Noninvasive Prediction of Significant Hepatic Injury in Treatment-Naïve Children with Chronic HBV Infection.

Background: Significant hepatic injury (SHI) in children with chronic hepatitis B virus (HBV) infection requires timely identification, yet liver biopsy is invasive and not routinely feasible. This study aimed to develop a non-invasive predictive model for SHI. Methods: This retrospective analysis included treatment-naïve children with chronic HBV infection undergoing liver biopsy. Participants were randomly split into 70% training and 30% independent testing cohorts prior to preprocessing. Missing data imputation and least absolute shrinkage and selection operator (LASSO)-based feature selection were performed, and nine machine learning (ML) models were developed and optimized using nested cross-validation (CV). The optimal logistic regression (LR) model was used to construct and validate a predictive nomogram. Results: A total of 246 eligible children were included, of whom 147 (59.8%) had SHI. LASSO regression identified five predictors for model development. Among the evaluated ML models, LR showed stable performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.799 (95% CI: 0.740-0.858) in the training folds, 0.781 (95% CI: 0.585-0.972) in the validation folds, and 0.707 (95% CI: 0.562-0.851) in the independent testing cohort. The nomogram showed moderate calibration and better clinical utility compared with alanine aminotransferase (ALT) alone across a broad range of threshold probabilities. Conclusion: We developed a non-invasive predictive model based on routine clinical data to detect SHI in treatment-naïve children with chronic HBV infection, which exhibited moderate predictive power and marginal improvement over ALT alone.