Evaluation of the Integrated Intervention for Dual Problems and Early Action Among Latino Immigrants With Co-occurring Mental Health and Substance Misuse Symptoms

Key Points Question Would a tailored behavioral health intervention reduce substance misuse and mental health symptoms, compared with enhanced usual care, in Latino immigrants with co-occurring mental health and substance misuse symptoms? Findings In this randomized clinical trial from 3 sites of 341 immigrants with co-occurring mental health and substance misuse symptoms, the primary outcome of substance misuse did not change in the intent-to-treat analysis. Patients who received the treatment statistically significantly experienced decreased mental health symptoms, compared with controls under enhanced usual care, and only participants with moderate to severe symptoms who received the intervention statistically significantly reduced their substance misuse. Meaning The intervention did not change drug misuse in a heterogeneous sample but did improve secondary mental health outcomes, a finding that might provide a path for treating Latino immigrants with co-occurring mental health symptoms whose symptoms are in the moderate-to-severe range.

others in their social network who might benefit from the program. Recruitment was conducted between September 2014 and May 2016 in Boston, Madrid and Barcelona. Approval was obtained from the institutional review boards of participating institutions. We held a series of presentations with Directors and staff in the clinics and community sites, to introduce our staff and the study.
We worked with the sites to determine optimal procedures for outreach and recruitment of patients.

E. SAMPLING, RECRUITMENT AND CONSENT
Research Assistants (RAs) approached potential participants in person, in the waiting room of clinics and community agencies. In certain sites, RAs took contact information for patients and followed up by phone to either screen or administer informed consent, depending on the permission granted by each institutional review board. In some cases, RAs called a list of patients, made available by clinical referral, directors of the site, or by access to electronic data.

E.1 Subject enrollment
Research staff first obtained informed consent from participants and then utilized the study screening to identify patients who were eligible for the study using the AC-OK short screener.
We asked participants' permission to contact them through their PCP and collect information of two close contacts that had had the most frequent or stable contact with the participant over the past six months. We collected an address to send intervention materials and follow up letters and establish contact with the participant.
Patients who were NOT eligible received the short screener and the CAT-MH interview only, and were compensated $20. Patients who were eligible received the short screener and scheduled for a follow up baseline interview. The interview included instruments designed to identify mental health disorders, substance use problems, and HIV risk behaviors, as well as socio-demographics, cultural, contextual and social factors, medication use, chronic conditions, disability in daily activities, health literacy, language proficiency, access to health services, migration history, assessment of illness management/recovery and mindfulness. At baseline, they were also administered the CAT-MH interview and a urine test for drug metabolites to determine drug use. They were administered a capacity to consent form to ensure they understood and could take part in the full trial. They were compensated $40-50 for the time spent in answering the assessment, depending on what assessment they were completing (see below).
Upon completion of baseline, participants were randomized into either the intervention or control condition, and began the trial. Additional interviews were administered by research assistants blind to study group at 2 months following baseline, 4 months, 6 months, and 12 months. At each interview a urine test was administered for drug metabolites.
Emergency protocol: We used an emergency protocol throughout the screening phase and in the trial. If participants responded affirmatively to questions in the screener or interview related to suicidal thoughts, they were administered the Paykel suicidality screener 5 after the interview. Massachusetts, study staff connected the patient by phone or in person to the BEST emergency services team, which performs an assessment of patient safety and provides referrals in the event immediate care is needed. In the two sites in Spain, participants where referred to clinical staff overseeing the study, who assessed and helped connect with emergency care if needed.
Participants with active suicidality could be contacted after 30 days for reassessment to see if they could safely participate in the trial if they were not actively receiving psychotherapy.

Certificate of Confidentiality:
We received a Certificate of Confidentiality from NIDA for the US-based participant in this project.

F.1 Research Assistants
We recruited bilingual and bicultural study staff who would be engaging, non-judgmental, persistent, and not easily discouraged. Given the sensitivity around substance use and mental health issues and potential concerns about immigration status and discrimination, we selected bilingual Spanish/English staff, who were also Latino and were sometimes born outside of the US or Spain, to build trust and familiarity with participants. In Barcelona, most of the study staff also spoke Catalan. All RAs received a comprehensive training to ensure they could conduct the interview effectively and made the participant feel comfortable. RAs followed 4 steps as part of their training 1) review of assessment questionnaire and materials, 2) remote training via WebEx, 3) conduct of at least two successful role play interviews of the sessions; and 4) Quality control of audio recording of first 2 patient interviews, including receiving detailed feedback from a study supervisor. During the study, all interviews were recorded and periodically evaluated under quality control procedures. A minimum of 15% (207 approximately) of all the interviews in the study were randomly selected, distributed to the quality control team and evaluated by the project coordinator of each site for feedback to the interviewer, and corrections if necessary.

F.2. Study Outcomes Measures
All participants in either condition took part in five research interviews (including baseline).
Assessments were accompanied by administration of a urine sample for biological confirmation of substance use. Interview data was audio-recorded using a digital audio recorder and responses were inputted into tablet computers. To collect research data in the tablets, we work with Dimagi, a health care technology company, to implement data collection of the screeners and interviews via CommCare technology. This technology was installed on tablets, and made available to research assistants serving as interviewers so all information was kept secured on Dimagi servers.
Eligible patients were randomized into the intervention and control condition after baseline.
They received follow up interviews and a urine screen at 2 months from baseline, 4 months from baseline, 6 months from baseline, and 12 months from baseline. They were compensated $30 (25€ in Spain) per follow up interview, or $20 if they completed only half (primarily the outcome measures). For follow up at 6 and twelve months we compensated $50 (30€ in Spain). We incorporated administration of the baseline and urine test to a random sample of patients who were NOT eligible, specifically to patients whose study ID ended with "7" but were not eligible.
This helped in the analysis phase to identify differences between patients who were part of the intervention and who did not screen in. They were administered the short screener, then the baseline interview, the CAT-MH as well as the urine test. In the event the interview took place by phone, we arranged for the participant to meet the interviewer at one of the study sites to administer the urine analysis before payment.
We collected the following measures as part of the study, including the outcomes. 9-item screening questionnaire to determine severity of depressive symptoms. Internal consistency was α= 0.85. Spanish version has good agreement with independent mental health professional diagnostic (k = 0.74; overall accuracy, 88%; sensitivity, 87%; specificity, 88%) 59 .

Post-traumatic Stress Disorder -PTSD Checklist for DSM-5 (PCL-5)
The PCL-5 is a 20-item questionnaire, corresponding to the DSM-5 symptom criteria for PTSD. We consider 33 as a cutoff given the latest data. The internal consistency was α= 0.94 61 Drug Abuse -Drug Abuse Screening Test (DAST-10) 10-item self-report instrument that has been condensed from the 28-item DAST. It is designed for clinical screening of substance use. The DAST-10 yields a quantitative index of the degree of consequences related to drug abuse. (α= 0.87) 62

Alcohol Abuse -Alcohol Use Disorders Identification Test -C (AUDIT-C)
World Health Organization screener for excessive drinking. 3question screen that evaluates lifetime and past 30 day alcohol use behaviors. 63 (α=0.78)

Alcohol -Addiction Severity Index Lite (ASI Lite) -Alcohol and Drug -Addiction Severity Index Lite (ASI Lite) -Drug
Shortened version of the Addiction Severity Index (ASI) which obtains lifetime information about problem behaviors, as well as problems within the previous 30 days. We used drug and alcohol questions only (α=0.84 for alcohol and α=0.70 for drugs) 65

HSCL-20 -Hopkins Symptom Checklist-20
The HSCL-20 is a symptom inventory for depression that indicates severity of psychological distress 66 . Internal consistency was α=0.94

Any positive in Urine Test
Using the DrugCheck for drug metabolites with a binary outcome (yes/no) for use of any of this five drug types: amphetamines, benzodiazepines, cocaine, methamphetamine, and marijuana

Smoking -Fagerstrom Test for Nicotine Dependence
The Fagerstrom Test is a standard instrument for assessing the intensity of physical addiction to nicotine, and designed to provide an ordinal measure of nicotine dependence related to cigarette smoking. It contains six items that evaluate the quantity of cigarette consumption, the compulsion to use, and dependence (α =0.75) 69 . Trauma Exposure -Brief Trauma Questionnaire (BTQ) 10-item self-report measure that examines experiences with potentially traumatic events that would meet Criterion A (serious injury/life threat/subjective response) for PTSD diagnosis per DSM-IV. It is derived from the Brief Trauma Interview 70

F.3. Quality control
The ILRP formed a quality control (QC) group to monitor the quality of the data collected over the course of this study. The group consisted of a quality supervisor (Ph. D), quality manager (M.S), quality coordinators (BFA, MSW) and interviewers. The role of the quality supervisor was to serve as a conduit between the quality manager and the Principal Investigator (PI) to convey questions or concerns. Quality coordinators oversaw reporting of any errors or questions the interviewers had while interviewing to the quality manager.
To ensure reliable data through quality control, interviews were assessed every three months during data collection from all three sites. To eliminate potential bias and capture a representative sample of the quality of the interviews, it was determined that every three months the quality coordinators would assess around 15% of all the interviews of the trial. This work entailed between 1-2 hours per interview and additional paperwork. We ran 9 batches of QC interviews. The first batch started on October 2014. We conducted QC on 578 out of 3564 screeners and interviews (16%). Moreover, additional to the 15% QC of all interviews, we worked on inter-site and cross-site reliability of feedback of cases. In addition to the fifteen percent of quality control data, we performed data cleaning activities on a regular basis to identify systematic errors, for example: duplicate identifiers, missing data and wrong skip patterns.
We created a qualitative standard report for the interviewers to report and correct the findings during QC. This document covers: gathering complete and correct information (participant ID, interviewer ID, date of interview and read all the questions, following the order and extra information), minimal noise and adequate pace (appropriate voice tone, adequate tone and without noise and minimal interruptions), ensure instruments work (identify tablet problems and Identify recording problems) and avoid mistakes (interviewer skips questions and interviewer finishes the interview "early").

F.4. HIV and STI testing (NIDA Supplement)
We conducted testing for HIV and sexually transmitted infections (Chlamydia and Gonorrhea) among enrollees of the clinical trial for both control and intervention condition. Testing was offered in person. In the intervention condition, clinicians and study staff offered testing at session 7, which focused on HIV/STI risk reduction. In the control arm, care managers and study staff offered participants testing at the second usual care call.

G.1. Treatment condition
The IIDEA intervention was designed for the treatment of depression, anxiety, PTSD, and substance use problems in Latino migrants over 18 years old. Substances of use include a range of drugs including stimulants (e.g., cocaine), opiates (oxycontin, heroin) and misuse of over the counter and prescription drugs (including benzodiazepines), as well as alcohol. Comorbid mental health disorder and substance use has been found to be a risk factor for HIV, and thus the intervention also addresses prevention of risky behaviors that can lead to HIV/STIs. Patients with severe substance use disorders, which include symptoms of physical withdrawal and tolerance (physical dependence), were not considered appropriate for this treatment until they had received medically needed treatment (e.g., medical detoxification and stabilization) to avoid medical complications that can arise due to discontinuing drug or alcohol use during treatment.
Study staff first took part in a cultural adaption process to make the manual more accessible to interviewing (MI) and motivational enhancement to prepare participants for treatment, but this is a general approach which can help throughout treatment. At each session, the clinician administers a series of questions to assess how the patient was doing in mental health and substance use symptoms. We also connected patients to an emergency responder if they displayed active suicidality based on these assessments.
To decrease barriers to engagement in the IIDEA intervention, we designed the program to be deliverable both in person and over the phone. To this end, we discussed important considerations for phone delivery during the training of study clinicians. For instance, we discussed the need to ask participants whether they were in a private place with no distractions, the importance of vocal intonation, referring to the participant manual, and verbal affirmations to convey empathy.
The manual integrates HIV prevention by addressing risky behaviors that can be related to substance use and mental health disorders. A NIDA supplement provided additional funding to best incorporate this component. Substance use is intricately linked to risk for HIV/STIs, not only through intravenous drug use but also by interfering with judgment, leading to risky sexual behaviors.

Study Clinicians
The IIDEA intervention was administered by clinicians with at least a Master level of training (e) respond to any concerns and doubts the clinicians raised.
Weekly discussion of the supervisors was carried out as part of the weekly teleconference.
During this time supervisors voiced concerns and questions about cases they were supervising, and the lead supervisors provided feedback and constructive criticism to facilitate deeper understanding of the components of the intervention. In addition, any adaptations to the treatment predicated on previous sessions were outlined.

G.3. EUC condition
Control patients continued their usual care routine with their primary care physician (PCP), if they had one. To address potential symptom attenuation, we contacted EUC participants 4 times to administer the same brief assessment used in the clinical sessions to see how the patient was doing in different areas. The control group was overseen by the care manager, who performed the EUC calls once every three weeks after the baseline interview. The care manager also assisted with the referral process to specialized substance use or mental health services, which was provided if the patient requested them.

H.1. Analytical Methods -Data Preparation
Some variables had missing data, due to drop-outs, missing assessments, lack of response to questions, or participant assessment of a question as not applicable. To account for missing data, we used multiple imputation methods in Stata version 14.2 via the mi impute chained command 77 . The multiple imputations were carried out in three steps: first, we imputed the missing data for all the variable considered, creating 20 imputed datasets that each consist of 341 participants with four follow-up assessments per participant. Second, we ran analysis on each individually imputed data set; third, we aggregated the individual estimate to obtain final estimates and adjusted standard errors for the uncertainty due to imputation 78 .Each imputation was carried out using the chained equations method 79 . In what follows, we refer to variables with missing data as incomplete variables and those with non-missing data as complete variables.
Each incomplete variable xj was specified as a conditional function (gj) given the set of all other variables used in the imputation, comprising both incomplete (x1,…,xm ) and complete variables (Z). We fitted the conditional model gj to generate the predicted values of xj using an iterative method. Specifically, each incomplete variable xj was iteratively estimated and in each iteration, the variable xj t was then updated to xj t+1 based on the conditional model. This updated variable was then used in the estimations of the other variables, following the conditional model specification: xj t+1 ~gj (xj│ xj t+1 ,…, xj-1 t+1 , xj+1 t ,…Z, ϕj), for j∈{1..m}, where ϕj j were parameters of the conditional model gj . These steps were repeated for all variables x1 ,…, xm and, after an initial burn-in phase, the procedure was stopped once convergence was reached. The variables used for imputation included outcome variables and their baseline measures, patient sociodemographics, clinical characteristics, and study design variables, such as site and intervention indicators, as well as dummy indicators for each follow-up assessment.
We used interval regressions to incorporate the theoretical bounds of the clinical outcome variables to increase the efficiency and accuracy of the imputation procedure, (e.g., we restricted the imputed values of PHQ-9 to lie within the 0 to 27 intervals). To impute the binary variables, (e.g. whether the urine drug test was positive), logistic regressions were used. Other conditional models were specified to be multiple linear regressions.
After running analysis on the twenty data sets we imputed via step 1, we combined these separate estimates from the different data sets to arrive at our final estimates.

Power Analyses:
We calculated power in multivariate regression models, assuming a squared multiple correlation of 0.20 between the treatment indicator and 20 other covariates, a type I error of 0.05, and a two-sided comparison. Using intervention effect sizes and standard deviations from previous studies, we expected our projected sample (180 treatment and 180 control) to provide 80% and 98% power to detect statistically significant treatment group reductions in the HSCL-20 and PCL-5, respectively. Based on the above-mentioned pilot study finding that brief CBT decreased the percent of calendar days with any drug use and any marijuana use, we expected 90% power to detect treatment group improvements in cannabis abstinence, and 99% power to detect treatment group improvements in abstinence from any drug.
No prior studies were identified to calculate power for detecting GAD-7 and ASI. We also assessed the power to test interaction effects between the intervention and individual-level cultural factors. Assuming we have 25 variables in the model and the R-squared of the model is .20 with the interaction effects contributing .02 to the explanatory power of the model, then we will have 83% power to detect a statistically significant interaction coefficient in the full sample of 360 patients.

H.2. Data Analysis
First, we compared distributions of baseline characteristics between participants who received the IIDEA intervention and participants in the enhanced usual care group (EUC), to assess the balance of the observed covariates. For each follow-up assessment, we compared the baseline covariates to detect any statistically significant baseline differences between those who completed the assessment versus those who did not.
Second, to assess the effect of the intervention, we conducted an intent-to-treat (ITT) analysis, using a multilevel, multivariate regression model to assess changes in outcome variables over time in the treatment and control groups. To account for the nature of longitudinal data, our ITT analysis was carried out by fitting multilevel mixed-effects models to allow for valid variance calculation and statistical inference. The multilevel models we used included random effects at the patient level to account for within-patient correlations, and robust clustered standard errors to account for within-clinic correlations due to patients nesting within the same clinic. Letting denote the outcome measured at time t after the baseline for patient I, we estimated the following model: where the patient-specific random intercept can be written as . We fitted linear models for continuous outcomes and logistic models for binary outcomes. Although scores on the ASI drug and ASI alcohol measures range from 0 to 1, participant scores on these measures were rescaled to a range of 0 to 100 prior to regression analyses (i.e., multiplied by 100). This adjustment was made to ensure meaningful regression estimates.
In the ITT analysis, individuals were assigned to the study arm to which they were randomized.
Thus, is equal to 1 if patient i is randomized to the intervention arm and 0 otherwise. is a continuous measure of time capturing differences in months between assessment t and assessment 1 . Because the intervention ended by the time of the 6-month assessment, we centered time variables at 6-month follow-up in the following way: equals to -4 for 2-month follow-up, -2 for 4-month follow-up, 0 for 6-month follow up, and 6 for 12month follow-up. To model the pattern of outcome changes over time, our primary analysis employed linear spline models to divide the time axis into two segments, and within each segment consider piecewise linear trends (e.g., having a different time trend before and after the intervention ended). Specially, we denote t* to be the month when the intervention stopped, i.e., 6 months after baseline and * to be the post-intervention time trend, which equals to * if * and 0 otherwise. This choice of linear spline models allows for the ability to test whether the time trends differ before and after the intervention was finished and that treatment effect could attenuate over time once participants did not receive more intervention. Since the time variable is centered at 6-month follow-up, the beta coefficient on can be interpreted as the treatment effect on outcome levels evaluated at the end of the intervention. That is, testing for the significance of tests the hypothesis that the treatment was more effective than EUC in reducing substance use disorder and/or mental health problems, as evaluated at 6 months after the baseline. Because a successful randomized control trial balances both known and unknown confounders between treatment and control groups, our primary ITT analyses did not control for other covariates. We tested the robustness of the results by further adjusting for current psychotropic medication use in a sensitivity analysis. The adjusted results did not differ from the main results.
Our study sample was comprised of a broad clinical population, with a substantial number of patients who only had mild symptoms in substance use and mental health problems. To test the hypothesis that the intervention would be more effective among those with moderate to severe symptoms, we extended model (1)  In the context of multisite randomized control trial, we examined whether the intervention was equally effective at the three sites. To do this, we added an intervention by site interaction to Model (1) and tested if the interaction term was statistically significant, with Boston as the reference group. We also estimated whether intervention effects were the same among patients who received the most sessions by telephone vs. in-person. We recategorized participants into four mutually exclusive groups: 1) patients in the control arm; 2) intervention patients who received zero sessions, 3) intervention patients who received most sessions by telephone, and 4) intervention patients who received most sessions in-person. When there was a tie between number of sessions received by telephone vs. in-person, we randomly assigned the patient into either group. Finally, we replaced the dummy-coded intervention variable with this multi-group variable to refit the ITT model.
In secondary analyses, we used dosage (defined as number of treatment sessions received) as the independent variable of interest, categorized as dosage equal to zero (control group), 0-3 sessions (inadequate treatment for the intervention group), 4 or more sessions (adequate treatment for the intervention group). While this analysis no longer relies on random assignment, it serves to provide confirmation of the results from the intent-to-treat analysis and provides further estimates of the magnitude of the intervention effects on the outcomes evaluated at 6 months after the baseline. Because the dosage analysis relies on actual treatment received, we used the original non-imputed sample for the analysis. To check the robustness of the dosage analysis, we conducted a third sensitivity analysis with an alternative categorization of treatment dosage. This alternative dosage variable has three categories: zero (control group), 0-5 sessions (intervention group) and 6 sessions or more (intervention group). We considered completion of treatment if the patient received 6 sessions or more, as 6 sessions cover the core components of the intervention (i.e. cognitive restructuring exercises, mindfulness practice, relapse prevention, etc.). Separating those with six or more sessions will allow us to examine the treatment effect for those who completed the core components of the intervention. We also performed separate sensitivity analyses to ensure the robustness of the results to alternative modeling strategies, estimation methods and how the missing data was handled.
Our first set of sensitivity analyses explored alternative methods for handling baseline response.
Our primary analysis was carried out through analysis of covariance, which analyzes postbaseline responses, and makes an adjustment for the baseline response by including it as a covariate. In the sensitivity analysis, we retained the baseline response as part of the outcome vector and assumed the group means were equal at baseline, as is appropriate in a randomized control trial. The analytical data was extended to a longitudinal dataset, consisting of 5 repeated measurements per person, with baseline assessment included as an additional repeated assessment. Next, we re-estimated the model (1) with this new dataset. Since the baseline was added as additional time point, the variable in this analysis, equals to -6 for baseline, -4, -2, 0 and 6 for the two, four, six and twelve-month assessments. We chose to present analysis of covariance as our primary results due to its potential efficiency gain. Because the baseline value has been obtained before any study intervention, i.e., the mean response that baseline is independent of treatment assignment, adjustment for baseline through covariance analysis will be more efficient, as it yields estimates of treatment effects with smaller standard errors 80 .
Our primary ITT analysis modeled the nonlinear trend by fitting piecewise linear trends before and after treatment completion. We explored alternative modeling strategy in sensitivity analysis where the post-treatment trend in the model (1) was replaced with a quadratic term of time trend to model the non-linear time trend. Linear spline models were finally chosen because they generally provide a flexible way to accommodate many non-linear trends that cannot be approximated by simple polynomials in time 80 .
Next, we checked the robustness of the results with respect to estimation method. Our primary ITT analysis used multilevel mixed-effect models to account for inter-participant correlation due to repeated measurement. Our sensitivity analysis instead used generalized estimating equations (GEE). GEE estimation will produce unbiased estimates if the underlying correlation structure is correctly specified. However, it has the limitation of being less efficient than a properly specified mixed model and does not allow weights to vary within clusters/panels 80 .
We also performed a sensitivity analysis using a propensity score weighting approach to balance the treatment and control groups on residual differences within each site after randomization. To do so, we estimated a logistic regression model of membership in the treatment group conditional on baseline measures and interactions between baseline measures and site, and generated the predicted probability (phat) of being assigned to the treatment group. In multivariate regression models, treatment group participants were given a weight of 1/phat whereas control group participants were given a weight of 1/1-phat to balance the groups on their predicted probability of being in the treatment group. Propensity score weights were then applied to all regression analysis. The weighted results remain similar to our main results.
Finally, we reran the dosage analysis with imputed data as a sensitivity check. In comparison to list-wise deletion, estimates of the imputed data from multiple imputation have slightly larger effect sizes (available from the authors upon request). This is because the intervention effects were greater among patients with moderate to severe symptomatology compared to those with mild symptomatology. Patients with missing observations, i.e., those who did not complete all follow-up assessments, had higher baseline drug symptom severity but where similar in baseline measures. Thus, by including these patients when using multiple imputation, the estimated intervention effects were amplified. Our primary analysis, which used non-imputed data and listwise deletion, provides a conservative estimate of the actual effect of the treatment dosage.

I.1. Risks and Discomforts
We did not anticipate noteworthyrisks associated with the proposed study. Minimal risks included the possibility of discomfort when discussing mental health, substance use, or confidential HIV/STI-related problems with the clinicians (for intervention arm) or during research assessments (for both intervention and EUC groups). We indicated to participants in the treatment condition that they could become upset in discussing their frustrations dealing with behavioral health concerns or seeking adequate care for dealing with these problems. Participants being administered screening or research assessment instruments could experience mild emotional discomfort in responding to sensitive questions in the interview.
Another possible risk was that some participants may feel uncomfortable answering certain questions or may feel a burden of answering questions. Respondents were told during the research assessments that they had the option of terminating the interview at any time or not answering specific questions. The interviewers were instructed to implement short breaks during the interview if the respondent became fatigued or commented about the length of the interview.
Every precaution was taken to maintain all rights and privacy protections. A data release agreement was signed by all investigators who worked with the data in any way. Because the goal of the study was to report findings based on aggregate data, all individual information obtained was held strictly confidential. These inherent risks are typically assessed to be low in comparison to the long-term potential benefits from this type of research. If patients endorsed suicidality in either of the research assessments, research assistants followed the emergency protocol described above.

I.2. Benefits:
Participants in the intervention arm could possibly achieve decreased substance use, improved depressive, anxiety, and/or PTSD symptoms, decreased HIV/STI risk behaviors, and/or improvements in global functioning. Participants in the intervention arm, as compared to the EUC condition, might be better able to recognize and self-manage their behavioral problems, as well as increase their ability to deal effectively with structural barriers (e.g., coping with cravings, stigma) that could impede them from entering and staying in behavioral treatment.
We hoped that the results of the study could provide evidence of the effectiveness of the IIDEA intervention for Latinos migrants with co-occurring disorders and decrease the unmet need for services. Participants randomized into the EUC group also benefited from support by research staff such as coordination of primary care appointments and referral to behavioral health services and/or HIV/STI testing.

J. MONITORING, QUALITY ASSURANCE AND ETHICS
Collection protocols were established to ensure accuracy and quality of the data obtained from all participant interviews by research assistants and treatment sessions by clinicians.
Research assistants have regular meetings and conference calls during which data quality is a primary issue. These calls provide a forum for RAs to discuss issues and concerns pertaining to data collection protocol as well as a time to provide feedback on completed sessions to clinicians. Training at the start of the program and ongoingly ensures RAs are prepared to skillfully conduct interviews and respond appropriately to patient needs. Quality of the clinicians' work with intervention participants has been explained above (see G.2. section).

J.1. Participant Confidentiality
We made make every effort to ensure that participant data was safely stored. Each participant in the study was assigned a unique ID number. All documents that included private information (i.e. names, addresses or telephone numbers) were separated from de-identified research materials. These items were stored by research staff in a locked file. Project staff were the only people with access to the locked filing cabinet and/or any information or files that linked the case number with any identifying information. No reports were made public using any names or identifying information. Computerized data was identified by ID number only. The deidentified information that was transferred between study sites was be stored on a secured central server.
As part of hospital and university network, all research sites protected research data per research regulations and site data-related regulations. Sites adhered to strict data safety guidelines, including implementing network firewalls, antivirus systems, internet screening, auditing systems and network intrusion detection systems. All research staff computers were password-protected and had standardized, full featured software and hardware configurations.
Only authorized research staff had access to the data. Emails containing potentially sensitive patient information were encrypted. Research staff identified patients by ID number in emails unless it is necessary to include patient's protected information (i.e., when updating primary care clinicians in a crisis).