Evaluating a Difference in Proportions in the Presence of Heterogeneity: An Application to Integrated Summaries of Safety

• Activity Number: 582
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): David Radley*+ and Michael Dallas and Adeniyi Adewale
• Companies: Merck Research Laboratories and Merck Research Laboratories and Merck Research Laboratories
• Keywords: Heterogeneity ; stratification ; binary data ; integrated summaries ; safety data

Abstract

In integrated summaries of safety (ISoS), there is often interest in the difference between a test group and a control group in the proportion of subjects reporting a safety outcome (occurrence of a specific adverse experience, for example). When combining evidence across multiple studies data are often simply pooled; however, heterogeneity may be introduced according to study characteristics. We simulated stratified binary data, in the presence of heterogeneity of varying degrees, and obtained the point and interval estimates of the overall risk difference using several methods. The methods considered include Mietinnen & Nurminen (MN) method for stratified data using various weighting schemes; a logistic mixed model with random effects for study; and a quasi-likelihood approach. Depending on the degree of heterogeneity present, we provide recommendations for the best method.