The COOL-AF 3-year prediction models for death, ischemic stroke, and major bleeding: a report from the COOL-AF registry

Abstract

Patients with atrial fibrillation (AF) are at increased risk of death, ischemic stroke/systemic embolism (SSE), and major bleeding. Most existing risk models focus on short-term outcomes and are derived mainly from Western populations. We aimed to develop and validate 3-year prediction models for these outcomes in an Asian AF cohort and to compare their performance with established risk scores. We analyzed the 3-year data from the prospective, nationwide COOL-AF registry, enrolling patients with non-valvular AF from 27 hospitals in Thailand. Multivariable Cox proportional hazards models were used to derive prediction models for death, SSE, and major bleeding. Model performance was assessed using discrimination (C-statistics), calibration, and reclassification metrics, and compared with GARFIELD, CHA₂DS₂-VASc, and HAS-BLED models. Among 3,405 patients, 380 deaths, 134 SSE events, and 199 major bleeding events occurred over a median follow-up of 35.9 months. The 3-year COOL-AF models showed good discrimination, with C-statistics of 0.728 (0.712–0.743) for death, 0.703 (0.687–0.718) for SSE, and 0.700 (0.685–0.716) for major bleeding, and demonstrated good calibration. Compared with existing scores, the COOL-AF models performed better than CHA₂DS₂-VASc for death and better than HAS-BLED for major bleeding, while showing comparable or superior performance to the GARFIELD models, particularly for bleeding risk. Reclassification indices further supported the incremental value of the COOL-AF models. In conclusion, the 3-year COOL-AF prediction models provide an acceptable levels of prediction, well-calibrated estimates of death, SSE, and major bleeding in Asian patients with AF and outperform commonly used clinical risk scores, supporting their use for long-term risk stratification and clinical decision-making.