Abstract / Summary
Abstract Background Severe fever with thrombocytopenia syndrome (SFTS) is a life-threatening viral disease in which fatal progression is associated with interactions between viral replication and dysregulated host immune responses. Whether markers representing these complementary biological processes can be integrated into an interpretable framework for mortality risk stratification remains unclear. We therefore developed and externally validated biologically informed prediction models incorporating viral, immune, inflammatory, and tissue-injury markers. Methods This multicenter retrospective cohort included 530 patients with laboratory-confirmed SFTS, comprising 372 patients in the training cohort and 158 patients from an independent center for external validation. Five biologically relevant predictors were selected: DBV RNA, IL-10, Hs-CRP, CD4/CD8 ratio, and CK. Logistic regression and XGBoost models were developed using the same predictor set and evaluated for discrimination, calibration, prediction error, clinical utility, and interpretability. Results Non-survivors showed higher DBV RNA, IL-10, Hs-CRP, and CK levels and lower circulating T-cell counts and CD4/CD8 ratios, indicating concurrent virological, inflammatory, and cellular immune abnormalities. In external validation, logistic regression and XGBoost achieved AUCs of 0.815 (95% CI, 0.730–0.900) and 0.848 (95% CI, 0.765–0.931), respectively. Logistic regression showed better calibration, with a lower Brier score than XGBoost (0.116 vs. 0.154), while both models provided clinical net benefit. Conclusion Integrating markers of viral burden, immune dysregulation, systemic inflammation, and tissue injury enabled reproducible mortality risk stratification across independent SFTS cohorts. These findings support a biologically informed framework for clinically interpretable risk assessment, while prospective studies with longitudinal immune profiling are needed to further validate this approach.