Predicting 28-day and 90-day mortality in microbiologically-confirmed sepsis: a retrospective cohort study

Sepsis patients with confirmed microbial infection represent a clinically important subgroup requiring specialized prognostic tools, though existing models fail to address their unique characteristics. A retrospective analysis was conducted using the MIMIC-IV repository, selecting adult sepsis cases fulfilling sepsis-3.0 criteria with microbiological confirmation. Primary endpoints were 28- and 90-day all-cause mortality. Analytical approaches included nonparametric comparative tests, stepwise survival regression modeling, and comprehensive sensitivity evaluations. Our analysis encompassed 5,694 patients with culture-confirmed sepsis, demonstrating mortality rates of 15.8% (28-day) and 17.5% (90-day), respectively. Multivariable regression identified significant mortality predictors (all p < 0.001), including age, respiratory rate, hemoglobin, lactate levels, vasoactive drug use, CCI, LODS score, SIRS score, SAPS III, and OASIS. These variables were incorporated into model 2 and model 4, both of which demonstrated strong predictive performance for 28-day and 90-day mortality with model 2 (AUC: 0.798; specificity: 0.765; sensitivity: 0.678) and model 4 (AUC: 0.795; specificity: 0.730; sensitivity: 0.708). Both models demonstrated consistent predictive performance in subgroup analyses. However, no significant association was observed between infection site, pathogenic microorganism type, and patient prognosis (p ≥ 0.05). These findings establish clinically applicable risk stratification tools specifically designed for culture-positive sepsis, potentially enabling more tailored therapeutic decisions for this vulnerable patient cohort. Not applicable.