Effects of Preference on Outcomes of Preventive Interventions among Ethnically Diverse Adolescents At-Risk of Depression

Abstract

Patient-centered care includes efforts to align treatment with patient preferences to improve outcomes and has not been studied in adolescent depression prevention. Within a school-based randomized trial, we examined the effects of offering a preference between two evidence-based preventive interventions for youth at risk of depression, Learning to BREATHE (L2B) and Interpersonal Therapy–Adolescent Skills Training. We examined the effects of 3 preference factors (assignment condition [preference vs. random], receipt of preferred program, and baseline program preference) on outcomes in a diverse sample of 111 adolescents (M age = 15.18 years, SD = .86): 81 (73%) girls, 45 (41%) White, 40 (36%) Asian American, 8 (7%) Latinx, 1 (1%) African American, and 17 (15%) multiracial or other race/ethnicity. Findings revealed little evidence that receiving a preferred intervention or being given a choice of interventions was linked to greater improvement or initial engagement. Further, analyses did not indicate that adolescents with baseline indications for a specific intervention would benefit more from that intervention; rather, adolescents with generally lower baseline functioning improved more regardless of the intervention received. However, receipt of L2B and a baseline preference for L2B were associated with greater improvements in about half of the outcomes examined, with effect sizes ranging from R2 = 0.04 to 0.14. There was little support for the need to match interventions to adolescent preferences in school- based prevention efforts. Rather, the more scalable mindfulness-based intervention had stronger effects than the interpersonal intervention and may hold promise for diverse adolescents.

Significant investment has gone into developing a wide variety of evidence-based preventive interventions (EBPIs) targeting putative risk and protective factors for depression in youth. Yet little research guides how to best select and implement EBPIs in ethnic minority communities where penetration has been low (Lee & Lau, 2014). “Patient-centered care” that increases the involvement of individuals in their intervention decisions may be relevant to adolescent prevention. Given that distinct treatments show similar effects in the treatment of depressive disorders (Cuijpers, van Straten, Andersson, & van Oppen, 2008), selection based on patient preference may promote engagement and effectiveness. Indeed, preference trials have found that depressed adults who receive their preferred treatment (i.e., medication vs. psychotherapy) experience better outcomes and show greater adherence (e.g., Lin et al., 2005; Raue, Schulberg, Heo, Klimstra, & Bruce, 2009). Yet meta-analyses and reviews of trials with adults with psychiatric disorders indicate that preference effects are small (Swift & Callahan, 2009) and the effect of offering a choice of alternative psychosocial interventions is unknown (Winter & Barber, 2013).

Thus far, research on treatment preference has been limited to adult patients with diagnosed psychiatric disorders. Questions remain about whether adolescents are sufficiently experienced to make informed and effective decisions about their health and mental health needs (Coyne, 2006). There is no data on the role of preference on effects of EBPIs for youth at risk of depression. In the present study, we examined the effects of preference on outcomes for ethnically diverse youth at risk. We conducted a randomized trial testing two EBPIs and the effect of offering adolescents a choice of intervention.

Evidence-Based Preventive Interventions for Adolescent Depression

A number of EBPIs have been established as efficacious for youth with elevated depression symptoms but who do not meet criteria for a mood disorder, and this literature suggests that indicated prevention appears to generate larger effects than universal prevention efforts (Stice, Shaw, Bohon, Marti, & Rohde, 2009). As more school districts move toward a tiered approach to mental health intervention (Mendez, 2016), research is needed to inform the selection of indicated preventive interventions (Tier 2). The majority of EBPIs shown to be effective in schools are based on principles of cognitive behavioral therapy focusing largely on behavioral activation and cognitive restructuring (Calear & Christensen, 2010). However, accumulating evidence supports the effectiveness of school-based mindfulness and interpersonally focused EBPIs for youth depression. Mindfulness-based interventions, including the Learning to BREATHE (L2B) program, reduce depression and anxiety symptoms and perceived stress and improve emotion regulation skills among at-risk students (Broderick & Metz, 2009; Fung et al., 2019). Interpersonal Therapy–Adolescent Skills Training (IPT-AST) teaches communication and interpersonal problem-solving skills and has been shown to reduce depressive symptoms and risk of depression onset among at-risk students (Young, Mufson, & Davies, 2006; Young, Mufson, & Gallop, 2010).

Prevention Targets for Adolescent Depression

In the current trial, we selected IPT-AST and L2B as two EBPIs of interest based on our previous research that identified risk and protective factors for the depressive symptoms in a culturally diverse school-based sample within a community partnered study (see next). We sought to test interventions that targeted distinct “intermediate outcomes” that could result in improvements on the “ultimate outcome” of depressive symptoms. Ultimate outcomes are the reasons for which treatment is undertaken and treatment is considered a success when they are achieved, whereas intermediate outcomes are outcomes that facilitate or lead to the obtainment of ultimate outcomes (Rosen & Proctor, 1981). These EBPIs targeted the candidate intermediate outcomes identified as potentially differentially salient across cultural groups. IPT-AST targeted interpersonal functioning and distress, and L2B targeted emotion acceptance and mindfulness in service of reducing vulnerability to depression.

First, interpersonal stress has widely been implicated as a risk factor for adolescent depression. Interpersonal vulnerability models implicate early family disruptions that interfere with the development of interpersonal skills resulting in deficits that generate relationship stressors (i.e., rejection or conflict), which prompt depressed mood, which then cyclically compromises interpersonal functioning and perpetuates depression (Rudolph, Flynn, & Abaied, 2008). Interpersonal stress is a salient and modifiable risk factor for depression symptoms across diverse adolescents from acculturating groups. Among Latinx and Asian American adolescents in immigrant families, interpersonal stress associated with family acculturation conflicts predicts increased internalizing symptoms (Nguyen, Kim, Weiss, Ngo, & Lau, 2018; Smokowski, Bacallao & Buchanan, 2009). IPT-AST aims to reduce interpersonal deficits and conflicts that increase depression risk and to promote skilled communication and positive relationships that protect against depression. Controlled trials have demonstrated that IPT-AST engages these intermediate outcomes of social functioning and interpersonal distress (Young, Kranzler, Gallop, & Mufson, 2012) in service of reducing depression risk (Young et al., 2006).

Second, maladaptive emotion regulation marked by emotion suppression and experiential avoidance is established as a risk factor for developing depressive symptoms. Emotion suppression is the active downregulation of emotion display or expression (Butler, Wilhelm, & Gross, 2006) and is prospectively linked to increased internalizing symptoms (Betts, Gullone, & Allen, 2009). Experiential avoidance is marked by attempts to disengage or draw attention away from emotion states and inner experiences (Greco, Lambert, & Baer, 2008). Avoidance and suppression may temporarily provide relief from negative affect but result in increased depression over time (Aldao, Nolen-Hoeksema, & Schweizer, 2010). Suppression and avoidance are targeted by mindfulness interventions that emphasize nonjudgmental awareness and healthy acceptance and engagement with emotions as transient, tolerable, and functional (Hayes & Feldman, 2004). The effects of the L2B intervention on anxiety and depression symptoms are mediated by reduced reliance on emotion suppression and experiential avoidance among Latinx and Asian American adolescents (Fung et al., 2019).

Need for Inclusive Prevention Trials

Inclusive prevention trials with diverse youth can promote scientific equity for depression prevention in underserved populations (Perrino et al., 2014) and provide experimental tests of the cultural generalizability of models of depression vulnerability. It is plausible that EBPIs may not generalize well across cultural groups if there is variation in the key vulnerability factors that cause and maintain depression. Inclusive relative effectiveness trials can shed light on whether diverse groups require attention to different intermediate outcomes to affect change in ultimate outcomes. Latinx adolescents are underrepresented in the evidence base, and Asian American adolescents have largely been excluded from the effectiveness literature (Piña, Polo, & Huey, 2019; Polo et al., 2019). There is some reason to question the generalizability of models of depression vulnerability and EBPI effects. For example, because of interdependent cultural values Asian American youth may rely on emotion suppression coping to a greater extent than Whites (e.g., Butler, Lee, & Gross, 2007); however, the associations between emotion suppression and depressive symptoms appear attenuated among Asian American youth compared to Whites (Tsai, Nguyen, Weiss, Ngo, & Lau, 2017). Controlled trials with diverse populations provide an opportunity to interrogate the role of putative modifiable risk factors for depression across groups.

The Current Study

The current study examined the role of preference in a school-based effectiveness trial using a two-stage trial design (Rücker, 1989) testing two EBPIs for adolescents at risk of depression. We examined the effects of EBPIs on (a) the ultimate outcomes of depression and anxiety symptoms, (b) the intermediate outcomes that were the targeted modifiable risk factor domains of emotion regulation (mindfulness, emotion suppression) and interpersonal stress (relationship problems, social support), and (c) initial treatment engagement. We had three research aims.

The first objective was to examine the effects of three distinct preference factors, which allowed us to separate different aspects of the preference process: (a) Does providing adolescents a choice among the two EBPIs influence outcomes? A primary mechanism through which patient preference is believed to operate is providing patients a sense of control over their care, resulting in increased motivation and investment (Winter & Barber, 2013). Because assignment condition reflects whether the adolescent is allowed to choose the EBPI, this factor may link to a sense of control over treatment. (b) Does receipt of one’s preferred EBPI influence outcomes? Preference has also been posited to enhance outcomes because patients may know what treatment is most likely to fit their needs (Swift & Callahan, 2009; Wittink, Morales, Cary, Gallo, & Bartels, 2013). The receipt of preferred program factor tests this possibility. (c) Does an adolescent’s baseline preference for one EBPI over the other predict outcomes? An individual’s baseline program preference for an EBPI, irrespective of the program received or how it was assigned, might predict outcomes if it reflects factors that relate to prognosis. For instance, adolescents who prefer IPT-AST may be high on social approach, which itself might predict better outcomes.

The second objective was to examine the relative effectiveness of the two EBPIs, apart from preference effects. Both EBPIs were selected on the basis of effectiveness in school-based trials, including trials with ethnically diverse samples. As such, we had no a priori hypotheses concerning the main effect of program. However, the third objective was to assess the extent to which adolescents’ baseline risk and demographic characteristics may moderate the effects of L2B and IPT-AST on our ultimate outcomes. Examining moderators of EBPI effects may identify characteristics that can help allocate treatments that would be most potent. We considered high baseline interpersonal distress as an indicator for benefitting most from IPT-AST, whereas high levels of emotion suppression might portend greater benefit from L2B. This objective aligned with personalized medicine, in which care decisions are based on predicted response as a function of individual differences (National Institute of Mental Health, 2015).

METHODS

Participants

Adolescents from two diverse public high schools in Southern California were recruited during two academic years (2014–15 and 2015–16). One school served 50% Asian American students, 14% Latinx students, and 31% White students. The second school served 10% Asian American, 19% Latinx, and 59% White students. At each school, one cohort of participants was recruited in the fall semester and one in the spring semester. To avoid overlap of cohorts, in 2014–15, 10thand 11th-grade students were recruited, and in 2015–16, 9thand 10th-grade students were recruited. Because there still was some risk of inclusion of the same students across the cohorts, we monitored for this within our recruitment tracking and confirmed that no participants were enrolled twice. Inclusion criteria included elevated depressive symptoms, with exclusion criteria including probable major depression, active suicidality, and developmental delay.

Figure 1 displays the CONSORT flow chart. A total of 623 adolescents, representing approximately 10% of students enrolled in the eligible grades in each cohort, obtained parental consent to participate in the study and completed an initial screening. Based on this initial screening, 196 (31%) screened positive for elevated depressive symptoms. Among this group, 53 students were subsequently determined ineligible based on our exclusion criteria (see below). Thus, 143 screened participants were determined to be eligible for the study and were randomized to an intervention condition. Post randomization, 32 students (22%) dropped out prior to participating in the baseline assessment or attending an intervention session, citing problems with scheduling or no longer being interested in participating in the study.

Figure 1.

CONSORT flow diagram indicating the numbers of participants in each stage of recruitment and participation, with (parentheses) indicating the numbers of participants who dropped from each cell in that stage.

Thus, 111 students participated in an intervention (M age = 15.18 years, SD = .86), 73% female (n = 81), 45 (41%) White, 40 (36%) Asian American, 8 (7%) Latinx, 1 (1%) African American, and 17 (15%) identified as multiracial or as a member of a racial/ethnic group not listed. Nearly all participants were born in the United States (n = 102, 92%), but 54 (49%) had at least one foreign-born parent. Nearly half of the participants reported that their father (n = 48, 43%) and mother (n = 54, 49%) had a college degree.

Procedures

Recruitment

In the 1st year of the study, school counselors and research staff visited classrooms to describe the study and interventions and distributed packets with parental consent forms and the initial screening instrument. Interested students returned the signed consent and completed screener to their classrooms over 1 week. Classrooms were given incentives (i.e., pizza lunch) if they returned all packets regardless of consent rates. In the 2nd year of the study, classroom teachers showed an informational video about the study and interventions and distributed a brochure describing the interventions and how to enroll. Students provided their assent (on paper or online) for the research team to contact their parents for consent. Research assistants called parents to discuss the study and parents completed an online consent form and the student was sent a link to complete the screener online. This study was approved by the Institutional Review Board at the University of California, Los Angeles.

Screening

The Mood and Feelings Questionnaire: Short Version (SMFQ; Angold & Costello, 1987) was the initial screener to identify adolescents with elevated depressive symptoms. There is no single established cutoff score for optimal sensitivity and specificity on the SMFQ. In the current trial, students who scored in the top 30% within their school cohort were considered positive screens, with cutoff scores ranging from 11 to 13 depending on the cohort. Prior studies have supported a cutoff as low as 10 (Kuo, Vander Stoep, & Stewart, 2005) and as high as 14 (McCarty, Violette, Duong, Cruz, & McCauley, 2013).

All 196 students scoring above their cohort cutoff were contacted for follow-up screening with a school counselor or research coordinator to confirm eligibility. The Patient Health Questionnaire (PHQ-9; Kroenke, Spitzer, & Williams, 2001) was administered to identify students with severe depression symptoms who would not be best served by a low-intensity EBPI. Previous prevention trials for adolescent depression have also excluded youth with probable (Seligman, Schulman, & Tryon, 2007) or assessed major depression (Gillham et al., 2007). We followed procedures used by McCarty et al. (2013), who used the PHQ-9 to identify probable major depression or active suicidal ideation for exclusion and referral to individual counseling. Of the 196 participants who screened positive on the SMFQ, 44 (22.4%) were screened out because of probable diagnostic-level depression or suicidality on the PHQ-9, four (2.0%) were ineligible because of developmental delays documented by school administrators, and six (3.1%) indicated they were no longer interested in the study.

Randomization and Program Assignment

After the secondary screening, eligible students were told they would be randomly assigned to (a) receive their preferred EBPI or (b) receive a randomly assigned EBPI: Based on this screening, you are eligible to participate in the Wellness group. Here are descriptions of the two programs we are offering. Please take your time and read it over. As part of the research, some students will get to pick the program that they feel is best for them, and other students will be randomly assigned to a program. Participants read a paragraph-long description of each EBPI, were given time to ask questions, and then stated their preference. The researcher opened an envelope with a card indicating their assignment condition, with either preference random assignment condition printed on the front. Random assignment cards listed the EBPI assignment on the reverse. Of the 143 eligible adolescents, 86 (60.1%) stated a preference for L2B and 57 (39.9%) for IPT-AST. Seventy-four participants were randomly assigned to the preference condition (36 had chosen IPTAST and 38 had chosen L2B) and 69 participants to the random assignment condition (34 were assigned to IPT-AST and 35 to L2B). Of the 143 adolescents allocated to intervention, 111 (78%) ultimately attended at least one intervention session, and of these, six (5%) formally discontinued by stating they no longer wished to attend. Regardless of assignment condition, 84 of 111 (76%) participants were ultimately assigned to their preferred EBPI.

Scheduling and Group Composition

Following the assignment of participants to either L2B or IPT, participants were informed of their group meeting schedule. Groups met after school or during the school day. When groups met during the day, they were offered during rotating class periods to minimize the time missed from a single course, and these were excused absences. At each school, there was one L2B group and one IPT group run concurrently each semester for a total of 16 groups across four semesters. The average group size was 7.0 in IPT (range = 5–9) and 8.25 in L2B (range = 5–13). All but one group had majority female composition (IPT: M = 65.3% female, range = 50%–88.9%; L2B: M = 74.1% female, range = 44.4%–100%).

Assessments

Participants completed three assessments: (a) pretreatment (baseline, 1 week before the intervention began), (b) posttreatment (the week following the last session of the intervention), and (c) 3-month posttreatment follow-up. Assessments were administered by the study personnel using online surveys in school computer labs after school hours. Adolescents received gift cards worth $20 at the baseline and post assessments and $25 at follow-up.

MEASURES

Screening Measures

Depression Symptoms

The 13-item SMFQ (Angold & Costello, 1987) was used to initially screen adolescents for elevated depressive symptoms. Respondents rated their experience of symptoms (e.g., “I felt miserable or unhappy”; “I found it hard to think properly or concentrate”) on a 3-point Likert-type scale: 0 (not true), 1 (sometimes true), 2 (true). The SMFQ has good internal consistency with adolescent samples as a screener in EBPI trials (e.g., Kuo, Stoep, & Stewart, 2005; McCarty et al., 2013).

Depression Severity

The PHQ-9 (Kroenke et al., 2001) was used to identify students with depression severity that indicates the need for treatment rather than prevention. Items map onto criteria for major depressive disorder and assess impairment and suicidal ideation. Participants who scored in the PHQ-9 range for probable major depression (above 16) or who reported active suicidal ideation were excluded and referred for counseling. The PHQ-9 has been used reliably with ethnically diverse adolescents (e.g., Fung et al., 2019).

Screening measures were hand scored by school and research personnel at school sites, and item-level data were not retained, so we are not able to report on internal consistency of these measures within our sample. However, both measures are widely used and have demonstrated reliability with ethnically diverse adolescents.

Ultimate Outcomes

Anxiety and Depression Symptoms

The Revised Child Anxiety and Depression Scales–Short Version (RCADS; Ebesutani et al., 2012) was administered to measure anxiety and depression symptoms as the ultimate outcomes. Respondents rated how often they experienced symptoms of anxiety (15 items; e.g., “I worry that I will suddenly get a scared feeling when there is nothing to be afraid of”) and depression (10 items; e.g., “I feel sad or empty”) on a 4-point Likert-type scale from 0 (never) to 3 (always). Internal consistency was good for both anxiety (α T1 = .82, T2 = .85, T3 = .86) and depression (α T1 = .82, T2 = .88, T3 = .88).

Intermediate Outcomes

We selected two measures to assess intermediate outcomes for each EBPI. For L2B, we assessed mindfulness/acceptance and emotion suppression previously shown to mediate effects of L2B (Fung et al., 2019). For IPT-AST, we assessed interpersonal problems and social support that are intervention targets (Young et al., 2012). These measures were used as (a) intermediate outcomes and (b) potential moderators or baseline indications that may predict EBPI response.

Acceptance/Mindfulness

The 25-item Child Acceptance and Mindfulness Measure (CAMM; Greco, Dew, & Baer, 2006) measured the extent to which participants endorsed mindfulness and acceptance in their daily lives, such as observing internal experiences (e.g., “I notice when my feelings begin to change”), acting with awareness (e.g., “I do many things at once”; reverse scored), and acceptance of emotions (e.g., “I tell myself that I shouldn’t feel the way I’m feeling”; reverse scored). Each item is rated on a 5-point scale from 0 (never true) to 4 (always true). The CAMM has shown good internal consistency in previous adolescent samples (Greco et al., 2006) and had adequate internal consistency in the current sample at post and follow-up assessments but was lower at baseline (α T1 = .57, T2 = .65, T3 = .72).

Emotion Suppression

The four-item Expressive Suppression subscale of the Emotion Regulation Questionnaire for Children and Adolescents (ERQ; Gullone & Taffe, 2012) was used to assess emotion suppression tendencies (e.g., “I keep my feelings to myself”). Each item is rated on a 5-point scale from 0 (strongly disagree) to 4 (strongly agree). Internal consistency for was acceptable in this sample (α T1 = .75, T2 = .74, T3 = .77).

Interpersonal Problems

The 32-item short version of the Inventory of Interpersonal Problems (IIP-32; Horowitz, Alden, Wiggins, & Pincus, 2000) was used to assess interpersonal behavioral inhibitions (“I find it hard to join in on groups”) and behavioral excesses (e.g., “I fight with other people too much”). Participants rated themselves on each interpersonal difficulty on a 5-point scale from 1 (not at all) to 5 (extremely). The IIP-32 had strong internal consistency in the current sample (α T1 = .85, T2 = .88, T3 = .90).

Social Support

Participants’ perceptions of family and friend social support was measured using the Multidimensional Scale of Perceived Social Support (MSPSS; Zimet, Dahlem, Zimet, & Farley, 1988). Participants rated their level of agreement with four statements on how much their family (e.g., “My family really tries to help me”) and friends (e.g., “I can count on my friends when things go wrong”) support them, on a 6-point scale from 1 (strongly disagree) to 6 (strongly agree). Internal consistency was high for Family Support (α T1 = .88, T2 = .89, T3 = .89) and Friend Support (α T1 = .90, T2 = .93, T3 = .90).

Treatment Engagement Outcome

Initial treatment engagement was indexed by tracking whether each participant attended at least one session (N = 111, 78%). Initial engagement was considered an important outcome given that indicated prevention trials in schools can have rates as low as 50% participation among students identified with elevated internalizing symptoms (e.g., Masia Warner, Fisher, Shrout, Rathor, & Klein, 2007; Young et al., 2006).

Demographic and Background Characteristics

Adolescents’ age, sex, race/ethnicity, immigrant generation status, and parental educational attainment were assessed on a demographic questionnaire. Because many participants could not confidently report family income, we assessed perceptions of financial hardship.

Family Financial Hardship

Eight items from the Responses to Stress Questionnaire (Connor-Smith, Compas, Wadsworth, Thomsen, & Saltzman, 2000) were used to assess experience of financial hardship over the past three months (e.g., “My family didn’t have enough money to pay the bills”). Participants rated their agreement on a 4-point scale from 0 (not true) to 3 (very true). The Financial Hardship scale has shown good internal consistency (Connor-Smith et al., 2000) and had acceptable internal consistency in the current sample (α = .69).

Interventions

IPT-AST (Young, Mufson, & Schueler, 2016) is a school based prevention program designed to improve interpersonal functioning to prevent depression onset. Participants first meet individually with a therapist to set their individual goals, identifying strengths and challenges in their key relationships. Groups begin with psychoeducation defining depression and prevention and discussing the links between moods and interpersonal interactions. The next stage focuses on developing communication skills through discussion, practice with other students in the group, and planned application of skills in targeted relationships outside group. Adolescents meet individually with a therapist to review progress on their goals mid-treatment. Thus, there were two individual and 10 group sessions in weekly 50 min sessions.

L2B (Broderick, Kabat-Zinn, & Kabat-Zinn, 2013) is a mindfulness-based intervention designed to improve stress management and adaptive emotion regulation. Each lesson includes an introduction to the topic, group activities and discussion, in-session mindfulness practice, and a homework assignment. The six themes of the curriculum are denoted by the acronym BREATHE: B—listen to your Body, R—Reflections are just thoughts, E—surf the waves of your Emotions, A—Attend to the inside and the outside, T—try Tenderness, and H—practice healthy Habits of mind. Two-week modules focus on one theme, with the end goal of becoming Empowered (the last E). L2B was designed for classroom use by school personnel without formal mental health training and involved 12 weekly 50 min sessions.

Interventionists, Training, and Adherence

Interventionists included school personnel (including fulltime staff [four master’s-level school psychologists, one speech and language therapist] and two school psychology interns) supplemented by two research staff (a postdoctoral fellow with a doctorate in social work, a study coordinator with a B.A. in psychology). Each group was co-led by a primary leader (full-time school staff or the postdoctoral fellow) and a coleader (a school psychology intern or the study coordinator). Therapist training for both EBPIs included a 2-day workshop with the developer, followed by weekly group supervision during the 1st year and biweekly group consultation in the 2nd year. For IPT-AST, supervision was provided by phone/video conference with a master trainer who reviewed session recordings and provided performance feedback for every other session. In the 2nd year, the same master trainer led consultation calls every other week without session review and performance feedback. For L2B, supervision was of lower intensity because the curriculum was designed to be delivered by teachers with minimal supervision and clinical oversight. In the 1st year, group leaders had weekly phone/videoconference supervision with the first author (who led previous L2B trials). In the 2nd year, group phone consultation was held every other week with the L2B developer.

All sessions were audio-recorded to assess adherence. For each EBPI, 20 sessions (20.8%) were coded for adherence. Independent teams coded L2B and IPT-AST session recordings for coverage of manualized components. Coders received an orientation on their respective EBPI, reviewed the manual, and received training on adherence coding procedures. Coders then rated gold standard sessions and received performance feedback, and they were considered reliable once they met 80% accuracy. Across both EBPIs, 20% of adherence checks were randomly selected for coding by a second rater. Inter-reliability was good for adherence scores with independent coder scores correlated at r = .83 for L2B and .95 for IPT-AST. For both L2B and IPT-AST, sessions were coded using the adherence checklist supplied by the EBPI developer. For both EBPIs, sessions were given 1 point for each activity listed in the checklist for that session that was observed. An adherence score was calculated by dividing the total number of adherence points credited by the total possible points for the session. The average session adherence score was 87.2% for L2B, and 93.0% for IPT-AST, indicating good coverage of components.

Statistical Analyses

To assess the effects of various factors on change in the outcome variables, we used mixed models with time as a random factor nested within subject, and the various predictors as fixed effects. Because assignment to program was only partially random (i.e., approximately 50% of were assigned based on preference), we used propensity scoring covariates to adjust for any baseline differences resulting from nonrandom assignment (Pan & Bai, 2015) that could bias treatment effect estimates by relations between pretreatment factors and conditions (Crano, Brewer, & Lac, 2015). Propensity analyses can also increase the power of the analyses by reducing between-group pretreatment error variance (Pan & Bai, 2015). Propensity scoring covariates were computed based on demographic variables including ethnic minority status, sex, immigrant family status, and financial hardship, and baseline values of intermediate and ultimate outcomes. Because the various analyses tested different independent variables (e.g., assignment condition; receipt of preferred program), propensity covariates were calculated for each model.

RESULTS

All analyses were intent-to-treat, with all available data included. Of the 143 participants eligible to participate, 100 (70%) provided data for all three time points, 122 (85%) provided data for at least one time point, and 21 (15%) participants were assessed as eligible but did not participate in any assessments. Thus, data analyses included 122 participants. Participant demographic characteristics are included in Table 1.

Table 1.

Participant Characteristics, by Program Assigned

	IPT-AST (n=49)		L2B (n=62)		Total sample (N=111)
Demographic Characteristics	n	(%)	n	(%)	n	(%)
Female	34	71%	47	76%	81	74%
Race/ethnicity
African American	1	2%	0	0%	1	1%
Asian American/Pac Islander	17	35%	23	37%	40	36%
Hispanic/Latinx	6	12%	2	3%	8	7%
White (Non-Hispanic)	21	43%	24	39%	45	41%
Other and Multiracial	4	8%	13	21%	17	15%
Family immigrant status1	26	53%	28	46%	54	49%
Baseline Measures	M	SD	M	SD	M	SD
Financial Hardship2	0.65	0.74	0.42	0.54	0.53	0.65
RCADS Anxiety	30.65	7.63	32.26	7.27	31.55	7.44
RCADS Depression	22.96	5.06	24.08	5.00	23.59	5.03
CAMM Acceptance/mindfulness	46.88	8.32	43.65	7.97	45.07	8.25
IIP Interpersonal problems	79.43	16.04	84.45	16.99	82.23	16.69
ERQ Emotion suppression	9.29	3.27	9.18	3.10	9.23	3.16
MSPSS Family support	3.66	1.25	3.46	1.20	3.55	1.22
MSPSS Friend support	4.51	1.18	4.05	1.29	4.25	1.26

Notes:

¹= Adolescent has at least one foreign-born parent.

²= The Financial Hardship measure runs on a Likert Scale of 0 (not true) to 3 (very true).

Research Question 1: Effects of Preference Variables

For the seven continuous dependent variables (i.e., RCADS anxiety, RCADS depression, CAMM acceptance/mindfulness, IIP interpersonal problems, ERQ emotion suppression, MSPSS family support and MSPSS friend support), we used a mixed model with time (the three assessments) as a random factor nested within subject and the propensity covariate in each model. The effects of interest were the interactions between assignment condition, receipt of preferred program, and program preference with time. For the engagement outcome, we used a generalized linear model with a binary distribution and a logit link function. Because there was a single binary endpoint, there was no time factor, and the effects of interest were the main effects of the independent variable.

RQ1a: Effects of Assignment Condition

We first examined whether participants who were assigned to preference versus random assignment differed across the outcomes. To evaluate the effectiveness of the propensity covariate adjustment, we tested for the demographic variables and the seven T1 scores on outcome variables as a function of assignment condition, including the propensity covariate in the model. Probability values for the effect of assignment condition in these models ranged from .95 to .99, indicating near complete adjustment. Across the seven continuous outcomes, the Time × Assignment Condition interaction was significant only for CAMM (see Table 2a). For participants in the random assignment condition, CAMM scores increased 3.33 points per time period (p < .0001), whereas the effect of time was not significant in the preference condition. There were no effects of assignment condition on engagement.

Table 2.

Research Question 1: Effects of program preference variables on ultimate, intermediate, and engagement outcomes

	RCADS Anx	RCADS Dep	CAMM	IIP	Emo Suppress	MSPSS Fam	MSPSS - Fri	Engagementd
A. Effects of Program Assignment condition
Time	F(1,116)=41.57****	F(1,116)=12.54***	F(1,116)=24.09****	F(1,116)=7.35**	F(1,116)=33.03****	F(1,107)=12.16***	F(1,107)=12.06***
Assignment Conditiona	F(1,99)=0.13	F(1,99)=0.26	F(1,99)=1.16	F(1,99)=1.74	F(1,99)=0.00	F(1,97)=0.01	F(1,97)=0.01	χ2(1)=0.47
Time X Assignment Conditiona	F(1,99)=0.73	F(1,99)=0.24	F(1,99)=6.64*	F(1,99)=2.30	F(1,99)=0.17	F(1,97)=0.39	F(1,97)=0.02
R2			.06
B. Effects of Receipt of Preferred Program
Time	F(1,116)=23.95****	F(1,116)=6.47*	F(1,116)=15.15***	F(1,116)=1.08	F(1,116)=23.52****	F(1,107)=5.87*	F(1,107)=5.53*
Preferred Programb	F(1,98)=0.35	F(1,98)=0.60	F(1,98)=0.02	F(1,98)=2.95	F(1,98)=0.07	F(1,97)=0.22	F(1,97)=0.13	χ2(1)=0.81
Time X Preferred Progb	F(1,98)=0.84	F(1,98)=0.62	F(1,98)=0.01	F(1,98)=5.43*	F(1,98)=0.01	F(1,97)=0.55	F(1,97)=0.85
R2				.05
C. Effects of Baseline Program Preference
Time	F(1,116)=33.45****	F(1,116)=10.41**	F(1,116)=14.78***	F(1,116)=7.28**	F(1,116)=29.88****	F(1,107)=10.92**	F(1,107)=7.72**
Baseline Preferencec	F(1,99)=0.81	F(1,99)=0.05	F(1,99)=2.33	F(1,99)=0.02	F(1,99)=0.04	F(1,97)=0.14	F(1,97)=2.10	χ2(1)=1.34
Time X Baseline Prefc	F(1,99)=5.55*	F(1,99)=1.09	F(1,99)=5.89*	F(1,99)=0.07	F(1,99)=0.41	F(1,97)=0.02	F(1,97)=5.42*
R2	.05		.06				.05

Notes: RCADS Anx = RCADS Anxiety Score. RCADS Dep = RCADS Depression Score. CAMM = Child Acceptance and Mindfulness Measure. IIP = Inventory of Interpersonal Problems. Emo Suppress = Expressive Suppression subscale of the Emotion Regulation Questionnaire for Children and Adolescents. MSPSS – Fam = Family Support subscale from the Multidimensional Scale of Perceived Social Support. MSPSS – Fri = Friends Support subscale from the Multidimensional Scale of Perceived Social Support.

^a= Reference group is Random Assignment (=0), Preference (=1)

^b= Reference group is Non-preferred Program Received (=0), Preferred Program Received (=1)

^c= Reference group is Baseline Preference for IPT-AST (=0), Baseline Preference for L2B (=1)

^d= Because Treatment Engagement is a binary, non-longitudinal factor, there are no parameter estimates for Time, and Time X (program preference variable).

^*= p<.05;

^**= p<.01;

^***= p<.001;

^****= p<.0001.

RQ1b: Effects of Receipt of Preferred Program

The next analyses examined whether participants who received the program for which they had stated a preference (IPT-AST or L2B) showed differential outcomes, regardless of assignment condition. For the seven continuous outcomes, the effects of interest were the Time × Receipt of Preferred Program interactions, and for the binary engagement outcome, the effect of interest was receipt of preferred program. Testing the independent variable with its propensity covariate, the minimum probability value for the effect of receipt of preferred program was .96. Across the eight outcomes, only the Time × Receipt of Preferred Program interaction for IIP was significant (see Table 2b). For individuals who received their preferred program, IIP scores decreased −4.10 points per time period, whereas for those who did not receive their preferred program, the effect of time did not differ significantly from 0. No other dependent variables showed significant effects.

RQ1c: Effects of Baseline Program Preference

We next examined whether participants who preferred IPT-AST had different outcomes than those who preferred L2B (irrespective of the program received), by testing the Time × Baseline Program Preference interaction. The minimum probability value for the effect of baseline program preference using the propensity covariate was .96, again indicating near complete adjustment. For three of the seven continuous outcomes (RCADS anxiety, CAMM, MSPSS friends support), the Time × Baseline Program Preference interaction was significant (see Table 2c), such that participants who preferred L2B showed significantly greater improvement over time than participants who preferred IPT-AST. For RCADS anxiety, adolescents who preferred IPT-AST showed a 1.06 point decrease per time unit (p < .05), whereas adolescents who preferred L2B showed a 2.50 point decrease per time unit (p < .0001). For CAMM and MSPSS friends support, for adolescents who preferred IPT-AST at baseline the effect of time was nonsignificant (i.e., did not differ from 0), whereas adolescents who preferred L2B showed a 2.96 decrease and a 0.26 increase in CAMM and MSPSS friends support, respectively. There was no significant effect of baseline program preference on the binary engagement outcome.

Research Question 2: Effects of Program

To examine relative effectiveness, we assessed the effects of program (L2B vs. IPT-AST) on outcomes. For the continuous outcomes, the effects of interest were the Time × Program interactions. For the engagement outcome, the effect of interest was the program effect. As shown in Table 3a, the Time × Program effect was significant for five of the eight outcomes. As Table 3b indicates, for each of the five significant Time × Program effects, the effect of time was larger and significant only for L2B as compared to IPT-AST. For instance, L2B participants’ anxiety scores decreased by 2.95 points on average per time period, whereas the change slope was −0.67 in the IPT-AST arm, which did not differ significantly from zero. Concerning the two outcomes with no Time × Program interaction, participants overall showed significant improvement across time: B(Time) = −.76, p < .0001 for emotion suppression; B(Time) = .19, p < .0005 for family support. Thus, significant improvement was observed for all continuous outcomes, with five of seven showing differential improvement between Programs.

Table 3.

Research Question 2: Effects for Potential Moderators of Ultimate Outcomes

	RCADS Anx	RCADS Dep	Engagement
T1 CAMM
Time	F(1,118)=44.04****	F(1,118)=13.85***
T1 CAMM	F(1,100)=11.55***	F(1,100)=20.32****	χ2(1)=0.09
Time X T1 CAMM	F(1,100)=3.49†	F(1,100)=6.76*
R2		.06
TI IPP
Time	F(1,118)=46.08****	F(1,118)=14.34***
T1 IPP	F(1,100)=40.16****	F(1,100)=18.21****	χ2(1)=0.06
Time X T1 IPP	F(1,100)=15.35***	F(1,100)=3.23†
R2	.13
T1 Emotion Suppression
Time	F(1,118)=43.40****	F(1,118)=13.30***
T1 Emo Suppression	F(1,100)=13.94***	F(1,100)=26.28****	χ2(1)=4.57*
Time X T1 Emo Suppress	F(1,100)=2.59	F(1,100)=4.52*
R2		.04	.04
TI Family Support
Time	F(1,118)=45.39****	F(1,118)=14.48***
T1 Family Supp	F(1,100)=10.78**	F(1,100)=23.69****	χ2(1)=2.66
Time X T1 Family Supp	F(1,100)=10.12**	F(1,100)=12.77***
R2	.09	.11
TI Friend Support
Time	F(1,118)=42.22****	F(1,118)=13.67***
T1 Friend Supp	F(1,100)=8.06**	F(1,100)=16.75****	χ2(1)=1.87
Time X T1 Friend Supp	F(1,100)=7.19**	F(1,100)=6.87*
R2	.07	.06

Note: RCADS Anx = RCADS Anxiety score. RCADS Dep = RCADS Depression score. CAMM = Acceptance and Mindfulness score. IIP = Inventory of Interpersonal Problems score. Friend Support = Friend subscale from the MSPSS. Family Support = Family subscale from the MSPSS. L2B = Learning to Breathe program. IPT-AST = Interpersonal Therapy- Adolescent Skills Training program.

^*= p<.05;

^**= p<.01;

^***= p<.001;

^****= p<.0001.

R² is effect size estimate for significant interactions.

Research Question 3: Moderators of EBPI Effects on Ultimate Outcomes and Engagement

The objective of Research Question 3 was to identify moderators of treatment effects (L2B vs. IPT-AST) on the ultimate outcomes and engagement. We used mixed models for the two continuous dependent variables (RCADS anxiety and depression), including time, program, and moderator (e.g., T1 CAMM) fully crossed models (i.e., including up to the three-way interactions), and the propensity covariate. The effects of interest were the Time × Program × Moderator interactions. For the engagement outcome, we used a generalized linear model with a binary distribution and a logit link function, and tested the Program × Moderator interaction.

The five potential moderators were the baseline values of the intermediate outcomes CAMM, ERQ emotion suppression, IIP, and MSPSS friends and family support. The minimum probability value for the effect of program using the propensity covariate was .97, indicating near complete adjustment. None of the 15 tests of Time × Program × Moderator interactions (three outcomes by five moderators) were significant, indicating that relative effectiveness of L2B versus IPT-AST did not vary as a function of these moderators. These tests evaluated the extent to which differential program effects (for L2B vs. IPTAST) occurred as a function of the various baseline indicators. We also tested whether these moderators influenced outcomes but not differentially by treatment by examining the Time × Moderator interactions in the same models. For the two continuous outcomes across the five potential moderators, seven of the 10 interactions were significant, with two additional interactions being marginally significant (see Table 4). For interpretation, we estimated the parameters for the relation between time and the outcome at −1 and +1 SD from the mean of the moderator. All significant interactions revealed the same pattern: Participants with less positive (lower CAMM, friend and family support) or more negative (higher IIP, emotion suppression) functioning at T1 showed greater improvement over time. For instance, for RCADS anxiety, the parameter estimate for time at +1 SD above the mean on IIP interpersonal problems was −3.20, whereas it was −0.81 at −1 SD from the IIP mean. In addition, one effect predicting engagement was significant (see Table 4) such that higher T1 emotion suppression was associated with increased engagement.

Table 4.

Effects of Program

A. Time X Program effects
	RCADS Anx	RCADS Dep	CAMM	IIP	Emo Suppress	MSPSS – Fam	MSPSS - Fri	Engagement
Time	F(1,116)=38.93****	F(1,116)=10.63**	F(1,116)=18.33****	F(1,116)=6.83*	F(1,116)=32.22****	F(1,107)=10.38**	F(1,107)=9.42**
Program	F(1,99)=2.31	F(1,99)=2.20	F(1,99)=1.63	F(1,99)=0.65	F(1,99)=0.04	F(1,97)=0.58	F(1,97)=3.12	χ2(1)=1.38
Time X Program	F(1,99)=15.33***	F(1,99)=10.07**	F(1,99)=11.79***	F(1,99)=4.01*	F(1,99)=0.74	F(1,97)=2.75	F(1,97)=16.10****
R2	.13	.09	.11	.04			.14

B. Simple effects of Program, on outcome variables with significant Time X Program effects
	RCADS Anx		RCADS Dep		CAMM		IIP		MSPSS - Fri
	IPT-AST	L2B	IPT-AST	L2B	IPT-AST	L2B	IPT-AST	L2B	IPT-AST	L2B
Beta (Time)	−0.67	−2.95****	−0.02	−1.78****	0.40	3.46****	−0.54	−4.61***	−.05	0.34****
C.I.	−1.51 to 0.17	−3.69 to −2.21	−0.82 to 0.78	−2.51 to −1.05	−0.68 to 1.48	2.17 to 4.75	−3.48 to 2.40	−7.22 to −2.00	−0.21 to 0.11	0.22 to 0.46

Note: RCADS Anx = RCADS Anxiety score. RCADS Dep = RCADS Depression score. CAMM = Child Acceptance and Mindfulness score. IIP = Inventory of Interpersonal Problems score. Emo Suppress = Expressive Suppression subscale of the Emotion Regulation Questionnaire for Children and Adolescents. MSPSS – Fam = Family Support subscale from the Multidimensional Scale of Perceived Social Support. MSPSS – Fri = Friends Support subscale from the Multidimensional Scale of Perceived Social Support.

^*= p<.05;

^**= p<.01;

^***= p<.001;

^****= p<.0001.

Beta estimate represents the number of units change in the dependent variable per Time unit. C.I. = 95% confidence interval.

DISCUSSION

The primary study objective was to assess the potential effects of adolescents’ EBPI preferences on outcomes. Although previous research has sometimes—but not consistently—found positive effects for preference (Swift & Callahan, 2009; Winter & Barber, 2013), we found little evidence that being given a choice to select one’s EBPI or receiving one’s preferred EBPI were associated with enhanced outcomes. Across eight ultimate, intermediate, and engagement outcomes, there was one significant effect for assignment condition and one for receipt of preferred program (two of 16). Within the two significant results, only one favored preference. These results are not highly discrepant with a review by Winter and Barber (2013), who found that slightly less than 25% of studies demonstrated significant preference effects on the ultimate outcomes, although more than 60% of studies did find significant effects on “indirect variables” (e.g., treatment satisfaction). In considering possible explanations for the lack of preference effects, it is important to consider how preference theoretically impacts outcomes. A hypothesized mechanism through which patient preference may operate is by providing a sense of control over the treatment process (Winter & Barber, 2013), resulting in increased motivation and investment. Our lack of findings may reflect the possibility that once adolescents began in an EBPI, regardless of their initial preference, the interventionists successfully engaged them and cultivated a sense of control over their mood and stress. It is also possible that the structural differences between the two EBPIs were minimal (compared to preference for medication vs. psychotherapy, the focus of prior research), hence the effects of our preference conditions were relatively minimal. The more clear distinction between medication versus “talk therapy” may afford a greater sense of agency than an option between two school-based group-delivered EBPIs.

It also has been suggested that providing clients with the opportunity to select a preferred treatment may be associated with enhanced outcomes because clients may know what intervention would work best for them (Swift & Callahan, 2009; Wittink et al., 2013). One explanation for our minimal effects of preference is that adolescents do not yet have the self-knowledge necessary to make predictions about which of the EBPIs would be most useful (Committee on Bioethics, 1995; Coyne, 2006). Further, some adolescents may have selected an EBPI based on their indicated needs, whereas others made choices based on strategies they are most comfortable with (e.g., sociotropic adolescents selecting IPT-AST). Heterogeneity in how adolescents arrived at preferences may explain the lack of systematic effects on outcomes. A qualitative examination of how adolescents select between options is an important future direction. Relatedly, although adolescents made a choice, we did not assess the strength of their preference. The difference between L2B and IPT-AST may appear minor (both are group skills-based EBPIs) and a weak or moderate preference may be insufficient to impact outcomes.

Offering a choice among two discrete treatment options represents only one variant of patient-centered care. A collaborative care approach more generally involves shared decision-making processes that involve ensuring that the patient fully understands the intervention and is helped to see its relevance and fit with his or her needs. There has been recent attention to the application of shared decision-making in adolescent mental health (Abrines-Jaume et al., 2016) with correlational evidence of benefits of shared decision-making processes on psychotherapy outcomes (Edbrooke-Childs et al., 2016). It may not be feasible in low resourced settings to offer choices of programs; it may be more actionable to enhance engagement by framing EBPIs as relevant to student and caregiver-identified needs and concerns (Yeh et al., 2019).

Aside from questions concerning the effects of receiving a preferred intervention on outcomes, our findings suggest that the baseline preference for a specific EBPI may itself be a marker of the likelihood of benefiting from any intervention. For three of the seven outcomes, adolescents who preferred L2B showed greater improvement than participants who preferred IPT-AST, irrespective of the program they received. The outcomes sensitive to this difference included an ultimate outcome (anxiety) and intermediate outcomes targeted by IPT-AST (friend support) and L2B (acceptance/mindfulness). It is unclear what may account for this effect; however, a baseline preference for L2B may reflect an adolescent’s insight into struggles with, or personal interest in, present moment awareness and nonjudgmental acceptance.

In terms of our second objective, we found that L2B was associated with greater rates of improvement than IPT-AST overall. Five of seven ultimate and intermediate outcomes showed significant effects favoring L2B, whereas no effects favored IPT-AST. L2B superiority was observed even for intermediate outcomes ostensibly more directly targeted by IPT-AST (interpersonal problems, friend support). Several possible explanations may underlie this apparent superiority of L2B. First, differences in implementation could account for differential effectiveness. Assessed adherence for both interventions was high, but we did not directly assess fidelity or therapist competence. IPT-AST is an intervention derived from a process-oriented psychotherapy model, which was then adapted as a skills-focused intervention that could be delivered in schools (Young, Mufson, & Schueler, 2016). It involves individualized assessment and goal-setting, didactic instruction and individually tailored skills training, and attention to group process. In comparison, L2B was designed to be implemented by educators without mental health training, with lessons and exercises that require little individualization (Broderick et al., 2013). Thus, although the time spent in training, supervision, and consultation was equivalent across the two programs, IPT-AST is a more complex intervention to deliver. Previous trials of IPT-AST have largely involved clinical psychologists and doctoral students as providers, with ongoing supervision by the developer team. Whereas our interventionists were primarily school counselors delivering the program within their broader job demands and resource constraints. A systematic review indicated that school-based EBPIs for youth depression are most effective when implemented by outside interventionists than school personnel (Calear & Christensen, 2010). Furthermore, diminished effects are observed with successive generations beyond the EBPI developer (McGrew, Bond, Dietzen, & Salyers, 1994) and toward representative practice conditions (Curtis, Ronan, & Borduin, 2004). Diminished fidelity of IPT-AST might explain the differential EBPI effects insofar as L2B may be relatively easier to implement. Our pre–post effect sizes for IPT-AST were small for the ultimate outcomes (d in the .20–.30 range), compared to very large pre–post effects in the developer-led trials (d in the 2.0–3.0 range; Young et al., 2006, 2010, 2012). In contrast, the pre–post effect sizes for L2B in the current study were somewhat larger (d in the .60 range) than those reported in developer-led L2B trials (d in the .30–.40 range; Bluth et al., 2016; Broderick & Metz, 2009; Metz et al., 2013).

A second potential explanation for the observed superiority of L2B is that it was a better fit for the needs of the sample. Our research in these schools indicated that limited emotion expression and strained family relationships predicted increased internalizing symptoms among Vietnamese American and White youth (Nguyen et al., 2018; Tsai et al., 2017), which suggests that the focus of IPT-AST could be useful. However, target skills related to self-expression and assertiveness may be difficult for some diverse adolescents from immigrant families. For example, values of family obligation lower the likelihood of seeking help for emotional problems and family stressors among Asian American youth (Guo, Nguyen, Weiss, Ngo, & Lau, 2015), and these values may also make the application of IPT-AST strategies less comfortable. However, we did not find moderation of EBPI effects by ethnic minority or immigrant family status, and previous IPT-AST trials produced large effects with high representation of ethnic minority youth (Young et al., 2006, 2010, 2012). It may be the case that L2B worked better in our sample with lower depression severity, whereas IPT-AST may be better suited for youth who need more intensive intervention that requires introspection and a willingness to share vulnerabilities in a group context.

In contrast to IPT-AST, L2B emphasizes adolescents making changes in secondary coping by adjusting the self in relation to stressful experiences and distressing emotions rather than acting more directly on the social environment. L2B does not require youth to change the nature of their interactions with others but instead focuses on enhancing present moment awareness and acceptance of emotions, thoughts, and bodily sensations—changes that may be perceived as more controllable and feasible. L2B was associated with self-reports of improved friend support and interpersonal skills to a greater extent than IPT-AST, although L2B did not produce stronger effects on emotion suppression or family support. Collectively the findings provided little support for the anticipated specificity of intermediate targets for the EBPIs.

Relatedly, within our third study objective, the lack of Program × Time × Moderator interactions suggested a non-specificity of EBPI effects relative to their intermediate outcome targets. Research on personalized medicine has found few significant moderators predicting differential response to alternative treatments. Research has yet to identify markers with “clinical utility to inform the choice between medication and psychotherapy, the selection of specific medication, or the selection of a specific psychotherapy” for depression (Simon & Perlis, 2010, p. 1445). The moderators examined in our study were selected because they are primary clinical targets (e.g., L2B targets mindful awareness, IPT-AST targets relationship stress). Our pattern of results fits with the proximal process treatment moderator pattern proposed by Weiss, Han, Tran, Gallop, and Ngo (2015). Proximal process moderators are intermediate outcomes targeted by the intervention that are causally proximal to the ultimate outcomes. Individuals with lower levels of proximal process factors targeted by the intervention are expected to profit more from the intervention, because the causes of their problems are targeted by the intervention. In the present study we found Time × Moderator effects with participants with lower baseline functioning in domains targeted by the interventions (i.e., lower mindfulness and social support, higher interpersonal problems and emotion suppression) showing greater improvement across EBPIs.

Several study limitations should be considered when interpreting our results. First, the rate of participation in the initial screening was low at 10%. However, given that the programs were introduced to students as prevention for depression, it would be expected that most students without mood problems would have little interest in participating. Second, we cannot estimate the absolute (vs. relative) effectiveness of the two EBPIs, because there was no control group. The within-group effects of L2B compared favorably with prior trials (in terms of pre–post change) and IPT-AST less so, but absolute effectiveness cannot be assessed. Third, all outcome measures were self-report, with potential for bias. However, given that the ultimate outcomes were affective symptoms, self-report is likely the most sensitive single informant approach (e.g., Cantwell, Lewinsohn, Rohde, & Seeley, 1997). Fourth, some prior research suggests that preference effects may be related more to engagement than to effectiveness (Winter & Barber, 2013). We assessed only initial engagement and were not able to include measures of attendance or observed participation, which might be more sensitive to preference. Fifth, in terms of implementation, we assessed adherence indexed by coverage of expected components, but we did not obtain other measures of integrity such as fidelity or competence, which may have differentiated delivery of the two EBPIs. Sixth, because this study followed recommendations that clinical trials should attempt to identify moderators of treatment effects as well as ultimate outcomes (National Institute of Mental Health, 2015), we conducted a number of analyses beyond the basic outcome analyses. We considered using alpha adjustments to reduce risk of Type I error; however, conventional corrections (e.g., Bonferroni) are known to be overly conservative (Perneger, 1998) and as such we elected not to use a correction to avert increased risk of Type II error. Finally, the two EBPIs were delivered concurrently in the schools, and students were the unit of randomization, so there was potential for contamination across programs, but this would have rendered a more conservative test of EBPI differences (Werner-Seidler, Perry, Calear, Newby, & Christensen, 2017).

Given these multiple limitations, replication will be important before reaching strong conclusions, but our results may have some potential implications. First, given limited resources, if a school system is interested in providing either of these EBPIs, it may be useful to identify adolescents who are vulnerable across proximal process factors and who may have a stronger treatment response. Second, (a) strong effects showing that L2B was associated with greater improvement than IPT-AST and (b) the lack preference effects suggest that if only a single program can be implemented, it may be most effective and feasible to provide L2B, a less intensive EBPI. Third, although requiring replication, our results suggest that EBPI assignment based on preferences may be unnecessary, as it did not appear to enhance outcomes.