Bleeding risks with direct oral anticoagulants versus aspirin: Comparison of intention-to-treat versus on-treatment meta-analyses

Background Clinical trials may inadvertently attenuate estimates of treatment-specific risks by using intention-to-treat (ITT) analyses. The impact of using ITT versus on-treatment data on pooled bleeding estimates comparing direct oral anticoagulants (DOACs) with aspirin is uncertain. Methods This study is a follow-up analysis of a previously published meta-analysis of randomized trials comparing DOACs with aspirin. Using the same set of trials, we performed parallel random-effects meta-analyses using ITT and on-treatment data. Outcomes were major bleeding and intracranial hemorrhage. We descriptively compared pooled risk ratios (RRs) and estimated absolute risk differences (ARDs). Results Eight trials were included. The proportion of patients with major bleeding was lower in on-treatment (DOAC, 2.0% vs. aspirin, 1.2%) than in ITT analyses (DOAC, 2.5% vs. aspirin: 1.8%). For major bleeding, pooled RRs point estimates were higher using on-treatment versus ITT analyses (RR 1.51, 95% CI 1.07–2.12 vs. RR 1.32, 95% CI 0.97–1.79), though estimated ARDs were similar (+6 vs. +6 per 1000 patients). For intracranial hemorrhage, pooled RRs (RR 1.13, 95% CI 0.62–2.07 vs. RR 1.06, 95% CI 0.63–1.80) and estimated ARDs (0 vs. +1 per 1000 patients) were similar between the on-treatment and ITT analyses. Conclusion In our meta-analyses comparing DOACs with aspirin, relative estimates of major bleeding varied according to analytic approach, with numerically higher RR point estimates with on-treatment over ITT analyses. Physicians should be aware that the choice of analytic approach may influence relative safety estimates, and trials should systematically publish data pertaining to changes in treatment.