A Bayesian network-based approach for identifying risk factors and predicting ischemic stroke in infective endocarditis patients.

Abstract

This study aimed to seek the risk factors and develop a predictive model for ischemic stroke (IS) in patients with infective endocarditis (IE) utilizing a Bayesian network (BN) approach. Data were obtained from the electronic medical records of all adult patients at three hospitals between 1 January 2018, and 31 December 2022. Two predictive models, logistic regression and BN, were used. Patients were randomly assigned to the training and test sets in a 7:3 ratio. We established a BN model with the training dataset and validated it with the testing dataset. The Bayesian network model was built by using the Tabu search algorithm. The areas under the receiver operating characteristic curve (AUCs), calibration curve, and decision curve were used to evaluate the prediction performance between the BN and logistic models. A total of 542 patients [mean (SD) age, 49.6 (15.3) years; 137 (25.3%) female] were enrolled, including 151 (27.9%) with IS and 391 (72.1%) without IS. Hyperlipidemia, hypertension, age, vegetation size (>10 mm), S. aureus infection, and early prosthetic valve IE were closely correlated with IS. The BN models outperformed the logistic regression in training and testing sets, with accuracies of 76.06% and 74.1%, AUC of 0.744 and 0.703, sensitivities of 25.93% and 20.93%, and specificities of 96.27% and 90.24%, respectively. The BN model is more efficient than the logistic regression model. Therefore, BN models may be suitable for the early diagnosis and prevention of IS in IE patients.