Journal of Clinical Hepatology
Metabolic-associated fatty liver disease, or MAFLD, is closely linked to obesity, abnormal blood lipids, high blood pressure and insulin resistance. Researchers investigated whether machine learning could combine routine health data and body-composition measurements to identify people at elevated risk.
The study included 2,168 adults attending a health-assessment centre, of whom 265 had MAFLD. The researchers collected 50 variables and divided the data into training and validation sets. They tested seven modelling approaches, including logistic regression, decision trees, random forests, gradient boosting, support vector machines and neural networks.
Ten factors were independently associated with MAFLD, including smoking, diastolic blood pressure, visceral fat area, waist-to-hip ratio, muscle-to-fat ratio, triglyceride-glucose index and gallstones.
The authors selected a random forest model for detailed interpretation. In the validation data, it achieved an area under the receiver operating characteristic curve of 0.796, with 81.01% sensitivity and 63.16% specificity. Its three most influential predictors were visceral fat area, waist-to-hip ratio and diastolic blood pressure.
The results show that where fat is stored—particularly around the abdominal organs—may be more informative than body weight alone when estimating fatty liver risk.
MAFLD can remain unnoticed until substantial liver injury has occurred. A model based on measurements already collected during health examinations could help identify people who may benefit from liver imaging, metabolic assessment or preventive support.
The prominence of visceral fat and waist-to-hip ratio reinforces the importance of fat distribution rather than relying only on body mass index. The model’s interpretation method also makes its predictions easier for clinicians to understand.
However, the model was developed and internally validated using data from one centre. Diet and physical activity were not included, and the participants may not represent other populations. External multicentre validation is needed before clinical use.
Objective: To investigate the value of predictive models established based on machine learning methods in predicting the risk of meta…
