Prediction and Evaluation of Machine Learning Algorithm for Prediction of Blood Transfusion during Cesarean Section and Analysis of Risk Factors of Hypothermia during Anesthesia Recovery
Wei Ren,Danmei Li,Jia Wang,Jinxi Zhang,Zhongliang Fu,Yu Yao
DOI: https://doi.org/10.1155/2022/8661324
IF: 2.809
2022-04-13
Computational and Mathematical Methods in Medicine
Abstract:Objective. To explore the application of machine learning algorithm in the prediction and evaluation of cesarean section, predicting the amount of blood transfusion during cesarean section and to analyze the risk factors of hypothermia during anesthesia recovery. Methods. (1)Through the hospital electronic medical record of medical system, a total of 600 parturients who underwent cesarean section in our hospital from June 2019 to December 2020 were included. The maternal age, admission time, diagnosis, and other case data were recorded. The routine method of cesarean section was intraspinal anesthesia, and general anesthesia was only used for patients’ strong demand, taboo, or failure of intraspinal anesthesia. According to the standard of intraoperative bleeding, the patients were divided into two groups: the obvious bleeding group (MH group, N = 154 ) and nonobvious hemorrhage group (NMH group, N = 446 ). The preoperative, intraoperative, and postoperative indexes of parturients in the two groups were analyzed and compared. Then, the risk factors of intraoperative bleeding were screened by logistic regression analysis with the occurrence of obvious bleeding as the dependent variable and the factors in the univariate analysis as independent variables. In order to further predict intraoperative blood transfusion, the standard cases of recesarean section and variables with possible clinical significance were included in the prediction model. Logistic regression, XGB, and ANN3 machine learning algorithms were used to construct the prediction model of intraoperative blood transfusion. The area under ROC curve (AUROC), accuracy, recall rate, and F 1 value were calculated and compared. (2) According to whether hypothermia occurred in the anesthesia recovery room, the patients were divided into two groups: the hypothermia group ( N = 244 ) and nonhypothermia group ( N = 356 ). The incidence of hypothermia was calculated, and the relevant clinical data were collected. On the basis of consulting the literatures, the factors probably related to hypothermia were collected and analyzed by univariate statistical analysis, and the statistically significant factors were analyzed by multifactor logistic regression analysis to screen the independent risk factors of hypothermia in anesthetic convalescent patients. Results. (1) First of all, we compared the basic data of the blood transfusion group and the nontransfusion group. The gestational age of the transfusion group was lower than that of the nontransfusion group, and the times of cesarean section and pregnancy in the transfusion group were higher than those of the non-transfusion group. Secondly, we compared the incidence of complications between the blood transfusion group and the nontransfusion group. The incidence of pregnancy complications was not significantly different between the two groups ( P > 0.05 ). The incidence of premature rupture of membranes in the nontransfusion group was higher than that in the transfusion group ( P < 0.05 ). There was no significant difference in the fetal umbilical cord around neck, amniotic fluid index, and fetal heart rate before operation in the blood transfusion group, but the thickness of uterine anterior wall and the levels of Hb, PT, FIB, and TT in the blood transfusion group were lower than those in the nontransfusion group, while the number of placenta previa and the levels of PLT and APTT in the blood transfusion group were higher than those in the nontransfusion group. The XGB prediction model finally got the 8 most important features, in the order of importance from high to low: preoperative Hb, operation time, anterior wall thickness of the lower segment of uterus, uterine weakness, preoperative fetal heart, placenta previa, ASA grade, and uterine contractile drugs. The higher the score, the greater the impact on the model. There was a linear correlation between the 8 features (including the correlation with the target blood transfusion). The indexes with strong correlation with blood transfusion included the placenta previa, ASA grade, operation time, uterine atony, and preoperative Hb. Placenta previa, ASA grade, operation time, and uterine atony were positively correlated with blood transfusion, while preoperative Hb was negatively correlated with blood transfusion. In order to further compare the prediction ability of the three machine learning methods, all the samples are randomly divided into two parts: the first 75% training set and the last 25% test set. Then, the three models are trained again on the training set, and at this time, the model does not come into contact with the samples in any test set. After the model training, the trained model was used to predict the test set, and the real blood transfusion status was compared with the predicted value, and the F 1 , accuracy, recall rate, and AUROC4 indicators were checked. In terms of training samples and test samples, the AUROC of XGB was higher than that of logistic regression, and the F 1 , accuracy, and recall rate of XGB of ANN were also slightly higher than those of logistic regression and ANN. Therefore, the performance of XGB algorithm is slightly better than that of logistic regression and ANN. (2) According to the univariate analysis of hypothermia during the recovery period of anesthesia, there were significant differences in ASA grade, mode of anesthesia, infusion volume, blood transfusion, and operation duration between the normal body temperature group and hypothermia group ( P < 0.05 ). Logistic regression analysis showed that ASA grade, anesthesia mode, infusion volume, blood transfusion, and operation duration were all risk factors of hypothermia during anesthesia recovery. Conclusion. In this study, three machine learning algorithms were used to analyze the large sample of clinical data and predict the results. It was found that five important predictive variables of blood transfusion during recesarean section were preoperative Hb, expected operation time, uterine weakness, placenta previa, and ASA grade. By comparing the three algorithms, the prediction effect of XGB may be more accurate than that of logistic regression and ANN. The model can provide accurate individual prediction for patients and has good prediction performance and has a good prospect of clinical application. Secondly, through the analysis of the risk factors of hypothermia during the recovery period of cesarean section, it is found that ASA grade, mode of anesthesia, amount of infusion, blood transfusion, and operation time are all risk factors of hypothermia during the recovery period of cesarean section. In line with this, the observation of this kind of patients should be strengthened during cesarean section.
mathematical & computational biology