Randomized Controlled Trial of Contingency Management for Stimulant Use in Community Mental Health Patients With Serious Mental Illness

Approximately 50% of adults with serious mental illness, such as schizophrenia spectrum, bipolar, and recurrent major depressive disorders, suffer from a substance use disorder at some point during their lives (1). Relative to people with only one of these conditions, individuals with co-occurring disorders have more severe substance use and psychiatric symptoms (2), poorer treatment adherence (3), and higher rates of homelessness (4), HIV infection (2), psychiatric hospitalization (5), emergency room use (6), and incarceration (7).

Despite the frequent co-occurrence of these disorders and associated detrimental outcomes, few individuals receive concurrent treatment for both conditions (8, 9). Integrated treatments delivered via individual (10, 11), group (12), case management (13), or multiple component (14) models have been associated with reductions in drug use. While these treatments have been associated with reduced psychiatric symptom severity in some studies, not all studies have obtained this result (15). Few individuals receive such treatments in community mental health centers, where most adults with co-occurring disorders receive care (8), for various reasons, including the cost of these interventions, organizational barriers, and the need for extensive training and adherence to these models (16–18).

Interventions that are less costly and easier to implement (e.g., do not require clinical staff, extensive training, supervision, and adherence ratings) are needed to improve outpatient treatment of co-occurring disorders. Contingency management is an evidence-based intervention in which individuals are provided with reinforcers (e.g., vouchers, prizes) based on abstinence from drugs. Contingency management is associated with the largest reductions in drug and alcohol use compared with all other psychosocial treatments (19). Contingency management has also demonstrated improved treatment retention and attendance (20, 21) and long-term efficacy (22, 23) in populations with psychiatric diagnoses below the threshold of serious mental illness. Initial evidence suggests similar improvements in treatment retention and abstinence for persons with serious mental illness (24, 25). Bellack and colleagues (14) observed reductions in drug use and hospitalizations, as well as increases in quality of life and financial management, in adults with co-occurring disorders who received a 6-month cognitive-behavioral group-based treatment that included contingencies for drug abstinence. These data suggest that contingency management may be an effective treatment approach for adults with co-occurring disorders. However, no adequately powered randomized controlled trial has been conducted to evaluate the efficacy of contingency management alone as an adjunct to treatment as usual for substance use disorders in seriously mentally ill outpatients.

Our primary aim in this study was to determine whether the addition of contingency management for psychostimulant drug abstinence would be successful in reducing stimulant use, as measured by urine drug tests and self-report, in persons with serious mental illness and stimulant dependence receiving treatment in a community mental health center. Our secondary aims were to determine whether contingency management was associated with reductions in use of other drugs and alcohol, drug use severity, psychiatric symptom severity, HIV risk behavior, and community problems (e.g., psychiatric hospitalizations and incarcerations). Stimulants were targeted because of the frequent abuse of these drugs by those with serious mental illness and the association between stimulant use and psychiatric symptom exacerbation (26).

Method

Participants

Participants were recruited from a multisite community mental health and addiction treatment agency in Seattle. To be eligible, participants had to have used stimulants during the past 30 days and had to meet Mini International Neuropsychiatric Interview criteria for methamphetamine, amphetamine, or cocaine dependence as well as criteria for schizophrenia, schizoaffective disorder, bipolar I or II disorder, or recurrent major depressive disorder. Exclusion criteria were presence of organic brain disorder, dementia, or medical disorders or psychiatric symptoms severe enough to compromise safe participation.

Overall, 205 individuals provided written informed consent for study participation. Of these, 201 completed the initial study assessment, and 197 met inclusion criteria (see the CONSORT diagram in the data supplement that accompanies the online edition of this article). Consistent with previous studies that used a noncontingent reinforcement control condition (27), the first five participants were assigned to the contingency management condition. Of the 192 individuals available for randomization, 176 returned for their second study visit, at which they were informed of their group assignment; these individuals are considered the intent-to-treat sample. Study procedures were approved by the University of Washington’s Human Subjects Division.

Design and Procedure

The study was a 3-month quasi-yoked randomized controlled trial of contingency management with a 3-month posttreatment follow-up period. Participants were randomly assigned to contingency management plus treatment as usual (N=91) or treatment as usual plus rewards for study participation only (noncontingent rewards; N=85). Randomization was conducted using the urn randomization procedure (28), balancing groups on gender, substance use severity, mood versus psychotic disorder, and psychiatric hospitalization in the past year.

Measures

Participants completed a structured psychiatric interview and study outcome measures. During the treatment phase, participants provided alcohol breath samples (Alco-Sensor III, Intoximeters, St. Louis) and urine samples for drug testing. Drugs were assessed using on-site immunoassays of amphetamine, methamphetamine, cocaine, marijuana, and opiate use (Integrated E-Z Split Key Cup, Innovacon, San Diego). Participants provided breath and urine samples three times a week (Monday, Wednesday, and Friday) and received prize draws as stipulated by their test results and experimental condition. During the follow-up period, participants provided breath and urine samples once a month. At weeks 4, 8, 12, 16, 20, and 24, participants completed other study outcome measures to assess days of substance use and substance use severity (the Addiction Severity Index–Lite Version [29]), psychiatric symptom severity (the Brief Symptom Inventory [30] and the Positive and Negative Syndrome Scale [31]), and HIV risk behavior (the HIV Risk Behavior Scale [32]). Participants were reimbursed for completing these interviews.

Community outcomes were gathered from administrative sources for the 3 months prior to randomization, the 3-month intervention period, and the 3-month follow-up period. Data included counts of outpatient mental health and chemical dependency visits, inpatient substance use and psychiatric treatment admissions and days, detoxification admissions, emergency department utilization, and incarcerations.

Treatment as Usual

Treatment as usual consisted of mental health, chemical dependency, housing, and vocational services. Most clients saw their case manager once a week, had access to psychiatric medication management, and could participate in various group treatments. Forty-six percent (N=42) and 54% (N=46) of individuals in the contingency management group and the noncontingent control group, respectively, received intensive outpatient group or individual substance abuse treatment during the study.

Treatment Conditions

Contingency management group.

Participants assigned to the contingency management condition received the variable magnitude of reinforcement procedure each time they tested negative for amphetamine, methamphetamine, and cocaine. This procedure is well researched (33) and involves making “draws” from a bowl of tokens representing different magnitudes of reinforcement. Fifty percent of the tokens read “good job” (no tangible reinforcer). The other 50% were associated with a tangible prize (41.8% read “small” [$1.00 value], 8% read “large” [$20.00 value], and 0.2% read “jumbo” [$80.00 value]). Participants began by earning one opportunity to engage in the reinforcement procedure for each urine sample that demonstrated abstinence. One additional opportunity to engage in the reinforcement procedure was earned for each week of continuous stimulant abstinence. Missing or drug-positive samples resulted in no delivery of reinforcement at that visit and a reset to one in the number of opportunities to engage in the reinforcement procedure when the next negative sample was submitted. After a reset, participants could return to the point in the escalation at which the reset occurred by providing three consecutive stimulant-negative samples. Participants were provided with one additional opportunity to engage in the reinforcement procedure during each visit if they submitted samples that demonstrated abstinence from alcohol, opioids, and marijuana.

Noncontingent control group.

Consistent with previous studies, participants assigned to the noncontingent control condition were quasi-yoked to participants in the contingency management condition (27) in order to equate the number of prize draws received between conditions while isolating the contingent nature of reinforcement for drug abstinence. To determine the number of prize draws received by individuals assigned to the noncontingent condition in the first week of the study, the first five individuals recruited to the study were assigned to the contingency management condition. These individuals were treated for 1 week, and their average number of prize draws was used to set the number of prize draws received by participants in the noncontingent group during their initial week of participation. For the remainder of the study, the number of prize draws of the noncontingent condition was equal (yoked) to the average number of draws earned by the contingency management group during the preceding week. The five individuals initially assigned to the contingency management condition were not included in the intent-to-treat sample.

Reinforcers and Earnings

Reinforcers were useful, supportive items, including shampoo, toothpaste, gift cards, microwave ovens, and electronics. The total average value of prizes earned was $150.30 (SD=130.94) for the contingency management condition and $201.48 (SD=343.32) for the noncontingent control condition (not statistically different).

Analysis

Chi-square tests for categorical variables and t tests for continuous variables (including outcome measures) revealed no significant differences between the groups in baseline demographic, clinical, or outcome variables. Analyses were conducted on data from the intent-to-treat sample. Generalized estimating equations were used for the analyses conducted on outcome measures that were assessed over time in conjunction with the robust maximum likelihood estimation procedure to protect against type I error (34). Analyses utilized bidirectional tests despite our hypotheses being unidirectional to further protect against type I error. Odds ratios with 95% confidence intervals (CIs) are presented for binary measures, and unstandardized regression coefficients with 95% CIs are presented for continuous measures. This method of analysis has been used in previous contingency management trials and is an effective and efficient method of analyzing outcomes across time nested within participants (20, 33). Generalized estimating equations were used to evaluate the significance of changes in outcomes over time by treatment condition.

Multiple imputation procedures were used to handle missing data in primary and secondary outcome analyses. This approach has significant advantages over single imputation or listwise deletion (35) or other techniques (36, 37) in conjunction with generalized estimation equation analyses and has frequently been used in psychiatric studies with similar levels of missing data (38, 39). Use of multiple imputations requires the assumption of “missing at random”—a more conservative approach compared with the default listwise deletion used in generalized estimating equation analysis, which assumes “missing completely at random.” Preliminary analyses identified 12 variables that predicted missingness due to treatment dropout. We used these variables during the imputation phase to help ensure that our “missing at random” assumption was tenable. While there is no test for whether missing data are truly “missing at random” as opposed to “missing not at random,” our inclusive strategy for auxiliary variables (i.e., variables used during the imputation but not the analysis phase) during the imputation phase made for a tenable assumption that data were “missing at random.” Multiple imputation procedures use a regression-based approach to fill in the missing values to produce multiple data sets. To maximize the efficiency of our standard errors, 50 data sets were analyzed for each analysis. Parameters and standard errors were combined using Rubin’s rules (35). Analyses were performed using Stata, version 11.2 (StataCorp, College Station, Tex.). We performed extensive sensitivity analyses to assess the relative stability of the effect of treatment on the primary outcome measure (simulant abstinence) across different missing data handling techniques. In addition to the multiple imputation analysis, we performed analyses that used listwise deletion and last observation carried forward on both the intent-to-treat sample and the treatment completers only. In addition, we performed latent growth curve modeling that utilized maximum likelihood using Diggle-Kenward selection and Wu-Carroll selection modeling. While we attempted to fit a basic pattern mixture model (36, 37), convergence proved difficult and we were not able to obtain parameter estimates.

Results

The demographic and clinical characteristics of the sample are summarized in Table 1. There were no statistically significant differences between groups.

Participants were considered treatment dropouts if they were absent from nine consecutive study appointments (i.e., 3 weeks) during the treatment phase. Significantly fewer participants in the contingency management group were retained throughout treatment compared with those in the control group (N=38 [42%] and N=55 [65%], respectively; χ²=9.8, df=1, p<0.05). Contingency management participants were retained for fewer weeks (mean=7.25; SD=4.25) than participants in the noncontingent control group (mean=9.33; SD=3.98). Dropout typically occurred during the first 4 weeks (contingency management group: N=34 [64%]; control group, N=19 [63%]).

Analyses conducted on data from the intent-to-treat sample revealed that participants in the contingency management group were 2.4 times (95% CI=1.9–3.0, p<0.05) as likely as those in the noncontingent group to submit a stimulant-negative urine sample during the treatment period (three urine tests submitted per week, for 12 weeks). The proportion of stimulant-abstinent participants (assessed by urine tests) in each group across the 12 weeks of the treatment period is displayed in Figure 1. The sensitivity analyses conducted on the intent-to-treat and treatment completer samples revealed a similar statistically significant group effect on stimulant abstinence. During the follow-up period, participants in the contingency management group were more likely than those in the noncontingent group to submit a stimulant-negative urine test (N=42 [46%] compared with N=30 [35%]; odds ratio=1.4, 95% CI=1.0–1.9, p<0.05 using multiple imputation procedures). However, significant group differences during the follow-up period were inconsistently observed in the sensitivity analyses.

Participants assigned to the contingency management condition reported significantly fewer days of stimulant use during the treatment period (β=2.70, 95% CI=0.91–4.31, p<0.05) and the follow-up period (β=2.16, 95% CI=0.18–3.24, p<0.05) compared with those in the noncontingent control condition. Table 2 provides descriptive statistics for outcome measures that demonstrated statistically significant group differences. Participants in the contingency management group reported fewer days of alcohol use than those in the noncontingent group during treatment (β=2.44, 95% CI=0.60–4.29, p<0.05), but not during follow-up. All other measures of other drug use and Addiction Severity Index composite scores did not differ between groups.

No group differences were observed in HIV risk-associated sexual behavior. Approximately 24% of the sample (contingency management group, N=21 [23%]; noncontingent control group, N=21 [25%]) reported injecting illicit drugs in the month prior to study entry. Participants in the contingency management group were less than one-third as likely (odds ratio=3.3, 95% CI=1.8–5.9, p<0.05) to report engaging in injection drug use during treatment compared with those in the noncontingent group; the groups did not differ significantly on this measure during the follow-up period.

Contingency management participants reported lower levels of psychiatric symptoms on the Brief Symptom Inventory compared with those in the noncontingent condition during treatment (β=0.25, 95% CI=0.08–0.43, p<0.05). They also had lower ratings on the excitement subscale of the Positive and Negative Syndrome Scale (β=0.86, 95% CI=0.11–1.60, p<0.05). The groups did not differ significantly on these measures during the follow-up period. One individual from each group was admitted for a psychiatric hospitalization during the 3 months preceding randomization (length of stay: contingency management group, 24 days; noncontingent control group, 6 days). During the 6 months following randomization, two participants (2%) in the contingency management condition and nine (10%) in the noncontingent condition were admitted for inpatient psychiatric care (χ²=5.4, df=1, p=0.02). The groups also differed in the total number of days hospitalized (contingency management group, 14 days; noncontingent control group, 152 days). The groups did not differ on other community outcomes.

Discussion

To our knowledge, this is the first large randomized controlled trial to investigate the efficacy of a contingency management intervention for drugs of abuse as an adjunct to treatment as usual in a typical outpatient setting. Participants who received the contingency management intervention were 2.4 times as likely as those in the control condition to submit a stimulant-negative urine sample during treatment. Contingency management also had a positive impact on substance use and psychiatric outcomes that were not the primary focus of the intervention. Relative to those assigned to the control condition, individuals who received contingency management experienced reductions in alcohol use, HIV risk behavior (injection drug use), psychiatric symptoms, and inpatient care. The reduced injection drug/HIV risk associated with contingency management in our sample is consistent with previous research in stimulant-abusing adults without serious mental illness (40). Reductions in injection drug use are of particular public health relevance given the relatively high comorbidity of stimulant and injection drug use (approximately 25%) observed in this sample.

Group differences in psychiatric symptoms were corroborated by differences in inpatient psychiatric utilization. Compared with participants in the control condition, those in the contingency management condition were one-fifth as likely to be admitted for a psychiatric hospitalization during treatment. Changes in substance use, psychiatric symptoms, and inpatient psychiatric care observed in this study were equivalent to those reported in previous studies of more comprehensive and costly psychosocial interventions that are currently the gold standard for evidence-based treatment for co-occurring disorders (10–13), suggesting that contingency management in combination with treatment as usual may be a viable alternative to these treatments.

Our data suggest that an effect of contingency management on stimulant abstinence persisted after treatment was discontinued. While results of multiple imputation analyses suggested higher levels of stimulant abstinence in the contingency management group relative to the noncontingent control group during the follow-up period, results of sensitivity analyses yielded inconsistent results (only the multiple imputations technique showed a statistically significant group difference). Lower levels of self-reported stimulant use were observed during the follow-up period by participants in the contingency management group relative to those in the control group. While these differences in self-reported stimulant use are consistent with previous studies in non-seriously mentally ill populations that demonstrated treatment effects up to 1 year after completion (22), this result should be interpreted with caution given the high level of missing data during follow-up.

Group differences in primary and secondary outcome measures were observed even though the contingency management dropout rate was somewhat higher (59%) than has been previously observed in stimulant-abusing populations (approximately 50%) (20). The higher dropout rate in this study likely reflects the psychiatric comorbidity in this sample and the fact that 66% of our sample was homeless. While the groups did not differ in psychiatric severity or homelessness, it is possible that lower-functioning individuals were more likely to consistently attend study sessions when provided with reinforcement for attendance, but not when the additional contingency of abstinence was added. Others have found that treatment completion is improved after patients are exposed to an initial period of noncontingent reinforcement (41). This and other approaches (e.g., providing higher-value rewards, such as housing [42], or adding contingency management to evidence-based treatments for mental illness) might improve treatment retention in this population.

This is the second study in a population with co-occurring disorders to find that contingency management can be delivered at a low cost. The cost of urine testing and reinforcers was $256 per participant for the entire treatment group ($864 for individuals with ≥8 weeks of abstinence). In this sample, individuals assigned to the contingency management condition experienced 138 fewer days of psychiatric hospitalization than those in the control condition. Although few participants in either group were hospitalized (two in the contingency management group and nine in the noncontingent control group) and differences in hospitalization rates may be due in part to chance, evidence from this study suggests that savings related to reductions in psychiatric hospitalization could offset the costs of contingency management.

Despite empirical support, potential cost savings, and characteristics that suggest that contingency management could be disseminated into clinical practice, it has not been fully utilized in clinical practice. The primary barriers to dissemination appear to be financial, rather than clinical or theoretical objections by clinicians (43). While the cost of delivering contingency management increases when individuals respond to the treatment (they receive more prizes), this increase is modest compared with savings in inpatient care demonstrated in this and other studies (14). While a number of innovative strategies have been explored to provide funding for contingency management reinforcers (e.g., use of donated funds/prizes, opportunities to work) (41), it is likely that contingency management will continue to be underutilized until payers provide funding for the costs of delivering this treatment. An example of this type of reform recently occurred within the Veterans Health Administration, where contingency management has been approved as a treatment for illicit drug use in veterans receiving intensive outpatient treatment (44).

The generalizability of our results may be limited because recruitment for the study occurred at one large treatment agency. Methodologically, the lower treatment completion rate in the contingency management condition relative to the noncontingent control condition resulted in group differences in rates of missing data. However, we used robust statistical methods (i.e., multiple imputation), which have been frequently used in psychiatric research where comparable levels of missing data were observed (38, 39), to account for missing data, and we conducted multiple sensitivity analyses to corroborate the findings of our multiple-imputation approach. Despite these consistent results, it is important to note that all imputation strategies bias study results, with some (e.g., single imputation) introducing more bias than others.

Despite these limitations, our results provide evidence that contingency management is an effective technique for reducing drug and alcohol use, HIV risk behavior, psychiatric symptoms, and rates of inpatient hospitalization in seriously mentally ill adults. If financial and other barriers to dissemination can be overcome, contingency management might be an effective adjunctive treatment for this population. Future research investigating the efficacy of contingency management in this population should focus on identifying strategies to improve treatment retention and exploring how contingency management might be optimally combined with other evidence-based interventions to further improve outcomes.