Developing and validating a prediction tool for identifying significant liver fibrosis in children with chronic...

Assessment of liver fibrosis is essential for monitoring disease progression, predicting prognosis, and initiating treatment in children with chronic hepatitis B (CHB). However, conventional non-invasive tests demonstrate limited predictive performance for children's liver fibrosis. This study aimed to develop and validate a machine learning (ML) approach to identify significant liver fibrosis in children with CHB.

Children with CHB who underwent liver biopsies at Hunan Provincial Children's Hospital (2015-2023) were enrolled and randomized (7:3) into training and validation sets. Six ML models were developed and validated for predicting significant liver fibrosis, and the final model was determined by the operating characteristic area under curve (AUC) and decision curve analysis. The performance of the final model was compared with fibrosis-4 index (FIB-4), aspartate aminotransferase to platelet ratio index (APRI), aspartate aminotransferase to alanine aminotransferase ratio index (AAR), and liver stiffness measurement (LSM) to evaluate its diagnostic performance.

Among 479 children, 18 % had significant liver fibrosis. Four predictors-Hepatitis B Surface Antigen, Hepatitis B e Antibody, γ-Glutamyl Transferase, and Alkaline Phosphatase-were included in the final model, which achieved an AUC of 0.85 (95 % CI: 0.79-0.90) in the training set and 0.81 (95 % CI: 0.73-0.89) in the validation set, outperforming APRI, FIB-4, AAR, and LSM (AUC range: 0.44-0.72). The new model is available online: https://kidfibrosis.streamlit.app.

This ML-based model provides a visual, personalized, and noninvasive tool for identifying significant liver fibrosis in children with CHB, facilitating early identification and disease surveillance of significant liver fibrosis.