Effectiveness of a Suicide Prevention Module for Adults in Substance Use Disorder Treatment

Key Points Question Does a secondary prevention module in community addiction treatment improve patients’ suicide knowledge, attitudes, and help-seeking compared with usual care? Findings In this stepped-wedge cluster-randomized clinical trial including 906 participants at 15 treatment sites, the Preventing Addiction Related Suicide (PARS) prevention module produced consistently greater improvements compared with usual care in suicide knowledge and a greater reduction in maladaptive attitudes about suicide across all time points, and greater improvement in help-seeking by 6 months. Meaning These findings suggest that PARS was effective in improving suicide prevention outcomes and has the potential for wide dissemination and implementation.

A. Background and Significance 1. Suicidal behavior is prevalent and costly in substance-abusing populations. Suicide and suicidal behaviors are over-represented in populations with substance use disorders (SUDs) compared to the general adult population. Recent reviews find that the risk of suicide is 14 times higher for people injecting drugs, 10 times for alcohol use disorders, and 17 times for polydrug users. 2,3 Clients receiving alcohol treatment are about 10 times more likely to endorse of a lifetime history of suicide attempts (43%) 4,5 compared to a nationally representative sample of adults (4.6%). 6 Moreover, prospective data shows that individuals in addiction treatment had five times the odds of suicide attempt over five years compared to those not in treatmen . 7 Suicidal behavior places a heavy burden on the health services system, resulting in more than 650,000 hospital visits and $2 billion in health care costs each year, while suicide deaths result in an annual economic burden of $44 billion. 13,14 Thus, consistent with NIDA's mission to identify and test populationlevel approaches for the prevention of drug-related problems, there is a need to develop and evaluate effective behavioral treatments that can be easily and widely implemented to reduce suicidal behaviors in substanceabusing populations. PARS is a prime candidate for such research, as it has shown promising results in Stage I research (i.e., intervention generation and refinement, pilot testing, and feasibility) conducted in community treatment settings (R21 DA026494). Funding the proposed Stage III research (i.e., "efficacy in the real world") trial would further evaluate PARS' effectiveness and utility in community settings.

Community addiction treatment is an ideal setting for targeting suicide risk in this high-risk group.
Every year, approximately 2.5 million people in the United States enter specialized addiction treatment programs. 11 By far, the most common modality of publicly funded addiction treatment available is group-based Intensive Outpatient Programs (IOP). 12 Thus, adding evidence-based, transportable suicide prevention strategies into the standard IOP treatment package has the potential to reach an enormous number of people who are at very high risk for suicide. Moreover, entering addiction treatment may represent a key window for intervention to reduce suicidal behaviors, as this transition is marked by high rates of suicidal thinking and behavior. Individuals often enter addiction treatment in the context of multiple increased risk factors for suicide: when substance use is out of control and/or is resulting in particularly severe impairment (e.g., marital or financial difficulties, severe depressive symptoms). 15 Between 10% and 40% of clients entering addiction treatment report suicidal ideation with a plan. 16,17 Roughly one out of every 25 clients entering addiction treatment report having made a suicide attempt in the 30-day period before treatment, 18 while one in four report having made a suicide attempt in the past year. 19 Moreover, suicide risk is known to be highly fluid, 20 and although most IOPs will screen patients for suicide risk at the outset of treatment, suicidal ideation and behaviors are likely to fluctuate over the course of treatment, particularly if high-risk situations such as relapse occur during treatment. 11 Thus, addiction IOPs will frequently treat people who are or have recently been suicidal, 18 and have the potential to directly intervene to reduce these problems. Clients with addiction also connect with each other during treatment, in twelve-step meetings, and in drug use. Improving accurate information and adaptive attitudes toward suicide prevention as well as how to effectively reduce risk and reach out for help may not only increase their access to care if suicidal but also increase access of their friends and family who are often also at risk.

Addiction treatment providers need additional training to prevent suicidal behavior.
Unfortunately, most chemical dependency counselors feel unprepared, inadequately trained, and uncomfortable addressing the issue of suicide. 21,22 Recently, there have been several efforts to respond to this need. For example, SAMHSA recently developed the Treatment Improvement Protocol number 50 (TIP50), 11 which provides best-practice guidelines for counselors and program administrators to effectively assess, manage and respond to acute suicide risk within addiction treatment programs. The TIP50 is associated with significantly increased staff self-efficacy, suicide-related knowledge, and suicide prevention behaviors. 23 Moreover, some states (e.g., Washington and Kentucky) have recently passed legislation requiring suicide prevention training for all human services personnel, including addictions counselors. 24 While these efforts are good first steps, they primarily focus on managing suicidal crises (e.g., assessing risk and determining when to refer the client to a higher level of service). Instead of aiming to treat acutely suicidal people, PARS aims to deliver an upstream prevention program to an at-risk population. This selected suicide prevention program thus has the potential to serve a dual purpose of providing prevention for clients, and providing ongoing education and training for the addiction treatment staff tasked with delivering the program. 4. Built-in therapist adherence and fidelity to model. One of the challenges of implementing suicide prevention programs in community treatment settings is the limited resources for ongoing training. PARS was developed to easily fit within the daily work and training models of community IOPs. PARS is delivered in a single IOP session (typically, a 3-hour group therapy session) using a detailed PowerPoint presentation (developed by the R21) to provide structure, psychoeducational content, and defined discussion periods. Clients are taught to recognize the warning signs for suicide in themselves and others and how to respond effectively by seeking support and treatment. Pilot testing showed that naïve community addiction counselors could be trained to competently and adherently deliver PARS in a single three-hour training session that utilized the same set of PARS PowerPoints that counselors later used in an IOP session with addictions clients. In this way, the PARS training model promotes faithful administration of the program without unduly burdening clinical and administrative staff. Counselors rated this model of training as very acceptable (62-69% "strongly agreed" that PARS would be acceptable and appropriate to addictions counselors, and would be beneficial to clients). Moreover, by repeatedly delivering PARS with each new IOP group cohort, counselors will be re-exposed to suicide-relevant risk assessment and help-seeking strategies, potentially continuing to improve their comfort and competence in discussing and determining suicide risk, and intervening when appropriate.

A selected prevention approach can expand impact and reach.
Although the empirical base for interventions to reduce suicide has rapidly developed in the last 30 years, many of these interventions have either aimed at 1) reducing suicide rates in an entire populations (universal prevention, e.g., suicide awareness media campaigns), limiting our ability to detect immediate and short-term effectiveness, or 2) only provided to individuals already presenting with acute suicidality (indicated prevention, e.g., individual cognitive-behavioral therapy for substance users with suicidal ideation or recent attempts 25 ), limiting their generalizability. In contrast to these approaches, a selected prevention strategy can have greater reach and impact by offering a compromise between these approaches. Selected prevention aims to intervene before suicidal thinking or behavior develops. Rather than targeting the entire population or the specific individual, selected prevention targets all individuals within a pre-selected high-risk population-in this case, adults presenting for addiction treatment. Because selected prevention assumes that not every individual receiving the program will experience the problem of interest (i.e., suicidal ideation or behavior), it aims to increase knowledge and attitudes that will facilitate help-seeking on behalf of the client as well as his or her peers, increasing the reach of PARS to include other individuals in the client's social network. Previous randomized controlled trials have shown that comparable, school based selected suicide prevention programs upon which PARS was based result in significantly increased rates of help-seeking on behalf of self and others 26 , as well as lower rates of suicidal thoughts, threats and attempts in high school students. [26][27][28][29] 6. Changing behavior through changes in beliefs about suicide has been effective. PARS was developed by adapting existing, empirically supported suicide prevention programs to fit substance use settings and populations. The three programs that informed PARS are: 1) Signs Of Suicide (SOS) 26,27 , 2) Counselors Care, Assess, Respond, Empower (C-CARE) [28][29][30] , and 3) Coping and Support Training (CAST). [28][29][30] Each of these programs aims to promote help seeking through two primary mechanisms. First, by providing education about warning signs for suicide, the programs are expected to increase recognition of depressive and suicide-related symptoms, which in turn is expected to promote help seeking. Second, by reducing stigma and promoting more adaptive attitudes toward suicide, the programs are expected to reduce barriers to help seeking. A broad definition of help seeking is encouraged, including not only referrals for a mental health professional but also seeking support from other resources (e.g., case managers, crisis lines, loved ones). Evidence from previous trials supports these mechanisms of change. For example, in a randomized controlled trial of the SOS program in high school students, reductions in suicide attempts in the treatment group were mediated by self-reported increases in suicide-related information and adaptive attitudes. 26 Specifically, more adaptive attitudes and more accurate knowledge of suicide risk factors were each uniquely and significantly related to lower probability of suicide attempts, and accounted for approximately 40% of the variance in treatment outcomes. An important next step is to examine whether a similar mechanism of change explains treatment effects in substance-using populations. 7. PARS was developed to be transportable, disseminable, and community-friendly. PARS is a psychoeducational behavioral treatment that aims to increase knowledge, attitudes, and behaviors that can promote recognition of and help-seeking for suicide risk among clients receiving addiction treatment. PARS is taught as a single three-hour module that is integrated within standard IOP therapy group treatment. From its inception, PARS was developed as a community-friendly suicide prevention program that can be easily incorporated into existing addiction treatment programs and agencies, consistent with NIDA's strategic priorities. All feasibility testing of PARS was conducted in community treatment programs. 1 Prior funding (R21 DA026494) enabled Stage I treatment development research, during which existing behavioral suicide prevention treatments were adapted and modified for substance abusing populations (Stage IA), and pilot and feasibility testing was conducted (Stage IB). The proposed project will move this promising line of research to Stage III by conducting a large-scale, experimental test of the effectiveness of PARS in real-world, community IOP settings. Moving from Stage I to Stage III is appropriate when the intervention was developed in community settings, when Stage I has produced promising findings as well as established methods to ensure fidelity of delivery and therapist training materials, and when promoting implementation is a major goal of the research. Consistent with NIDA's research priorities, our proposal includes explicit examination of the putative mechanisms of behavioral change that were highlighted in Stage I research. In sum, the proposed research takes a critical next step by evaluating the real-world efficacy of a behavioral treatment that has the potential to reduce a prevalent, important, and costly behavior that is a major cause of death for individuals with SUD. 31 B. Innovation 1. Innovative use of IOPs to reduce risk of suicide: PARS is an innovative approach that allows addiction treatment agencies to act as key players who can reduce and possibly prevent suicide in one of the most highrisk for suicide populations in the USA. 8 To our knowledge, this proposal is one of the first studies to use a randomized design to examine the effectiveness of suicide prevention within addiction IOPs. 2. Innovative dissemination and health services impact: PARS is innovative in being developed and evaluated in "real-world" clinical sites that provide both public and private addiction services. PARS was designed based on clinical and administrative input to not only fit the recovery philosophy and clinical approach of IOP settings but also to fit into the billing and employee models of community addiction agencies. This provides high likelihood that PARS could be readily implemented, allowing PARS to reach millions of people who seek treatment every year, 11 should it prove to be effective. 3. Use of innovative research design: The goal of this Stage III trial is to evaluate the efficacy of PARS while being responsive to the unique challenges of conducting randomized, controlled research in real-world, community treatment settings. In line with Stage II trials, we strive for rigorous research methods that maintain adequate internal validity. This study is therefore unique in that it combines a rigorous, randomized design with a disseminable suicide prevention program that targets a very high-risk population. In striking a balance between these two goals, we propose to use a stepped wedge randomized trial design. This design allows for a sequential roll-out of PARS to all of the community agencies enrolled in this study by the end of the trial, but promotes internal validity by randomly assigning each site to a "step" which will determine the timing of its transition from control (i.e., TAU) to treatment (i.e., PARS). Thus, while each client experiences either PARS or TAU only once, clinics differ in their exposure to PARS based on the timing of their transition to PARS in the stepped wedge design. This design has a number of benefits over the traditional RCT. Like cluster randomization, it allows randomization at the site rather than individual level which fits with a Stage III evaluation of the intervention as implemented in community treatment. Furthermore, stepped wedge trials facilitate the examination of dose-response or delay effects at the clinic level by modelling the association between the time clinics spent in the PARS phase and the effectiveness of PARS on clients. Finally, a stepped wedge design facilitates complete implementation of PARS among our 15 sites, enhancing the clinical practice benefits of this trial for our community partners. 4. Explicit examination of treatment outcome mediators: Because the proposed research directly examines potential mediators of treatment effectiveness (i.e., accurate information and adaptive attitudes mediating improved help-seeking behavior), this study can guide future efforts to improve the PARS program. Identifying mechanisms of treatment action may allow us to refine PARS; making it more potent and targeted by increasing the emphasis on challenging maladaptive attitudes, increasing factual knowledge, or both. 5. Built-in counselor re-training: As noted above, PARS could represent an innovative training strategy for counselors who are consistently updated in these competencies by repeatedly using skills taught through PARS training with each new IOP group cohort. Few studies have examined the effects of repeated training exposure on suicide prevention effectiveness for clients, thus our exploratory aim examining dose effects may make an important contribution to this literature by testing whether there are effects of repeated exposure to PARS at the clinic level on client outcomes.
As part of a NIDA R21, a pre-post pilot study of PARS was conducted with clients attending group-based IOP addiction treatment at one of three publicly funded addiction treatment agencies in Washington State. 1 All agencies were members of the NIDA Clinical Trials Network (CTN) and were in urban areas. Prior to PARS, none of the agencies included a suicide module in their IOP programs (for more detail on IOP in Washington State, see Design section C.3). Seventy-nine clients were approached, of whom 78 consented to participate. as feasible and acceptable. Sites represent urban and rural areas in Washington as well as representing sites primarily funded by private insurance and self-pay and those primarily paid by Medicaid.

Inclusion and Exclusion Criteria
Recent suicidal behavior or suicidal ideation is not an exclusion criterion in this study. (While it was in our pilot study, no participant was excluded on this basis.) The scope of chemical dependency counselor's practice in Washington State is to screen and refer acute suicidality to other professionals -often this means outside of the IOP program if a licensed mental health counselor or psychiatric provider is not available. Suicidality is not a static phenomenon -it waxes and wanes for the suicidal individual depending on their internal state and external circumstances. In addition, many individuals do not disclose their suicidality or their suicidality is not acute in while in IOP although it may have been before or after. These individuals are standardly part of IOP programs and they are part of the mission of PARS as are those who have never been suicidal but are at high risk by virtue of substance abuse history severe enough to be in treatment. Thus, all addiction clients regardless of suicidal thoughts or attempt history will be included in the study if they are in treatment at the time of recruitment. We have reviewed with our agency partners the expected incidence of acute suicide risk during a given IOP step. They report this is a rare event. Out of over 200 clients in 15 sites receiving IOP at any given time, they estimate that that 1-3 clients (~1%)are acutely suicidal and may have been referred out.

C5. Recruitment and Consent
The sites in this study vary from 1-3 IOP groups running simultaneously. Therefore, based on a consensus of the research team and community agency for that site, one IOP group in each site will be identified as the best fit for this study based on likely longevity of the counselor over the course of the study and engagement of that counselor with the study and interest in learning PARS. If this identifies more than one group, the decision will include the group with maximum attendance. As noted above, all counselors at the site will be trained in PARS (standard PARS training model) and multiple counselors will be trained to fidelity to assure PARS is administered on schedule to the identified group. Counselors may or may not administer PARS in the other IOP groups as decided by that site. This will not impact the study treatment.
Potential client eligibility will be determined by their IOP counselors. All clients will be asked at the end of a standard treatment group in their IOP if they would like to volunteer for a study that involves staying after that group (or participating before or after one of the other IOP groups that week, if that day is not convenient). If interested, the study will be described to them by the research staff. Interested individuals will be asked to sign the consent form and then complete the baseline assessment.

C.5. Measures
Primary outcome assessments will include knowledge and attitudes toward suicide, and participants' helpseeking behavior for themselves and their family and friends. Exploratory outcomes will be suicidal ideation, threats, and behavior as measured by the Suicidal Behavior Questionnaire -Revised. Potential moderators will also be assessed including measures of drug and alcohol use, depression, physical and mental health functioning, and demographic characteristics. To promote compliance with follow-ups, outcome measures will be delivered via brief, online questionnaires or text message surveys with an option to complete questions via telephone call if participants prefer this. To keep the follow-up measures short, we selected a subset of items used in our initial pilot study that demonstrated optimal psychometric properties. (See Appendix for copies of non-standard measures) Accurate Information about suicide. The PARS Suicide Knowledge Scale, 1 which was adapted from the Staff Suicide Prevention Survey 45 in our previous pilot trial, was condensed to 6 well-performing items that closely map onto the content of PARS. The Knowledge scale assesses factual understanding of warning signs, Inclusion Criteria 1. Enrolled client in one of the community treatment settings 2. Over 18 years of age 3. Ability to understand written and spoken English Exclusion Criteria 1. Any clinical medical/psychiatric condition, severity of that condition, or life situation that in the opinion of the counselors or Drs. Comtois or Ries would compromise safe and voluntary study participation (e.g., psychosis, custody conflict). time window since the previous assessment.) The SBQ-R has excellent internal consistency ( = .83), sensitivity and specificity (.80 and .91, respectively), and convergence with the SHBQ (r = .77, p < .01). 46,47 Drug and alcohol use. To assess baseline substance use problems, participants will complete the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST), 48 an 8-item screening tool developed by the World Health Organization to examine lifetime and recent (past three months) substance use, problems resulting from substance use, and risk of current or future harm from substance use. The ASSIST demonstrates acceptable to excellent test-retest reliability (r = .58 to .90), rater agreement (kappas = .61 to .78) and acceptability in substance users. To assess alcohol use problems at baseline, participants will also complete the Alcohol Use Disorders Identification Test (AUDIT) self-report, 49 a 10-item questionnaire that examines frequency, severity, and negative consequences of alcohol use. The AUDIT has demonstrated excellent internal consistency ( = .80 to .98), sensitivity (.74 to .92) and specificity (.71 to .94) in detecting hazardous drinking and alcohol related problems (e.g., hospital admissions, social problems, outpatient treatment, medical disorders). [49][50][51] At each follow-up, participants will be asked to report on how many of the past 30 days they used drugs or alcohol.
Depression, Physical and Mental Health. Two screening measures will be used to assess covariates and potential moderators (depression, physical health, mental health) of PARS outcomes. The two-item Patient Health Questionnaire (PHQ-2 52-54 ) assesses recent depressive symptoms, and has excellent sensitivity (.77 to .86) and specificity (.78 to .95) in detecting major depression. 52,53 The Short-Form 12 Health Survey (SF-12) consists of 12 items assessing overall physical and mental health, and demonstrates excellent test-retest reliability over two weeks (r = .76 to .89) and convergence with other health measures. 55

Treatment engagement will be measured by a combination of three metrics (a) % sessions attended, (b) graduated from program or terminated, and (c) clinician rating of participant's engagement in IOP treatment. Discharge plan for each client will be determined by a query of the IOP medical record as well as clinician report. (See PARS IOP Discharge Ratings in Appendix).
PARS Acceptability. Addictions counselors will complete the PARS Counselor Acceptability Scale, 1 a 13item survey that was developed in our R21 trial to measure acceptability, ease, and perceived effectiveness of incorporating PARS into day-to-day IOP procedures where it had excellent internal consistency ( = .99).

C.6. Procedures
Baseline Assessment will be conducted by in-person group administration as described in Recruitment above. If the individual does not have time or requests an individual assessment, this will be offered at a later time or in a different room at the same time. Assessments will be completed on tablet computers.
Post-Intervention Assessment will be conducted after the PARS or TAU group session and include just the information and attitudes items as too little time has passed for other behavioral measures to be relevant.
Follow-Up Assessments will be conducted 1, 3, and 6 months after baseline assessment. Assessments will be completed using online assessments using their smart phone, text message responses, or a computer. If a participant prefers or has any technical problems, follow-up assessments can be conducted by phone. Dr. Comtois' current trials use these modalities to complete questionnaire follow-up assessments with over 500 Soldiers and Marines. Online methods are extremely effective with a very mobile population such as Service Members who move while active duty and as they separate from the military and also for those in addiction treatment who are often moving residences or homeless.
In Drs. Comtois' and Ries' ongoing trial of caring contacts via text message with suicidal active duty military personnel, we are utilizing a text messaging platform designed and maintained by that is capable of administering surveys via text message. We have demonstrated through extensive testing and implementation that the system is fully capable of obtaining follow-up survey responses via text message. The system records every text "question" sent by the system reliably as a success or failure (i.e., the SMS "succeeded" in delivery to the participants' mobile phone, or the SMS "failed" to go through). This function is highly reliable and our team is well-practiced in addressing any issues with undeliverable text messages. Our group uses Research Electronic Data Capture (REDCap) online surveys to reliably collect online survey data via any device with an internet connection (e.g., smartphone, tablet, desktop computer). As described in the Human Subjects section, both proposed data collection systems are HIPAA compliant and designed for clinical and research applications. 56 (See Letter of Support from team.) Steps to Prevent Participant Attrition. Our goal is for all participants to complete all outcome assessments regardless of treatment participation. Informed consent will include a separate tracking consent form on which participants choose tracking strategies they consent to have used (e.g., obtaining forwarding address from post office, checking social media) and provide alternative contacts to whom the study can reach out in case the participant moves or changes contact information. We have used this tracking consent form for over 20 years of research. It finds an excellent balance between obtaining detailed information and allowing participants to only provide information they are comfortable providing. In our previous studies, we have achieved 80% follow-up in the six months following an ER admission for self-injury in a past study 32 and an 82% six month follow-up rate for our current study of suicidal Marines and Soldiers. In addition, we have conducted a literature review of methods to achieve high subject retention in substance abuse studies. [57][58][59] While we already use most of these procedures, we will follow all the recommendations to minimize attrition in this study.
Participants will be reimbursed (in a choice of gift cards) $30 for their time for each baseline, $20 for posttreatment, and follow-up assessments in increments of $10 for each 10 questions completed. To minimize attrition, participants will also be offered an addition incentive to be paid at the final 6-month assessment (or end of the 6-month assessment window if they do not complete it). The additional incentive (also in gift cards) will be $20 for completing 2 of the 3 outcome assessments and $30 for completing all outcome assessments.

C.7. Data Analytic Plan
Data screening and preliminary analyses. First, exploratory data analysis will be performed to characterize the distributional characteristics of all variables. Frequency distributions and plots will be examined to identify out-of-range values and to assess data distributions. To assess scale reliability and validity, item structure will be examined using Cronbach's alpha and factor analysis. Descriptive statistics, stratified by site and study condition, will be examined to characterize sample characteristics and assess randomization. Baseline characteristics including socio-economic status, gender, suicidality, or substance use found to vary significantly between study conditions will be included as covariates in subsequent outcome analyses. Plots of mean outcomes by assessment point, site, and study condition will be examined to characterize outcome trajectories over time and inform the parameterization of time in the longitudinal models. Because differential attrition can compromise inference about experimental effects, we will compare dropouts and completers on demographic characteristics and available outcome variables.
Primary outcome analyses. In the proposed stepped wedge design, 15 addiction treatment sites that initially provide TAU (Step 1) will be randomized to implement PARS beginning at one of five ensuing steps (Steps 2-6), 3 sites at a time. Participants recruited at each step will be assessed at points: baseline, post-treatment (information and attitudes only), 1, 3, and 6 months. The primary outcome analyses will be intent-to-treat analyses that include all participants, including those with incomplete follow-up data. The primary outcomes will be 1) knowledge about suicide, 2) maladaptive attitudes towards suicide, and 3) help-seeking behaviors. Each outcome will be evaluated in a separate generalized linear mixed model (GLMM) 60 to evaluate the effect of PARS program vs. Treatment As Usual on change in each outcome over time. The GLMM approach accounts for within-individual and between-group variance and is well-suited for the analyses of multisite, repeated measures data. A logit link function will be used to model dichotomous outcomes and a Gaussian link function when the outcome is relatively normally distributed.

In this model, each outcome is regressed on Step, Setting (Rural vs. Urban), Treatment (PARS vs. TAU), Time (in months), and the Treatment by Time interaction. The effect of Step will be coded into contrasts of
Step 1 against each subsequent step (i.e., Step 2 vs 1, Step 3 vs 1, etc.). The effect of Time will be coded into contrasts of baseline (BL) against each post-baseline assessment (i.e., BL vs. 1 month, BL vs. 3, BL vs. 6 months). To account for the cluster-randomized design, a random effect for site (u 0s ) will be included to model the correlation of individuals from the same site. To account for the repeated assessments, a random effect for participant (r 0is ) will be included to model the correlation of observations within the same individual. The statistical test of the treatment effect will be the magnitude and statistical significance of the Treatment by Time interactions. Comparable analyses will be use for exploratory aims with suicidality and substance use.
Since this study will be carried out at 15 sites, there is the potential for site-specific effects. We anticipate intervention effects to be comparable across all clinics since they are all serving similar client populations and have counselors with comparable levels of experience and training. To verify that treatment effects are comparable across site, preliminary outcome analyses will incorporate indicator variables for site and their interactions with the treatment effects. A likelihood ratio test will be used to compare the models with and without site-specific effects. If the likelihood ratio tests reveal differences in treatment effect by site, these site-specific parameters will be retained in the final analyses, otherwise they will be excluded.

Secondary Analyses
Moderation Analyses. We will examine whether individual-level factors (e.g., demographic characteristics, drug and alcohol use, depression, physical and mental health functioning, treatment engagement) moderate the effect of the PARS program. To evaluate these individual-level moderators, the primary outcome analyses above will be extended to include the main effect of the moderator and its interactions with Treatment, Time, and the Treatment by Time interactions. Each moderator will first be evaluated in a separate model to ascertain whether any differences in treatment effects were associated with individual level factors. A follow-up model will evaluate the moderators simultaneously in a single model. The statistical test of moderation will be the three-way Treatment by Time by Moderator interactions.
Mediation Analyses. Secondary mediation analyses will examine whether baseline to postintervention 1) increases in knowledge about suicide and 2) decreases in maladaptive attitudes about suicide will mediate the effect of PARS vs. TAU help-seeking behaviors. The mediation analyses will be conducted using multilevel structural equation modeling. 61 Robust standard error estimates and corrected model fit statistics (e.g., Satorra-Bentler statistic) will be utilized to accommodate nonnormally-distributed dependent variables in the model. 62 Each mediator will first be evaluated in a separate model. Figure 2 illustrates the basic autoregressive regression model 63 that will be used to assess longitudinal mediation of the effect of intervention on 1-, 3-, and 6-month help-seeking behavior. An indicator variable for Step will be included as a covariate if the effect of step is statistically significant in the primary outcome analyses. The model will incorporate random effects for site and individual to account for the clustered design (i.e., repeated measures nested within individuals, nested within site). The mediated effect (a 1 *b 5 ) and its confidence interval will be obtained using bootstrap resampling, as recommended for testing indirect effects. 64 (Note, arrows on upper right side of mediator and outcome boxes represent error terms.) Dose Response Analyses. Exploratory analyses will evaluate whether Time (i.e., number of steps) in the intervention phase is associated with improved outcomes in successive steps. The following regression equation depicts the basic dose response model for each outcome: OUTCOME t>0,is = b 0 + b 1 BL_OUTCOME is + b 2 TAU_TIME is + b 3 PARS_TIME is + u 0s + r 0is + e tis , where t indexes the assessment, i indexes the individual, and s indexes the site. In this model, Time will be centered at the beginning of the intervention phase, such that Time = 0 will correspond to the step in which PARS was implemented at a particular site. Time will be coded as a piecewise linear function representing two phases: 1) the TAU phase: the change over time in outcomes during the control period and the 2) PARS phase: a "deflection" term representing the change in slope from the TAU phase following the implementation of PARS. To account for the cluster-randomized design, a random effect for site (u 0s ) will be included to model the correlation of individuals from the same site. To account for the repeated assessments, a random effect for participant (r 0is ) will be included to model the correlation of observations within the same individual. The statistical test of the dose-response effect will be the magnitude and statistical significance of the slope of Time during the intervention period.
Power Analysis. We base our power analysis on a stepped wedge design 35 evaluating the difference in the rate of help-seeking behavior at month 1 (post-intervention) among those receiving PARS vs. TAU. Based on our pilot study of PARS, we anticipate a 15.6% baseline rate of help seeking behavior, which we expect to increase to 28.1% at post-intervention (1-month) follow-up. Based on the proposed stepped wedge design in which 10 individuals are recruited per site in each of the 6 steps (N = 900), we will have 89% power to detect the 12.5% difference in the rate of help-seeking behavior. Assuming 25% fewer participants (N = 675, 7.5 individuals per site per step), due to under-recruitment and/or attrition, there will still be 80% power to detect an intervention effect. This represents a conservative estimate of statistical power as the full longitudinal model will leverage data from all participants, including those with incomplete follow-up data.
Missing data. Missing data may occur in several ways. First, missing data may occur due to item nonresponse. When missing data is limited to only a few items on a measure, we will prorate total scores for a measure by taking an average score on the measure and multiplying it by the total number of items in the scale. Missing data can also occur from attrition due to missed assessments or dropout from the study. Prior to performing any outcome analyses, we will evaluate the amount, reasons, and patterns of missing data. If the