuation decisions for any given individual in the study sample were unknown.

Analysis

A longitudinal data set was constructed with repeated person-month observations, in which each individual had 24 observations centered on the start of the index treatment episode, with 12 months of observations before and after initiation of the index treatment event. Generalized estimating equations (31) were used to fit the multivariate repeated-measures logistic regression models and estimate the differential changes in the odds of crisis-driven health care utilization and criminal reoffending between the two study groups before and after the intervention.

We also conducted subgroup analyses stratified by primary psychiatric diagnosis, on the basis of an expectation that MAT could have differential effectiveness across psychiatric disorders with varying average levels of severity, symptoms, and disability and given the possibility that many of the individuals with a diagnosis of major depression may have had less serious mental illness that was largely secondary to their substance use disorder. Finally, to determine whether adherence to the medication regimen for the major psychiatric disorder was a necessary condition for MAT to exert beneficial effects, we examined the moderating effect of medication adherence in the subsamples of individuals with schizophrenia and bipolar disorder, stratifying each diagnostic subsample by high compared with low medication possession ratio for psychiatric medications.

All analyses were conducted with the SAS 9.4 PROC GENMOD procedure (SAS Institute, Cary, N.C.).

Results

Demographic and clinical characteristics of the two study groups are presented in Table 1. The MAT group (N=896) was slightly older on average and more likely to be Caucasian compared with the comparison group (N=4,847), suggesting a racial/ethnic disparity in MAT use in this sample. Furthermore, persons in the MAT group were more likely to have a psychotic disorder (schizophrenia or bipolar disorder) compared with those in the comparison group. In both groups, more than two-thirds were men. The distributions for treatment utilization and criminal justice involvement in the 12 months leading up to the index treatment episode are summarized in Table 2. The MAT group had significantly more intensive utilization of both outpatient and inpatient mental health and substance abuse treatment than the comparison group during the preindex period. Additionally, persons in the MAT group had lower psychiatric medication adherence than those in the comparison group. Furthermore, persons in the MAT group had a significantly lower prevalence of any jail time during the pretreatment period but were more likely to be on probation compared with persons in the comparison group.

In the MAT group, the majority of persons received treatment with acamprosate for alcohol dependence, during both the index treatment episode (52.46%) and the full 12-month follow-up period, which accounted for both the index treatment episode and any subsequent MAT episodes (56.14%) (Table 3). Naltrexone was the second most common medication used for alcohol dependence in the MAT group, with 39.29% of persons using the drug for their index treatment episode and 43.53% using the drug at some point during the full 12-month follow-up period. Less than 10% of the sample received treatment with disulfiram.

MAT was not, on average, associated with reduced criminal recidivism compared with non-MAT treatment. For arrests, there was no statistically significant difference in change in the odds of arrest between the pre- and posttreatment periods across the two study groups (odds ratio=0.99, 95% confidence interval [CI]=0.86–1.15). The repeated-measures regression models demonstrated modestly higher relative odds of incarceration from the pre- to posttreatment periods in the MAT group compared with the comparison group (Table 4) (odds ratio=1.50, 95% CI=1.28–1.77). This was driven by the comparison group having a significant reduction in the incarceration rate during the posttreatment period (pretreatment mean, 53.63 days [SD=90.42]; posttreatment mean, 39.63 days [SD=76.59]), while the incarceration rate for the MAT group remained stable (pretreatment mean, 30.32 days [SD=63.50]; posttreatment mean, 32.34 days [SD=69.11]). Because jail time can be a function of sanctions for probation violations or prior charges, these differences may reflect baseline differences in the samples.

However, MAT was associated with strong beneficial changes in treatment utilization. The odds of mental health hospitalizations were significantly decreased in the MAT group compared with the comparison group (odds ratio=0.72, 95% CI=0.57–0.92) (Table 5), and the odds of emergency department/crisis visits after initiating the index treatment episode was also decreased in the MAT cohort (odds ratio=0.86, 95% CI=0.75–0.98). The mean number of jail days during months without hospitalizations was relatively stable among persons receiving MAT (pretreatment mean, 2.78 days [SD=8.20]; posttreatment mean=2.84 days [SD=8.37]) but decreased in the comparison group (pretreatment mean=4.61 days [SD=10.30]; posttreatment mean=3.40 days [SD=9.05]), indicating that reductions in mental health hospitalizations were not a function of transinstitutionalization to incarceration. MAT also appeared to be associated with improved adherence to psychotropic medication: the odds of having a good medication possession ratio in the posttreatment period was increased in the MAT group compared with the comparison group (odds ratio=1.57, 95% CI=1.28–1.93).

Model results were consistent among the stratified subsample of adults with schizophrenia, in which the MAT group had no differences in arrests (including felony arrests) and somewhat worse outcomes for jail time (odds ratio=1.38, 95% CI=1.04–1.83) but significantly better service-related outcomes, including lower odds of mental health hospitalization (odds ratio=0.69, 95% CI=0.48–0.98) and emergency department/crisis visits (odds ratio=0.80, 95% CI=0.65–0.98), as well as increased odds of good adherence to psychiatric medications during the posttreatment period (odds ratio=1.46, 95% CI=1.10–1.95) (for further details, see the online supplement). However, among the subgroup of adults with bipolar disorder, MAT was associated with reduced odds of felony arrest after initiating treatment (odds ratio=0.50, 95% CI=0.29–0.86) as well as improvements in the odds of good psychiatric medication adherence (odds ratio=1.65, 95% CI=1.21–2.24) (for further details, see the online supplement).

We also examined the moderating effect of psychiatric medication adherence, with the expectation that individuals with good medication adherence would be more likely to fully benefit from the therapeutic effects of MAT. The unfavorable jail outcomes in the MAT group among persons with schizophrenia were only evident for those with poor medication adherence (odds ratio=1.64, 95% CI=1.13–2.37) (for further details, see the online supplement). Among the subgroup with bipolar disorder, those who had good medication adherence benefited from MAT in the same way as the total sample and the schizophrenia subgroup (i.e., having lower odds of mental health hospitalization [odds ratio=0.43, 95% CI=0.24–0.78] and emergency department/crisis visits [odds ratio=0.60, 95% CI=0.39–0.91] in the follow-up period). Likewise, the association of MAT with decreased odds of felony arrests among individuals with bipolar disorder held only for those with good psychiatric medication adherence (odds ratio=0.18, 95% CI=0.05–0.64) (for further details, see the online supplement).

To explore the influence of changing prescribing patterns over time, we conducted a set of sensitivity analyses, including the year of index treatment episode as a covariate in the model. All model outcomes were statistically consistent with our original set of models that did not include the index year as a covariate, indicating that evolving prescribing patterns did not independently drive the observed treatment effect.

Discussion

Our study results indicate that MAT had strong clinical benefits for adults with serious mental illness and co-occurring alcohol dependence, including reductions in psychiatric hospitalizations, emergency department visits, and improved adherence to psychiatric medications. To the extent that crisis-driven treatment utilization is a proxy for the incidence of mental health or substance abuse crises, reductions in such treatment use directly benefit patients who are spared the experience and may also translate to considerable public behavioral health care savings given the high costs of hospitalization and emergency department care. The beneficial effect observed in the MAT group may, to some extent, be more broadly attributable to engagement in other outpatient substance abuse treatment services at the time of MAT initiation, although the analytic models did control for use of other treatment services, which helps distinguish the MAT-specific effect.

MAT was not associated with reduced recidivism in the total study sample, which may have been related to the greater number of persons in the MAT group who were on probation during the follow-up period. Community supervision places individuals under greater legal scrutiny and, especially among those with co-occurring disorders, at higher risk of incarceration due to probation violations (32). These are not circumstances that we could investigate in the administrative records, but they may at least partly explain the lack of decrease in the incarceration risk among persons receiving MAT. Furthermore, because the incarceration data are not directly linked to the arrest data, we could not distinguish the seriousness of the offenses that led to the incarceration, nor the extent to which psychiatric and/or alcohol use disorders were influencing factors.

The criminal justice outcomes associated with MAT may also reflect the complex, multidetermined causes of criminal offending among the study population. There are many examples in the literature of only partial effectiveness of interventions for adults with mental illness who are at risk for criminal justice involvement, in which the intervention is associated with improved clinical engagement and functioning but not reductions in the risk of offending (33–36). The lack of treatment effect on the risk of criminal offending may be due to adverse characteristics of the social environment that increase the risk in this population (37) or to limited attention to independent criminogenic risk factors that propel offending behavior. For the subset whose offending was tied exclusively to intoxication, treating the underlying alcohol dependence could resolve the offending behavior almost entirely, while for others with long-standing antisocial behavior patterns, treating the alcohol dependence would not alone successfully address the offending risk. However, one challenge in attributing criminal offenses to substance use is knowing that many offenses themselves are not substance-related in nature (e.g., assault) and that alcohol may have been an important contributing factor. However, it is possible that our comparison group had greater latitude for improvement in their incarceration risk once they initiated community-based treatment, because they had higher baseline rates of incarceration, possibly due to worsening of mental illness or alcohol use disorder.

One subgroup in our study population did appear to uniquely benefit from MAT via reduced offending risk: in persons diagnosed with bipolar disorder, the odds of felony arrest conviction were reduced by one-half, and for persons with good psychiatric medication adherence, the odds of felony arrest conviction were reduced by 80%. Adults with bipolar disorder are more likely than adults with schizophrenia to engage in criminal behavior, especially if they have co-occurring substance use disorders (13, 38–40). Thus, for this particular diagnostic group, treating alcohol dependence with MAT was associated with a reduced risk for serious offending, including by a demonstrated moderating benefit of good adherence to psychiatric medications—and without special attention to criminogenic risk factors.

Furthermore, for persons with bipolar disorder, it appeared that the clinical benefits of MAT were actualized for individuals with good psychiatric medication adherence in the 12 months after initiating MAT. For this subgroup with good medication adherence, MAT was associated with substantial reductions in mental health hospitalizations and emergency department visits. By contrast, the bipolar disorder subgroup with poor psychiatric medication adherence experienced no benefit from MAT for either clinical or offending outcomes. These findings suggest that for adults with bipolar disorder, the potential benefits of MAT are realized only when individuals are adherent to their other psychotropic medications. One limitation of these findings is that a high medication possession ratio could be a proxy for good adherence with MAT as well, or with treatment in general, or other unidentified selection effects. Additionally, it is possible that initiation of MAT was a proxy for willingness to take medications and potentially a stronger marker of medication adherence and associated improvements in crisis service utilization than was individuals’ level of adherence to psychiatric medications. As a result, these interpretations should be regarded with some caution.

There are other important limitations to consider when interpreting our study findings. The administrative data for psychiatric diagnoses were not as reliable as they would have been if they were directly assessed with structured diagnostic interviews. However, the Connecticut Department of Mental Health and Addiction Services assesses and records psychiatric diagnoses at inpatient and outpatient admissions and reviews diagnoses at least once every 6 months to increase accuracy. The medication possession ratio serves as a validated near proxy for medication adherence but is not equivalent to evidence of actually taking the medication. Furthermore, while crisis-driven service utilization is an important part of a person’s functional status, the administrative data lack measures on alcohol consumption or other functional status measures. Thus, we could not determine the extent to which potentially greater illness severity (and thereby greater potential for improvement) in the MAT group contributed to the observed MAT effect on outcomes. However, given the severity of comorbid disorders in this population with chronic, relapsing disorders, it is unlikely that differential severity by study group would account for the full treatment effect as it could in a more diverse sample.

The quasi-experimental design yields less definitive findings than a randomized controlled trial. Additionally, the two study groups were nonequivalent on several dimensions, with the MAT group having more clinically severe symptoms, as indicated by higher rates of both outpatient and inpatient service use and emergency department visits than the comparison group. The group differences at baseline yielded poorly matched samples when a propensity analysis approach was attempted. Thus, rather than use the unstable propensity sample and violate the assumption that the two groups were similar at baseline, we opted to use all 24 months of observed data in the analyses for both groups and measure the difference in the rate of change for outcomes of interest in both study groups with study period-by-study group interaction terms.

Many adults with serious mental illness and co-occurring alcohol dependence have relatively good access to MAT compared with adults without serious mental illness who could benefit from such intervention but face treatment barriers. Adults with serious mental illness are more likely to be engaged in psychiatric treatment than adults with alcohol dependence without serious mental illness, most of whom do not receive any behavioral health care. This active treatment engagement facilitates the addition of MAT to a person’s treatment regimen, especially the medications used for treating alcohol dependence, which do not require any special licensure. Additionally, adults with serious mental illness are more likely to have Medicaid coverage than adults with substance use disorders without serious mental illness (who do not qualify for Medicaid-eligible disability status), thereby reducing their out-of-pocket medication costs. The Medicaid formularies in most states include the current evidence-based medications for treating alcohol dependence, including extended-release naltrexone.

To reduce offending rates in this population of alcohol-dependent adults with serious mental illness, MAT may need to be paired with other interventions that directly address criminogenic risk. However, the demonstrated clinical benefits of MAT alone are compelling and should inform practice. Important next steps would be to identify opportunities to increase MAT prescribing for this population and to identify and reduce racial and ethnic disparities. Significant reductions in psychiatric hospitalizations and emergency department visits, along with improvements in psychiatric medication adherence associated with MAT, suggest that evidence-based medications for treating alcohol dependence among adults with serious mental illness can significantly improve their clinical functioning and should be considered more systematically during assessments of their treatment needs.