Étude d'efficacité CHOICES-TEEN

This CHOICES-TEEN intervention study will use a Phase II Behavioral Treatment Trial to employ a single blind randomized design with an attention control (AC) group to assess the efficacy of the CHOICES-TEEN intervention. Young women, 14-19 years of age, entering the Harris County Juvenile Probation (HCJP) system's probation and field diversion and community probation program will be eligible for screening into the study. The investigators anticipate recruiting N=435 with 92% retention based on prior experience, yielding a total sample size of N=400, stratified by program, with 200 randomized to the CHOICES-TEEN intervention (plus Standard Care; CT) or the Attention only group (AC) using urn randomization. Both groups will be assessed at 3-, 6- and 9-month follow up. Eligibility will be determined based on the following inclusion/exclusion information. This efficacy trial will: (1) Test the efficacy of CHOICES-TEEN (CT) compared with attentional control (AC) on reducing the risk of substance-exposed pregnancy (SEP) and HIV/STI among high-risk female youth involved with the juvenile justice system by reducing alcohol use, increasing marijuana cessation, reducing pregnancy risk, and increasing condom use; (2) Test the efficacy of CT, compared to an attentional control condition, in increasing cognitive self-regulation abilities; (3) Test proposed intervention mediators/mechanisms of action for CT overall and by race/ethnicity; and (4) Test the moderating effect of initial readiness to change on risk of SEP and risk of HIV/STI.

Female adolescents between the ages of 14-19 will be recruited for eligibility screening from the aforementioned community probation program. Voluntarily referred youth will be screened for eligibility in the study after obtaining parental permission and youth assent. All youth enrolled in the study must be identified as being at risk for substance-exposed pregnancy and HIV/STI. Eligible youth who provide written informed consent (and parents who provide written permission) will then be randomized to the CHOICES-TEEN intervention or the Attentional Control Condition.

The investigators anticipate recruiting N=435 with 92% retention based on prior experience with similar studies, yielding a final sample size for analyses of N=400. Randomization, stratified by program, will result in n=200 participants per condition with participants clustered within k=4 forensic programs. Investigators assume a conservative ICC =0.20 due to clustering. Absolute risk reductions in risk of SEP range from 14.8% to 25.1% based on Project CHOICES, Project CHOICES Plus and our pilot CHOICES-TEEN. For the purposes of sample size justification, investigators will assume N=435 randomized in 1:1 fashion (minimum 400 completers), stratified by program, and ICC = 0.20 and a conservative estimate of an ARR=15% for reduced risk of SEP and HIV/STI. Finally, investigators stipulate that if the posterior probability that there is an effect of treatment (Odd Ratio>1.0) is greater than 0.75 and that the median treatment effect estimate exceeds an Odds Ratio=1.5, this constitutes sufficient evidence to warrant subsequent investigation. M=1000 Monte Carlo simulations, using a normal approximation to the posterior indicates that under the preceding assumptions the proposed design will identify an effect of treatment 81.9% of the time.

Data analyses. The data analytic strategy will use generalized linear mixed and structural equation modeling (SAS 9.4, R v. 3.4, Stan,v. 2.17 and MPlus v. 8.3) for both continuous and discrete outcomes. All analyses will be conducted on an intention-to-treat basis. To address missingness, Bayesian approaches will implement joint modeling of observed outcomes and the missing data which is robust to ignorable missingness (i.e., MCAR and MAR). Sensitivity analyses will evaluate robustness of analytic conclusions to missing data. Non-ignorable missing data patterns (i.e., MNAR) will be addressed through pattern-mixture modeling methods.Specification of diffuse, neutral priors will reflect the initial uncertainty regarding effect sizes. For all generalized linear mixed models, priors for regression coefficients will be specified as ~Normal (µ=0, σ2=1 x 106) (for non-normal outcomes this refers to the prior for the coefficient within the link function), level one error variances will be specified as ~Inverse Gamma (shape=0.001, scale=0.001). Choice of prior distribution for level two variances will follow Gelman's recommendations. Bayesian Structural Equation Modeling (BSEM) prior specification will adapt recommendations from Muthén and Asparouhov 230. Priors for the comparison of proportions will be specified as ~Beta (α=0.5, β=0.5). Similar procedures will be used in secondary analyses to investigate subgroups of youth using specific substances (i.e. alcohol and marijuana), as well as intervention effects as a function of baseline readiness to change as a potential moderator. Mediational modeling will examine the degree to which putative mechanisms of behavioral change transmit the effects of the intervention on the specified outcomes. BSEM will investigate mediation of treatment effects due to CT on SEP and HIV/STI risk at 9 months by hypothesized mechanisms (processes of change, cognitive self-regulation, and confidence and temptation) measured at 3 months utilizing MPlus v. 8.3. Examination of the posterior distribution of the indirect effects will evaluate the probability that mediational effects exist.

Specific Data Analyses - Hypothesis 1: CT, compared to Attentional Control (AC) AC will be associated with reduced risk of SEP and HIV/STI at 9-months post intake. The primary outcome is reduced risk of SEP and HIV/STI at 9 months, however at each time point (3-, 6-, and 9-month) multilevel logistic models will evaluate the risk of SEP and HIV/STI as a function of treatment condition while addressing clustering due to forensic program assignment. At each time point generalized linear multilevel models will evaluate the presence/absence of risk drinking, presence/absence of marijuana use, presence/absence of vaginal intercourse without effective contraception, and presence/absence of vaginal or anal intercourse without condom use as a function of treatment condition while addressing clustering due to forensic program assignment.

Hypothesis 2: Compared to AC, CT will improve cognitive self-regulation abilities at 3-, 6-, and 9-month post intake as measured by self-report self-regulation measures. At each time point (3- and 9-month) generalized multilevel linear models will evaluate self-regulation as a function of treatment condition while addressing clustering due to forensic program assignment.

Hypothesis 3: The processes of change, confidence and temptation, and cognitive self- regulation, for each risk behavior at 3-months will mediate the effect of treatment on SEP risk and HIV/STI risk at 9-months post intake for CT. Multilevel Bayesian structural equation modeling (M-BSEM) will evaluate the degree to which processes of change, cognitive self-regulation, confidence and temptation measured at 3 months follow-up mediate the effect of treatment on SEP and HIV/STI risk at 9 month follow-up. Multilevel elements will address clustering as a function of forensic program assignment. Multigroup analyses testing the mediation models will find invariance between Non-Hispanic Black, Hispanic, and Non-Hispanic White.

Hypothesis 4: Female youth with low baseline readiness to change risk behavior will have less risk of SEP and HIV/STI at 3-, 6- and 9-months post intake in the CT intervention condition, designed to increase motivation and goal striving, than female youth with low baseline readiness to change risk behavior in the AC condition. At each time point (3-, 6-, and 9-month) multilevel logistic models will evaluate the risk of SEP and HIV/STI as a function of treatment condition, baseline readiness and the interaction of treatment and baseline readiness. These models will use the approach advocated by Simon and Dixon.

Sample Size. The investigators anticipate recruiting N=435 with 92% retention based on our experience with Project CHOICES, CHOICES Plus, and CP-T yielding a final sample size for analyses of N=400. Randomization, stratified by program, will result in n=200 participants per condition with participants clustered within k=4 forensic program assignments. Investigators assume a conservative ICC = 0.20 due to clustering. Absolute risk reductions in risk of SEP range from 14.8% to 25.1% based on Project CHOICES, Project CHOICES Plus and our pilot CHOICES-TEEN. For the purposes of sample size justification, investigators will assume N=435 randomized in 1:1 fashion (minimum 400 completers), stratified by program, and ICC = 0.20 and a conservative estimate of an ARR=15% for reduced risk of SEP and HIV/STI. Finally, investigators stipulate that if the posterior probability that there is an effect of treatment (Odd Ratio>1.0) is greater than 0.75 and that the median treatment effect estimate exceeds an Odds Ratio=1.5, this constitutes sufficient evidence to warrant subsequent investigation. M=1000 Monte Carlo simulations, using a normal approximation to the posterior indicates that under the preceding assumptions the proposed design will identify an effect of treatment 81.9% of the time.