A new score with superior stroke risk prediction in atrial fibrillation: entropy-based information gain approaches in a large nationwide cohort

Background: Risk scores have been used to assess stroke risk in atrial fibrillation (AF) for reducing ischemic stroke and bleeding risk. Information gain ratio (IGR) is an entropy-based parameter that shows which clinical score is more informative for prediction of the clinical endpoint. Objective: Herein, we aimed to generate and validate a stroke risk score based on the TuRkish Atrial Fibrillation (TRAF) data. Methods: We used a split-sample approach to develop and internally validate the new stroke risk score. Based on multivariate logistic regression analysis, we generated CHADS-F in the anticoagulation naïve TRAF cohort (274,631 patients). CHADS-F stands for Cardiac failure (1 point), hypertension (1 point), age (≥ 65–69 = 1 point, ≥ 70–74 = 2 points ≥ 75 = 3 points), diabetes (1 point), stroke (2 points), and older female (1 point) (≥ 65). External validation was performed in the “Global Anticoagulant Registry in the FIELD-Atrial Fibrillation (GARFIELD-AF)” Registry. Informative capacity and accuracy of the CHADS-F score was compared with CHADS₂ and CHA₂DS₂-VASc scores. Results: In anticoagulation naïve cohort, CHADS-F (IGR for all cohort: 0.7526) outperforms both the CHADS₂ (IGR for all cohort: 0.6340) and CHA₂DS₂-VASc (IGR for all cohort: 0.6969) in terms of the IGR for ischemic stroke and systemic embolism. Receiver operating characteristic curves revealed highest accuracy for the CHADS-F score [area under curve for CHADS-F: 0.743, CHADS₂: 0.722, and CHA₂DS₂-VASc: 0.722]. CHADS-F had good discriminative abilities at predicting clinical endpoints in the GARFIELD-AF registry. Conclusion: The CHADS-F score had higher informative capacity and accuracy than the current CHADS₂ and CHA₂DS₂-VASc scores for predicting stroke and systemic embolism.