Abstract
Background: Direct oral anticoagulants (DOAC) have become standard therapy for thrombosis prevention since 2008. Meta‐analyses (MAs) of randomized controlled trials (RCT) are considered high‐level evidence supporting clinical guidelines. However, methodological biases may yield false‐positive findings, particularly through inadequate hypothesis formulation, multiple testing without alpha‐risk correction, and selective outcome reporting. No study has systematically assessed the risk of false positives in MA evaluating DOAC through hypothetico‐deductive reasoning. Objective: This study aimed to assess methodological features relevant to confirmatory inference in MA of RCTs evaluating DOACs, including hypothesis specification and multiplicity management, and to quantify nominal exposure to type I error under specified assumptions. Methods: Following a pre‐specified protocol, we conducted a meta‐epidemiological study using embase and PubMed databases. We included MA of RCT evaluating four EMA‐approved DOAC. Two independent authors performed study selection. For each MA, we calculated the nominal family‐wise probability of at least one type I error under the global null hypothesis, assuming independent tests. We assessed the presence of directional hypotheses, primary outcome definition, pre‐specified protocols, sensitivity analyses, and spin in conclusions. Results: Ninety‐three MA were included (mean 34 118 patients; seven RCT per MA). The median nominal family‐wise type I error probability was 40%, and only 2.2% of MAs reported alpha‐risk management, 26.9% had pre‐specified protocols, and 14.0% reported directional hypotheses. Only 5.4% combined a clearly specified directional hypothesis with a single primary objective and well‐defined outcome. Spin was present in 30.1% of conclusions. Conclusion: Most MAs showed substantial uncorrected nominal exposure to multiplicity; this estimate is a theoretical indicator and not the probability that a published conclusion is false. Many MAs also lacked explicit hypotheses and frequently exhibited spin. Strengthening reliability requires integrating genuine hypothetico‐deductive approaches and adapting methodological guidelines.

This publication has 33 references indexed in Scilit: