Development and validation of a machine learning-based model for identifying liver fibrosis in individuals with prior Schistosoma japonicum infection: a step toward precision management

BACKGROUND: Liver fibrosis caused by Schistosoma japonicum may continue to progress even following successful praziquantel chemotherapy. Although liver biopsy remains the gold standard for diagnosis of liver fibrosis, its invasiveness limits clinical applications. Conventional non-invasive indices, such as aspartate aminotransferase to platelet ratio index (APRI) and fibrosis-4 (FIB-4) index, often show suboptimal performance in schistosomiasis-related cases. This study aimed to develop and validate a machine learning (ML)-based model using epidemiological and laboratory data to identify liver fibrosis in individuals with prior S. japonicum infection, as a step toward precision management. METHODS: Data were obtained from the Jiangsu S. japonicum Infection Cohort (Jiangsu Province, China), involving 6158 participants with a documented history of infection who completed follow-up assessments during 2021-2022. Modeling variables were selected using LASSO regression and variance inflation factor analysis. Five ML models (k-nearest neighbor, logistic regression, support vector machine, decision tree, and extreme gradient boosting (XGBoost)) were evaluated. Model performance was compared against APRI and FIB-4 using the area under the receiver operating characteristic curve (AUC), decision curve analysis (DCA), and shapley additive explanations (SHAP) for interpretability. RESULTS: In the validation set, the XGBoost model demonstrated the highest diagnostic performance with an AUC of 0.854 (95% confidence interval (CI): 0.826-0.881), outperforming other ML models (AUC range: 0.609-0.776). This was significantly superior to FIB-4 (AUC = 0.514) and APRI (AUC = 0.516) (P < 0.001). DCA confirmed that the XGBoost model provided a substantial clinical net benefit across a broad range of threshold probabilities. According to Shapley additive explanations analysis, variables such as alcohol intake (mean SHAP value = 0.401), gamma-glutamyl transferase (0.295), and triglycerides (0.292) were the most influential predictors of liver fibrosis risk. CONCLUSIONS: The XGBoost-based model offers a robust, non-invasive tool for identifying liver fibrosis in individuals with prior S. japonicum infection, with alcohol intake, gamma-glutamyl transferase, and triglycerides identified as the critical predictors. By outperforming traditional indices and leveraging routinely available data, this model represents a promising advancement toward precision management of individuals with prior S. japonicum infection, offering a scalable approach for early intervention, risk‑stratified assessment, and monitoring of hepatic morbidity in post‑transmission settings.