Issues in the design and analysis of controlled clinical trials

Several issues that pertain to the analysis of data from repeated measurements designs are addressed. The necessity for an assumption concerning uniformity of correlations (symmetry) can be avoided by calculating linear and/or nonlinear trend scores. In the usual repeated measurements ANOVA the baseline covariate does not affect within-subject sums of squares, such as that for the groups x times interaction. However, the separately calculated trend scores have the advantage of accounting for most of that interaction and permitting covariance-correction for baseline differences. Chance baseline differences in a repeated measurements design otherwise can contribute materially to probabilities of erroneous conclusions. A conservative criterion for use of one-sided tests of significance is proposed.