Predicting Mortality Risk Among Non-Cardiac Surgical Patients in the Surgical Intensive Care Unit: A Retrospective Study Based on the MIMIC-IV Database

Background Accurate prediction of mortality risk in non-cardiac surgical patients is critically important for informing clinical decision-making and resource allocation. This study aims to develop a predictive model utilizing deep learning and machine learning to assess mortality risk in this patient population. Methods Clinical indicators and electrocardiogram (ECG) signals were extracted from non-cardiac surgery patients in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Five machine learning models were developed to estimate 30-day mortality risk: logistic regression (LR), decision tree (DT), light gradient boosting machine (LGBM), extreme gradient boosting (XGBoost), and support vector machine (SVM). To enhance analysis scope, two additional models, backpropagation neural network (BPNN) and recurrent neural network (RNN), were constructed and their performance compared to initial models. SHAP was employed to analyze the optimal model, identifying the most influential risk factors from both global and local perspectives. Results Among 4843 MIMIC-IV patients, 526 (10.8%) died within 30 days after non-cardiac surgery. The LGBM model surpassed other machine learning and deep learning models, attaining the highest scores for accuracy, precision, recall, F1-score, and area under the ROC curve (AUC), which were 0.949, 0.925, 0.983, 0.95, and 0.97, respectively. Compared to the traditional revised cardiac risk index (RCRI) score, the LGBM model significantly improved classification accuracy. SHAP analysis revealed that preoperative INR, bicarbonate, BUN, and creatinine levels were the four key variables influencing the performance of the LGBM model. Conclusion The LGBM model provides a new, convenient approach for the prognosis and assessment of non-cardiac surgical patients. This tool has the potential to offer effective decision support for clinicians in their risk assessment and clinical decision-making processes.