Cardiac arrest remains a significant cause of mortality worldwide. Identifying factors associated with in-hospital mortality can improve patient care and outcomes. A retrospective analysis was performed, including 1845 patients diagnosed with cardiac arrest from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and 52 patients from the Affiliated Hospital of Yunnan University (YNU). The study evaluated variables such as demographic characteristics, laboratory test results, and comorbidities. Prediction models for in-hospital mortality following cardiac arrest were constructed using five machine learning algorithms: logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), and light gradient boosting machine (LightGBM). The predictive performance of these models was evaluated using the area under the receiver operating characteristic curve (AUC), as well as metrics including accuracy, precision, recall, and the F1 score. Among the five machine learning algorithms evaluated, the XGBoost model demonstrated the highest performance. The XGBoost model achieved an AUC of 0.828, with an accuracy of 0.737, a precision of 0.709, an F1 score of 0.713, a recall of 0.717, a positive predictive value (PPV) of 0.709, and a negative predictive value (NPV) of 0.767. External validation of the XGBoost model using data from the Affiliated Hospital of Yunnan University yielded an AUC of 0.845, further supporting its predictive reliability. SHAP (SHapley Additive exPlanations) analysis identified the five variables with the greatest impact on in-hospital mortality in patients with cardiac arrest as mechanical ventilation, age, minimum lactate value, maximum arterial blood gas oxygen partial pressure, and urine output. The XGBoost model enabled good prediction of in-hospital mortality in ICU admission cardiac arrest patients, which may be widely used in clinical decision-making. Not applicable.
Jia et al. (Sat,) studied this question.