Scholarlink

Background <p>The coexistence of chronic hepatitis B (CHB) and nonalcoholic fatty liver disease (NAFLD) accelerates liver fibrosis progression, but effective noninvasive tools for fibrosis risk assessment in this specific population are lacking. This study aimed to develop and validate an explainable machine learning (ML) model to predict significant liver fibrosis in patients with comorbid CHB and NAFLD.</p> Methods <p>This retrospective study analyzed 376 patients with CHB and NAFLD. Significant fibrosis was defined as a liver stiffness measurement (LSM) ≥ 7.3&#xa0;kPa via transient elastography (FibroScan). Predictors were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariable logistic regression. The cohort was split into training (70%) and test (30%) sets. Using the training set, we developed eight predictive models, including logistic regression, random forest, XGBoost, LightGBM, support vector machine, multilayer perceptron (MLP), k-nearest neighbors, and naive Bayes. Hyperparameters were tuned in the training set using grid search with five-fold stratified cross-validation, with selection based on the mean cross-validated AUC. The final model was then refit on the entire training set using the optimal hyperparameters and subsequently evaluated in an independent held-out test set.Discrimination, calibration, and clinical utility were assessed using ROC analysis (AUC), calibration curves, the Brier score, and decision curve analysis. The best model was interpreted using SHapley Additive exPlanations (SHAP) and compared against the APRI and FIB-4 indices. In addition, we conducted stratified analyses within the full dataset by CAP, BMI, and HBV DNA status to assess the robustness of the MLP model across subgroups.</p> Results <p>Among 376 patients with CHB and NAFLD, 201 were classified as having significant liver fibrosis. A least absolute shrinkage and LASSO regression followed by multivariable logistic regression identified six independent predictors: age, body mass index (BMI), platelet count, total bilirubin, aspartate aminotransferase (AST), and γ-glutamyltransferase (GGT). OOF internal validation using five-fold stratified cross-validation in the training set indicated that the MLP model yielded the best performance (AUC = 0.741). In the independent test set, MLP again achieved the highest AUC (0.801) and showed strong overall classification performance (accuracy = 0.761; F1-score = 0.773) with the lowest Brier score (0.183). Decision curve analysis further suggested that MLP provided a higher net benefit across a wide range of clinically relevant threshold probabilities. SHAP analysis indicated that predictions were most strongly influenced by GGT, AST, PLT, BMI, TBIL, and age. The model significantly outperformed APRI (AUC = 0.691) and FIB-4 (AUC = 0.604) (DeLong test, both <i>P</i> &lt; 0.05). Subgroup analyses indicated that MLP maintained relatively stable discrimination across subgroups (AUC approximately 0.75–0.87) and generally consistent performance across key classification metrics.</p> Conclusions <p>An MLP-based prediction model incorporating explainable machine-learning techniques may be useful for estimating liver fibrosis risk in patients with CHB and NAFLD. As a noninvasive and interpretable tool, it has potential value for clinical risk stratification; however, its robustness and generalizability should be further confirmed in external cohorts from diverse populations and multicenter studies.</p> Trial registration <p>Not applicable.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An interpretable machine learning model for prediction of significant liver fibrosis in comorbid chronic hepatitis B and nonalcoholic fatty liver disease: a retrospective development and validation study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月16日

查看全文
  • Jiaping Gu,
  • Jiale Chai

摘要

Background

The coexistence of chronic hepatitis B (CHB) and nonalcoholic fatty liver disease (NAFLD) accelerates liver fibrosis progression, but effective noninvasive tools for fibrosis risk assessment in this specific population are lacking. This study aimed to develop and validate an explainable machine learning (ML) model to predict significant liver fibrosis in patients with comorbid CHB and NAFLD.

Methods

This retrospective study analyzed 376 patients with CHB and NAFLD. Significant fibrosis was defined as a liver stiffness measurement (LSM) ≥ 7.3 kPa via transient elastography (FibroScan). Predictors were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariable logistic regression. The cohort was split into training (70%) and test (30%) sets. Using the training set, we developed eight predictive models, including logistic regression, random forest, XGBoost, LightGBM, support vector machine, multilayer perceptron (MLP), k-nearest neighbors, and naive Bayes. Hyperparameters were tuned in the training set using grid search with five-fold stratified cross-validation, with selection based on the mean cross-validated AUC. The final model was then refit on the entire training set using the optimal hyperparameters and subsequently evaluated in an independent held-out test set.Discrimination, calibration, and clinical utility were assessed using ROC analysis (AUC), calibration curves, the Brier score, and decision curve analysis. The best model was interpreted using SHapley Additive exPlanations (SHAP) and compared against the APRI and FIB-4 indices. In addition, we conducted stratified analyses within the full dataset by CAP, BMI, and HBV DNA status to assess the robustness of the MLP model across subgroups.

Results

Among 376 patients with CHB and NAFLD, 201 were classified as having significant liver fibrosis. A least absolute shrinkage and LASSO regression followed by multivariable logistic regression identified six independent predictors: age, body mass index (BMI), platelet count, total bilirubin, aspartate aminotransferase (AST), and γ-glutamyltransferase (GGT). OOF internal validation using five-fold stratified cross-validation in the training set indicated that the MLP model yielded the best performance (AUC = 0.741). In the independent test set, MLP again achieved the highest AUC (0.801) and showed strong overall classification performance (accuracy = 0.761; F1-score = 0.773) with the lowest Brier score (0.183). Decision curve analysis further suggested that MLP provided a higher net benefit across a wide range of clinically relevant threshold probabilities. SHAP analysis indicated that predictions were most strongly influenced by GGT, AST, PLT, BMI, TBIL, and age. The model significantly outperformed APRI (AUC = 0.691) and FIB-4 (AUC = 0.604) (DeLong test, both P < 0.05). Subgroup analyses indicated that MLP maintained relatively stable discrimination across subgroups (AUC approximately 0.75–0.87) and generally consistent performance across key classification metrics.

Conclusions

An MLP-based prediction model incorporating explainable machine-learning techniques may be useful for estimating liver fibrosis risk in patients with CHB and NAFLD. As a noninvasive and interpretable tool, it has potential value for clinical risk stratification; however, its robustness and generalizability should be further confirmed in external cohorts from diverse populations and multicenter studies.

Trial registration

Not applicable.