Statins alongside anticoagulation in venous thromboembolism: a signal, not an answer

Design

Bayesian random-effects meta-analysis of 28 studies: 17 observational and 11 randomised controlled trials

Population

Adults receiving anticoagulation for venous thromboembolism, with or without concomitant lipid-lowering therapy

Primary outcome

Recurrent venous thromboembolism, with post-thrombotic syndrome and major bleeding

Effect

Recurrent venous thromboembolism relative risk 0.73 (95% credible interval 0.58-1.05); post-thrombotic syndrome 0.83 (0.40-1.59); major bleeding 0.85 (0.76-0.96); C-reactive protein reduced

Whether adding a lipid-lowering drug to anticoagulation reduces recurrent venous thromboembolism has been circling for years, driven by the anti-inflammatory hypothesis. This Bayesian meta-analysis pooled 28 studies — 17 observational and 11 randomised — comparing the combination with anticoagulation alone.

For recurrent venous thromboembolism the pooled relative risk was 0.73, with a credible interval from 0.58 to 1.05 and a 97% posterior probability of benefit. Post-thrombotic syndrome showed a 17% lower risk but with an interval spanning 0.40 to 1.59. Major bleeding was 15% lower (0.76 to 0.96, 99% posterior probability of benefit) — a finding that is easier to attribute to who gets prescribed a statin than to any effect of the drug on bleeding. C-reactive protein fell.

The Bayesian framing deserves attention because it changes how the result reads. A 97% posterior probability of benefit sounds more definitive than a credible interval that includes no effect, and both describe the same data. With most of the evidence observational, confounding by indication is the obvious alternative explanation: patients suitable for long-term statin therapy differ from those who are not, in ways that predict both recurrence and bleeding.

• Do not add a statin for the purpose of preventing recurrent venous thromboembolism
• Continue statins that are already indicated; nothing here argues for stopping
• Note that most included studies were observational, with confounding by indication likely
• The lower bleeding risk is more plausibly a marker of healthier patients than a drug effect
• Await the randomised evidence the authors themselves call for

Why it matters

It shows how a Bayesian presentation can make an inconclusive result read as a positive one.

Don't overread it

A high posterior probability of benefit in a mostly observational meta-analysis is not evidence of efficacy, and the credible interval includes no effect.

The statistics, in plain English

A Bayesian credible interval running from 0.58 to 1.05 includes the possibility of no benefit, and the accompanying 97% posterior probability is simply the proportion of that interval below 1.0 — a different presentation of the same uncertainty, not stronger evidence. Pooling 17 observational studies with 11 randomised ones means the observational data dominate, and the direction of confounding by indication would produce exactly this result in the absence of any true effect.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.