Abstract
BACKGROUND: Validated risk prediction scores for incident heart failure (HF) in patients with atrial fibrillation are lacking. We aimed to externally validate the the HF prediction risk score derived from three large control randomized trials RE-LY, AVERROES, and ACTIVE-A (REACT-HF) score and assess potential improvements by incorporating biomarkers. METHODS: We included 2599 patients with atrial fibrillation without prior HF from the Swiss-AF (Swiss Atrial Fibrillation) and BEAT-AF (Basel Atrial Fibrillation) cohorts. We estimated the C statistics of the REACT-HF score with Cox proportional hazards models and improved prediction by adding hs-CRP (high-sensitivity C-reactive protein), NT-proBNP (N-terminal pro-B-type natriuretic peptide), and high-sensitivity troponin T. The primary outcome was incident HF hospitalization within 2 years. Secondary outcomes included cardiovascular death, a composite of incident HF hospitalization and cardiovascular death, and all-cause death. RESULTS: The mean age was 70.2±10.3 years, 29.1% were women, and 54.8% had paroxysmal atrial fibrillation. Across risk score quintiles, the incidence rates per 100 patient-years increased for the primary outcome (0.27, 0.54, 1.00, 2.24, 5.49), cardiovascular death (0.00, 0.11, 0.10, 0.91, 2.01), the composite of cardiovascular death and first HF hospitalization (0.27, 0.65, 1.10, 3.09, 6.86), and all-cause death (0.00, 0.65, 0.40, 1.56, 4.10). The estimated C statistic (95% CI) for the primary outcome was 0.76 (0.72-0.81). C statistics for the secondary outcomes were consistent. The biomarker-enhanced model, including hs-CRP and NT-proBNP, improved the C statistic to 0.84 (0.80-0.87). CONCLUSIONS: In this external validation, the REACT-HF risk score demonstrated good discrimination for predicting the first HF hospitalization within 2 years of follow-up. The addition of NT-proBNP and hs-CRP further improved the score. The REACT-HF score may help identify patients with atrial fibrillation at risk for HF, aiding in preventive therapy.