Developing a prediction model for in-hospital mortality in sepsis patients with gastrointestinal bleeding using the MIMIC-IV database

Background: Sepsis associated with gastrointestinal hemorrhage is a critical condition in ICU patients, significantly impacting mortality rates. This study aimed to develop a predictive model for in-hospital death risk in sepsis patients with gastrointestinal bleeding, improving treatment strategies and resource allocation. Methods: In a retrospective investigation of patients with sepsis and gastrointestinal bleeding, we gathered information from the MIMIC-IV database, including key demographics, comorbidities, laboratory indicators, and therapies. The dataset was split 70:30 for model development and validation. The Least Absolute Shrinkage and Selection Operator (LASSO) approach was used to select features, and multivariate logistic regression was then used to create a prognostic model. A nomogram was created to visualize predictive outcomes. Model performance was evaluated using calibration curve, receiver operating characteristic (ROC) curve, clinical impact curve (CIC), and decision curve analysis (DCA). Results: Nine significant predictors of in-hospital mortality were identified: APS III score, prothrombin time, body temperature, activated partial thromboplastin time, respiratory rate, vasopressor use, acute kidney injury, non-invasive ventilation, and malignancy. Area beneath the ROC curve for the training and testing groups The values are 0.8266 (95% CI: 0.8018-0.8515) and 0.7961 (95% CI: 0.7577-0.8345), respectively. Our model outperformed the APS III score in terms of ROC curve discrimination and demonstrated greater net benefit on the DCA curve. The CIC showed strong concordance between predicted and actual in-hospital death rates when the predicted probability exceeded 70%. Conclusion: We developed a robust predictive framework for assessing in-hospital death risk in sepsis patients with gastrointestinal hemorrhage. Early intervention based on identified risk factors could improve patient survival rates.