OBJECTIVES

To develop and validate a risk prediction model for alcoholic fatty liver disease (AFLD) based on clinical characteristics and liver stiffness measurements.

METHODS

This retrospective cohort study included 148 AFLD patients and 148 healthy controls from a tertiary hospital between January 2018 and December 2023. Participants underwent biochemical tests (lipid profile, liver function, uric acid) and liver stiffness measurements using Elastography Protocol for Quantification (Elast PQ). The external validation cohort, was from another hospital, with data collected from May 2019 to December 2023. It included 90 patients diagnosed with AFLD and 90 healthy controls. Machine learning methods (random forest, support vector machine, logistic regression) were employed to compare model performance. Logistic regression was used to identify predictive factors. Model performance was evaluated using Receiver Operating Characteristic (ROC) curve analysis, confusion matrices, calibration curves, and Decision Curve Analysis (DCA).

RESULTS

Univariate analysis revealed significant associations between body mass index (BMI), alcohol consumption, blood lipids, and liver function with AFLD (P < 0.001). Multivariate analysis identified high-aensity lipoprotein (HDL) (P = 0.041), alanine aminotransferase (ALT) (P = 0.007), and Elast PQ (P = 0.038) as independent risk factors. The logistic regression model showed an area under the curve (AUC) of 0.81 in the training set, 0.67 in the validation set, and 0.79 in the external validation cohort. The optimal cutoff value of 0.403 maximized sensitivity (0.62) and specificity (0.69), with an accuracy of 0.66. DCA indicated a high clinical net benefit. The risk prediction score enables rapid AFLD risk assessment and demonstrates strong predictive ability.

CONCLUSIONS

The AFLD risk prediction model, based on clinical features and liver stiffness assessment, exhibits strong predictive power and significant clinical value for early diagnosis and management.