Abstract:
|
A meta-analysis involves combining data from multiple studies to achieve higher statistical power for the measure of interest. A weighted average of the measure of interest is typically computed using weights based on the inverse of the study-specific variances (INVAR) or the study-specific sample sizes (SSIZE). When the measure of interest is the difference between success proportions on 2 treatments, the INVAR and SSIZE weights are optimal if the differences between proportions are constant across studies on the proportion and logit scales, respectively. If a scale exists on which treatment differences are constant across studies, the optimal weights can be derived. However the existence of such a scale is not known a priori, so an optimal weighting strategy cannot be pre-specified. We propose an alternative testing strategy based on the simultaneous use of 3 test statistics, each involving a weighted average of within study differences between proportions. The optimal weights for no treatment by study interaction on 3 common scales, i.e., proportion, logit and log will be used. We will demonstrate the power advantages of this procedure over existing procedures via simulation.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.