Why the study?
Accurate mortality risk prediction for elderly patients undergoing non-cardiac surgery is critical to help surgeons make better perioperative clinical decisions.
Do machine learning models improve 30-day mortality risk prediction compared to the RCRI score in geriatric patients undergoing non-cardiac surgery in the ICU?
Do machine learning models improve 30-day mortality risk prediction compared to the RCRI score in geriatric patients undergoing non-cardiac surgery in the ICU?
A CatBoost machine learning model provides highly accurate 30-day mortality risk prediction for critically ill elderly patients undergoing non-cardiac surgery, outperforming the conventional RCRI score.
Answers are grounded in this paper and cite their sources.
May aid perioperative risk stratification in elderly non-cardiac surgery patients; leaves open need for external validation before clinical use.
BACKGROUND: The prediction of mortality for elderly patients undergoing non-cardiac surgeries is a vital research area, as accurate risk assessment can help surgeons make better clinical decisions during the perioperative period. This study aims to build a mortality risk prediction model for surgical intensive care unit (ICU) patients aged 65 and older undergoing non-cardiac surgery. METHODS: Data was obtained from 1960 patients who underwent non-cardiac surgery from the medical information mart for intensive care IV (MIMIC-IV) database. The least absolute shrinkage selection operator (LASSO) regularization algorithm and the extreme gradient boosting (XGBoost) for feature importance evaluation were used to screen important predictors. Five predictive models were established: categorical boosting (CatBoost), logistic regression (LR), decision tree (DT), random forest (RF), and support vector machine (SVM). External validation was performed utilizing data from 153 patients in the MIMIC-III database. Finally, shapley additive explanations (SHAP) was utilized for a personalized analysis of the models. RESULTS: Among the five predictive models developed in this study, the CatBoost model demonstrated superior overall performance in both the test data set (AUC = 0.96, F1 = 0.90) and the external validation data set (AUC = 0.98, F1 = 0.91). The decision curve analysis showed that the model offers a beneficial net benefit. The CatBoost model showed significant enhancements in classification accuracy when compared to the conventional revised cardiac risk index (RCRI) score. SHAP analysis revealed that anion gap, age, prothrombin time (PT), and weight were the four key variables influencing the predictive performance of the CatBoost model. CONCLUSIONS: This study demonstrates the potential of machine learning methods for early prediction of outcomes in critically ill elderly patients undergoing non-cardiac surgery. A web-based application was developed, which could serve as an effective tool for clinicians in their risk assessment and clinical decision-making processes.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: