Integration of metabolic parameters with SOFA score for mortality risk assessment in elderly sepsis: a derivation study

BACKGROUND: Sepsis mortality remains elevated among elderly populations, yet existing risk stratification tools demonstrate suboptimal predictive accuracy. We explored whether incorporating metabolic parameters into the Sequential Organ Failure Assessment (SOFA) score could enhance prognostic discrimination in this vulnerable cohort. METHODS: This retrospective cohort study analyzed 4,056 elderly sepsis patients (age ≥ 65 years) admitted to intensive care units between 2008 and 2019, identified through ICD coding from the MIMIC-IV database. Random forest-based feature selection and SHAP value analysis identified lactate and bicarbonate as influential mortality predictors. We constructed three nested logistic regression models: Model 1 (SOFA alone), Model 2 (SOFA plus lactate), and Model 3 (SOFA plus lactate and bicarbonate). Multiple imputation addressed missing data. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), integrated discrimination improvement (IDI), calibration curves, and decision curve analysis. Restricted cubic spline regression characterized nonlinear mortality associations. RESULTS: Twenty-eight-day mortality occurred in 1,259 patients (31%). Model 3 achieved superior discrimination (AUC 0.785, 95% CI 0.770-0.801) compared to SOFA alone (AUC 0.738, 95% CI 0.722-0.754; difference 0.048, P < 0.001). Compared with Model 1, Model 3 demonstrated substantial reclassification improvement (NRI 0.816, 95% CI 0.733-0.893; IDI 0.077, 95% CI 0.072-0.082; both P < 0.001), with balanced gains in event (NRI + 0.401) and non-event (NRI - 0.415) classification. Sensitivity increased from 0.32 to 0.45 while maintaining comparable specificity (0.91 vs. 0.90). Restricted cubic spline analysis revealed nonlinear threshold effects, with mortality risk accelerating sharply beyond SOFA scores of approximately 8 points. Decision curve analysis confirmed clinical utility across relevant threshold probabilities. CONCLUSIONS: Integrating lactate and bicarbonate measurements with SOFA scoring demonstrated improved mortality discrimination in elderly sepsis patients. These hypothesis-generating findings from a single-center derivation cohort require prospective multicenter validation before clinical implementation can be considered. CLINICAL TRIAL NUMBER: Not applicable.