Final random-forest-based models outperformed all publicly available risk scores on internal and external test sets.