Abstract
BACKGROUND: Patients with acute gastrointestinal bleeding (AGIB) carries a significant risk of sepsis, particularly in intensive care units. We aimed to develop a validated predictive model for sepsis risk stratification to guide clinical management. METHODS: This multicenter study (Jan, 2020-Jul, 2024) in China included 1449 patients with AGIB. Participants were enrolled into a retrospective training (n = 878) cohort, and prospective internal validation (n = 187; prospectively enrolled from lead center) and external validation (n = 384; from three independent tertiary hospitals) cohorts. We excluded patients with hospitalization <24 h or pre-existing infection or sepsis. A multivariable logistic regression-based nomogram to predict risk of sepsis (defined per Sepsis-3 criteria) was developed and evaluated using area under the receiver operating characteristic curve (AUC), calibration curves, decision curve analysis (DCA), and SHapley Additive exPlanations (SHAP). Predictive performance was compared against established scoring systems-Glasgow Blatchford Score (GBS), Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ), and Sequential Organ Failure Assessment (SOFA)-using discrimination metrics to differentiate high- and low-risk patients. A real-time clinical warning system was implemented. FINDINGS: Among 1449 patients (68.7% male; median age 65 [IQR 54-73.5]), 223 (15.4%) developed sepsis, linked to higher mortality (23.7% vs. 6.8%, p < 0.001). Key predictors included chronic kidney disease (CKD), elevated respiratory rate (RR), neutrophil-to-lymphocyte ratio (NLR), C-reactive protein (CRP), creatinine (Cr), activated partial thromboplastin time (APTT), and fibrinogen (FIB). This model outperformed some traditional scores, with AUCs of 0.827 (95% CI: 0.759-0.888), 0.836 (95% CI: 0.776-0.896) and 0.884 (95% CI: 0.816-0.952) in training, internal, and external validation sets, respectively. Calibration curves and DCA confirmed accuracy (showed excellent agreement between actual sepsis occurrences and nomogram-estimated probabilities, with mean square error [MSE] of 0.00094, 0.00791, and 0.00045 for the training, internal validation, and external validation sets, respectively) and clinical utility. SHAP analysis highlighted key predictors. An online platform enabled real-time risk monitoring. INTERPRETATION: This validated model effectively identifies patients with AGIB at high risk for sepsis, addressing a critical unmet need in emergency care. Its integration with dynamic warning systems will optimize risk stratification, facilitate preemptive management, and improve clinical outcomes. FUNDING: The National Key R&D Program of China, the National Natural Science Foundation Project of China, the Knowledge Innovation Program of Wuhan, and the Cross Innovation Talent Project of Renmin Hospital of Wuhan University.