Estimates of Intraclass Correlation Coefficients (ICCS) from a Group-Randomized HIV Prevention Trial in an HIV-Positive Population in Sub-Saharan Africa

• Activity Number: 140
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jun Zhang*+ and Sherri L. Pals and Amy Medley and Catherine Nichols
• Companies: ICF Macro and Centers for Disease Control and Prevention and Centers for Disease Control and Prevention and Centers for Disease Control and Prevention
• Keywords: Group-randomized trials ; intraclass correlation coefficient ; HIV/AIDS

Abstract

Group-randomized trials (GRTs) are often used to test the efficacy of structural interventions or when individual randomization is not feasible. This design requires an estimate of the expected intraclass correlation coefficient (ICC) to estimate the sample size needed, but few ICC estimates from GRTs in HIV/AIDS have been published. This study will evaluate the effectiveness of an HIV prevention intervention among 3,547 HIV+ patients in HIV Care and Treatment clinics in Kenya, Namibia and Tanzania. We used baseline data from the study to estimate ICCs for 3 dichotomous variables related to unprotected vaginal sex in the past 3 months, taking HIV medications (antiretroviral and/or prophylaxis) and missed HIV medications in the past 30 days. The ICC for unprotected vaginal sex was 0.000005, and ICCs for missed medication and on medications were 0.036 and 0.184, respectively. With 200 patients per clinic, the ICC for missed medications would result in variance of at least 8 times as large as expected with independent observations. Our analyses contribute to knowledge regarding ICCs for outcome variables related to HIV/AIDS interventions and can be useful in designing future trials.