P126. Enhancing Surgical Outcomes: A Machine Learning Model to Anticipate Stroke After Hemiarch Surgery
Presented During: POD Abstracts (Available for viewing in the exhibit hall for the duration of the meeting)
University of Colorado, Anschutz Medical Center, Aurora, CO
Denver, CO
United States - Contact Me
Nicolas Chanes is a current PGY-1 surgical resident at the University of Colorado. He obtained his medical degree in 2023 from Louisiana State University School of Medicine in New Orleans. He is an active member of the Aortic Surgery Research Laboratory in the Department of Cardiothoracic Surgery at the University of Colorado. As a data scientist, he has specific interest in utilizing artificial intelligence to enhance outcomes in cardiothoracic surgery.
Thursday, April 25, 2024: 5:38 PM - 7:00 PM
Sheraton Times Square
Room: Central Park
Description
Objective:
With the development of machine learning comes the opportunity to better predict risk factors for postoperative morbidity. Although surgical and cerebral perfusion techniques have improved, postoperative stroke remains a devastating outcome after hemiarch surgery. To better predict at risk patients, we developed a machine learning algorithm to assess preoperative and operative risk factors associated with postoperative stroke following hemiarch surgery.
Methods:
We identified a total of 602 adult patients who underwent hemiarch surgery between June 2009 to October 2022 from our single institution prospectively maintained database. These patients were randomly divided into training (70%) and testing (30%) sets and various eXtreme gradient boosting (XGBoost) models were constructed to predict postoperative stroke in the cardiothoracic intensive care unit (CTICU). We considered 64 input parameters from the index hospitalization which were comprised of 24 demographic characteristics as well as 8 preoperative and 32 intraoperative variables. Our model underwent hyperparameter fine-tuning with 10-fold cross-validation at each iteration, leading to the development of the final model. We employed various evaluation metrics to assess model performance, including accuracy, Brier score, and area under the receiver operating characteristic curve (AUC-ROC). Additionally, we employed a SHapley Additive exPlanation (SHAP) beeswarm plot to elucidate the impact of individual features on the predictions generated by the XGBoost model.
Results:
Postoperative stroke occurred in 5.1% of patients (31 cases) following hemiarch surgery. The final XGBoost model showcased a cross-validation accuracy of 96% and exhibited excellent calibration, indicated by the low Brier score of 0.04. The predictor also displayed robust performance on the test dataset, attaining an accuracy rate of 96%. Our best performing postoperative stroke prediction model achieved an AUC-ROC of 0.80 on the training set and an AUC-ROC of 0.81 on the testing set. The SHAP beeswarm plot helped explain the complex decision-making process of our XGBoost model and provided insights into 20 of the key features that significantly influence model prediction. Elevated stroke risk was linked to factors such as female sex, higher intraoperative cryoprecipitate administration, older age, reduced nadir bladder temperature, and a history of CT surgery. Patients at a reduced risk of postoperative stroke exhibited characteristics such as aortic aneurysm at presentation, lower BMI, and underwent elective aortic surgery.
Conclusions:
Our model demonstrated excellent accuracy in predicting postoperative stroke after hemiarch surgery. Reduced stroke occurrences at higher nadir bladder temperatures could imply improved myocardial protection with normothermic cardioplegia in certain patients undergoing aortic procedures. Further research using a broader machine learning model is necessary to understand specific risk factors, including why females exhibit a higher incidence of stroke. Additionally, the apparent protective effect of aortic aneurysms without dissection warrants investigation, which may be due to a diminished inflammatory response stemming from less extensive intraoperative tissue handling.
Authors
Nicolas Chanes (1), Adam Carroll (1), Michael Kirsch (1), Bo Chang Wu (1), Muhammad Aftab (1), T. Brett Reece (1)
Institutions
(1) University of Colorado Anschutz, Denver, CO
With the development of machine learning comes the opportunity to better predict risk factors for postoperative morbidity. Although surgical and cerebral perfusion techniques have improved, postoperative stroke remains a devastating outcome after hemiarch surgery. To better predict at risk patients, we developed a machine learning algorithm to assess preoperative and operative risk factors associated with postoperative stroke following hemiarch surgery.
Methods:
We identified a total of 602 adult patients who underwent hemiarch surgery between June 2009 to October 2022 from our single institution prospectively maintained database. These patients were randomly divided into training (70%) and testing (30%) sets and various eXtreme gradient boosting (XGBoost) models were constructed to predict postoperative stroke in the cardiothoracic intensive care unit (CTICU). We considered 64 input parameters from the index hospitalization which were comprised of 24 demographic characteristics as well as 8 preoperative and 32 intraoperative variables. Our model underwent hyperparameter fine-tuning with 10-fold cross-validation at each iteration, leading to the development of the final model. We employed various evaluation metrics to assess model performance, including accuracy, Brier score, and area under the receiver operating characteristic curve (AUC-ROC). Additionally, we employed a SHapley Additive exPlanation (SHAP) beeswarm plot to elucidate the impact of individual features on the predictions generated by the XGBoost model.
Results:
Postoperative stroke occurred in 5.1% of patients (31 cases) following hemiarch surgery. The final XGBoost model showcased a cross-validation accuracy of 96% and exhibited excellent calibration, indicated by the low Brier score of 0.04. The predictor also displayed robust performance on the test dataset, attaining an accuracy rate of 96%. Our best performing postoperative stroke prediction model achieved an AUC-ROC of 0.80 on the training set and an AUC-ROC of 0.81 on the testing set. The SHAP beeswarm plot helped explain the complex decision-making process of our XGBoost model and provided insights into 20 of the key features that significantly influence model prediction. Elevated stroke risk was linked to factors such as female sex, higher intraoperative cryoprecipitate administration, older age, reduced nadir bladder temperature, and a history of CT surgery. Patients at a reduced risk of postoperative stroke exhibited characteristics such as aortic aneurysm at presentation, lower BMI, and underwent elective aortic surgery.
Conclusions:
Our model demonstrated excellent accuracy in predicting postoperative stroke after hemiarch surgery. Reduced stroke occurrences at higher nadir bladder temperatures could imply improved myocardial protection with normothermic cardioplegia in certain patients undergoing aortic procedures. Further research using a broader machine learning model is necessary to understand specific risk factors, including why females exhibit a higher incidence of stroke. Additionally, the apparent protective effect of aortic aneurysms without dissection warrants investigation, which may be due to a diminished inflammatory response stemming from less extensive intraoperative tissue handling.
Authors
Nicolas Chanes (1), Adam Carroll (1), Michael Kirsch (1), Bo Chang Wu (1), Muhammad Aftab (1), T. Brett Reece (1)
Institutions
(1) University of Colorado Anschutz, Denver, CO
Presentation Duration
PODS will be on display in the exhibit hall for the duration of the meeting during exhibit hall hours. PODS will also be available for viewing on the meeting website. There is no formal presentation associated with your POD, but we encourage you to visit the PODS area during breaks to connect with those viewing.