PCV15 Bias-Corrected Indirect Comparison Meta-Analysis of New Oral Anticoagulant in Atrial Fibrillation

Abstract

Meta-epidemiological studies have shown that open-label trials were potentially biased and, in average, overestimate treatment effect magnitude by about 14% compared to double-blind trials. In network meta-analysis where the two types of trials are present, this bias could influence indirect comparisons and favor the less rigorous trials. The objective was to estimate the relative efficacy of apixaban compared to other new anticoagulants (NOAC) in stroke and systemic embolism prevention in atrial fibrillation (AF) using indirect comparison meta-analysis and taking into account the bias of open-label trial such RELY (dabigatran). We performed a systematic literature review to identify all RCTs evaluating NOAC in AF. Adjusted indirect comparisons and bias correction were conducted using methods derived from those described by Bucher et al. and Welton et al., respectively. The magnitude of the bias induced in treatment effect estimation by open-label studies was estimated using meta-epidemiological studies from Juni et al. This bias correction was unnecessary for rivaroxaban results given that ROCKET AF was double-blinded. No statistically significant difference in the composite endpoint of stroke and systemic embolism was found between apixaban versus dabigatran 150mg/d (odds-ratio: 1.03, 95%CI [0.75;1.42]), between apixaban versus dabigatran 110mg/d (0.75 95%CI [0.55;1.02]), and between apixaban versus rivaroxaban (0.90 95%CI [0.70;1.15]). For ischemic strokes, indirect comparisons for apixaban versus comparators were: 1.09 [0.75;1.58] versus dabigatran 150mg BID, 0.72 [0.51;1.04] versus dabigatran 110mg BID and 1.03 [0.75;1.42] versus rivaroxaban. After correction for the potential bias of the phase 3 trial of dabigatran due to the open-label design, indirect comparisons of the apixaban to dabigatran showed no significant difference in terms of stroke or systemic embolism and ischemic stroke for the 2 doses. Further research on a AF specific bias estimation should be undertaken to better understand the magnitude of the possible effect size overestimation in RELY open-label study.