External validation in a Chinese cohort of a nomogram developed using MIMIC-IV for predicting sepsis-associated delirium in elderly ICU patients

Background : Sepsis-associated delirium (SAD) is a common acute brain dysfunction in elderly patients in the intensive care unit (ICU), which significantly increases the length of hospital stay, medical costs, and the risk of death. Despite the availability of multiple delirium prediction tools, there are few models specific to the elderly septic population, and most lack external validation. Methods : This retrospective cohort study enrolled 5034 elderly ICU patients with sepsis. A prediction model was constructed based on the MIMIC-IV database and externally validated using 281 patients admitted to the First Affiliated Hospital of Jilin University between January 2019 and November 2024. A workflow was developed using R software. Candidate predictors were first identified using the LASSO regression method, then incorporated into a multivariate logistic regression model and visualized as a nomogram. Patients were randomly divided into a training set and an internal validation set in a 6:4 ratio. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), the Hosmer-Lemeshow test, calibration plots, Brier scores, and decision curve analysis (DCA). Results : The overall incidence of delirium was 46.44%. Seven variables were ultimately included in the final model: body temperature, SOFA score, hemoglobin level, serum sodium concentration, history of neurological disease, mechanical ventilation, and midazolam use. The model demonstrated good discriminatory performance, with area under the receiver operating characteristic curve (AUC) values of 0.794 (95% CI: 0.777–0.810) in the training set, 0.784 (95% CI: 0.763–0.804) in the internal validation set, and 0.815 (95% CI: 0.765–0.864) in the external validation set. The Hosmer–Lemeshow goodness-of-fit test showed no significant deviation between predicted and observed outcomes (P > 0.05), indicating good calibration. The Brier scores were 0.185, 0.188, and 0.176 for the training, internal validation, and external validation sets, respectively. Decision curve analysis (DCA) further confirmed the model’s potential clinical utility. Conclusion: The SAD risk prediction model developed in this study features a simple structure and relies on readily available clinical variables. It demonstrated favorable discrimination and calibration in the external validation cohort, suggesting its potential utility as a practical tool for early detection and targeted intervention of delirium in elderly ICU patients with sepsis.