Randomization-Based Methods for Treatment Comparisons for Dichotomous Outcomes for Multiple Anatomical Regions

Abstract

Some clinical trials in ophthalmology or dermatology randomize patients into two or more treatment groups, and observations are made for one or more anatomical regions (or sites) for each patient. In this case, patients are primary sampling units, and their responses for multiple anatomical sites are usually correlated. Useful randomization-based extensions of the Cochran–Mantel–Haenszel method are discussed here for such studies; and aspects of their application are illustrated with two examples. The first example is a randomized ophthalmology clinical trial with the number of treated eyes as strata and a dichotomous response criterion. The second example is a randomized dermatology clinical trial with three strata for the number of treated anatomical regions among the face, scalp, or chest, one covariable (as disease severity at baseline), and a dichotomous outcome. For these examples, the results are interpretable through the proportions of treated sites with favorable outcome for each treatment and their difference between treatments.