Parameters for sample size estimation from a group-randomized HIV prevention trial in HIV clinics in sub-Saharan...

Sample size calculations for a group-randomized trial (GRT) require an estimate of the expected intraclass correlation coefficient (ICC). However, few ICC estimates from GRTs in HIV/AIDS research have been published, leaving investigators with little data on which to base expectations. We used data from a multi-country study to estimate ICCs for variables related to physical and mental health and HIV risk behaviors. ICCs for perceptions of physical and mental health tended to be higher than those for HIV risk behavior variables, which were higher than ICCs for CD4 count. Covariate adjustment for country and socio-demographic variables reduced most ICC estimates. For risk behavior variables, adjustment for country and socio-demographic variables reduced ICC estimates by as much as 84 %. Variability in ICC estimates has important implications for study design, as a larger ICC reduces power. ICC estimates presented in this analysis will allow more precise sample size estimates for future GRTs.