Risk Factors Analysis And Risk Prediction Modeling For Postpartum Hemorrhage

Postpartum hemorrhage is a serious complication in obstetrics,and is the main cause of maternal death in China.Postpartum hemorrhage can bring a range of physical hazards to the mother,and may even endanger the mother’s life.To explore the risk factors of postpartum hemorrhage,to take effective measures to prevent and treat postpartum hemorrhage and to control the incidence of postpartum hemorrhage have become the main research direction of postpartum hemorrhage.This thesis is focused on exploring the risk factors of postpartum hemorrhage in three aspects of pregnant women:epidemiology,biochemistry and placenta,with a retrospective data set of pregnant women in Beijing maternity hospital.Descriptive analysis is conducted firstly on the overall data to understand the incidence of postpartum hemorrhage under various factors.The results show that there is significant difference in the prevalence of postpartum hemorrhage among different pre-pregnancy BMI,and there is also significant difference in the prevalence of postpartum hemorrhage among different age groups.The retrospective data were collected from 560 lying-in women,23 variables were recorded for each woman.After data preprocessing,several classification models are established based on the data set.Firstly,the adaptive Lasso and SCAD methods are used to select variables,and the selected variables are used to perform stepwise logistic regression to obtain the optimal regression model.Then a random forest model is constructed with the optimally chosen parameters,and the importance order of variables is obtained.Finally,the prediction model of postpartum hemorrhage based on neural network is established by determining the number of hidden layers and neurons.With these models,we find that the risk factors for postpartum hemorrhage include Age,pre-pregnancy BMI,number of births,abnormal hemoglobin,abnormal platelets,multiple pregnancy,history of abdominal surgery,macrosomia,uterine and ovarian fibroids,pelvic soft birth canal abnormalities,gestational diabetes,gestational hypertension,placenta praevia and polyhydramnios.According to the sensitivity,specificity,F₁value and AUC value as evaluation indicators,the advantages and disadvantages of the three classification models are analyzed.When sensitivity,F₁value and AUC are the evaluation indexes,the fitting effect of neural network model is the best.The prediction model of postpartum hemorrhage based on random forest has the highest accuracy when the specificity is used as the evaluation index.Based on the analysis of the risk factors for postpartum hemorrhage,it is suggested to control the occurrence of postpartum hemorrhage clinically from three aspects:prenatal publicity,prenatal guidance and postpartum monitoring,so as to reduce the morbidity and mortality of postpartum hemorrhage.