Using Optimal Survival Tree Model for AF Event-Free Survival Time Prediction

This study presents a methodology to acquire, integrate, and analyze clinical data based on an innovative application of the Optimal Survival Tree (OST) algorithm. It has been tested on a clinical dataset of 4114 patients with a follow-up of 59.0 ± 19.3 months, collected in 10 years at the University of Catanzaro Medical Hospital. Tests confirmed the capacity of the OST-based methodology to predict four profiles of 10-year atrial fibrillation risk with good results. Results have been compared with those obtained by using (i) Classification and Regression Tree (CART), (ii) Conditional Inference Tree (cTree), and (iii) Random Forest (RF). Performances for OST reported an AUC of 0.794 and a Brier Score of 0.131 whereas CART, C-Tree and RF reported an AUC of 0.764, 0.766 and 0.804, and a Brier Score of 0.137, 0.156 and 0.131, respectively, proving the efficacy of the proposed methodology.