A comparison of multimodal biomarkers for chronic hepatitis B assessment using recursive feature elimination

An effective assessment of liver fibrosis in patients with chronic hepatitis B (CHB) is highly desired because it is important not only for clinical courses prediction, but also for the determination of antiviral therapy schemes. In recent years, various approaches for liver biopsies analysis have been highlighted, such as elastography techniques and serum markers, due to their properties of non-invasiveness. The aim of this study is to determine the best biomarkers or their combination by comparing multimodal biomarkers (ultrasound elastography parameters, biochemical hematologic parameters, and clinical parameters) for fibrosis assessment in chronic hepatitis B using a support vector machine combined with recursive feature elimination (RFE-SVM) approach. Results revealed that biomarkers from ultrasound elastography techniques achieved better prediction performance than others in the assessment of significant fibrosis (≥ F2) and cirrhosis (F4), and the best prediction performance were (1) ≥ F2: AUC = 0.902, ACC = 86.697%; (2) F4: AUC = 0.976, ACC = 90.364%. The findings are useful in guiding biomarkers selection and features optimization and in simplifying the prediction system for evaluation of liver fibrosis stage.