Machine learning analysis of factors influencing ischemic stroke after coronary artery bypass grafting

Abstract

Abstract Aims Ischemic Stroke (IS) is a devastating complication of coronary bypass (CABG) surgery, significantly increasing mortality, morbidity, cost, and the need for long-term care, and reducing the quality of life. With the aim of a learning-machine method, we investigated factors influencing the occurrence of IS after CABG. Methods and results We employed a statistical learning method – random forests – to examine which of various variables had the greatest impact on postoperative IS. A dataset including 16,255 consecutive patients undergoing isolated CABG between 1997 and 2017 In one Institution was examined. Along with demographic and clinical variables technical-related factors were included, encompassing surgical technique, number of touches on the aorta, cardiopulmonary bypass, total aortic clamping, use of the side-biting clamp, or number of proximal anastomoses. A ranking score based on the average percent increase in mean squared error and obtained for all variables was employed to quantify the importance of any variable in predicting IS. A total of 641 strokes occurred (3.9%). Total aortic clamp showed the highest incidence of stroke (n=570, 88.95% of total). Total aortic clamp had the highest score among all variables followed by previous stroke and carotid artery disease >50%. In contrast, other variables related to surgical technique showed low ranking scores in predicting IS. In patients undergoing CABG with the use of total aortic, a previous brain ischemic insult as well as age >75 years increase the risk of IS (fIGURE) Conclusions The use of total aortic clamp is the strongest predictor of stroke. A clamp-less technique is recommended whenever possible especially in older patients and those who experienced preoperative stroke or who had significant carotid stenosis. Importance of variables in predicting IS Funding Acknowledgement Type of funding source: None