Clinical studies in which multiple outcomes are evaluated pose methodological dilemmas. One may test for group differences for each outcome measure separately, and then adjust the alpha level with a Bonferonni type correction, or attempt to combine these assessments in a variety of ways to maintain the experimentwise type I error at size alpha. Two general approaches for combining outcome assessment were considered: composite outcome measures (outcomes within an individual were summarized and the summary measure used in the analysis); and global test statistics (combining the individual test statistics into one global test). Global test statistics are linear combinations of the dependent individual test statistics (GLS was used here). Dichotomous outcome responses in models with several covariates were considered. Quasilikelihood methods were used to estimate correlations between individual test statistics. Simulations are being conducted to compare the global test procedures with the composite outcome measure method.
