To the Editor.
—The article by Drs Brophy and Joseph1 is a valuable contribution to the realistic appraisal of clinical trial evidence. Apart from the Bayesian perspective, however, the article highlights a critical problem with the traditional approach to hypothesis testing. The difficulty stems from the tradition of testing a null hypothesis of "no difference" when the more clinically relevant issue is whether there is adequate evidence to reject a hypothesis of "no sufficiently large difference." This is a particularly important distinction when one is comparing two therapies that have different costs, side effects, or other negative consequences.The Global Utilization of Streptokinase and Tissue Plasminogen Activator in Occluded Arteries (GUSTO) trial is a perfect example. A clinician or clinical policymaker may, for example, prefer tissue-type plasminogen activator (t-PA) only if it reduces mortality at least 1% more than streptokinase (SK), given the higher complication rate and greater expense
Avins AL. Bayesian Analysis and the GUSTO Trial. JAMA. 1995;274(11):873. doi:10.1001/jama.1995.03530110035027