P260. Predictive Risk Score for Cerebral Malperfusion in Type A Aortic Dissection Utilizing a Machine Learning Model
Presented During: POD Abstracts (Available for viewing in the exhibit hall for the duration of the meeting)
University of Texas Health Science Center at Houston (UTHealth Houston)
Pearland, TX
United States - Contact Me
I am currently a general surgery resident at UT Houston. I am in year 2 of a 2-year research sabbatical and also in a Masters program in Clinical Research. I am working in cardiothoracic surgery research under the guidance of Dr. Anthony Estrera and Dr. Danny Ramzy. I am doing congenital heart surgery research with Dr. Damien LaPar. My current interests are in adult structural heart and aortic surgery. I am also an active duty officer in the United States Army.
Thursday, April 25, 2024: 5:38 PM - 7:00 PM
Sheraton Times Square
Room: Central Park
Description
Objectives:
Cerebral malperfusion due to acute type A aortic dissection can lead to severe neurologic deficits and even operative death. Early detection of risk for this sequela is imperative to optimize clinical decision-making and improve patient outcomes. Machine learning models have been shown to synthesize large, granular databases efficiently and accurately. We hypothesized that a machine learning model would be able to produce a novel predictive risk score for cerebral malperfusion to aid in early risk assessment.
Methods:
We retrospectively reviewed all patients undergoing surgical management for acute type A aortic dissection from 2001 to 2020 at our institution. We focused on readily available admission data that was retrieved upon initial assessment by the surgical team (i.e., prior medical/surgical history, admission laboratory values, admission imaging). Any missing data was imputed with a missRanger imputation model (RStudio). All data was then analyzed in a random forest regression machine learning model to identify key predictor variables for cerebral malperfusion (primary outcome). Risk scores were calculated for these variables using a generalized regression model. Predictive probabilities were then obtained, and the strength of the model was evaluated by the area under the curve, sensitivity/specificity, and negative/positive predictive values.
Results:
We identified 650 patients who underwent surgical repair for acute type A aortic dissection during the study period. Median age was 58.0 years old (interquartile range: 47.0-69.0). 183 (28.1%) were female. There were 119 patients (18.3%) who were diagnosed with cerebral malperfusion preoperatively. Our model determined 8 key predictor variables (score) for cerebral malperfusion: presenting comatose (2), presenting with altered mental status (2), hemiparesis on presentation (4), concomitant celiac malperfusion (1), concomitant renal malperfusion (1), prior history of stroke (1), prior history of transient ischemic attack (1), and dissection extending into either common carotid artery (1). This model had an area under the curve of 0.866 (0.798-0.933, p<0.001). For patients with a score of 1 or less, the sensitivity was 95.0%. With a score of 4 or more, the specificity was 97.0%. Table illustrates the sensitivities, specificities, positive predictive values, and negative predictive values for the different score ranges. Figure demonstrates the receiver operative curve for the model.
Conclusion:
A novel predictive risk score was developed using machine learning models to aid in the early detection of cerebral malperfusion. This model uses readily available information, is easy to calculate, and can aid the surgical team in clinical decision-making.
Authors
Alexander Mills (1), Akiko Tanaka (1), Yuki Ikeno (1), Lucas Ribe (1), Harleen Sandhu (1), Charles Miller (1), Charles Green (1), Danny Ramzy (1), Anthony Estrera (1)
Institutions
(1) McGovern Medical School at UTHealth, Houston, TX
Cerebral malperfusion due to acute type A aortic dissection can lead to severe neurologic deficits and even operative death. Early detection of risk for this sequela is imperative to optimize clinical decision-making and improve patient outcomes. Machine learning models have been shown to synthesize large, granular databases efficiently and accurately. We hypothesized that a machine learning model would be able to produce a novel predictive risk score for cerebral malperfusion to aid in early risk assessment.
Methods:
We retrospectively reviewed all patients undergoing surgical management for acute type A aortic dissection from 2001 to 2020 at our institution. We focused on readily available admission data that was retrieved upon initial assessment by the surgical team (i.e., prior medical/surgical history, admission laboratory values, admission imaging). Any missing data was imputed with a missRanger imputation model (RStudio). All data was then analyzed in a random forest regression machine learning model to identify key predictor variables for cerebral malperfusion (primary outcome). Risk scores were calculated for these variables using a generalized regression model. Predictive probabilities were then obtained, and the strength of the model was evaluated by the area under the curve, sensitivity/specificity, and negative/positive predictive values.
Results:
We identified 650 patients who underwent surgical repair for acute type A aortic dissection during the study period. Median age was 58.0 years old (interquartile range: 47.0-69.0). 183 (28.1%) were female. There were 119 patients (18.3%) who were diagnosed with cerebral malperfusion preoperatively. Our model determined 8 key predictor variables (score) for cerebral malperfusion: presenting comatose (2), presenting with altered mental status (2), hemiparesis on presentation (4), concomitant celiac malperfusion (1), concomitant renal malperfusion (1), prior history of stroke (1), prior history of transient ischemic attack (1), and dissection extending into either common carotid artery (1). This model had an area under the curve of 0.866 (0.798-0.933, p<0.001). For patients with a score of 1 or less, the sensitivity was 95.0%. With a score of 4 or more, the specificity was 97.0%. Table illustrates the sensitivities, specificities, positive predictive values, and negative predictive values for the different score ranges. Figure demonstrates the receiver operative curve for the model.
Conclusion:
A novel predictive risk score was developed using machine learning models to aid in the early detection of cerebral malperfusion. This model uses readily available information, is easy to calculate, and can aid the surgical team in clinical decision-making.
Authors
Alexander Mills (1), Akiko Tanaka (1), Yuki Ikeno (1), Lucas Ribe (1), Harleen Sandhu (1), Charles Miller (1), Charles Green (1), Danny Ramzy (1), Anthony Estrera (1)
Institutions
(1) McGovern Medical School at UTHealth, Houston, TX
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.