Development and validation of a noninvasive liver fibrosis screening model incorporating GP73 in a health examination population

This study aimed to investigate the association between serum Golgi protein 73 (GP73) and liver fibrosis and developed a noninvasive screening model for a health-examination population. This cross-sectional study included 758 participants without previously diagnosed liver disease. Liver fibrosis was defined as liver stiffness measurement ≥ 7.3 kPa. Participants were randomly divided into training and testing cohorts at a ratio of 7:3. Least absolute shrinkage and selection operator (LASSO) regression and multivariable logistic regression were performed to identify independent predictors and construct a nomogram. Random forest (RF) and extreme gradient boosting (XGBoost) models were developed for comparison. Discrimination, calibration, threshold-based performance, and clinical utility were evaluated, with FIB-4 and APRI serving as conventional comparators. Age, AST, GP73, and BMI were independent predictors. The logistic regression model achieved area under the curve (AUC) values of 0.768 and 0.788 in the training and testing cohorts, respectively. Testing-cohort AUCs were 0.789 for RF and 0.773 for XGBoost, with no significant pairwise differences after Holm adjustment. The models demonstrated acceptable calibration and positive net clinical benefit. GP73 was higher in participants with liver fibrosis but correlated only weakly with other clinical indicators. A model incorporating age, AST, GP73, and BMI showed acceptable performance for liver fibrosis screening. GP73 may provide complementary value when combined with routinely available clinical indicators.