Development and internal validation of a routine metabolic screening model for current hyperuricemia among gas station workers in Guangxi, China: a cross-sectional study

Abstract Background Serum uric acid (SUA) measurement is the direct basis for identifying current hyperuricemia (HUA). We developed a non-SUA screening model for situations in which routine occupational health information is already available but a contemporaneous SUA result is temporarily unavailable, delayed, or requires separate testing. Methods This cross-sectional study included 894 workers from 11 gas stations and related operational sites in Nanning, Guangxi, China. Current HUA was defined as same-examination SUA > 420 µmol/L in both sexes. Twelve prespecified non-SUA candidate variables were considered; total cholesterol was excluded because of high redundancy and multicollinearity with LDL-C, leaving 11 candidates for least absolute shrinkage and selection operator (LASSO) screening. Internal validation used 100 repetitions of site-grouped five-fold cross-validation. Model performance was assessed using out-of-fold (OOF) predictions, discrimination, calibration, decision curve analysis, candidate screening thresholds, and bootstrap optimism correction. Results Among 894 participants, 169 (18.9%) had current HUA. Sex, waist circumference, triglycerides (TG), systolic blood pressure (SBP), and fasting blood glucose (FBG) met the prespecified selection-stability threshold and comprised the final model. The mean area under the receiver operating characteristic curve (AUC) across repeated site-grouped cross-validation was 0.858 (SD 0.004), and the pooled OOF AUC was 0.866 (95% CI, 0.838–0.893). The pooled OOF Brier score was 0.108, the calibration intercept was − 0.064 (95% CI, -0.322 to 0.192), and the calibration slope was 0.920 (95% CI, 0.791–1.077). At a 10% candidate probability threshold, sensitivity was 92.9%, specificity was 64.6%, and the negative predictive value was 97.5%. Conclusions The five-variable model estimates the probability of current HUA and may support prioritization of confirmatory SUA testing when a contemporaneous SUA result is unavailable. It does not replace direct SUA measurement or predict future incident HUA. Independent external validation is required before broader implementation.