Evaluation of Thiw he Gluw 'Akapi Substance Use Prevention Program

Description

Study design overview

All 13 middle schools on the reservation will participate in this study by providing space for program groups to meet in the evenings and allowing recruitment activities at school events. A total of 480 families will be recruited into 6 cohorts across two years (spring, summer and fall). In each community, families with a child 10-12 years of age will be recruited to participate in the study. Half of recruited families will be randomly assigned to receive the Thiwhe Gluw'akapi (TG) substance use prevention program and half to the comparison group, which will receive the Woyute Wate (WW) healthy eating and exercise program. Randomization will be explained to families as part of the recruitment and consent process. Given the potential for contamination across groups with this design, the investigators will carefully examine outcomes across groups and statistically control for contamination effects as necessary. Alternative designs, including randomization at the school/community level, were rejected due to greater concerns about cross-community variability.

Recruitment and retention strategies

Schools (e.g., registration, family nights, parent-teacher conferences, holiday events) will be the main points of recruitment but additional advertising and promotion will take place at community events such as powwows, health fairs, and basketball tournaments. Written materials about the program will be distributed, along with promotional items (e.g., water bottles, magnets, and lanyards with the project name and logo). The investigators will also advertise on local community radio, which has wide reach on the reservation. Because TG has been implemented at 12 of the 13 middle schools on the reservation as a part of the adaptation study and has been well received by both families and school staff, the investigators often have inquiries about opportunities for future participation, and anticipate a positive response to recruitment efforts.

Facilitators will call families weekly to touch base with them and remind them about upcoming sessions. The investigators provide meals to families at the start of each program session, giving them opportunities to connect with other families and facilitators, fostering relationships, and making it easier for busy families to get to the sessions. Child care and transportation will be provided, as needed.

Approaches to ensuring fidelity

All facilitators will be trained in TG by a master SFP 10-14 trainer from Iowa State University (Beth Fleming) and the master TG trainer (Dr. Alicia Mousseau, who led the program adaptation). Weekly supervision and preparation for sessions before and during implementation will be provided onsite by local staff certified as TG trainers, who will also train new facilitators as needed due to staff turnover. These on-site trainers will also conduct regular fidelity checks (observations during sessions using standardized fidelity checklists) and provide feedback to facilitators.

Data collection schedule and estimated sample sizes across waves

In each family, up to 2 eligible youth (aged 10-12) and 1 parent/caregiver will be eligible to participate in the research and take study surveys. Sample size estimates for each wave are estimates of the number of families (F), youth (Y), adults (A), and total sample (T). Ranges for youth and total reflect that 1-2 youth per family are eligible to complete surveys. Estimates include an attrition rate of 2.5% every 6 months, based on past experience with families in this community.

1. Baseline - 1-2 weeks before first session: F=480; Y=480-960; A=480; T=960-1440
2. Post-intervention -1 week after last session: F=480; Y=480-960; A=480; T=960-1440 6 follow-up surveys:
3. 6-month: F=468; Y=468-936; A=468; T=936-1404
4. 12-month: F=456; Y=456-912; A=456; T=912-1368
5. 18-month: F=445; Y=445-890; A=445; T=890-1335
6. 24-month: F=434; Y=434-868; A=434; T=868-1302
7. 30-month (excluding final cohort): F=370; Y=370-740; A=370; T=740-1110
8. 36-month (excluding final 2 cohorts): F=206; Y=206-412; A=206; T=412-618

Data collection procedures

All surveys will be administered online, using REDCap survey software. The baseline survey will be collected during an in-person meeting, after consent is obtained, using iPads provided by research staff. Post-intervention surveys will also be administered in person. For follow-up surveys, participants will be sent unique links to access their surveys on their own devices (e.g., computers, phones, tablets) or at a location of their choice (e.g., tribal college center). They will also be able to request to have research staff meet with them to collect these follow-up surveys in person and provide a project iPad for survey completion. Surveys collected via REDCap will be immediately uploaded to a secure cloud storage. REDCap is a secure, web-based application designed to support data capture for research studies, providing: 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources. REDCap is hosted by the University of Colorado Anschutz Medical Campus and supported by the Colorado Clinical and Translational Sciences Institute.

Analytic Plan

Three analytic approaches will be used: (1) discrete-time time survival analysis (DTSA), (2) mixed model analyses (MM), and (3) latent growth curve modeling (LGCM). Intention to treat principles will be applied in all analyses. Participant sex will be included as a covariate or separate models will be estimated by gender to explore differential effect patterns. Sandwich-estimators (type=complex in Mplus) and random effects models (in SAS Proc Mixed) will adjust for clustering at the family (and/or school) level. Full information maximum likelihood (FIML) procedures will be used to provide unbiased and efficient estimates.

Discrete-time survival analyses (DTSA) will be used to estimate the probability (hazard) for initiation of substance use among youth who have not used substances prior to enrollment in the study. Initial DTSA analyses will compare the fit of models assuming proportional odds (i.e., constant intervention effect across waves) either with or without frailty (residual variance) to non-proportional odds (i.e., variable intervention effect across waves). Comparisons will be based on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). Exposure to TG will be included as a time-invariant covariate to estimate treatment effects on risk of initiation. Mplus Monte Carlo analyses were used to determine power requirements for DTSA models. Effect-size estimates from earlier etiological work in this reservation community were used (hazard age 10-13, alcohol =.14; marijuana = .13, tobacco = .31). Estimated power to detect a moderate treatment effect - 20% reduction in risk of initiation - exceeded 80% across all effect-size specifications tested.

Mixed model analyses (MM) will be used to examine effects on all outcomes, including primary outcomes (youth substance use and suicide, adult substance use) and secondary outcomes (proximal mediators of risk for primary outcomes). MM analyses will account for correlated observations within participants over time and examine intervention effects averaged across the entire post-program data collection period and at each post-program assessment point individually. Wave of data collection and Condition (TG or WW) X Wave interaction will be fixed effects. Time will be a repeated measures factor and study participant and family cluster will be included as random effects. After verifying that randomization yielded baseline equivalence across Conditions, the investigators will use either linear (ordinal/continuous outcomes) or generalized linear mixed models (dichotomous outcomes) to test Condition X Wave interactions. The overall difference between TG and WW groups across the follow-up period compared to baseline will be obtained by calculating the average of all 7 Condition X Wave interaction effects. These analyses will test hypotheses about immediate and sustained effects of intervention exposures on outcomes. Power estimates for MM analyses were calculated using GLIMMPSE software, specifying estimates of substance use at each wave derived from previous work with this population. Analyses indicate that the projected sample size will be sufficient to detect a small to moderate intervention effect - 20% reduction in substance use across time - with power greater than .80.

The final analytic approach will follow Muthn & Curran, estimating linear latent growth curve models (LGCM) to assess the effect of TG on outcome trajectories. The investigators will estimate the 'normative' trajectory of each outcome in the WW (comparison) group and the comparable trajectory in the TG (intervention) group. In the third step, the investigators will compare these trajectories and to determine whether there is a statistically significant change in the outcome trajectory related to TG intervention exposure. Finally, the investigators will evaluate the interaction between intervention exposure and initial level of the outcome to explore differential intervention effectiveness as a function of baseline levels for each outcome. Mplus Monte Carlo analyses were used to estimate power requirements for LGCM models and effect-size estimates were drawn from earlier etiological research. The power to detect slopes ranging from .14 to 1.4, were tested, estimating intercepts low (.05) consistent with our earlier findings and the young age of our study sample, setting alpha at .05, within each Condition group (TG treatment and WW comparison). Power estimates exceeded .90 to detect slope effects in all models.

Details