A Generalized Global Rank Test for Multiple, Possibly Censored, Outcomes

• Activity Number: 282
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Ritesh Ramchandani*+ and Dianne M. Finkelstein and David Alan Schoenfeld
• Companies: Harvard and Massachusetts General Hospital/Harvard University and Massachusetts General Hospital
• Keywords: multiple outcomes ; global ; nonparametric ; censored

Abstract

In clinical studies, a single outcome does not always adequately capture the effect of an intervention, so other outcomes are often considered as well. The design, analysis, and interpretation of studies in the presence of multiple outcomes like these can be difficult. We propose a general nonparametric scoring method for combining two or more endpoints into a single summary statistic for efficacy. This general test encompasses other global or composite rank tests that have been previously proposed (e.g. O'Brien 1984), and can accommodate censored observations. First we score each pair of subjects with respect to each outcome, and then reduce the dimension of the multiple pairwise scores to get a composite score for the pair of subjects. A rank-sum type test on the composite scores is then performed. The method of dimension reduction is investigator chosen, and examples of various tests are provided. We discuss which tests are best suited to different situations, and propose methods for choosing outcome weights to improve the power of certain tests under specific alternative hypotheses.