Independent predictors of stroke in patients with atrial fibrillation: a systematic review.

BACKGROUND

Absolute stroke rates vary widely among patients with nonvalvular atrial fibrillation. To balance the benefits and risks of chronic antithrombotic prophylaxis, it is important to estimate the absolute risk of stroke for individual patients.

METHODS

Systematic review of studies using multivariate regression techniques to identify independent risk factors for stroke in patients with atrial fibrillation was conducted, and reports of absolute stroke rates in subgroups of patients with these risk factors collected. A summary estimate of the relative risk associated with each independent risk factor was calculated using maximum likelihood methods.

RESULTS

Seven studies (including six entirely independent cohorts) were identified. Prior stroke/TIA (relative risk 2.5, 95% CI 1.8 to 3.5), increasing age (relative risk 1.5 per decade, 95% CI 1.3 to 1.7), a history of hypertension (relative risk 2.0, 95% CI 1.6 to 2.5), and diabetes mellitus (relative risk 1.7, 95% CI 1.4 to 2.0) were the strongest, most consistent independent risk factors. Observed absolute stroke rates for nonanticoagulated patients with single independent risk factors were in the range of 6 to 9% per year for prior stroke/TIA, 1.5 to 3% per year for history of hypertension, 1.5 to 3% per year for age >75, and 2.0 to 3.5% per year for diabetes. Female sex was inconsistently associated with stroke risk, whereas the evidence was inconclusive that either heart failure or coronary artery disease is independently predictive of stroke.

CONCLUSIONS

Four clinical features (prior stroke/TIA, advancing age, hypertension, diabetes) are consistent independent risk factors for stroke in atrial fibrillation patients. Prior stroke/TIA is the most powerful risk factor and reliably confers a high stroke risk (>5% per year, averaging 10% per year). Absolute stroke rates associated with other individual risk factors are difficult to precisely estimate from available data.