Machine learning-based predictive and risk analysis using real-world data with blood biomarkers for hepatitis B
Yuemin Nan1, Suxian Zhao1, Xiaoxiao Zhang1
1Department of Traditional and Western Medical Hepatology, Third Hospital of Hebei Medical University, Hebei Provincial Key Laboratory of liver fibrosis in chronic liver diseases, Shijiazhuang, China.
Abstract:
Hepatitis B Virus (HBV) infection may lead to various liver diseases such as cirrhosis, end-stage liver complications, and Hepatocellular carcinoma (HCC). Patients with existing cirrhosis or severe fibrosis have an increased chance of developing HCC. Consequently, lifetime observation is currently advised. This study gathered real-world electronic health record (EHR) data from the China Registry of Hepatitis B (CR-HepB) database. A collection of 396 patients with HBV infection at different stages were obtained, including 1) patients with a sustained virological response (SVR), 2) patients with HBV chronic infection and without further development, 3) patients with cirrhosis, and 4) patients with HCC. Each patient has been monitored periodically, yielding multiple visit records, each is described using forty blood biomarkers. These records can be utilized to train predictive models. Specifically, we develop three machine learning (ML)-based models for three learning tasks, including 1) an SVR risk model for HBV patients via a survival analysis model, 2) a risk model to encode the progression from HBV, cirrhosis and HCC using dimension reduction and clustering techniques, and 3) a classifier to detect HCC using the visit records with high accuracy (over 95%). Our study shows the potential of offering a comprehensive understanding of HBV progression via predictive analysis and identifies the most indicative blood biomarkers, which may serve as biomarkers that can be used for immunotherapy.
