Cardiac Arrest, Patient Characteristics and Prognosis: a Machine Learning Approach

Background Cardiac arrest is a severe medical emergency with poor prognosis. This study aimed to analyze the clinical characteristics of cardiac arrest patients and explore the key factors influencing their outcomes. Additionally, we applied machine learning methods to evaluate the performance of different models in predicting return of spontaneous circulation (ROSC), with the goal of optimizing strategies for managing cardiac arrest.Material and methods We comprehensively assessed the demographic characteristics, physiological parameters, and laboratory results of 748 cardiac arrest patients, and compared the differences between the ROSC and non-ROSC groups. We applied LASSO regression analysis to identify the key variables predictive of ROSC. Furthermore, we evaluated the performance of various machine learning models, including GBDT and LGBM, in ROSC prediction, including calibration, decision curve analysis, and ROC curves.Results Patients in the ROSC group were younger and predominately male. They had more normal blood pressure, temperature, and oxygen saturation, as well as less severe organ dysfunction, LASSO regression analysis identified age, WBC, and lactate as key predictors of ROSC. Among the machine learning models, GBDT and LGBM exhibited the best performance, with superior alibration, decision curve analysis, and ROC curves compared.Conclusions This study identified key clinical factors influencing the prognosis of cardiac arrest patients, and it identified machine learning models that were superior for predicting ROSC.