Machine Learning for Individualized Clinical Risk Prediction and Prevention

The rising burden of healthcare costs suggests that the healthcare system could benefit from novel methods that allow for continuous learning to provide more individualized care at lower costs and with improved outcomes. This thesis work presents the s3m learning PIC (synergistic, structured, scientific/clinical machine learning for prediction, interpretation, and communication) toolkit to address the challenges hindering the successful implementation of learning healthcare systems. In this dissertation, we introduce a s3m learning random forest approach for clinical risk prediction with survival, longitudinal, and multivariate (SLAM) data. Then, we present methods for assessing model performance and visualizing the results of dynamic models. Afterwards, we describe methods for clinically-relevant visualizations to communicate s3m learning predictions. Although we specifically illustrate our methods with sudden cardiac arrest (SCA) prediction, our approaches generalize more broadly to other medical applications and beyond