Predictive Study Of Apnea During Recovery From Postoperative Anesthesia

Objective: Patients who spend the recovery period in postanesthesia care unit(PACU)are prone to various respiratory complications,in which apnea events that can’t be corrected in time will lead to serious dangerous symptoms such as severe bradycardia,cardiac arrest,and tissue hypoxia.If the possibility of apnea in the PACU can be predicted before the patient enters the PACU after surgery,this will strengthen the monitoring of such patients in the PACU and help correct the apnea event in time.This study proposed to use clinical electronic medical record data,based on Logistic regression method to develop a clinical predictive model(CPM)to predict whether apnea events occurred in patients recovering from postoperative anesthesia.Methods: Using patient’s recovering from postoperative anesthesia continuous breathing sound signal collected by the autonomous microphone and the nasal airflow pressure signal collected by the nasal catheter for apnea recognition and manual verification.The clinical electronic medical record data of relevant patients were collected to extract candidate variables that might be associated with the occurrence of apnea.The variables statistically significant after univariate analysis were put into stepwise regression for multi-factor adjustment and the binary Logistic regression prediction model was established,and compared with the model established by the prediction variables screened by Lasso method.Leave-One-Out Cross Validation(LOOCV)was used to verify the model.Comparing the prediction model established by Logistic regression with the model established using machine learning methods such as SVM and RF.Model performance evaluation mainly includes three aspects: discrimination,calibration and clinical decision making ability.Results: The study included data of 158 patients,of which the incidence of apnea during postoperative anesthesia recovery was 33.5%(53 cases).After single-factor regression analysis(P<0.2)of 9 candidate variables,it was found that age,operation methods,ASA,BMI,analgesic methods,anesthesia methods,and gender were statistically significant.Using the above 7 variables to do stepwise regression and modeling,determining the optimal model according to the AIC criteria.The finally established binary Logistic regression model includes age,BMI,surgical methods,analgesic methods,ASA and gender.The validated ROC value of the Logistic regression model is equal to 0.7297(95%CI [0.646,0.81]),which was better than that of variable modeling screened by Lasso method.And the ROC value of the model established using SVM and RF was respectively equal to 0.7048(95%CI [0.621,0.789])and 0.6909(95%CI [0.605,0.777]).Conclusion: Using the clinical electronic medical record data combined with the clinical prediction model established by the binary Logistic regression method,it can predict whether the patients in the postoperative anesthesia recovery period will have an apnea event.The predictive model will assist medical staff in predicting the possibility of apnea during the recovery period with patients who have not yet entered PACU,thereby strengthening the monitoring of such patients.If apnea occurs,it can be corrected in time to ensure that patients can pass the postoperative anesthesia recovery period steadily and safely.With the aid of the predictive model,the limited energy of medical staff can maximize the value,improve the work efficiency of medical staff,and ensure high-quality medical results.