Prediction Of Adverse Cardiac Complications In Pregnant Women With Congenital Heart Disease Based On Machine Learning

With the advancement of medical technology,an increasing number of Congenital Heart Disease(CHD)patients are able to survive into reproductive age and have normal pregnancies.However,the physiological changes brought by pregnancy can exacerbate the cardiac burden and increase the risk of adverse cardiac complications in CHD patients.Adverse cardiac complications include heart failure,arrhythmia,peripartum cardiomyopathy,etc.,which seriously endanger the lives of CHD patients with pregnancy.Therefore,it is necessary to assess the risk of adverse cardiac complications in CHD patients with pregnancy,and to detect and intervene early in order to improve patient prognosis.However,currently there is a lack of relevant research on the predictive models of adverse cardiac complications in CHD patients with pregnancy.In this study,we used data from CHD patients with pregnancy from multiple medical centers as samples and used various machine learning algorithms to establish predictive models for adverse cardiac complications in CHD patients with pregnancy,which can effectively provide clinical reference and assist doctors in the diagnosis and treatment of patients.The steps for establishing prediction models for adverse cardiac events in CHD patients with pregnancy are as follows:Firstly,the raw data were preprocessed,including missing value imputation using multiple imputation and median imputation,and feature selection using Least Absolute Selection and Shrinkage Operator(Lasso)to lay a data foundation for subsequent modeling.Then,based on five machine learning algorithms,namely,Logistic Regression,Random Forest,Gradient Boosting Decision Tree(GBDT),Extreme Gradient Boosting(XGBoost),and Light Gradient Boosting Machine(LightGBM),predicting models for adverse cardiac events in CHD patients with pregnancy were constructed,and parameter tuning was performed using grid search.The predictive abilities of the models were compared and analyzed using balanced F Score(F1-score),the area under the receiver operator characteristic(ROC)curve(AUC),and Matthews Correlation Coefficient(MCC),and the models were explained using the SHAP(Shapley Additive exPlanations)interpretation tool.Finally,the models were fused using the Stacking algorithm to obtain a new model.The study found that all five machine learning models had good predictive performance for whether CHD patients would experience adverse cardiac events,with the GBDT,XGBoost,LightGBM,random forest,and logistic regression models showing decreasing predictive ability in that order.Through SHAP analysis,the specific effects of clinical features on predictions for both the overall sample and individual samples could be determined,providing data support for clinical diagnosis and treatment strategies for patients.Using the GBDT and XGBoost models,which had better predictive performance,as base models and logistic regression as the meta-model,a new model was obtained through model fusion using Stacking.The F1-score,AUC,and MCC of this fusion model were higher than those of the other five machine learning models.That is,the fusion model had better predictive performance for adverse cardiac events in CHD patients with pregnancy.