Establishment And Verification Of Anesthetic Drug Pharmacodynamic Model Based On Machine Learning

Objective: In recent years,the combined application of anesthetic drugs in clinical practice has become more and more common.In order to provide more effective guidance for clinical administration,the research on the interaction of anesthetics and its pharmacodynamics has become more and more in-depth.At present,response surface models are mostly used to analyze the pharmacodynamic interaction of anesthetic drugs,but different response surface models have different effects for different drugs and drug effects,which may cause a lot of inconvenience in the process of clinical practical application.This study will use machine learning to establish a model for the combined efficacy of anesthetic drugs,which will provide a certain theoretical basis for future research on machine learning in anesthesia pharmacodynamics,and it will also provide more effective and more reasonable guidance for the anesthesiologist’s clinical administration.Method: In this paper,the four drug effects(LREI,IVD,LREIR and RC)involved in the process of propofol and remifentanil in the esophagus implantation process obtained from the previous study are respectively corresponding to the optimal response surface model using the out-of-bag verification of bagging method and leave-one-out crossvalidation were verified,and the results were compared with the bagging method(based on BP neural network,support vector machine,decision tree,K nearest neighbor,naive bayes,Ada Boost)and leave-one-out cross-validation(including BP neural network,support vector machine,random forest,K nearest neighbor,naive bayes,Ada Boost and super learner)used in this article.Results: The result of out-of-bag validation of the bagging method: for the efficacy of LREI,the AUC of verification set of the reduce Greco model and the SVM-based bagging method were 0.8797 and 0.868,respectively,and the accuracy were 0.8185 and0.8092,respectively.The two methods did not show any significant difference in their ability to predict the probability of occurrence of LREI;for IVD efficacy,the AUC of verification set of the Minto model and the SVM-based bagging method were 0.8777 and0.8763,respectively,and the accuracy were 0.796 and 0.7925,respectively.There was no significant difference in the ability to predict the probability of IVD;for the efficacy of LREIR,the AUC of the validation set of the Minto model and the SVM-based bagging method were 0.8646 and 0.8659,and the accuracy were 0.7809 and 0.7911,respectively,The two methods did not show any significant difference in their ability to predict the probability of occurrence of LREIR;for RC efficacy,the AUC of the validation set of the Minto model and the SVM-based bagging method were 0.8731 and 0.8697,respectively,and the accuracy were 0.7898 and 0.7705 respectively,and they did not show any significant difference in their ability to predict the probability of occurrence of RC.The result of leave-one-out cross-validation was similar to the result of out-of-bag validation of the bagging method.Conclusion: There is no significant difference compare the bagging ensemble method based on the support vector machine and the support vector machine leave-one method for the four pharmacological effects with the previous optimal response surface model in AUC and accuracy.For the five response surface models and the various machine learning models proposed in the article,from the perspective of model generalization performance,prediction accuracy and prediction error,various drug effects established by SVM(LREI,IVD,LREI and RC)all have relatively stable and excellent performance and are universal,that is,for various drug effect modeling,the SVM method can be given priority for modeling.