Sample sizes for designed studies with correlated binary data

Abstract

Summary. The paper considers comparative studies in which subjects use more than one product or receive more than one treatment. The paper is focused mainly on the comparison of products, including the possibility of a large number of products. The data to be analysed are on a binary variable that is observed by each subject for each product used. The calculation of the number of subjects or sample size is illustrated for a variety of basic study designs, assuming that the data may be correlated and the analysis of the data uses generalized estimating equations methodology. The sample sizes use asymptotic theory developed by Liu and Liang. A simulation study that evaluates some of the resulting sample sizes suggests that these need to be increased slightly to achieve the nominal power.