PATHway: Intervention optimization of a prevention program for adolescents at–risk for depression in the primary care setting

Abstract

With as many as 13% of adolescents diagnosed with depressive disorders each year, prevention of depressive disorders has become a key priority for the National Institute of Mental Health (NIMH). Currently, we have no widely available interventions to prevent these disorders. To address this need, we developed a multi-health system collaboration to develop and evaluate the primary care based technology “behavioral vaccine,” Competent Adulthood Transition with Cognitive-Behavioral Humanistic and Interpersonal Therapy (CATCH-IT). The full CATCH-IT program demonstrated evidence of efficacy in prevention of depressive episodes in clinical trials. However, CATCH-IT became larger and more complex across trials, creating issues with adherence and scalability.

We will use a multiphase optimization strategy approach to optimize CATCH-IT. The theoretically grounded components of CATCH-IT include: behavioral activation, cognitive-behavioral therapy, interpersonal psychotherapy, and parent program. We will use a 4-factor (2×2×2×2) fully crossed factorial design with N=16 cells (25 per cell, after allowing 15% dropout) to evaluate the contribution of each component. Eligible at-risk youth will be high school students 13 through 18 years old, with subsyndromal symptoms of depression. The study design will enable us to eliminate non-contributing components while preserving efficacy and to optimize CATCH-IT by strengthening tolerability and scalability by reducing resource use. By reducing resource use, we anticipate satisfaction and acceptability will also increase, preparing the way for an implementation trial.

1. Introduction

As many as 7–13% of adolescents living in the United States (US) experience minor or major depressive episodes each year [1]. Prevention of mental health disorders is a key priority for the National Institute of Mental Health (NIMH) [2], but we lack public health interventions to prevent onset of depressive episodes in adolescence [3]. The American Psychological Association strongly recommends the development and implementation of preventive interventions in medical settings [4]. Yet, few programs exist [5], although most adolescents see their primary care physician at least once a year and are receptive to counseling from their physician [6]. A primary care-based intervention may be more acceptable to adolescents than specialty mental health services and would provide the psycho-social guidance many seek [6, 7, 8].

To address these needs, we developed the primary care-based technology “behavioral vaccine” [9], Competent Adulthood Transition with Cognitive- Behavioral Humanistic and Interpersonal Therapy (CATCH-IT). CATCH-IT has demonstrated low time requirements [10, 11], high cultural acceptability [12, 13], low implementation costs [14, 15], and evidence for efficacy [16, 17]. Implementation studies within 8 different health systems demonstrate health systems’ ability to implement the model with REACH (percent of at-risk youth who complete intervention) of 18% to 60% of those at risk for depressive episodes [14, 18].

Like many interventions, CATCH-IT has become larger and more complex across trials [16, 17], and users have become less adherent. Additionally, primary care practices reported “scalability” issues (resource allocation, referral usage). While evidence of efficacy developed over several trials, we saw evidence of declines in number of modules completed (decreased from > 8 modules to < 4 modules) and REACH in primary care declined from > 40% to < 20% [14, 16, 18, 19].

We will use a Multiphase Optimization Strategy (MOST) [20] approach to optimize CATCH-IT for depression prevention, using an indicated prevention perspective (i.e., preventing the onset of a depressive episode in adolescents who present with subthreshold symptoms). MOST uses a systematic analytic approach and a factorial randomized clinical trial design to simultaneously address efficacy, tolerability, and scalability. In contrast to traditional RCT designs, the MOST strategy identifies specific components of the behavioral intervention that are active. This strategy allows an intervention to be dismantled and streamlined, to improve its REACH, adherence, and sustainability [21]. While MOST has generally been used to prospectively develop an initial intervention, we use MOST for a similar purpose of optimization, in order to selectively retain effective modules and thereby strengthen adherence and scalability.

1.1. Intervention

The CATCH-IT intervention combines several theoretical models to target vulnerability factors (Figure 1) [22, 23, 24, 25]. The intervention uses motivational interviewing (MI), goal-setting, and text reminders to enhance change processes and advance adolescent participants through the stages of change to reduce vulnerability and increase protective factors. The Internet program guides participants through exercises that allow them to learn behavioral activation, cognitive-restructuring, and interpersonal skills, to help reduce negative cognitions before they become too severe, thereby averting the development of full depressive episodes. In addition, given that families are a substantial source of vulnerability and protection regarding future incidence of depression in adolescents, CATCH-IT includes modules that provides parents/caregivers with psychoeducational information about adolescent depression. Because family change may promote resilience in adolescents [26, 27], it follows that an important aim for depression prevention programs is to help change maladaptive aspects of the family environment.

Figure 1:

CATCH-It Theoretical Design

CATCH IT helps to address five key barriers in primary care preventive mental health. First, to address cost, distribution difficulty, and low acceptability of face-to-face interventions, CATCH-IT [28, 29, 30, 31] delivers a high fidelity [28, 32, 33, 34, 35] intervention via the Internet and primary care in teen-friendly language [36]. Second, to address lack of motivation and engagement [22, 29, 37], CATCH-IT uses a social marketing [38] and MI strategy [39, 40]. Third, to target the epidemiologic context, CATCH-IT addresses multiple etiological elements acting either in concert or in combination, including negative cognitions [41], poorer social skills [42], stressful events, subsyndromal depressive symptoms [43], and the absence of protective factors (e.g., high self-esteem, coping skills) by engaging participants with distinct behavior change programs [23, 24, 25]. Finally, to address the lack of cultural relevance and the absence of multi-modal learning opportunities in other interventions, CATCH-IT uses a multi-channel learning process and culturally relevant lessons, stories, and music, based on face-to-face manual-based interventions [28, 32, 33, 34, 35] of demonstrated efficacy [22, 29, 33, 44].

CATCH-IT also makes use of new engagement models for by integrating program use reminder texts to adolescents with engaging questions and responses aligned with the internet site content. The program also offers multiple types of learning options (e.g., visual, auditory, and comic artwork). In addition, CATCH-IT incorporates gaming and social media elements, which are preferred by adolescents [39]. Original cartoon figures, short videos, and choose-your-adventure type modelling of behavioral strategies are used to create a more dynamic and interactive experience for teens (see Figure 2).

Figure 2:

Identifying Thoughts module from the CATCH-IT intervention

While the Internet has proved to be a promising modality [45, 46, 47], use of mental health promotion Internet sites appears to be limited to those highly motivated to receive treatment [48, 49]. Self-directed interventions may have higher completion rates and greater symptom reduction when combined with minimal amounts of face-to-face contact. Evidence exists for the value of MI, which uses increased quantity and quality of motivation to strengthen internal rationale and enhance behavior change [50]. The value of MI has been demonstrated in multiple studies with moderate to large effect sizes [47, 51, 52, 53]. Based on the Trans-theoretical Model of Change [54], CATCH-IT also uses MI, goal setting, and telephone coaching to enhance the quality and quantity of motivation to prevent depression.

1.2. Aims

To date, most preventive interventions have been assembled as complex packages [55]. Observational models have not shown successful elucidation of mechanisms or pathways related to specific components of these interventions [56]. The factorial design will enable us to determine the impact of different theoretical models, alone or in combination, on the prevention of depressive episodes. Additionally, the technology-based delivery provides a practical deployment, monitoring and fidelity assessment of a full factorial design of content elements.

Effects from four components of the CATCH-IT intervention (Table 1) will be examined individually and in combination: the Behavioral Activation (BA) adolescent component; the Cognitive Behavioral Therapy (CBT) adolescent component; the Interpersonal Therapy (IPT) adolescent component; and the Parent Program (PP) component. Our primary focus is to identify intervention components that reduce the likelihood of developing a depressive episode, as determined by the Mini-International Neuropsychiatric Interview for Children and Adolescents (MINI Kid) diagnostic tool. We will also examine other secondary outcomes, such as depression symptoms and/or symptoms of other disorders, vulnerability and protective factors, resiliency, cognitive and family/peer tolerability (adolescent/family), and scalability in primary care. We hope to answer the following questions:

Table 1.

Intervention Components

Component Name	Module Content and Exercises	Behavioral Target
Informational or base component
Psychoeducation (adolescent)	Recognition of need for treatment for depressive episode	Stigma, lack of knowledge of symptoms and treatment options/benefits
Psychoeducation (parent)	Treatment education Access to depression treatment program	Increase likelihood of treatment Self-directed CBT and IPT to change behaviors
Motivational Interview Component	MI at time 0 and 8 weeks Texts at 2 and 4 weeks and if stop using Internet site for 10 days	Low motivation for prevention
Internet Preventive Component
Behavioral Activation (BA)	Event scheduling Practicing active behaviors (Behavioral Activation) BA; Community involvement	Loss of response contingent Reinforcement Low levels of pro-social activities
Cognitive Behavioral Therapy (CBT)	Identifying and countering pessimistic automatic thoughts, general beliefs and hopelessness Problem solving (Cognitive Behavioral Psychotherapy); Flexibility/humor/persistence,	Cognitive distortions Pessimistic cognitive style/content Poor coping skills Inflexible responses
Interpersonal Therapy/Resiliency (IPT)	Improving communication skills, coping transitions, conflict resolution, engaging new networks	Lack of social support Social skills deficits Lack of peer support
Parent Training Intervention (PP)	Activism Connectedness Affect recognition	Cultivating strengths Encourage discussion, behavioral activation, resiliency behaviors and expression of emotion

1. Does the BA component improve outcomes relative to no BA?;

2. Does the CBT component improve outcomes relative to no CBT;

3. Does the IPT component improve outcomes relative to no IPT?;

4. Does the PP component improve outcomes relative to no PP?

We hypothesize that completing the BA, CBT, IPT, or PP component on its own will improve outcomes compared to completing no components, suggesting an intervention with fewer components or even just one may be as beneficial for participants as completing the full program.

We will also examine interactions between components/factors using the factorial design, and interactions between components, baseline characteristics, and outcomes (moderation), and specific pathways between baseline characteristics, components, and intermediary psychological processes and outcomes (mediation). This study design (Figure 3) offers an opportunity to explore multiple interactions. We hope to determine which mediators (dose, symptoms, protective and vulnerability factors) and moderators (mediating variables, baseline characteristics, implementation factors) will affect outcomes. We hypothesize that demographic, psychological (comorbid symptoms, trauma, rumination, cognitive styles), and familial vulnerability or protective factors will interact with components.

Figure 3:

Study Design

2. Methods

The PATHway study was approved by the Chicago Area Institutional Review Board (CHAIRb). All adult parents, guardians, and adolescents aged 18 provide informed written consent prior to participating. Adolescents ages 13–17 provide informed assent prior to participating. PATHway is registered at ClinicalTrials.gov (#NCT05203198).

2.1. Study design

This study uses MOST [20] to comprehensively evaluate whether one component of CATCH-IT demonstrates an equivalent effect to combinations of other components in terms of efficacy, while demonstrating superior adolescent/family tolerability and scalability over a 12-month follow-up. The sample will include N=400 adolescents from online social media recruitment and from participating health systems (Advocate Aurora Health, Lurie Children’s Hospital, NorthShore University Health System, University of Illinois Hospital and Health Sciences System, University of Texas Southwestern Medical Center, University of Chicago Medicine, Lawndale Christian Medical Center, and Duke University Medical Center). Using a 2 × 2 × 2 × 2 factorial design, 400 adolescents will be randomized into the different cells after stratification by gender to evaluate the contribution of each component (Table 3).

Table 3:

Factorial Design Groups

Experimental Condition	BA	CBT	IPT	PP
1	−	−	−	−
2	−	−	+	−
3	−	+	−	−
4	−	+	+	−
5	+	−	−	−
6	+	−	+	−
7	+	+	−	−
8	+	+	+	−
9	−	−	−	+
10	−	−	+	+
11	−	+	−	+
12	−	+	+	+
13	+	−	−	+
14	+	−	+	+
15	+	+	−	+
16	+	+	+	+

Note: Teens in groups 1–15 will be given access to the full CATCH-IT program upon completion of the 12-month follow-up checkpoint.

2.2. Setting and recruitment

To obtain a sample that encompasses diverse populations in rural, suburban, and urban settings, participants will be identified using a public health screening model through in-person, informational, and targeted social media recruiting.

When visiting participating health systems, patients aged 13–18 are offered information about the study and given an eligibility screening for symptoms of depression. Potentially eligible adolescents are also identified through chart reviews and contacted by clinical staff to obtain consent for contact by the research team. We are focusing on patient-centered methods of engagement developed in prior studies, including empathic screening, clinical referrals, emails, MyChart messages, after-visit summaries, social media, and letters [56]. Potential participants are directed to visit their primary care clinic or contact the study team for an eligibility screening.

We will also employ a nation-wide public health social media campaign to recruit non-primary care-based participants, using approaches commonly used in other studies of adolescent mental health. Prior studies have found Internet recruitment to be successful in not only recruiting but also in retaining at-risk youth for remote, longitudinal behavioral interventions and to be more cost-effective than in-person recruiting [57]. This recruitment strategy will reduce time burden on potential participants and allow our study to capture a wider range of participants, including non-white, low SES, and rural adolescents.

2.3. Screening and Eligibility

The Patient Health Questionnaire (PHQ)-9 is used to determine initial patient eligibility. In clinics where the PHQ-9 is administered as part of routine care, any PHQ-9 from the past month serves as the eligibility screener. In clinics where the PHQ-9 is not regularly administered or was collected more than a month ago, clinic staff follow up to gauge interest and refer potential participants to the Boston Call Center (BCC) for a phone screen. In clinics where the PHQ-2 is administered, scores of 1–2 are considered potentially eligible and potential participants are referred to the BCC. Patients may also call the BCC directly.

Interested adolescents who interact with targeted social media advertisements will be directed to fill out an online form with both their information and contact information for a parent/guardian. The BCC will reach out to obtain parent permission and teen assent to conduct a phone screen. Adolescents aged 18 will be provided information about the research study, document agreement to the screening process, and then complete the PHQ-8 to assess eligibility.

After initial eligibility has been determined, research staff obtains consent from the eligible adolescent and obtains permission and consent for participation from the parent (either in-person or virtually by telehealth). The study team then conducts a mental health assessment by phone to determine full eligibility. In response to the COVID-19 pandemic, our protocol was adapted to enable remote implementation of study activities.

2.4. Inclusion criteria

To participate in the study, adolescents must be 13–18 years old at time of enrollment; show elevated depressive symptoms (PHQ-8/9 score of 5–18); have one caregiver who will participate in the intervention; and express willingness to participate in the longitudinal assessments for 12 months. Adolescents can be included if they have past, but not current, depression.

2.5. Exclusion criteria

Adolescents will be excluded if they have a current diagnosis of major depressive disorder (MDD; determined by assessment with the MINI Kid); have a lifetime DSM-5 [58] diagnosis of schizophrenia or bipolar affective disorder; have a current DSM-5 diagnosis of extreme drug or alcohol abuse (determined to have drug or alcohol use that interferes with current academic or social functioning); are currently engaged in individual treatment for a mood disorder; are currently using medication therapy for depression, anxiety, or other internalizing disorders; are currently engaged in a cognitive-behavioral group or therapy; have any past psychiatric hospitalizations; have any past self-harm attempt with moderate or greater lethality; have current, active suicidal thoughts; do not speak/read English; have less than sixth-grade reading level based on parental report, cognitive or intellectual impairment, or developmental disability; participating parent/guardian does not speak English or Spanish; participating guardian has less than sixth-grade reading level based on self-report, cognitive or intellectual impairment, or developmental disability.

2.6. Intervention fidelity checks

The CATCH-IT intervention includes two, 15-minute MI phone calls conducted by a research staff member, first at baseline and then again at two months post enrollment. All enrolled teens will receive these calls, regardless of randomization. These calls encourage the adolescent to weigh the pros and cons of program usage. MI calls are recorded and uploaded to a HIPAA compliant server. The research team will review one-tenth of the MIs for fidelity to the MI model and provide feedback.

2.7. Safety protocols

All participants, whether potential or enrolled, are assessed for symptom severity. Research staff are trained to take appropriate steps regarding severe psychopathology and suicidality, including assessment of intent, plan, possible method, safety planning, and discussion of findings with the participant, appropriate family member for teens under 18, senior clinician, or principal investigator. Families are also provided with referral information. For imminent crises, appropriate action is taken immediately, including escorting and/or providing medical transportation to an emergency department. For risky but not imminent events, we contact the clinically responsible investigators (i.e., site principal investigators, co-investigators, or senior clinicians) for consultation. Enrolled participants who show signs of escalating mental health problems during the study are not dis-enrolled for pursuing therapy outside of the study.

3. Data collection and outcomes

3.1. Data procedures

All research staff are trained on the study protocol, informed consent, interviewing skills, mental health assessment, suicide assessment and response protocol, participant contact (in-person, texting, and phone call etiquette), and protocol for compensation. Data collected by the research team is not entered into the participants’ medical records, except if a participant develops a major depressive episode, reports suicidal ideation and/or behavior, or reports a child abuse or neglect issue that requires mandated reporting.

3.2. Data management

Centralized databases have been developed to collect, process, and manage data related to screening, enrollment, intervention delivery, and follow-up checkpoints. The data management process includes the de-identification of the data, implementation of the unique ID for each participant, and downloading the CATCH-IT web data to a secure server, followed by cleaning, documentation, and construction of master files for all data collection. The research team utilizes REDCap as the primary tool to collect and manage data, track participation in intervention and research activities, and document communication with participants.

3.3. Assessment time-points

All adolescents, regardless of which components they have been assigned to, complete phone and self-assessments at baseline, two, six, and 12 months. Caregiver assessments occur at baseline, six, and 12 months. Self-assessments are completed online but, if preferred by the participant, can be completed over the phone with research staff. If adolescents report severe depressive symptoms consistent with a major depressive episode, suicidal ideation/behavior, or an issue that requires mandated reporting during the phone interview, appropriate caregivers and/or site staff are notified within 48 hours.

3.4. Compensation

Adolescents and caregivers are compensated for their participation over 12 months ($150 for adolescents and $50 for caregivers). Payment is distributed through Greenphire, which utilizes re-loadable MasterCard debit cards, called ClinCards, and a clinical trial payment software.

3.5. Measures

Data collected measure the following domains: socio-demographic, anthropometric, depressive, and other mental disorder symptoms, cognition, resiliency, relationships, development, and vulnerability (Table 2). Data generated from participants’ use of the CATCH-IT program includes number of logins, specific pages or modules accessed and completed, data entered into site activities, and the date and time modules were started and completed, which will specifically identify teens who completed the program in an unreasonably short amount of time. In addition to the Usefulness, Satisfaction, and Ease of Use questionnaire, the team also collects data from the two MI calls.

Table 2.

Assessments and Instruments

Domain	Form	Months
		0	2	6	12
Baseline Characteristics	From EMR: Height, Weight, Blood Pressure; Self-report, online survey: Children with Special Health Care Needs Screener (CSHCN, 5-item), race, ethnicity, gender identity, religion, school, employment, Holliston at-Home Questionnaire (40-item).	A*P		AP	AP
Component/Factor Dose from CATCH-IT website:
CATCH-IT Component Dose	Number of web pages and modules completed, number characters typed in, completed exercises, and log files of clicks/actions patterns.	Continueus
Vulnerability and Protective Factors
Depressive and other mental disorder symptoms (self-report)	Online Survey: Center for Epidemiological Studies-Depression Scale (CES-D, 20-item), Patient Health Questionnaire-Adolescent (PHQ-A, 8 or 9-item, 4 follow-up items), Disruptive Behavior Disorders Adolescent (DBD-A, 41-item), Screen for Child AnxietyRelated Disorders (SCARED, 41-item), Car, Relax, Alone, Forget, Friends, Trouble Screening Test (CRAFFT, 6-item), Child PTSD Symptom Scale (24-item).	AP*	AP*		AP*
Resiliency (self-report)	Online Survey: Connor-Davidson Resilience Scale (CD-RISC, 10-item), The Socio-Cultural Relevance Scale (10-item).	A		A	A
Cognition (self-report)	Online Survey: Children’s Cognitive Style Questionnaire (CCSQ 30-item), Children’s Response Style Scale (CRSS, 10-item), The Dysfunctional Attitude Scale (DAS, 9-item).	A	A		A
Family and Peer (self-report)	Online Survey: UCLA Life Events Scale (19-item), Child Report of Parental Behavior Inventory (CRPBI, 23-item), Perceived Social Support from Friends (PSS-fr, 20-item).	A	A		A
Mental Disorder, Functional and Developmental Outcomes
Depressive and mental disorder episodes (self-report)	Phone interview by BCC Staff: Mini-International Neuropsychiatric Interview for Children and Adolescents (MINI Kid).	A		A	A
Function and development (self-report)	Online Survey: Social Adjustment Scale (SAS-SR, 23-items), Pediatric Quality of Life and Enjoyment and Satisfaction Questionnaire (PQLESQ, 13-item).	A		A	A
Implementation Factors
Adolescent “tolerability” (self-report)	Online Survey: Usefulness, Satisfaction, and Ease of Use Questionnaire (USE, 20-item), Trans-Theoretical Model Scale (10-item), resource use (time as “cost:”).	A	A		A

A=adolescent, P=parent

^*CESD only

3.6. Sample size determination

The determination of sample size for this study used the power calculation method of Rochon, which addresses two-group repeated measures designs [59]. The 2⁴ factorial design employed involves 4 completely crossed factors (BA, CBT, IPT, PP) and thus the same power calculation applies to each factor being tested. There are 4 waves of data collection. For each factor, we wish to test the omnibus G(2) × T(4) interaction that compares outcome trajectories between groups with a 3 d.f. F-test, with correlation among cross-time residuals ρ_t=0.70 a total sample size of 320 provides power > 0.80 to detect a “smallish” effect size ES = 0.30 at Type I error α = 0.05. Allowing for 20% dropout, a total sample size of 320/0.80 = 400 is obtained. Note that sample size suffices for all single d.f. G × T contrasts as well (adjusted for multiplicity). These can be used to study whether change from baseline has reached significance at the 2-, 6- or 12-month follow-up waves, or to determine whether the trajectory contains linear, quadratic or cubic growth curve components. Regarding the 2⁴=16 cells of the factorial design, we will allocate 25 subjects to each cell and retain an average of 20, yielding the final 16 × 20 = 320 sample size. The planned final sample size of 320 is ample for mediation analyses according to Fritz and MacKinnon [60]. For single mediators, we will test using the bias-corrected bootstrap procedure. A joint test of multiple simultaneous mediators also enjoys ample power.

3.7. Randomization of participants

A randomization equal-allocation table is created in R by the project statistician and imported into the REDCap randomization module. Randomization assignment is stratified by gender, giving each participant an equal chance of assignment to one of the 16 conditions shown in Table 3. To reduce the potential impact of the order of presentation within adolescent CATCH-IT, adolescents assigned to conditions receiving 2 or 3 adolescent components (BA, CBT, IPT) are further randomized to give them an equal chance to receive the components in each possible order (e.g., for Condition 4, CBT first then IPT, or IPT first then CBT).

3.8. Analytic plan

The analysis of a MOST design is accomplished via repeated measures analysis of variance; the modern approach is to employ linear mixed models (LMM) or a generalized LMM (GLMM) with a random intercept term (and possibly other random effects), which are available in the SAS and R statistical packages [61, 62]. The time effect will be represented as growth curves using orthogonal polynomials (linear and quadratic, adjusted for unequal intervals between times). A vector of fixed baseline covariates, including variables used for stratification and other predictors, will also be included. Initiation of therapy after randomization will be included as a covariate as well. Certain variables will be examined as moderators by incorporating their interactions with the 4 factors into the model. In MOST, it is standard to employ effect coding of the factors, where +1’s and −1’s following the sign pattern shown in Table 3 are used [63]. Except for an inevitable slight correlation induced by unequal cell n’s owing to dropout, the factors are orthogonal and provide separate tests of each factor and its interaction with time. In sum, the model tests whether each factor exhibits an overall mean response difference between groups, whether the response exhibits an overall trend, and whether the time trends in response differ by group.

Further information can be extracted via expansion of the foregoing analysis. It is of great interest for MOST to determine whether the co-occurrences among the 4 factors yield greater or lesser benefits. This information is obtained by including 4-choose-2 = 6 factor interactions in the model; 3-way interactions of these terms with time will be included as well, to reveal how factor interactions change over time. We will employ multiplicity-corrected contrasts on factor × time and factor interaction × time cell means to test the time course of factor impacts. It might be possible to select program factors that yield a relatively immediate preventive effect.

One outcome, incidence of a depressive episode during the study, will be analyzed using a logistic regression model that includes the factors, factor interactions, and baseline covariates (not time). This model is cross-sectional but offers the same information for MOST as the models described above.

The mechanisms of action of the CATCH-IT components will be examined using mediation analysis [64]. For outcomes measured at 6 months, mediators measured at 2 months can be tested; for outcomes measured at 12 months, mediators measured at 2 months or 6 months can be tested. We will employ a structural equation model (SEM) approach to mediation modeling, and will use bias-corrected bootstrap estimation, an approach with superior power, to test coefficients and their product. When several mediators are tested in parallel, multiplicity will be controlled using the sequentially rejective Holm-Bonferroni procedure [65].

3.9. Treatment of missing data and sensitivity analysis

All participants are followed, and reasons for drop-out are recorded per CONSORT standards [66]. We anticipate that missing data prior to end-of-study or dropout will be missing at random (MAR) [67]. Under these circumstances, use of all available data without imputation (i.e., full information maximum likelihood) is nearly unbiased in mixed model analyses, even nonlinear ones such logistic GEE, which offers consistent estimators [68, 69]. If item-missing or missed-wave missing data exceeds 10%, we will consider imputation using fully conditional specification (FCS), available in SAS and R, as it has the versatility to handle various levels of measurement [70, 71]. In general, we will apply generalized linear mixed models (GLMM) for subject-specific trajectory analyses but shift to generalized estimating equations (GEE) procedures for population-averaged marginal comparisons between groups [61, 72]. We acknowledge that the GEE approach is not generally able to accept data that is not missing completely at random (MCAR), so that multiple imputation techniques must be employed prior to GEE for MAR and missing not at random (MNAR) data.

Our statistical approach always includes sensitivity analyses, to assess whether the conclusions are dependent on small assumptions. Results obtained with available data and imputed data will be compared. We also rely on robust standard errors, for example sandwich type variance calculations.

We plan to conduct our analyses under the intention-to-treat (ITT) rubric [73], which in turn requires that we address treatment of missing data and sensitivity to the MAR data assumption (as against not missing at random, or NMAR) [67]. We will employ both maximum likelihood [68], and multiple imputation using the fully conditional specification (FCS, also called chained equations), as implemented in R or SAS PROC MI [74]. The sensitivity analysis will employ the latter procedure’s SCALE and SHIFT parameters for testing a range of NMAR imputations to determine whether there is any impact on analytic conclusions.

4. Discussion

A primary care-based intervention, such as CATCH-IT, may be more acceptable than specialty mental health services to adolescents and would provide the psycho-social guidance many seek [7, 8]. Like many interventions, CATCH-IT became larger and more complex across several trials [16, 18] Thus, as trials moved further into the community, “reach” and “dose” declined as adolescents responded to the “additional work” of the intervention, being less adherent and undermining the end goal of public health impact [14, 16, 18, 19]. The optimization work provided by this study will help to reduce resource use while also emphasizing tolerability and scalability, preparing the way for an implementation trial and eventual dissemination.

Abbreviations:

NIMH

National Institute of Mental Health

CATCH-IT

Competent Adulthood Transition with Cognitive-Behavioral Humanistic and Interpersonal Therapy

MOST

Multiphase Optimization Strategy

BA

Behavioral Activation

CBT

Cognitive Behavioral Therapy

IPT

Interpersonal Therapy

PP

Parent Program

MINI Kid

Mini-International Neuropsychiatric Interview for Children and Adolescents

MI

Motivational Interviewing

PHQ

Patient Health Questionnaire

BCC

Boston Call Center

MDD

Major Depressive Disorder