Discover Temporal Dynamics of Biomarkers in Predictive Modeling with Longitudinal Data

In the longitudinal study measurements of various biomarkers are taken from the same set of patients repeatedly over long periods. The longitudinal data provides important temporal information about the development of diseases. When applying standard data mining techniques to perform biomarker analysis and build predictive models to study diseases, typically the important temporal relationship among the measurements of the same biomarker is not considered. In this paper we present a novel method for predictive modeling leveraging the temporal information in the longitudinal patients data. Specifically, we propose a temporal dynamics model featured by a structural sparsity regularization designed according to the temporal structures of the longitudinal data. The model simultaneously identifies important biomarkers and discovers their temporal dynamics. We design experiments to show that the proposed method can better capture the temporal dynamics of the biomarkers and discuss related issues when applying the proposed method in healthcare analysis.