Statistical Issues in the Design and Analysis of Ulcer Healing and Recurrence Studies

Abstract

Statistical considerations are discussed for a randomized parallel groups study to compare treatments for the healing of ulcers during a specified time period of dosing and for the avoidance of subsequent ulcer recurrence (in patients with healing) during a follow-up period with no medication. For this study, randomization enables valid comparisons of treatments for the rates of healing during the dosing period and for the cumulative rates of being ulcer-free (ie, healing and no recurrence subsequent to healing) during the combined dosing and follow-up periods. Appropriate methods of analysis include Mantel-Haenszel tests and logistic regression for dichotomous outcomes and their life table counterparts for time to event outcomes. For recurrence rates among patients with healing, the basis for comparisons among treatments is unclear because of potential lack of similarity of treatment groups for risk factors for recurrence at the beginning of the follow-up period. This difficulty can be addressed by interpreting recurrence rates within treatment groups as descriptive for corresponding populations with healing. Moreover, such descriptions can involve statistical models which account for the effects of risk factors. Consideration is additionally given to sample size determination and other aspects of the design for a healing and recurrence study.