Novel HEP-FIB score combined with machine learning models for non-invasive prediction of significant hepatic fibrosis in chronic hepatitis B patients: a single-center, internally validated study

In a retrospective, single-center study of 382 patients with chronic hepatitis B, a newly developed HEP-FIB score showed strong performance for identifying significant hepatic fibrosis, while an XGBoost model achieved the best predictive performance among 10 machine-learning approaches evaluated. HEP-FIB produced an area under the receiver operating characteristic curve of 0.892 (95% CI, 0.859–0.925), and the machine-learning model achieved an AUC of 0.914 (95% CI, 0.857–0.971) in the validation dataset. The study authors said external validation is required before the score can be recommended for routine clinical use.

Researchers used least absolute shrinkage and selection operator regression followed by multivariable logistic regression to identify independent predictors of fibrosis. They then developed HEP-FIB from routinely available laboratory parameters, including platelet count, liver-function measures and lipid biomarkers. The score’s diagnostic performance was assessed with ROC-curve analysis, bootstrap internal validation and calibration analysis, and was compared with established non-invasive indices: APRI, FIB-4 and GPR. Ten machine-learning algorithms were also constructed and evaluated using measures of discrimination, classification performance, calibration and overall predictive ability. Decision curve analysis was used separately for HEP-FIB and the machine-learning models to examine clinical utility.

At an optimal HEP-FIB cutoff of 4 or higher, the score had a sensitivity of 78%, specificity of 89% and overall accuracy of 88%. Its AUC was higher than those for APRI, FIB-4 and GPR, which were 0.724, 0.654 and 0.773, respectively; all three comparisons had p values below 0.001 by the DeLong test. Bootstrap validation produced an optimism-corrected AUC of 0.881. Calibration analysis indicated good agreement between predicted and observed fibrosis probabilities, while decision curve analysis favored HEP-FIB over the default strategies assessed.

XGBoost performed best among the machine-learning approaches in the validation dataset, with an AUC of 0.914, accuracy of 0.88, sensitivity of 0.72, specificity of 0.93, balanced accuracy of 0.83 and Matthews correlation coefficient of 0.74. Its Brier score was 0.095, and the reported false-positive and false-negative rates were 0.062 and 0.275, respectively. The machine-learning models showed a marginal performance advantage over HEP-FIB and greater net clinical benefit than test-all and test-none strategies in decision curve analysis. These findings come from a retrospective cohort at one center and were supported only by internal bootstrap validation. The study did not establish performance in independent or diverse populations, so the results do not yet establish routine clinical use or replacement of existing fibrosis-assessment methods.

Subject of Research: Internal validation of HEP-FIB and machine-learning models for non-invasive prediction of significant hepatic fibrosis in chronic hepatitis B

Article Title: Novel HEP-FIB score combined with machine learning models for non-invasive prediction of significant hepatic fibrosis in chronic hepatitis B patients: a single-center, internally validated study

Article References: Bashir, A., Arora, R., Mehrotra, D., Bala, M., Parry, A. H., & Iqball, A. (2026). Novel HEP-FIB score combined with machine learning models for non-invasive prediction of significant hepatic fibrosis in chronic hepatitis B patients: a single-center, internally validated study. BMC Infectious Diseases. https://doi.org/10.1186/s12879-026-14508-8

Image Credits: AI Generated

DOI: 10.1186/s12879-026-14508-8

Keywords: chronic hepatitis B, hepatic fibrosis, HEP-FIB, machine learning, XGBoost, non-invasive testing, diagnostic accuracy, internal validation, clinical utility, Novel, score, combined