Study On The Risk Prediction Model Of Lung Complications After Computer Assisted Cardiac Surgery

Objective:Artificial Intelligence(AI)is a frontier subject in the field of computer science,and Machine Learning(ML)is its core field.Machine learning techniques are increasingly being used in clinical medicine and are increasingly associated with anesthesiology to support the development of better and more effective predictive models,It is helpful and necessary for clinicians to accurately predict the occurrence of Pulmonary Complications after cardiac surgery.Postoperative Pulmonary Complications(PPCs)can be identified in advance by using various patient indicators.And instruct clinicians to treat patients in the perioperative period to reduce or prevent the occurrence of PPCs.The purpose of this study was to establish a risk prediction model for PPCs based on machine learning algorithm and verify its effectiveness,screen and analyze risk factors for PPCs,and evaluate the risk of PPCs in patients undergoing cardiac surgery.Methods:We analyzed data from an observational cohort study that prospectively collected information on 490 patients undergoing cardiac surgery using the Do-Care anesthesia system and the EMRS electronic case system,based on inclusion and exclusion criteria.Among them,there were 42 preoperative indicators,25 intraoperative indicators,and 18 postoperative indicators,a total of 85 indicators.The collected patient information was taken as a complete feature set.The collected data set was randomly divided into the training data set(75%)and the test data set(25%),and the proportion of positive and negative samples in the training set and the test set was kept equal for model development.Based on intelligent machine learning algorithm,six kinds of machine learning models including logistic regression(Log R),support vector machine(SVM),random forest(RF),deep neural network(DNN),gradient lifting decision tree(GBDT)and extreme Gradient lifting tree(XGBoost)are established to learn and verify the data of feature set.The performance of different models was compared based on accuracy rate,accuracy rate,recall rate,F1-score and area under the curve(AUC).Results:We compare the performance of six machine learning models: The AUC of Log R(0.62,95%CI0.63,0.71),SVM(0.62,95%CI0.60,0.64),RF’s AUC(0.65,95%CI0.62,0.68),DNN’s AUC(0.85,95%CI0.83,0.87),GBDT’s AUC(0.67,95%CI0.63,0.71),XGBoost’s AUC(0.56,95%CI0.52,0.60).The area under the curve(AUC)of deep neural network(DNN)is higher,which indicates that the deep neural network(DNN)has better efficiency and can better fit the experimental data.Conclusion:The risk prediction model based on machine learning algorithm can be used to predict PPCs in cardiac surgery.Among the six machine learning models established in this study,the AUC of deep neural network(DNN)(0.85,95%CI0.83,0.87)showed better model performance compared with the other five models,which can identify the high-risk population of PPCs and has a certain predictive effect.