Mortality risk associated with venous thromboembolism: a systematic review and Bayesian meta-analysis

Background: Venous thromboembolism is associated with increased mortality risk in some populations, but how frequently it is a direct cause of death is unclear. We used data from venous thromboembolism prevention trials to evaluate the causal effect of venous thromboembolism reduction on mortality. Methods: We did a systematic review and meta-analysis of randomised controlled trials (RCTs) evaluating venous thromboembolism prevention. We searched MEDLINE, Embase, PubMed, and Web of Science starting from Jan 1, 1993, to March 19, 2018. We included studies of patients who were at elevated risk of venous thromboembolism and were randomly assigned to either anticoagulant or antiplatelet therapy versus placebo or no treatment. We excluded studies with an active control agent (which might mitigate the lethality of venous thromboembolism) and those for which mortality data were unavailable. We modelled heterogeneity in a Bayesian framework, taking overall mortality as a primary endpoint, and pulmonary embolism, fatal pulmonary embolism, and major bleeding as secondary endpoints. We focused our analyses on studies reporting statistically significant effects of prevention on venous thromboembolism endpoints. We report treatment effects as median risk ratios (RRs), wherein a null effect equals 1, with 95% credible intervals (CrIs). This meta-analysis was registered with PROSPERO, CRD42018089697. Findings: From 4229 studies screened, we identified 86 eligible RCTs; 52, with data from over 70 000 patients, were positive, with significantly increased venous thromboembolism risk in patients in control groups versus treatment groups (RR 2·74, 95% CrI 2·32–3·31, p<0·0001). The meta-analysis established that the causal effect of venous thromboembolism prevention on mortality was null (control group mortality was 3391 [9·8%] of 34 537 patients; treatment group mortality was 3498 [9·8%] of 35 795 patients [RR 1·01, 95% CrI 0·97–1·06; p=0·58]) with low heterogeneity (τ 0·02, 95% CrI 0·00–0·07, p=0·89). Patients in control groups had more pulmonary embolism (RR 2·22, 95% CrI 1·78–2·89, p<0·0001) and fatal pulmonary embolism (1·58, 1·14–2·19, p=0·01), but less major bleeding (0·60, 0·47–0·75, p<0·0001) than those in treatment groups. A meta-analysis with the additional 34 negative studies yielded similar results for all endpoints except fatal pulmonary embolism, where evidence of an effect was weaker (1·42, 1·05–1·91, p=0·02). Interpretation: The perception that venous thromboembolism is a common cause of mortality should be revised considering the null effect of venous thromboembolism prevention on mortality. Our findings call into question the use of composite endpoints in venous thromboembolism-prevention trials and provide rationale for de-escalation trials. Funding: None.