Predicting Metabolic Dysfunction-Associated Fatty Liver Disease Phenotypes Among Adults: 2-Stage Contrastive Learning...

Area of Science:

• Hepatology and data science
• Machine learning applications in medicine
• Personalized medicine and risk prediction

Background:

• Metabolic dysfunction-associated fatty liver disease (MAFLD) is a significant cause of chronic liver disease, potentially progressing to fibrosis or cancer.
• Current analytical methods often fail to capture the multisystemic nature, intraphenotype variability, and dynamic progression of MAFLD.
• These limitations impede accurate risk stratification and personalized intervention strategies for MAFLD patients.

Purpose of the Study:

• To develop and validate a novel, multi-stage contrastive learning-based method for predicting MAFLD phenotypes in adults.
• To leverage multiview contrastive learning to model individual heterogeneities and relationships within clinical and survey data.
• To enhance clinical decision-making and facilitate personalized care for MAFLD.

Main Methods:

• Utilized demographic, clinical, lifestyle, and genetic family history data from 4408 adults.
• Employed a two-stage contrastive learning approach to model complementary data views and individual patient representations.
• Evaluated predictive performance against eight benchmark methods using recall, precision, F1-score, and AUC, with Shapley additive explanations for interpretability.

Main Results:

• The proposed contrastive learning method significantly outperformed all benchmark methods in MAFLD phenotype prediction.
• Achieved substantial improvements in F1-score: 32.8% for nondiabetic MAFLD and 30.4% for diabetic MAFLD compared to the best benchmarks.
• Demonstrated the clinical utility of integrating diverse data sources for predicting MAFLD phenotypes.

Conclusions:

• The novel contrastive learning method offers a viable and effective approach for MAFLD phenotype prediction.
• This method enhances the identification of at-risk adults compared to existing data-driven analytics.
• The findings support improved clinical decision-making, patient-centric care, and management of MAFLD.