META ANALYSIS OF BINARY OUTCOMES DATA IN CLINICAL TRIALS

摘要

The primary objective of this paper is to present different methods for combining the results of several investigations involving binary outcomes. Typically the studies involve comparison of a treatment with control. However, in this study we focus on only non-comparative binary outcomes. Towards this end, we include methods based on different effect sizes, which are functions of the success rate of the attribute of interest, namely, (i) success rate p, (ii) log of success rate ln(p), (iii) log of odds ratio ln(), and (iv) variance stabilizing transformation of the success rate sin~(-1) (). These meta analytic methods are illustrated using two different data sets: a data set based on pharmacoeconomic applications using antifungal onychomycosis lacquers, where the success rate of ciclopirox lacquer is relatively large and another data set involving cancer incidence, which is a rare event. The meta analytic methods behave differently depending on which type of data is used. When the data set involves rare events, application of the variance stabilizing transformation can be carried out without adjusting the success rate and its standard error, where as the other methods do require that adjustment, and hence the variance stabilizing transformation is preferred for the meta analysis.