Design and analysis of phase III trials with ordered outcome scales: the concept of the sliding dichotomy

Abstract

The conventional approach to the analysis of a Phase III trial in head injury or stroke takes an ordered scale measuring functional outcome and collapses the scale to a binary outcome of favorable versus unfavorable. This discards potentially relevant information which limits statistical power and moreover is not in accord with clinical practice. We propose an alternative approach where a favorable outcome is defined as better than would be expected, taking account of each individual patient's baseline prognosis. This is illustrated through a worked example based on data from a Phase III trial in head injury. The approach is also compared with the proportional odds model, which is another statistical approach that can exploit an ordered outcome scale. The approach raises issues of clinical, statistical, and regulatory importance, and we initiate what we believe needs to become a widespread debate amongst the community involved in clinical research in head injury and stroke.