P-Value Interpretation and Alpha Allocation in Clinical Trials

Abstract

PURPOSE: Although much value has been placed on type I error event probabilities in clinical trials, interpretive difficulties often arise that are directly related to clinical trial complexity. Deviations of the trial execution from its protocol, the presence of multiple treatment arms, and the inclusion of multiple end points complicate the interpretation of an experiment’s reported alpha level. The purpose of this manuscript is to formulate the discussion of P values (and power for studies showing no significant differences) on the basis of the event whose relative frequency they represent. METHODS: Experimental discordance (discrepancies between the protocol’s directives and the experiment’s execution) is linked to difficulty in alpha and beta interpretation. Mild experimental discordance leads to an acceptable adjustment for alpha or beta, while severe discordance results in their corruption. RESULTS: Finally, guidelines are provided for allocating type I error among a collection of end points in a prospectively designed, randomized controlled clinical trial. CONCLUSIONS: When considering secondary end point inclusion in clinical trials, investigators should increase the sample size to preserve the type I error rates at acceptable levels.