Short lookback periods causing exaggerated stroke risk estimates in atrial fibrillation may expose patients to unnecessary anticoagulant treatment

Purpose The purpose was to investigate how different lookback periods in observational registry studies affect estimates of stroke risk in patients with atrial fibrillation and stroke risk score CHA2DS2‐VASc 1, a level where the appreciated risk is likely to affect decisions about oral anticoagulation. Methods All 354 854 individuals in Sweden with a hospital diagnosis of atrial fibrillation during 2010‐2016 were included. At least 13 years of observational data prior to inclusion was available for all patients. The prevalence of hypertension, heart failure, diabetes, previous thromboembolism, and vascular disease was estimated from data with different lookback periods. The incident stroke rates at CHA2DS2‐VASc score 1 was then assessed using data with successively longer lookback periods. Results Depending on duration of lookback period, the proportion of patients with heart failure varied 2.7 times, thromboembolism 3.7 times, hypertension 4.0 times, and diabetes and vascular disease both approximately 4.5 times. During follow‐up, 22 237 patients suffered an ischaemic stroke. The estimated risk without anticoagulant treatment at CHA2DS2‐VASc score 1 was 51% higher if the scores had been calculated with the shortest lookback period than if all information had been used. Conclusions Short lookback periods underestimate comorbidity, cause high‐risk patients to be misclassified as low risk, and overestimate stroke risk at CHA2DS2‐VASc 1. This may lead to unnecessary anticoagulant treatment of true low‐risk patients. Transparency regarding lookback periods is essential for interpretation and comparison of registry studies.