Cutting through the statistical fog: understanding and evaluating non-inferiority trials

Abstract

Every year, results from many important randomised, controlled trials are published. Knowing the elements of trial design and having the skills to critically read and incorporate results are important to medical practitioners. The goal of this article is to help physicians determine the validity of trial conclusions to improve patient care through more informed medical decision making. This article includes a review of 162 randomised, controlled non-inferiority (n = 116) and equivalence (n = 46) hypothesis studies as well as the larger Stroke Prevention using Oral Thrombin Inhibitor in atrial Fibrillation V study and the Ongoing Telmisartan Alone and in Combination with Ramipril Global Endpoint Trial. Evaluation of data from small and large trials uncovers significant flaws in design and models employed and uncertainty about calculations of statistical measures. As one example of questionable study design, discussion includes a large (n = 3922), double-blind, randomised, multicentre trial comparing the efficacy of ximelagatran with warfarin for prevention of stroke and systemic embolism in patients with non-valvular atrial fibrillation and additional stroke risk factors. Investigators concluded that ximelagatran was effective compared with well-controlled warfarin for prevention of thromboembolism. However, deficiencies in design, as well as concerns about liver toxicity, resulted in the rejection of the drug by the US Food and Drug Administration. Many trials fail to follow good design principles, resulting in conclusions of questionable validity. Well-designed non-inferiority trials can provide valuable data and demonstrate efficacy for beneficial new therapies. Objectives and primary end-points must be clearly stated and rigorous standards met for sample size, establishing the margin, patient characteristics and adherence to protocol.