A Simple Laboratory-based Machine Learning Model Accurately Predicts Advanced Liver Fibrosis in Metabolic...

Area of Science:

• Hepatology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) affects nearly a quarter of the global population.
• Current diagnosis relies on invasive liver biopsy, limiting widespread screening.
• There is a critical need for noninvasive methods to assess liver fibrosis in MASLD.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for assessing advanced liver fibrosis in MASLD patients.
• To compare the performance of the developed ML model against established noninvasive scores (FIB-4 and APRI).

Main Methods:

• Retrospective analysis of electronic medical records from biopsy-proven MASLD patients.
• Development of fourteen ML models using demographic, clinical, and liver biopsy data.
• Comparison of the best performing ML model (FAET) against FIB-4 and APRI for advanced fibrosis detection.

Main Results:

• The FAET model, utilizing 13 features, achieved 84% accuracy, 0.82 AUC, 0.77 F1 score, 0.81 NPV, and 0.91 PPV.
• FAET demonstrated superior performance over FIB-4 and APRI, significantly improving accuracy, AUC, F1 score, NPV, and PPV.
• A simplified version using five features achieved 82% accuracy and 0.80 AUC on an external test set, enabling a web-based tool.

Conclusions:

• The FAET model represents a significant advancement in the noninvasive assessment of advanced fibrosis in MASLD.
• FAET shows potential for routine clinical use in MASLD patients pending further validation.
• This ML approach offers a more accurate and accessible diagnostic alternative to invasive liver biopsy.