Best (but oft forgotten) practices: testing for treatment effects in randomized trials by separate analyses of changes from baseline in each group is a misleading approach

摘要

Researchers often analyze randomized trials and other comparative studies by separate analysis of changes from baseline in each parallel group. This may be the only analysis presented or it may be in addition to the direct comparison of allocated groups. We illustrate this by reference to 3 recently published nutritional trials. We show why this method of analysis may be highly misleading and may produce type I errors far greater than the 5% that we expect. We recommend direct comparison of means between groups with the use of baseline as a covariate if required.