Some non-parametric and categorical data analyses for a change-over design study and discussion of apparent carry-over effects.

This paper addresses the statistical analysis of data from a pair of two period change-over design studies from two centres. These studies compared an active treatment with placebo for the relief of heartburn. The principal response variable was whether or not the patient experienced relief for heartburn from a symptom-provoking meal within 15 minutes of the initial dose of treatment. Other response variables were an ordinal response classification for the pattern of relief for up to two doses and a composite measure for time to relief. Categorical data methods and non-parametric rank methods for the analysis of such responses are described. A result of interest is the significance of (treatment X period) interaction (i.e. carry-over effects). It is interpreted as a tendency for less effective placebo treatment in the period following the active period compared with the first period; in contrast, the active treatment was equally effective in both periods. Discussion centres on the implications of this finding for conclusions concerning the comparison of the active treatment with placebo.