Machine Learning Prediction of 3-Year Mortality in Patients Undergoing Percutaneous Coronary Intervention for Coronary Artery Disease
Author(s)
Lejia Hu, MS, XUAN ZHANG, MD PhD PhD, Hal Yapici, MBA, MPH, MD, Kunal J. Lodaya, MD, Sibyl H. Munson, PhD, Fabian Thomas William D'Souza, MD, MSurg, FRCS, MBA, Weiqi Jiao, ScM, Hayden W. Hyatt, PhD, Rahul Rajkumar, MD, MPH, Nicholas Bettencourt, BS.
Boston Strategic Partners, Inc., Boston, MA, USA.
Boston Strategic Partners, Inc., Boston, MA, USA.
Presentation Documents
OBJECTIVES: This study aims to predict 3-year mortality among patients undergoing percutaneous coronary intervention (PCI) for coronary artery disease (CAD) using the National COVID Cohort Collaborative (N3C) database. By leveraging machine learning models, it identifies key predictors of mortality and explores potential regional disparities in outcomes.
METHODS: Data from patients who underwent PCI for CAD from 2017 to 2021 in the N3C database were extracted, including variables such as age, sex, comorbidities (e.g., chronic pulmonary disease, myocardial infarction, renal dysfunction), medication use (insulin, diuretics etc.), and other individual items in The European System for Cardiac Operative Risk Evaluation (EuroSCORE) II composite. Machine learning models were applied to predict mortality within three years post-surgery, with the Gradient Boosting Classifier ultimately selected for its superior performance. Feature importance analysis identified the top predictors and regional differences in mortality rates were further examined.
RESULTS: The selected population (n = 7,505) had a mean age of 65 with 35% males and a 3-year mortality rate of 15%. 52% of the selected patients resided in the South, 28% in the Midwest, and 20% in the Northeast, West, and other places. Among the 22 factors analyzed, renal impairment, age, region, pulmonary hypertension, and index year emerged as the most important factors in predicting mortality. The Gradient Boosting Classifier showed an overall accuracy of 0.86 (precision: 0.86; recall: 0.98). Mortality rates varied significantly across regions: Midwest (20.02%), Northeast (14.66%), South (27.04%), West (15.38%), and other regions (6.62%).
CONCLUSIONS: This study highlights the utility of machine learning in identifying key predictors of mortality among patients undergoing PCI for CAD treatment, with regional differences playing a significant role. Higher mortality rates in the South and Midwest suggest potential healthcare disparities requiring further investigation. These insights can guide interventions to improve PCI outcomes and address regional inequities in care.
METHODS: Data from patients who underwent PCI for CAD from 2017 to 2021 in the N3C database were extracted, including variables such as age, sex, comorbidities (e.g., chronic pulmonary disease, myocardial infarction, renal dysfunction), medication use (insulin, diuretics etc.), and other individual items in The European System for Cardiac Operative Risk Evaluation (EuroSCORE) II composite. Machine learning models were applied to predict mortality within three years post-surgery, with the Gradient Boosting Classifier ultimately selected for its superior performance. Feature importance analysis identified the top predictors and regional differences in mortality rates were further examined.
RESULTS: The selected population (n = 7,505) had a mean age of 65 with 35% males and a 3-year mortality rate of 15%. 52% of the selected patients resided in the South, 28% in the Midwest, and 20% in the Northeast, West, and other places. Among the 22 factors analyzed, renal impairment, age, region, pulmonary hypertension, and index year emerged as the most important factors in predicting mortality. The Gradient Boosting Classifier showed an overall accuracy of 0.86 (precision: 0.86; recall: 0.98). Mortality rates varied significantly across regions: Midwest (20.02%), Northeast (14.66%), South (27.04%), West (15.38%), and other regions (6.62%).
CONCLUSIONS: This study highlights the utility of machine learning in identifying key predictors of mortality among patients undergoing PCI for CAD treatment, with regional differences playing a significant role. Higher mortality rates in the South and Midwest suggest potential healthcare disparities requiring further investigation. These insights can guide interventions to improve PCI outcomes and address regional inequities in care.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR18
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory)