Phenotyping of risk factors and prediction of inhospital mortality in patients with coronary artery disease after coronary artery bypass grafting based on explainable artificial intelligence methods

• Published in: Rossiĭskiĭ kardiologicheskiĭ zhurnal Vol. 28; no. 4; p. 5302
• Main Authors: Geltser, B. I., Shakhgeldyan, K. I., Rublev, V. Yu, Domzhalov, I. G., Tsivanyuk, M. M., Shekunova, O. I.
• Format: Journal Article
• Language: English; Russian
• Published: FIRMA «SILICEA» LLC 01-05-2023
• Subjects: coronary artery bypass grafting; explainable artificial intelligence; inhospital mortality; predictive models; risk factor phenotyping
• ISSN: 1560-4071; 2618-7620
• DOI: 10.15829/1560-4071-2023-5302

Aim
. To develop predictive models of inhospital mortality (IHM) in patients with coronary artery disease after coronary artery bypass grafting (CABG), taking into account the results of phenotyping of preoperative risk factors.
Material and methods
. This retrospective study was conducted based on the data of 999 electronic health records of patients (805 men, 194 women) aged 35 to 81 years with a median (Me) of 63 years who underwent on-pump elective isolated CABG. Two groups of patients were distinguished, the first of which was represented by 63 (6,3%) patients who died in the hospital during the first 30 days after CABG, the second — 936 (93,7%) with a favorable outcome. Preoperative clinical and functional status was assessed using 102 factors. Chi-squares, Fisher, Mann-Whitney methods were used for data processing and analysis. Threshold values of predictors were determined by methods, including maximizing the ratio of true positive IHM cases to false positive ones. Multivariate logistic regression (MLR) was used to develop predictive models. Model accuracy was assessed using 3 following metrics: area under the ROC curve (AUC), sensitivity (Sens), and specificity (Spec).
Results
. An analysis of preoperative status of patients made it possible to identify 28 risk factors for IHM, combined into 7 phenotypes. The latter formed the feature space of IHM prognostic model, in which each feature demonstrates the patient’s compliance with a certain risk factor phenotype. The author’s MLR model had high quality metrics (AUC-0,91; Sen-0,9 and Spec-0,85).
Conclusion
. The developed data processing and analysis algorithm ensured high quality of preoperative risk factors identification and IHM prediction after CABG. Prospects for further research on this issue are related to the improvement of explainable artificial intelligence technologies, which allow developing infor­mation systems for managing clinical practice risks.