Directional baseline differences and type I error probabilities in randomized clinical trials.

Abstract

Adequate correction for baseline differences that occur in the same or opposite direction from inferred treatment effects in a simple randomized experimental design is addressed by Monte Carlo simulation. Results confirm that the analysis of covariance (ANOVA) can provide appropriate correction whether baseline means differ significantly by chance in the same or opposite direction from the inferred treatment effect. The results from analysis of simple pre-post difference scores are highly dependent on the direction of chance baseline deviations in relation to the directional treatment effect that would be inferred from rejection of the null hypothesis. The attempt to correct for baseline differences by expressing outcome as percentage of baseline entails directional bias that depends on the location of the zero point, the direction of change, and the level of correlation between baseline and follow-up measurements.