Veterans who (1) are entering an MH RRTP, at either of the 2 VA health care systems; (2) were arrested and charged or released from incarceration in the past 12 months; and (3) have a COD diagnosis in their medical record will be eligible for participation. Diagnoses will be confirmed during screening using the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition [56]. The only exclusion criterion is cognitive impairment, which will be assessed using the Montreal Cognitive Assessment’s section on Orientation. Veterans who are unable to correctly answer the date and location items will be deemed too cognitively impaired to understand the study’s procedures, participate meaningfully in the study, and respond to interviews, and therefore will be ineligible.

To ensure sufficient variability in our outcomes and increase the power to test study hypotheses, power analyses were calculated to determine the sample size needed to obtain a medium-to-large effect size (based on the 3 open pilots with MISSION-CJ) with 80% power [57]. To account for having veterans at 2 sites, power analyses using G*Power were based on having 4 cells (2 conditions per site) [58]. A sample size of 180 (90 in each condition) will provide 91% power to detect a medium-large effect size at an α of .05 (2-tailed test). On the basis of an N of 180 and of .05, we would have 80% power to detect a Cohen f=0.25, a medium size effect [57]. Thus, after accounting for attenuation and 20% attrition [31,59,60], we will recruit 226 veterans (113 per condition). The sample will be stratified by site such that 113 veterans (with rounding) will be recruited separately from each VA site, with 57 veterans (with rounding) entering each condition at each site. The sites were selected because they have MH RRTPs that serve a high volume of veterans with criminal legal involvement and a COD; have criminal legal system outreach specialists, which will maximize recruitment by increasing the pool of veterans that could be referred to the MH RRTPs; and provide geographic variability, which will increase the generalizability of the findings.

MH RRTP staff will screen veterans during their admissions process to determine their eligibility. Potentially eligible veterans will be contacted by a research assistant who will explain the purpose of the study, procedures, and risks and benefits of participation. Veterans will be informed that they will be assigned to either an average of 2 hours of additional programming via a peer support group with community outreach care (EUC) or an intervention which provides an average of 2 hours of additional programming weekly, delivered by a MISSION-CJ treatment team. Veterans will be asked to complete an in-person baseline assessment and 2 follow-up assessments (in-person or via phone) 6 months and 15 months (ie, approximately 1 year post-MH RRTP discharge for those not irregularly discharged) later. Veterans will be compensated US $25 for completion of the baseline, US $50 for the 6-month assessments, and US $75 for the 15-month assessments.

Veterans will be randomized to either MH RRTP care as usual (described in the previous section), with MISSION-CJ or EUC (refer to Table 1 for a comparison of the conditions and Figure 1 for a schematic flow diagram). EUC was chosen to provide an active control that was more ethically and clinically appropriate while still maintaining enough distinction between conditions to adequately test the added value of MISSION-CJ. Both conditions will be rolling entry.

At the outset of the study, the case manager and peer support specialist teams delivering MISSION-CJ will receive training on the curriculum, whereas the peer support specialist delivering EUC will receive training on the MISSION Peer Support curriculum as well as how to perform community outreach guidance. The training will be kept separate for each condition but will use the same format. The staff in both groups will first watch a 1.5-hour webinar on the MISSION-CJ or the EUC curriculum. This will be followed by a 1-day comprehensive training on the components of the MISSION-CJ or the EUC curriculum led by the MISSION developers. The MISSION-CJ training will review the components of the intervention, the risk-need-responsivity model, and the implementation materials. The EUC training will cover the MISSION Peer Support curriculum, community outreach, and the implementation materials. Furthermore, staff will have access to implementation materials (ie, the MISSION-CJ Treatment Manual or the MISSION Peer Support Enhanced Service Delivery Manual; Pinals, DA, unpublished manual, April 2014 and Smelson, DA, unpublished manual, August 2019).

Trained research assistants will administer a battery of assessments (Table 2) with veterans at 3 time points: baseline and at 6 months and 15 months after baseline. All data are entered and managed in REDCap (Research Electronic Data Capture; Vanderbilt University) [61,62]. Each assessment includes questions regarding substance use and mental health history, criminal history and recidivism risk, housing, employment, medication adherence, mutual help groups attendance, antisocial attitudes, affiliations with antisocial peers and prosocial peers, community involvement, and treatment services received.

Health care tracking (mediator) will be conducted weekly and in 2 ways. First, self-reported treatment services received will be tracked via the MISSION Treatment Service Tracking Sheet, which includes detailed logs that quantify the amount and type of services received before and after the MH RRTP and is collected via REDCap. This includes all possible services offered within the MISSION-CJ and the EUC, as well as the community linkages. Regarding linkages, the measure quantifies the number of referrals, referral type, and client engagement. The MISSION-CJ and EUC providers will be prompted weekly to send their logs. The MISSION-CJ services received will be tracked via the MISSION-CJ Fidelity Scale. This scale consists of 27 items (scored yes or no) assessing adherence to the DRT and peer components (Pinals, DA, unpublished manual, April 2014). The fidelity for the MISSION Peer Support curriculum as part of the EUC will consist of a 13-item scale (scored yes or no), which also assesses model delivery adherence (Pinals, DA, unpublished manual, April 2014 and Smelson, DA, unpublished manual, August 2019). Both scales have demonstrated good interrater reliability (Pinals, DA, unpublished manual, April 2014 and Smelson, DA, unpublished manual, August 2019). For MISSION-CJ, we will also use the Critical Time Intervention Fidelity Scale, which includes 20 items rated on a 5-point scale, ranging from “not implemented” to “ideally implemented.” Item ratings can be combined to compute an overall fidelity score [87,88]. Second, additional data will be captured from the VA’s Corporate Data Warehouse (CDW) [89]. These data include MH RRTP completion status, which will be defined as a regular discharge from the program. Data on substance use and mental health continuing care use from the previous assessment will be defined as the number of sessions in each setting that the clients attend throughout the follow-up period, which will also be captured via the CDW. The CDW data includes both outpatient encounter records and information on inpatient and residential care services used. We will extract all health care encounters from the CDW that occur in a 12-month period before, and ending 15 months after, the study enrollment for all veterans.

For the qualitative evaluation, we will interview 50% (12/24) of the veterans from each site who consented to participate in the study and were randomized to the MISSION-CJ group. Half of these veterans will be drawn from those considered high engagers (ie, completed a full dose, 6 months) of MISSION-CJ, and the other half will be drawn from those considered low engagers (ie, dropped out or completed half or less than the full dose, ≤3 months). The dose of the MISSION-CJ will be determined by the number of sessions attended by the veteran. We will also interview 50% (7/14) of the VA care providers (eg, nurses, psychiatrists, and social workers) from each site.

The RE-AIM Planning Tool will be used to conduct semistructured interviews with the MH RRTP veterans and care providers at each site, using 2 different sets of interview guides [90]. The RE-AIM framework highlights 5 domains to evaluate an intervention’s potential for implementation and widespread impact: reach (how to reach the target population with the intervention), effectiveness (how to know the intervention is effective), adoption (organizational support), implementation (fidelity of intervention’s delivery), and maintenance (sustainability) [91]. The RE-AIM Planning Tool asks questions to veterans and VA care providers within each domain regarding issues that should be considered when planning to implement an intervention (refer to Table 3 for sample items). Notably, we will supplement qualitative data with quantitative fidelity data to assess reach and adoption.

Before any outcome analysis, we will examine whether veterans assigned to the MISSION-CJ and the EUC conditions are comparable on sociodemographics (eg, age, race and ethnicity, marital status, education, and income) and baseline measures of the outcomes and mediators using independent group t tests and chi-square tests for continuous and categorical variables, respectively.

To test the effectiveness on the primary (ie, recidivism, risk for recidivism, substance use, and mental health) and secondary (ie, housing, employment status, and medication adherence) outcomes, we will use generalized linear mixed effects regression models to compare the EUC and MISSION-CJ conditions on outcomes. For each outcome, we will estimate a generalized linear mixed effects regression model that specifies the appropriate distribution of the outcome variable in the regression model (ie, normal, binomial, or Poisson) and uses the appropriate link and variance functions [92]. All models will include main effects for time (ie, baseline and 6- and 15-month follow-up), treatment (ie, MISSION-CJ vs EUC), and study sites, and will account for clustering of repeated measures over time within individuals using a random effect for veterans. We will assess the effectiveness of the intervention using a treatment x time interaction term, which will estimate the change in the outcome measure of interest across the 3 assessment points for the MISSION-CJ group as compared to the EUC group. All models will also be adjusted for all baseline characteristics to increase the precision of intervention effect estimates. Because the MISSION-CJ and EUC conditions will be standardized and monitored for fidelity, we do not expect variations in outcomes across sites. However, to address the possibility of contamination, we will ask the participants in the EUC group at the 6- and 15-month follow-up assessments whether they observe the MISSION-CJ information being shared in the MH RRTP. The records of the participants in the EUC group who say “yes” will be flagged, and the contamination-adjusted intent-to-treat analyses will be calculated (ie, the effect of treatment assignment on outcomes is adjusted by the percentage of participants assigned to the EUC condition who may have received the treatment) [93].

To minimize missing data, we will train research assistants on the importance of completing each item while respecting the veteran’s rights to refuse answers. On the basis of previous studies, we expect less than 5% missing data on each variable at each assessment, in addition to any attrition. We will obtain as high a follow-up rate as possible at each assessment. In contrast to the data on the MH RRTP completion and substance use or mental health continuing care use (which will be collected primarily via the VA administrative data at the 6- and 15-month follow-ups), and the data on recidivism (which will be collected as described previously), the data on recidivism risk and health-related outcomes at the follow-ups will be collected by our study team, and will therefore have more missing data due to expected attrition. The strengths of the generalized linear mixed effects regression model include its ability to use all available data, including data on respondents who have not provided data at each assessment. All the veterans who enter the study will be included in generalized linear mixed effects regression model analyses. However, we will also conduct analyses that include only those veterans who attended at least the first DRT and peer sessions. We will use a model-based multiple imputation procedure and impute missing values using least squares regression imputation [94]. We will select a set of measures associated with the variable at issue and use it in a series of iterated least squares regression models to generate a predicted value for the variable being imputed, and then we will substitute the missing value with the predicted value [95].

We will also examine whether and to what extent, reductions in antisocial attitudes, reductions in affiliations with antisocial peers, increases in affiliations with prosocial peers, MH RRTP completion, substance use or mental health continuing care engagement, mutual help group attendance, and community integration mediate the relationship between treatment group and recidivism, risk of recidivism and health-related outcomes at 6- and 15-month follow-ups [13,96]. To test each mediator, we will estimate a series of regression models that correspond to a hypothesized causal sequence among (1) MISSION-CJ, (2) one of our hypothesized mediators, and (3) better outcomes. In each model, we will control for relevant covariates (eg, baseline values for sociodemographic and the dependent variable). In the first model, we will regress the outcome variable of interest (eg, recidivism and the Level of Service Inventory–Revised [LSI-R] total scores) on the MISSION-CJ treatment condition. In the second model, we will regress the potential mediator of interest on the MISSION-CJ treatment condition. In the third model, we will regress the outcome of interest (eg, recidivism and the LSI-R total scores) on the potential mediator. If the coefficient for the MISSION-CJ treatment condition or a potential mediator is significant in all cases, we will proceed. In the final model, we will regress the outcome of interest (eg, recidivism and the LSI-R total scores) on the MISSION-CJ treatment condition and the potential mediator. If the coefficient for the potential mediator is significant and the coefficient for the MISSION-CJ dummy variable on the outcome is reduced, we will conclude that mediation is supported and will evaluate whether the indirect effect is significant.

Deidentified audio files of the veteran and VA care provider interviews will be transcribed using VA Microsoft Teams and analyzed using NVivo (Lumivero) [97]. The research staff will review transcripts of these interviews, taking detailed notes using the RE-AIM Planning Tool structure to guide the analysis. Interviewers will record supplementary observations of initial impressions from the interview. Using an a priori approach informed by AIM combined with thematic analysis, research staff will jointly code several interviews to craft a codebook based on RE-AIM. Coding will involve close engagement with the data, multiple readings, and collaborative iterative defining of the codes [98-101]. We will then organize the codes using RE-AIM into broader themes. We will synthesize the codes into content categories and organize themes for final data interpretation. Drawing on our NVivo database and with team member feedback, we will create tables to summarize and categorize the barriers and facilitators of the MISSION-CJ implementation at the system, treatment care provider, and patient levels, including potential solutions and potential-to-leverage columns that contain evidence-based implementation tools to be considered for more widespread MISSION-CJ use. These analyses will reduce and display the data. With the tables, we will identify a small set of themes within each domain of the RE-AIM framework that signify triangulated findings (eg, factors that would facilitate and maintain use of MISSION-CJ that fit within the clinical and administrative infrastructure of a given facility). We aim to capture in-depth information on the patient, treatment provider, and contextual factors that facilitate or hinder the successful implementation of MISSION-CJ. We will use such factors to inform post hoc interpretation of the findings and guide the development of action plans to be tested in subsequent implementation projects. If hypotheses regarding the effectiveness of MISSION-CJ (vs EUC) are not supported, we will use this analytic approach to identify barriers in delivering MISSION-CJ during the RCT and to determine what modifications could be made to maximize effectiveness.