Adverse Outcomes Prediction for Congenital Heart Surgery: A Machine Learning Approach

In contrast to traditional risk assessment methods (logistic regression), which assume that risk factors interact linearly and additively, the non-linear machine learning methodology of Optimal Classification Trees provides superior power for predicting risks after congenital heart surgery, with the advantage over other machine learning methods of logical interpretability. This methodology also allows estimation of individual patient risk, based on aggregate database data, and may facilitate decision–making and quality improvements in congenital heart surgery.