Conformal Risk Prediction for Non-Alcoholic Fatty Liver Disease Using Gradient Boosting with Distribution-Free Coverages
Researchers have developed a machine-learning framework called Method for predicting the risk of non-alcoholic fatty liver disease (NAFLD). This framework utilizes gradient-boosted decision trees combined with conformal prediction to provide calibrated risk estimates with guaranteed coverage levels. Method demonstrated strong performance, achieving an AUROC of 0.912 internally and 0.891 externally, outperforming other established models. The selected features, such as waist circumference and BMI, align with known metabolic risk factors, and the risk stratification effectively separates individuals by their progression rates. AI
IMPACT Provides a novel, calibrated ML approach for disease risk prediction, potentially improving clinical decision-making and patient stratification.