Study On The Risk And Prediction Model Of Birth Defectscaused By Peri-pregnancy/pregnancy Medication

Objective: This study focused on the population of peri-pregnancy / pregnancy medication,explored the situation of drug exposure and perinatal birth defects in this population,analyzed the risk factors of birth defects,and constructed an interpretable birth defect risk prediction model based on machine learning,in order to provide an intelligent auxiliary tool for early clinical assessment of the risk of birth defects caused by peri-pregnancy / pregnancy medication.Methods: Firstly,by reviewing the meta-analysis of the influencing factors of birth defects,the information variables needed to fully understand the real world data were included in the model study.Then,the medical data of obstetric discharged patients and gestational drug consultation outpatients in Sichuan Provincial People ’s Hospital from January 2019 to July 2022 were collected,and the cases were screened according to the inclusion and exclusion criteria and initially processed to form a basic data set.A retrospective study was conducted to investigate the use of drugs during peri-pregnancy/ pregnancy and the incidence of birth defects in perinatal infants.Univariate and multivariate logistic regression analysis were used to analyze the risk factors of birth defects.Then,the machine learning algorithm was used to establish a birth defect risk prediction model,evaluate and verify the performance of the model,and select the optimal model.Finally,the Shapley Additive Explanations(SHAP)algorithm is used to explain and analyze the optimal model.Results: 1.The results of literature research show that the influencing factors of birth defects mainly include three aspects: environmental factors,maternal demographic information,maternal disease and behavior,and 26 influencing factors are extracted into the real world data extraction table.2.A total of 1023 subjects and 53 characteristic variables were included.3.A total of 74 perinatal infants had birth defects,and the incidence of birth defects was 7.23 %;multiple pregnancy,family history of birth defects,gestational diabetes,sex hormones and gonadotropins drugs,and dermatological drugs are risk factors for birth defects.4.Among the birth defect risk prediction models constructed,the GS＿XGBoost model based on the top 10 feature variables of the feature importance score is the best model,and the AUC of the model is 0.8884.5.The interpretation and characteristic analysis of GS＿XGBoost model based on SHAP algorithm showed that drug exposure period,sex hormone and gonadotropin,number of pregnancies and abnormal pregnancy history were predicted as positive influencing factors for birth defects.The four characteristic combinations of adrenocortical hormone and drug exposure period,sex hormone and gonadotropin and drug exposure period,age at reproductive time and drug exposure period had greater interaction effect on predicted birth defects.Conclusion: This study analyzed the risk factors of birth defects.The constructed risk prediction model of birth defects has good performance and can distinguish the risk of birth defects.The SHAP algorithm is introduced to realize the ’ visualization ’ of the prediction model,and the analysis of the influencing factors of individual birth defects is realized.It provides a new methodological reference and auxiliary tool for clinical early assessment of the risk of birth defects caused by peri-pregnancy / pregnancy medication.