AutoSCORE

Risk scores are crucial for clinical risk stratification and resource allocation, with point-based scores being favored for their interpretability. However, the development of such scoring models has been challenging and underexplored, particularly in leveraging electronic health records. This study introduces AutoScore, a machine learning-based framework that automates the creation of interpretable point-based clinical scores, demonstrating its effectiveness in mortality prediction using data from the Beth Israel Deaconess Medical Center and showing comparable performance to traditional models with fewer variables and greater interpretability.