Predicting bleeding risk in frail older patients with atrial fibrillation using data from the FRAIL-AF trial

Abstract Background Bleeding remains an important complication in anticoagulated patients with atrial fibrillation (AF). Hereto, multiple scores have been developed to predict bleeding risk and subsequently identify a subset of patients in whom additional effort may be warranted to lower the risk of bleeding. Bleeding risk is particularly high in frail older individuals with AF. Yet to what extent existing scores can accurately identify patients at high risk in this population is unknown. Purpose To validate existing bleeding risk scores for predicting bleeding in frail older AF patients using oral anticoagulants. Methods We used data from the FRAIL-AF trial that randomized frail AF patients ≥75 years of age between continuing vitamin K antagonist (VKA) treatment or switching to non-VKA oral anticoagulant (NOAC) treatment. The outcome was major or clinically relevant non-major bleeding. We validated 4 existing bleeding risk scores: ORBIT, ATRIA, AF-BLEED and (the recently derived) DOAC-score. We used cause specific cox proportional hazard models with all-cause mortality as a competing risk, with the number of points on each score as a univariate variable to calculate discrimination by means of the c-statistic for each score. Next, to assess clinical utility, we estimated the incidence bleeding rates for patients classified as low-risk and high-risk by each respective score. Results In total 661 patients continued VKA treatment, and 663 switched to any NOAC. An additional 66 patients, originally excluded from randomization due to an estimated glomerular filtration rate (eGFR) ≤30 ml/min/1.73 m2 but followed observationally were included in the current analysis. The median age was 82.3 years (interquartile range (IQR) 78.7 – 86.6), 61.4% were male. Median eGFR was 61 (IQR 49 – 57), 54% had hypertension, 14% a previous bleeding, and 1.2% liver disease. In total 172 first bleedings occurred. Discrimination of all risk scores was poor, both in VKA and in NOAC users. Table 1 shows the c-statistics across the different risk scores; ranging from 0.54 to 0.56. The observed risk of bleeding in the low-risk category varied widely, see Table 2. Conclusion Existing bleeding risk scores are unable to accurately predict the bleeding risk in frail older patients with AF. Novel risk tools are needed for this growing population of vulnerable AF patients at high risk of bleeding.