Discovering linear biosignatures for treatment response based on maximizing Kullback-Leibler Divergence in linear mixed-effect models
Date:
Abstract: It is a continuing challenge in mental health research to identify patients who respond to treatment, since the treated and untreated patients often have similar outcomes on average. Precision medicine approaches, that consider an individual’s personal information, often produce treatment decision rules that are quite complicated. In this talk, I provide an approach to precision medicine to estimate a linear combination of patient baseline characteristics, i.e., a “biosignature”, defined to maximize the Kullback-Leibler Divergence between a treatment and control distribution. I will describe an algorithm to estimate the biosignature and illustrate the approach via a simulation study and a depression clinical trial.