30-day mortality risk model for critically ill coronary artery bypass grafting patients based on ICU routine data

Existing risk scoring systems have limitations in predicting the mortality risk in critically ill patients receiving CABG. This study intends to construct and validate a simple yet effective model for forecasting 30-day mortality in patients receiving CABG using routine assessments from the intensive care unit (ICU). Patients who underwent CABG from MIMIC-IV were randomly split into training (60%) and validation (40%) sets. The Boruta algorithm was leveraged to determine the optimal predictive variables. The contribution of individual predictors to the model was appraised through the calculation of SHAP values. A nomogram was constructed utilizing multivariable logistic regression. The predictive performance of the model was appraised via the area under the receiver operating characteristic curve (AUROC), integrated discrimination improvement (IDI), and net reclassification improvement (NRI). The clinical net benefit was estimated through decision curve analysis (DCA). Among 2,767 patients, eight predictors were selected. The model significantly outperformed the Sequential Organ Failure Assessment (SOFA) and the Simplified Acute Physiology Score II (SAPS II) in both training (AUROC=0.882, 95%CI:0.833–0.931) and validation (AUROC=0.880, 95%CI:0.816–0.943) sets. Moreover, both the IDI and NRI were greater than zero. Furthermore, the DCA demonstrated greater clinical net benefit than comparators at threshold probabilities of 0.1–0.6. The prediction model incorporates eight readily available variables and demonstrates superior performance in forecasting 30-day mortality in patients following CABG compared to SOFA and SAPS II. This model serves as a practical tool for risk stratification and early intervention in critically ill patients.