Study design and procedure
Project WISDOM was a cluster randomized, controlled, hybrid type III effectiveness-implementation trial designed to test the effects of LOCI versus training and technical assistance only on MBC fidelity in outpatient mental health clinics serving youth. Details of the trial and primary implementation and clinical outcomes are reported elsewhere [
12]. The trial enrolled 21 clinics serving youth in Idaho, Oregon, and Nevada, USA. Clinics were eligible if they were not actively implementing a digital MBC system and if they employed three or more clinicians delivering psychotherapy to youth (ages 4–18 years). Using covariate constrained randomization, clinics were randomly assigned to one of two parallel arms: (1) LOCI plus training and technical assistance in MBC or (2) training and technical assistance in MBC only. Clinic-level randomization aligned with the scope of the LOCI strategy and prevented contamination of outcomes at the clinician and patient levels. Clinic leaders could not be naïve to condition; however, clinicians and caregivers of youth were naïve to condition.
Following baseline assessments and randomization of clinics, executives and first-level leaders in the LOCI condition began participating in the LOCI implementation strategy. One month later, clinicians who worked with youths in both conditions received training to implement an evidence-based digital MBC system called the Outcome Questionnaire-Analyst (OQ-A; see below for details; [
49,
50]). Following the initial OQ-A training, clinics in both conditions received two booster trainings and ongoing OQ-A technical assistance from the OQ-A purveyor organization until the trial’s conclusion.
To assess LOCI’s effects on its targeted mechanisms of change, clinicians who served youth in participating clinics were asked to complete web-based assessments evaluating their clinic’s leadership and clinic implementation climate for MBC at five time points: baseline (T1; following randomization of clinics but prior to initiation of LOCI or OQ-A training), 4-month post-baseline (T2), 8-month post-baseline (T3), 12-month post-baseline (T4; coinciding with the conclusion of LOCI), and 18-month post-baseline (T5; 6 months after LOCI concluded). Surveys were administered from October 2019 to May 2021. Clinic leaders provided the research team with rosters and emails of all youth-serving clinicians at each time point. Confidential survey links were distributed by the research team directly to clinicians via email. Clinicians received a small financial incentive for completion of each assessment (i.e., gift card to a national retailer) based on an escalating structure (US $30, US $30, US $45, US $50, US $55).
The primary implementation outcome of MBC fidelity was assessed for new youth outpatients who initiated treatment in the 12 months following clinician training in the MBC system. Upon intake to services, parents/caregivers of new youth patients were presented with study information requesting their consent for contact by the research team. Caregivers who agreed were contacted by research staff via telephone to complete screening, informed consent, and baseline measures (if eligible). After study entry, caregivers completed assessments reporting on the youth’s treatment participation (i.e., number of sessions) and symptoms monthly for 6 months following the youth’s baseline. Assessments were completed regardless of the youth’s continued participation in treatment, unless the caregiver formally withdrew (
n = 7). Caregivers received a US $15 gift card to a national retailer for completion of each assessment. Enrollment and collection of follow-up data for youth occurred from January 2020 to July 2021. The CONSORT and Stari guidelines were used to report the results of this mediation analysis within the larger trial [
51,
52].
Participants
All licensed clinicians who worked with youth in participating clinics at each time point were eligible to participate in web-based surveys of clinic leadership and climate. This broad inclusion criterion ensured a full picture of clinic leadership and climate at each time point.
Inclusion criteria for youth were intentionally broad to reflect the trial’s pragmatic nature and the applicability of MBC to a wide range of mental health diagnoses. Eligible youth were new patients (i.e., no psychotherapy at the clinic in the prior 12 months), ages 4 to 17 years, who had been diagnosed by clinic staff with an Axis I DSM disorder deemed appropriate for outpatient treatment at the clinic; it was not required that youths be assigned to clinicians who completed surveys. Youths were excluded if they initiated treatment more than 7 days before the informed consent interview. Electronic informed consent was obtained from all participants. The Boise State University Institutional Review Board provided oversight for the trial (protocol no. 041‐SB19‐081) which was prospectively registered at ClinicalTrials.gov (identifier: NCT04096274).
Clinical intervention: digital measurement-based care
The OQ-A is a digital MBC system shown to improve the effectiveness of mental health services in over a dozen clinical trials across four countries [
49,
53]. OQ-A measures are sensitive to change upon weekly administration and designed to detect treatment progress regardless of treatment protocol, patient diagnosis, or clinician discipline [
53]. In this study, clinicians had access to parent- and youth-report forms of the Youth Outcomes Questionnaire 30.2 [
54,
55] and the Treatment Support Measure [
56,
57]. Measures were completed by caregivers and/or youth electronically (via tablet or phone). Administration typically took 3–5 min. Measures were automatically scored by the OQ-A system, and feedback reports were generated within seconds. Feedback included a graph of change in the youth’s symptoms, critical items (e.g., feelings of aggression), and a color-coded alert, generated by an empirical algorithm, indicating whether the youth was making expected progress or was at risk of negative treatment outcome.
Clinicians were instructed to administer a youth symptom measure to the caregiver and/or youth at each session, review the feedback within 7 days of the session, and use the feedback to guide clinical decision-making. Clinicians were encouraged to discuss feedback with the caregiver and/or youth when they believed it was clinically appropriate and to administer a Treatment Support Measure if a youth was identified as high risk for negative outcome. Consistent with prior MBC trials, clinicians were not given specific guidance on how to respond to feedback; instead, they were advised to use their clinical skills in partnership with patients and clinical supervisors.
Leadership and Organizational Change for Implementation (LOCI)
Details of the LOCI implementation strategy are available elsewhere [
11,
12]. Briefly, LOCI was implemented in quarterly cycles over 12 months. During each cycle, (1) executives and first-level leaders within LOCI clinics attended monthly organizational strategy meetings to review data and to develop clinic-wide policies, procedures, and practices to support OQ-A implementation, and (2) first-level leaders attended leadership development trainings (5 days total) and participated in brief (~ 15 min) weekly coaching calls, designed to enhance their leadership skills. Once per month, individual coaching calls were replaced by group coaching calls with all other first-level leaders in the LOCI condition.
To support enrollment in the study, clinic leaders in the training and technical assistance only condition were offered access to four, professionally produced, web-based, general leadership seminars (1 h each). Seminars covered general leadership topics like giving effective feedback and leading change. The seminars were made available immediately after the OQ-A training.
Clinicians assessed the extent to which their first-level leaders exhibited transformational leadership behaviors using the Multifactor Leadership Questionnaire (MLQ) [
62,
63]. The MLQ is a widely used measure that has demonstrated excellent psychometric properties [
64] and is associated with implementation climate for EBP as well as clinicians’ attitudes toward, and use of, EBPs [
65‐
68]. Responses were made on a 5-point scale (“not at all” to “frequently, if not always”). Consistent with prior studies, we used the 20-item transformational leadership total score, calculated as the mean of four subscales: idealized influence, inspirational motivation, intellectual stimulation, and individual consideration.
Data aggregation
Best practice guidelines [
73‐
76] recommend clinician ratings of first-level leadership and clinic implementation climate be aggregated and analyzed at the clinic level. To justify aggregation, guidelines recommend that researchers test the level of inter-rater agreement among clinicians within each clinic to confirm there is evidence of shared experience. We used the
rwg(j) statistic [
77] to assess inter-rater agreement among clinicians within each clinic. Across all clinics and all waves, average values of
rwg(j) were above the recommended cutoff of 0.7 [
78,
79] for implementation leadership (
M = 0.82,
SD = 0.27), transformational leadership (
M = 0.87,
SD = 0.24), and clinic implementation climate (
M = 0.94,
SD = 0.10).
Data analysis
All analyses used an intent-to-treat approach. To test LOCI’s effects on growth in first-level leaders’ implementation leadership (H1), transformational leadership (H2), and clinic implementation climate (H3) for Aim 1, we used three-level linear mixed-effects regression models [
82,
83] with random effects addressing the nesting of repeated observations (level 1) within clinicians (level 2) within clinics (level 3). Separate models were estimated for each outcome. At level 1, observations of leadership and climate collected from clinicians at each time point were modeled using a piecewise growth function that captured differences in change from baseline to each time point across conditions [
84]. Implementation condition and clinic covariates were entered at level 3. Models were estimated using the mixed command in Stata 17.0 [
85] under full maximum likelihood estimation, which accounts for missing data on the outcomes, assuming data are missing at random. Effect sizes were calculated as the standardized mean difference in
change (i.e., difference in differences) from baseline to each time point (i.e., Cohen’s
d) using formulas by Feingold [
86]. Cohen suggested values of
d could be interpreted as small (0.2), medium (0.5), and large (0.8) [
87].
Aim 2 tested the hypotheses that experimentally induced improvement in first-level leaders’ implementation leadership (H4), and transformational leadership (H5) by T
2, would mediate LOCI’s effect on improvement in clinic implementation climate by T
4. These mediation hypotheses were tested using the multilevel causal mediation approach by Imai et al. [
81], implemented in the R “mediation” package [
88]. To align our analytic approach with our theoretical model, we estimated a 2–2-1 mediation model in which the primary antecedent (LOCI) and mediator (clinic-level aggregate leadership scores) entered the model at level 2 (i.e., the clinic level), and the outcome (clinician ratings of implementation climate) entered at level 1, representing latent clinic means [
89]. Separate models were estimated for each type of leadership because Imai’s approach does not accommodate simultaneous mediators [
81]. The inclusion of baseline values for the mediator (i.e., leadership) and outcome (i.e., climate) in each model modified the interpretation of the effects so that they represented the effect of LOCI on
change in leadership from T
1 to T
2 and of change in leadership on change in climate from T
1 to T
4. To stabilize the effect estimates, we set the number of analytic simulations for the direct and indirect effects to 10,000. These analyses produced estimates of LOCI’s indirect and direct effects on T
4 implementation climate, as well as the proportion of LOCI’s total effect on implementation climate that was mediated by improvement in leadership (i.e., proportion mediated =
pm). Indirect effects indicate the extent to which LOCI influenced T
4 implementation climate through its effect on T
2 leadership (i.e., mediation). Direct effects indicate the residual (remaining) effect of LOCI on T
4 implementation climate that was
not explained by change in T
2 leadership. The
pm statistic is an effect size measure indicating
how much of LOCI’s effect on implementation climate was explained by change in leadership.
Aim 3 tested the hypothesis that improvement in clinic implementation climate from T
1 to T
4 would mediate LOCI’s effect on MBC fidelity during the same time period (H6). The nested data structure was accommodated using a 2–2-1 model in which the primary antecedent (LOCI) and mediator (aggregate clinic-level T
4 implementation climate scores) occurred at level 2 (i.e., clinic level) and the outcome (MBC fidelity) occurred at level 1 (i.e., youth level). Note that the inclusion of baseline values of clinic implementation climate in this model modified the interpretation of the effects so that they represent the effect of LOCI on
change in climate from T
1 to T
4 and of change in climate on fidelity during the same time period. To address the events/trials nature of the MBC fidelity index, a generalized linear mixed-effects model with random clinic intercepts, a binomial response distribution, and a logit link function was used in the second step of the mediation analysis [
82]. In total, 18 clinics enrolled a total of 234 youth, all of whom had MBC fidelity data; however, one clinic was missing ratings of T
4 implementation climate, resulting in a sample of 17 clinics and 231 youth for this analysis. A sensitivity analysis based on mean imputation of the missing T
4 implementation climate value yielded the same inferential conclusions. A priori statistical power analyses conducted with the PowerUp! macro [
90,
91] indicated the trial had power of 0.74–0.90 to detect minimally meaningful mediation effect sizes depending on observed intraclass correlation coefficients and variance explained by covariates.