Incorporating Insulin Resistance Biomarkers into Machine Learning Models Enhances Diagnostic Accuracy for Metabolic...

Area of Science:

• Hepatology
• Medical Informatics
• Biomarkers

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) is the leading cause of chronic liver disease globally.
• Early MASLD is often asymptomatic, leading to underdiagnosis.
• Insulin resistance (IR) is a key factor in MASLD development, but its role in diagnostic models needs further investigation.

Purpose of the Study:

• To investigate the incremental value of IR-related biomarkers in machine learning (ML) diagnostic models for MASLD.
• To assess the impact of incorporating IR biomarkers on the accuracy and clinical utility of MASLD detection.

Main Methods:

• A retrospective study of 18,535 adults, with and without MASLD.
• Development of ML models using demographic, clinical, biochemical, and IR-related biomarkers (e.g., TyG indices, METS-IR).
• Evaluation of model performance using AUC, sensitivity, specificity, F1-score, and decision curve analysis (DCA).

Main Results:

• All ML models showed significant improvement in AUC after including IR biomarkers (P < 0.05).
• The extremely randomized trees (ERT) model with IR biomarkers achieved the highest AUC (0.901) in the validation set.
• DCA indicated improved net clinical benefit with IR biomarker incorporation; ERT model showed optimal discrimination and utility.

Conclusions:

• Incorporating IR-related biomarkers into ML models enhances MASLD diagnostic accuracy and clinical utility.
• These biomarkers offer incremental diagnostic value for MASLD screening in the general population.
• ML models with IR biomarkers can serve as effective decision-support tools for MASLD detection without increasing costs or patient burden.