Structured Rule Discovery from Heterogeneous Longitudinal Data for Complex Disease

Abstract

Many complex diseases manifest heterogeneous degenerative disease progression processes that impose enormous challenges for accurate disease prognosis and effective intervention. The emerging “big” data collected from the population with predisposed disease risk brings us motivation to translate it into accurate prognosis and effective risk monitoring. We propose a structured sparse rule discovery method to identify risk-predictive patterns from heterogenous longitudinal. By extending the existing RuleFit framework, we have developed an analysis pipeline to derive risk-predictive patterns from complex data. The results in a de-identified Type 1 Diabetes dataset have shown promising predictive performances.