Early detection and risk assessment for chronic disease with irregular longitudinal data analysis

Abstract

Early detection and risk assessment of complex chronic disease based on longitudinal clinical data is helpful for doctors to make early diagnosis and monitor the disease progression. Disease diagnosis with computer-aided methods has been extensively studied. However, early detection and contemporaneous risk assessment based on partially labeled irregular longitudinal measurements is relatively unexplored. In this paper, we propose a flexible mixed-kernel framework for training a contemporaneous disease risk detector to predict the onset of disease and monitor the disease progression. Moreover, we address the label insufficiency problem by identifying the pattern of disease-induced progression over time with longitudinal data. Our method is based on a Structured Output Support Vector Machine (SOSVM), extended to longitudinal data analysis. Extensive experiments are conducted on several datasets of varying complexity, including the contemporaneous risk assessment with simulated irregular longitudinal data; the identification of the onset of Type 1 Diabetes (T1D) with irregularly sampled longitudinal RNA-Seq gene expression dataset; as well as the monitoring of the drug long-term effects on patients using longitudinal RNA-Seq dataset containing missing time points, demonstrating that our method enhances the accuracy in both early diagnosis and risk estimation with partially labeled irregular longitudinal clinical data.

Keywords: Early diagnosis; Longitudinal measurements; Machine learning; Risk monitoring; Structured output; Support Vector Machine.