When does the failure to find a difference mean that there is none?

Abstract

When a published clinical report concludes that there is no difference between two or more groups of patients, the reader must discern between two alternative possibilities. Either there is truly no difference in outcome between the groups or the study failed to detect an actual difference. The application of sample size and power calculations to outcome variables in a clinical study can provide essential insight into resolving this ambiguity. While one can never prove that there is no difference between groups of patients, these techniques can estimate the likelihood that a particular study will detect or fail to detect a difference of specified size.