Group-Randomized Trials (GRT) in HIV/AIDS Research: How Often Is the Intraclass Correlation Ignored in Design and Analysis?

• Activity Number: 455
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ryan E. Wiegand*+ and Sherri L. Pals and David Murray
• Companies: CDC and CDC and The Ohio State University
• Keywords: Group randomized trials ; Clustered data ; HIV/AIDS

Abstract

A popular study design in HIV/AIDS research is the delivery of an intervention to an intact group of participants. In these group-randomized trials (GRTs), groups instead of individuals are randomly assigned to the study conditions. Ignoring the intraclass correlation coefficient (ICC) may cause studies to be underpowered and result in type I errors in analyses. We present results from an appraisal of the sample size and statistical methods used to analyze HIV/AIDS-focused GRTs found in 26 research articles published from 2005 to 2009. Evidence was provided in 15% of articles of accounting for ICCs in power calculations. Despite 96% of articles containing a significant finding, only 39% of articles appropriately accounted for ICCs in all statistical analyses. These results indicate continued dissemination of information on the dangers of the failure to account for ICCs is needed.