Machine learning-driven predictive model for 28-day mortality risk in patients with diabetic retinopathy complicated with sepsis: dual-database validation in MIMIC-IV and eICU-CRD

Clinical observations suggest that patients with diabetic retinopathy (DR) may have worse prognosis after sepsis, and it is hypothesized that systemic microvascular dysfunction underlying DR may accelerate infection progression and organ failure. However, there is currently a lack of prognostic prediction tools specifically for this comorbid population. Retrospective data of DR patients meeting Sepsis-3 criteria (2008–2022) were extracted from MIMIC-IV. The least absolute shrinkage and selection operator (LASSO) regression and Boruta algorithm were combined for key variable screening. MIMIC-IV data were split into training and internal validation sets by 7:3 stratified random sampling, and eICU data were used as external validation set. Seven machine learning algorithms were applied to construct models using data collected within the first 24 h of ICU admission, with the prediction landmark set at 24 h after ICU admission. Models were compared via nested cross-validation within the training set to select the optimal algorithm, and performance was evaluated by area under the curve (AUC), calibration curves, and decision curve analysis (DCA) in internal and external validation cohorts. The optimal model was interpreted using SHapley Additive exPlanations (SHAP). A total of 525 patients from MIMIC-IV (28-day mortality 12.95%) and 86 from eICU (28-day mortality 33.72%) were included. The final model contained 4 predictors. The AUC values of the training set, MIMIC-IV internal validation set and eICU external validation set were 0.870, 0.805 and 0.833 respectively. Both validation sets showed favorable calibration and positive clinical net benefits within the threshold probability range of 0.0–1.0. The model based on MIMIC-IV can accurately predict 28-day mortality risk after the 24-hour landmark in patients with DR complicated with sepsis using clinical data within the first 24 h of ICU admission. External validation confirms its good generalization ability, providing a reliable quantitative reference for ophthalmology-intensive care unit (ICU) collaborative intervention. The web-based application could be available at https://chentaiyu-mortality-prediction.hf.space/ . Not applicable.