| Objectives:With the implementation of the“two-child policy”in China,the number of women undergoing repeat cesarean delivery had increased year by year.Compared with first cesarean delivery,the probability of intraoperative hemorrhage and transfusion for repeat cesarean delivery was significantly increased due to more complications,such as placenta previa,placental adhesion,uterine scar,etc.Therefore,we conducted a retrospective analysis with a large sample of clinical data to identify the effect of maternity factors and anesthesia methods on the risk of intraoperative hemorrhage and construct a prediction model of intraoperative transfusion for repeat cesarean delivery,then to compare the prediction performance of logistic regression,extreme gradient boosting(XGB)and artificial neural network(ANN).Methods:1.By searching the electronic medical record system,the clinical medical records of2442 women undergoing repeat cesarean delivery in our hospital from October 2015 to October 2017 were collected.Inclusion criteria:20-45 years old,BMI 18-40 kg/m~2,ASA?-?.Intraspinal anesthesia was routinely used in cesarean section,general anaesthesia was virtually exclusively used in emergency situations,when intraspinal anaesthesia techniques had failed or were contraindicated.According to the criteria of intraoperative hemorrhage,patients were divided into marked hemorrhage group(group MH,n=494)and non-marked hemorrhage group(group NMH,n=1948).The preoperative,intraoperative and postoperative data of the two groups were compared.Then the occurrence of marked hemorrhage was taken as the dependent variable,and the factors with P<0.05 in univariate analysis were included in Logistic regression analysis as independent variables to analyze the risk factors of intraoperative hemorrhage.2.To clarify the effect of anesthesia methods on parturients and fetuses,propensity score matching(PSM)was used to eliminate confounders,then the incidence of marked hemorrhage,neonatal asphyxia and hospital stay were compared between the general anesthesia group(group GA,n=141)and the non-general anesthesia group(group NGA,n=141).3.In order to further predict intraoperative transfusion,standard cases and variables with possible clinical significance were brought into the establishment of prediction model.Logistic regression,XGB and ANN were used to build the prediction model of intraoperative transfusion.The area under receiver operating characteristic curve(AUROC),precision,recall(sensitivity)and F1 score of the three models were calculated,analyzed and compared.Results:1.The logistic regression results showed that the risk factors of intraoperative hemorrhage for repeat cesarean delivery were placenta previa[odds ratio(OR)=38.269,95%confidence interval(95%CI)15.970-91.706,P<0.001],uterine atony(OR=10.047,95%CI6.155-16.399,P<0.001),placental adhesion(OR=5.045,95%CI 3.146-8.089,P<0.001),prenatal anemia(OR=4.082,95%CI 2.485~6.707,P<0.001),general anesthesia(OR=2.922,95%CI 1.521-5.614,reference group:non-general anesthesia,P<0.001),ASA?~?(OR=2.699,95%CI 1.539~4.735,reference group:ASA?~?,P<0.001),history of abortion(OR=2.420,95%CI 1.034~5.661,P=0.042)and abdominal/pelvic adhesions(OR=2.124,95%CI 1.478~3.052,P<0.001).2.7 influencing factors including abortion history,prenatal anemia,ASA classification,placenta previa,placental adhesion,abdominal/pelvic adhesions and uterine atony were brought into PSM,and 141 pairs of women with general anesthesia and non-general anesthesia were matched.The PSM showed that patients in group GA had higher morbidity of marked intraoperative hemorrhage(46.1%vs 31.9%,P=0.015),higher incidence of neonatal asphyxia at 1 minutes after delivery(9.9%vs 2.1%,P=0.012)and longer hospital stay(4.1±3.2d vs 3.3±2.9d,P=0.039)compared to group NGA.3.A total of 2525 patients were enrolled in the prediction model,of whom 332(13.1%)received intraoperative transfusion.The final model had five important predictors,including predelivery hemoglobin,length of operation,uterine atony,placenta previa and ASA classification.To compare the prediction effect,further predictive verification was carried out on training samples and test samples and we found the AUROC of XGB were0.904 and 0.886,higher than 0.868 and 0.878 of logistic regression,0.882 and 0.884 of ANN.The precision,recall and F1 of XGB were also better than logistic regression and ANN.Conclusions:1.Risk factors of intraoperative hemorrhage for repeat cesarean delivery included placenta previa,uterine atony,placental adhesion,prenatal anemia,general anesthesia,ASA?~?,history of abortion and abdominal/pelvic adhesions.2.Compared with non-general anesthesia,general anesthesia could increase the risk of intraoperative hemorrhage,the morbidity of neonatal asphyxia and the hospital stay.3.The predictors of intraoperative transfusion for repeat cesarean delivery were predelivery hemoglobin,estimated length of operation,uterine atony,placenta previa and ASA classification.By predicting intraoperative transfusion with logistic regression,XGB and ANN,we found XGB may be better than logistic regression and ANN. |
PDF Full Text Request