A Machine Learning Approach to Predicting Nonalcoholic Fatty Liver Disease Using Non-Invasive Health Indicators

Nonalcoholic fatty liver disease (NAFLD) is a widespread liver condition associated with obesity, type 2 diabetes, and metabolic syndrome, impacting over a quarter of the global population. This study aims to develop a machine learning model for predicting NAFLD, using only demographic, vital sign, and laboratory data from the National Health and Nutrition Examination Survey (NHANES) 2017–2020 cycle to enable accessible, non-invasive screening options.

Adults aged 18 years and older with relevant health data were included, excluding individuals with high alcohol consumption or certain liver-related conditions. NAFLD was defined using a Controlled Attenuation Parameter (CAP) threshold of ≥302 dB/m based on non-invasive measurements. An extensive data-driven approach was applied to enhance model reliability and address data challenges, focusing on understanding the significance of various health indicators while ensuring model interpretability and accuracy.

The model effectively identified NAFLD prevalence within a diverse U.S. cohort, highlighting key health indicators associated with increased disease risk. The results demonstrate strong potential for accurately identifying NAFLD using commonly available health data, supporting the model’s applicability in large-scale, real-world settings.

This study emphasizes the value of a non-image-based, data-driven approach to developing scalable screening tools for NAFLD. By utilizing routinely collected clinical and demographic information, the model aids in early detection and risk stratification for NAFLD, addressing diagnostic gaps and advancing targeted public health interventions, particularly in underserved areas.