Construction And Validation Of Fetal Outcome Prediction Model Based On Artificial Intelligence Algorithm For Twin-to-twin Transfusion Syndrome Treated By Fetoscopic Laser Surgery

Objective:Based on the Lasso regression to screen the factors influencing fetal outcomes in cases of Twin-to-twin transfusion syndrome(TTTS)treated with fetoscopic laser surgery(FLS),and to establish artificial intelligence algorithm prediction models for fetal outcomes.The results were compared with the results of conventional analysis methods to seek the best prediction method for the prognosis of fetoscopic surgery.Methods:Retrospective analysis of 159 TTTS cases treated by FLS from January 2010 to December 2021 at Shengjing Hospital of China Medical University,which were divided into a training set(127 cases)and a test set(32 cases)according to the ratio(8:2).The Lasso algorithm was used to screen the factors affecting fetal outcome in TTTS cases treated with FLS,and the K-nearest neighbor(KNN),random forest(RF)and extreme gradient boosting tree(XGBOOST)algorithms were applied to construct the fetal prognosis prediction models after FLS surgery.The traditional prediction model was also developed based on logistic regression analysis.Finally,the predicted values of each model were evaluated and compared by accuracy,sensitivity,specificity,Cohen’s kappa coefficient,receiver operating characteristic(ROC)area under the curve(AUC)and Net Reclassification Index(NRI).Results:(1)In the traditional parameter analysis,single factor and logistic multi factor analysis screened out that:the time from onset to operation,preoperative cervical status,preoperative uterine contraction,Quintero stage,Recipient’s MOM value of middle cerebral artery,Donor’s MOM value of middle cerebral artery,and amniotic fluid loss rate(ml/min)were the key factors affecting the postoperative fetal outcomes(P<0.05).The traditional logistic regression prediction model was established according to the above factors.The test set verified that the accuracy of the model was 68.75%,the AUC was 0.789,the 95%CI was(0.634,0.945),and the sensitivity and specificity were 69.23%and 68.42%,respectively.(2)According to Lasso algorithm,in addition to the above 7 influencing factors,placental location and gestational age of operation are also closely related to fetal outcome.Among the artificial intelligence algorithms,the accuracy,sensitivity,specificity,AUC and Cohen’s kappa coefficients of the KNN algorithm model based on Lasso algorithm were 87.50%,63.64%,100%,0.955(95%CI:0.8378～0.9992),0.9292,respectively.Compared with other prediction models,the comprehensive prediction value of the model established by the KNN algorithm has the best performance.At the same time,compared with the traditional logistic regression prediction model,the NRI under the KNN algorithm is 0.285(Z=2.449,P<0.001),which has a statistically significant difference,indicating that the prediction ability of the KNN algorithm is better than the traditional logistic regression model,and the proportion of correct classification is improved by 28.5%.Conclusion:The fetal outcomes in FLS-treated TTTS cases is associated with multiple factors,The KNN algorithm can be used to establish a reliable prediction model of fetal outcome after fetal endoscopy,which can provide a new method for scientifically predicting fetal outcome after FLS.