Development of a survival model for predicting short-term mortality in patients with gut-derived sepsis: a MIMIC-IV study

Gut-derived sepsis, a challenging subtype associated with gastrointestinal barrier dysfunction and often lacking a clear infectious focus, remains relatively understudied with regard to prognostic risk factors. This study aimed to identify independent predictors and develop a survival prediction model for 28-day mortality in patients with this phenotype using the MIMIC-IV database. This retrospective cohort study used the MIMIC-IV database (2008–2019). Adult ICU patients meeting Sepsis-3 criteria with early gastrointestinal manifestations and no other identifiable infection source were included. The primary outcome was 28-day all-cause mortality. Missing data were handled using multiple imputation (10 datasets). After multicollinearity assessment and univariate Cox screening, candidate predictors were evaluated in multivariable Cox models, with model selection based on pooled Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Harrell’s C-index. Model performance was assessed using time-dependent AUROC, calibration, decision curve analysis, and selection-aware bootstrap internal validation. A nomogram was developed for individualized risk prediction. Among 226 patients, 52 (23.0%) died within 28 days. The final model included six predictors: age, mean arterial pressure, INR, APTT, ALP, and lactate. The model demonstrated good discrimination (AUROC 0.831, 95% CI 0.762–0.899; C-index 0.809), outperforming the SOFA score (AUROC 0.710, P = 0.009) and SOFA plus lactate (AUROC 0.704, P = 0.004). Calibration showed good agreement between predicted and observed outcomes, while decision curve analysis indicated favorable clinical utility. Selection-aware bootstrap internal validation yielded a mean optimism of 0.025 and an optimism-corrected AUROC of 0.806. Risk stratification based on the nomogram showed significantly lower survival in the high-risk group (log-rank P < 0.0001). A six-predictor prognostic model demonstrated good discrimination and potential clinical utility for predicting short-term mortality in gut-derived sepsis. This model may facilitate early risk stratification in ICU patients, although external validation is warranted. Not applicable.