This R03 research project features secondary analyses of datasets from four federally-funded longitudinal group randomized trials (GRTs) of school-based HIV/STI prevention interventions to provide the field with needed estimates of intraclass correlation coefficients (ICCs) - a parameter critical for determining the sample size necessary for estimating intervention effects with adequate power and precision when planning GRTs. Studies have shown that ICC estimates vary widely depending on the outcome, population, setting, and study design, and small changes in ICC estimates can have large effects on study power and precision (Murray, 1998;Donner and Klar, 2000;Rotondi and Donner, 2009;Resnicow et al., 2010). Consequently, in the past decade researchers have been urged to publish estimates for a wider variety of outcomes, populations and settings so that future trials can be planned more efficiently (Murray et al., 2004;Resnicow et al., 2010;Murray and Blitstein, 2003;Pals et al., 2009). This project would employ 2-level (students nested within schools) and 3-level (students nested within classrooms nested within schools) multilevel models to analyze cross sections of each dataset to provide the field with the first set of school-level ICC estimates specific to planning cross- sectional and pretest/posttest GRTs of school-based HIV prevention interventions focused on sexual risk- taking outcomes. In addition, recent advances in methodology and software provide the potential to refine sample size determinations when planning longitudinal GRTs by using estimates of ICC or variance component parameters from 3-level repeated measures models. Yet very few estimates of these parameters have been published in any health risk prevention field (Murray and Blitstein, 2003;Murray et al., 2006;Murray et al., 2007;Moonseong and Leon, 2008;Spybrook et al., 2009). This study would use repeated measures multilevel models to obtain the first estimates of these parameters for planning longitudinal GRTs of school- based HIV prevention interventions. In all cases, models with and without adjustment for covariates would be analyzed to contribute information on the extent to which adjustment for covariates may reduce ICCs and needed sample size in school-based GRTs of HIV/STI prevention interventions. The effects of the various ICC and variance component estimates on the sample size needed to detect commonly seen effect sizes also would be examined. The results of this study would be made available through peer-reviewed articles and conference presentations in order to benefit researchers planning future school-based HIV/STI prevention studies. Efficient planning of these trials is critical given the importance of schools as a venue for delivering prevention interventions to adolescents and the costs associated with implementing school-based GRTs.
Schools are an important venue for HIV/STI prevention interventions because most youth attend schools, making them efficient sites for program delivery. Rigorous evaluation of these interventions is critical to ensure public health resources are applied to programs that have meaningful effects on the rates of HIV and other STIs in youth. The gold standard for rigorous evaluation of school-based HIV prevention interventions is the group randomized trial (GRT), but planning such trials with the sample size (number of schools/students) necessary for estimating intervention effects with adequate power and precision requires estimates of the intraclass correlation coefficient (ICC) parameter from studies involving similar outcomes, populations and settings. This project would provide the first ICC parameter estimates specific to HIV/STI-related behavioral and psychosocial outcomes across a variety of different school settings and populations so that researchers in the field could select the most appropriate ICCs for efficiently sizing these costly studies.