Novel Machine-Learning Models Outperform Noninvasive Tests in Metabolic Dysfunction-Associated Steatohepatitis With...

Area of Science:

• Hepatology and Gastroenterology
• Biomarker Discovery
• Medical Imaging and Diagnostics

Background:

• Noninvasive tests (NITs) lack systematic evaluation for detecting metabolic dysfunction-associated steatohepatitis (MASH) with fibrosis stages F2-F3, a key group for emerging treatments.
• Current diagnostic gaps hinder patient identification for MASH therapies and clinical trials.
• Cirrhosis (F4) detection also requires optimized noninvasive methods.

Purpose of the Study:

• To assess existing and novel noninvasive tests (NITs) for detecting MASH with fibrosis stages F2-F3.
• To evaluate the performance of NITs in excluding cirrhosis (F4).
• To develop and validate machine learning (ML) models for improved MASH fibrosis detection.

Main Methods:

• Analyzed a cohort of 905 biopsy-proven individuals, including healthy controls.
• Evaluated 28 biomarker-, imaging-, or algorithm-based NITs for MASH fibrosis stages F2-F4.
• Developed and validated machine learning models using routine variables and hormonal markers via categorical gradient boosting.

Main Results:

• Existing NITs showed suboptimal performance for F2-F3 detection (FAST AUC 0.67).
• Novel ML models achieved AUCs up to 0.86 for F2-F3 detection with improved accuracy (87%) and positive predictive value (65%).
• For cirrhosis (F4), ML models reached a mean AUC of 0.93, outperforming conventional tools like AGILE4+.

Conclusions:

• Novel ML models significantly enhance the detection of MASH F2-F3 and cirrhosis compared to existing NITs.
• These ML tools address a critical diagnostic gap for FDA-defined MASH treatment windows.
• The developed models show promise for patient selection in MASH clinical trials and therapy enrollment.