Development of a machine learning-based noninvasive diagnostic model for liver fibrosis in metabolic-associated steatotic liver disease.

BACKGROUND

To address the lack of simple tools for assessing fibrosis in metabolic dysfunction-associated steatotic liver disease (MASLD), this study develops a machine learning-based diagnostic model to identify patients at high risk of significant fibrosis and advanced fibrosis.

METHODS

Data from biopsy-proven MASLD patients were randomly divided into training and validation sets. Variables were then selected, and models using Logistic, Support Vector Machine, and eXtreme Gradient Boosting (XGBoost) were compared with identify the optimal model. Finally, Shapley Additive Explanations-based interpretability analysis was applied to explain the best-performing model. The DeLong test was applied to compare the new model with the aspartate aminotransferase to platelet ratio index (APRI) and fibrosis-4 index (FIB-4) models.

RESULTS

The XGBoost models demonstrated robust accuracy. For the significant fibrosis outcome, the model incorporating age, total cholesterol, glycosylated hemoglobin (HbA1c), and albumin/globulin ratio achieved an area under the receiver operating characteristic curve (AUC) of 0.74 in the training set and 0.72 in the testing set. Regarding the advanced fibrosis outcome, the model including age, total cholesterol, HbA1c, platelets, white cell count, and albumin/globulin ratio yielded an AUC of 0.78 in the training set and 0.74 in the testing set. Decision curve analysis curves confirmed clinical utility, and performance surpassed APRI and FIB-4 ( P < 0.05).

CONCLUSION

This study developed a novel noninvasive diagnostic model for MASLD using simple and easily accessible variables, which demonstrates superior performance compared with traditional serological models.