Trials and tribulations of non-inferiority: the ximelagatran experience

Ximelagatran is a novel oral direct thrombin inhibitor that offers a number of advantages over the standard treatment, warfarin, in patients with atrial fibrillation. Two large clinical trials, one open-label (Stroke Prevention Using Oral Thrombin Inhibitor in Atrial Fibrillation [SPORTIF] III), one double-blind (SPORTIF V), have compared the efficacy and safety of fixed-dose ximelagatran without anticoagulation monitoring with dose-adjusted warfarin using a non-inferiority design. On the basis of the results, the investigators concluded that ximelagatran was just as effective as warfarin in preventing stroke or systemic embolism (the primary end point), because the pre-specified non-inferiority criterion was met. Reanalysis of the data with rather conservative interpretive criteria, however, revealed a number of deficiencies: 1) an unreasonably generous margin that was potentially biased toward non-inferiority, given the low baseline event rate of warfarin; 2) the inappropriateness of the analytical method used to estimate the non-inferiority margin; 3) a lack of confidence that ximelagatran retains at least 50% of warfarin's effect (a prerequisite to the establishment of non-inferiority); 4) significant heterogeneity in the magnitude of efficacy observed in the two trials; and 5) safety concerns regarding increased liver toxicity with ximelagatran without a significant offsetting advantage in major bleeding. This imbalance in the benefit-risk profile materially undermines the investigators' claim of non-inferiority of ximelagatran and led the Food and Drug Administration to reject the sponsor's application for ximelagatran. Despite published conclusions to the contrary, we conclude that ximelagatran has not been shown to be non-inferior to warfarin. Such determinations of non-inferiority are highly dependent on the underlying assumptions, and graphical sensitivity analyses make this dependence explicit.