Abstract:
|
In a crossover design with continuous outcomes (e.g., blood pressure), baseline and post-baseline responses are obtained in each treatment period. The baselines can be utilized as covariate(s) in an analysis of covariance (ANCOVA) to increase the precision of the treatment effect estimate. Previous authors have noted that the potential efficiency gain from using baselines depends on the joint covariance structure of all the baseline and post-baseline responses. We show how the underlying covariance structure can be leveraged to find an optimal combination of the baselines so as to minimize the theoretical variance of the ANCOVA-based estimated treatment effect. We do this for balanced 2x2, 3x3, and 4x4 crossovers under four commonly seen covariance structures. We also develop an adaptive method in which first a suitable covariance structure for the given dataset is selected via AICC values, and then the corresponding optimal baseline covariate combination is used in the ANCOVA. We show that, relative to previously published methods, the proposed method leads to sizable gains in power, while maintaining the nominal type I error rate.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.