Statistical controversies in clinical research: limitations of open-label studies assessing antiangiogenic therapies with regard to evaluation of vascular adverse drug events-a meta-analysis

Background: Previous meta-analyses have shown paradoxical increased risk of bleeding and thrombotic events in patients receiving antiangiogenics (AA) that may be simply explained by the studies design included. By a meta-epidemiological approach, we aim to investigate the impact of double-blind (DB) and open-label study designs on the risks of bleeding, venous thrombotic events (VTE) and arterial thrombotic events (ATE) in cancer patients treated with AA. Materials and methods: We searched Medline, Cochrane, ClinicalTrials.gov databases and proceedings of major oncology congresses for clinical trials published from January 2003 to January 2016. Randomized clinical trials that assigned patients with solid cancers to AA or control groups were eligible for inclusion. Combined odds ratios (ORs) for the risks of bleeding events, VTE and ATE were calculated for open and DB trials. Estimation bias of the treatment effect was determined by the ratio of OR, by dividing the OR values obtained in open-label trials by those obtained in DB trials. Results: The literature-based meta-analysis included 166 trials (72 024 patients). For bleeding events, comparison of AA versus control yielded an overall OR of 2.41 [95% confidence interval (95% CI) 2.12-2.73; P<0.001], but this risk was overestimated by 1.68 (95% CI 1.33-2.13) in open-label studies. Concerning VTE, the OR was 1.19 (95% CI 1.04-1.35; P=0.012) overall with AA, but this effect disappears when considering only DB trials (OR 0.99, 95% CI 0.83-1.17). The corresponding ratio of OR showed a significant overestimation of 1.53 (95% CI 1.19-1.96) in open-label trials. For ATE, an OR of 1.59 (95% CI 1.30-1.94; P<0.001) was observed, associated with a significant overestimation of 1.65 (95% CI 1.13-2.43) in open-label trials. Conclusions: Open-label studies overestimated the risk of vascular adverse events with AA by at least 50%. Meta-analyses assessing adverse drug events should therefore be restricted to DB randomized trials.