|
Abstract:
|
Personalized medicine, or precision medicine, is a medical model that uses disease subtypes, genetic makers, and other patient-level factors to develop customized treatment with desirable benefit/risk profiles for a given patient. In recent years, various statistical methodologies have been developed in this domain but there are still many open questions. We propose a novel tree-based ensemble method, a random forest of modified interaction trees (RFMIT), to generate predictive importance scores for covariates, and directly predict treatment effects for each individual patient with prediction intervals. We compared our methods with selected existing ones. Our method can be used to select predictive biomarkers, visualize treatment effects, generate predictive models, and easily incorporate a clinically meaningful difference for treatment decision or enrichment design.
|
