Abstract
BACKGROUND: The integration of serum biomarkers and gene polymorphisms may enhance early prognostic assessment in sepsis. Early and accurate prediction of outcomes is crucial for optimizing treatment strategies and improving survival. However, the clinical utility of combining genetic markers with conventional inflammatory indicators remains insufficiently validated. METHODS: In this retrospective cohort (n = 81; July 2022-July 2024), patients were grouped by 28-day outcome. Candidate genes (TLR4, PPARγ, IL-12B, IL-27) were shortlisted from GSE54514. After data standardization and complete-case handling of < 5% missingness, ten-fold cross-validated LASSO selected predictors for multivariable logistic regression. Model performance was assessed by ROC AUC, Hosmer - Lemeshow (H - L) test, calibration plots, bootstrap optimism-correction (1,000 resamples), and decision-curve analysis (DCA). RESULTS: 6 predictors were retained - PCT, CRP, lactate (LAC), lactate clearance rate (LCR), TLR4 rs4986790, and PPARγ rs1801282. The nomogram achieved AUC 0.885 (95% CI 0.812-0.943) with sensitivity 88.6% and specificity 73.9%; calibration was good (H - L χ(2) = 9.191, p = 0.156). Bootstrap-corrected AUC was 0.872, with calibration slope 0.97 and Brier score 0.165. DCA indicated higher net benefit than SOFA or APACHE II across 0.05-0.40 threshold probabilities. In benchmarking, the integrative model outperformed SOFA and APACHE II (ΔAUC 0.129 and 0.104; DeLong p = 0.004 and 0.009) and showed higher net benefit on decision-curve analysis across 0.05-0.40 threshold probabilities. CONCLUSION: This integrative biomarker-genotype model demonstrated strong internal performance and potential clinical utility for individualized risk stratification in sepsis. The results support combining genetic susceptibility and inflammatory biomarkers for enhanced prognostic precision, although external and multi-ethnic validation remains warranted before widespread adoption.