Prediction Of Early Postoperative Delirium Based On Machine Learning

Objective:Postoperative delirium(POD)is a temporary mental disorder after surgery,especially in patients who underwent heart surgery.It affects the recovery of surgical patients and often leads to dementia,prolongs time of staying at hospital,increases medical costs,and even causes death.Forknowing a patient’s risk factors of postoperative delirium will guide preventive interventions timely,which will help reduce the burden and negative consequences associated with postoperative delirium.In this study,several machine learning prediction models were developed to predict patients at high risk of delirium after cardiac surgery and their performance was compared.Methods:367 patients in Affiliated Hospital of Traditional Chinese Medicine of Southwest Medical University were retrospectively involved in the research.After single factor analysis,21 influencing factors with statistical significance on postoperative delirium were selected and included in machine learning.10 fold cross validation was used to randomly divide the data set in ratio of 9: 1 for model training and testing.Five models including Random forest(RF),Support vector machine(SVM),Radial basis kernel neural network(RBFNN),K-nearest Neighbor(KNN)and Kernel ridge regression(KRR)were created to predict postoperative delirium.Accuracy(ACC),sensitivity(SN),specificity(SPE),and Matthews coefficient(MCC)were used to evaluate the performance of the model,and the area under the receiver operating characteristic curve(AUC)was compared.Results:Among 367 patients,105 patients suffered postoperative delirium.The incidence of delirium was 28.6%.After comparing the performance of the five models,the random forest model obtained the best effect in ACC(87.99%),SN(69.27%),SPE(95.38%),MCC(70.00%)and AUC(0.9202),which was far superior to the other four models.Support vector machine,radial basis kernel neural network,kernel ridge regression,the three models are in the middle.The K nearest neighbor effect was the worst among the five models,and the proportions of ACC,SN,SPE,MCC and AUC were 75.20%,35.09%,91.21%,31.54% and 0.7330,respectively.Conclusion:1.Delirium is very common in patients underwent cardiac surgery.2.Random Forest,support vector machine,radial basis kernel neural network,kernel ridge regression and K nearest neighbor models are feasible for predicting POD after cardiac surgery.3.After model parameter optimization,the POD recognition effect of random forest,kernel ridge regression,K nearest neighbor,support vector machine and radial basis kernel neural network was improved compared with the default parameters.4.Random Forest shows the best machine learning efficiency,which is suitable for the data of this study.In clinical practice,random forest model can be adopted to identify POD after cardiac surgery.