In Hellenic journal of cardiology : HJC = Hellenike kardiologike epitheorese
BACKGROUND : This study sought to develop and validate a risk score for predicting mortality in patients with AF after a hospitalization for cardiac reasons.
METHODS : The new risk score was derived by a prospective cohort of hospitalized patients with concurrent AF. The outcome measures were all-cause and cardiovascular mortality. Random forest was used for variable selection. A risk points model with predictor variables was developed by weighted Cox-regression coefficients and was internally validated by bootstrapping.
RESULTS : In total, 1130 patients with AF were included. During a median follow-up of 2 years, 346 (30.6%) patients died, 250 of whom had a cardiovascular cause of death. N-terminal pro-B-type natriuretic peptide and high-sensitivity troponin-T were the most important predictors of mortality, followed by indexed left atrial volume, history and type of heart failure, age, history of diabetes mellitus, and intraventricular conduction delay, all forming the BASIC-AF risk score (Biomarkers, Age, ultraSound, Intraventricular conduction delay, Clinical history). The score had good discrimination for all-cause (c-index=0.85, 95% CI 0.82-0.88) and cardiovascular death (c-index=0.84, 95% CI 0.81-0.87). The predicted probability of mortality varied more than 50-fold across deciles and adjusted well to observed mortality rates. Decision curve analysis revealed a significant net benefit of using the BASIC-AF risk score to predict the risk of death, compared with other existing risk schemes.
CONCLUSIONS : We developed and internally validated a well-performing novel risk score for predicting death in patients with AF. The BASIC-AF risk score included routinely assessed parameters, selected through machine learning algorithms, and may assist in tailored risk stratification and management of these patients.
Samaras Athanasios, Kartas Anastasios, Akrivos Evangelos, Fotos George, Dividis George, Vasdeki Dimitra, Vrana Elena, Rampidis Georgios, Karvounis Haralambos, Giannakoulas George, Tzikas Apostolos
atrial fibrillation, biomarkers, left atrial volume, machine learning, mortality, risk score