Abstract 12825: Merging Machine Learning and Patient Preference: A Contemporary, Comprehensive, Patient-Centered Tool for Risk Prediction Prior to Percutaneous Coronary Intervention

• Published in: Circulation (New York, N.Y.) N.Y.), 2022-11, Vol.146 (Suppl_1), p.A12825-A12825
• Main Authors: Hamilton, David E, Albright, Jeremy, Seth, Milan, Sukul, Devraj, Gurm, Hitinder S
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.146.suppl_1.12825

Introduction Predicting personalized risk for adverse events following percutaneous coronary intervention (PCI) remains critical in weighing treatment options, employing risk-mitigation strategies, and enhancing shared decision-making. Hypothesis We hypothesized that by employing machine learning models, we would be able to use pre-procedural factors to predict common post-PCI complications. Methods A patient panel ranked the importance of common PCI complications. A separate group of 66 adults underwent a semiquantitative survey assessing for a preferred list of outcomes and model display. Our cohort included 71,963 PCI procedures performed at 48 hospitals in Michigan between 4/1/2018 and 9/30/2020 in the BMC2 registry divided into two groups for training (75%) and validation (25%). XGBoost and random forest models were created to predict the risk of in-hospital outcomes. Results The overall rate of mortality was 1.76% (n=1,264), AKI 2.64% (n=1,712), new need for dialysis 0.41% (n=295), major bleeding 0.89% (n=631), and transfusion 2.43% (n=1,746). The XGBoost model was selected as the final model and demonstrated excellent discrimination and calibration for all outcomes including mortality (AUC0.950 [95%CI 0.939-0.962]), AKI (AUC0.889 [95%CI 0.874-0.904]), dialysis (AUC0.949 [95%CI 0.928-0.969]), transfusion (AUC0.918 [95%CI 0.905-0.931]), and major bleeding (AUC0.892 [95%CI 0.861-0.922]). Survey subjects preferred a comprehensive list of individually reported post-procedure outcomes. Conclusion Using common pre-procedural risk factors, we designed an XGBoost machine learning model that accurately predicts individualized post-PCI outcomes. Utilizing patient feedback, we created a patient-centered tool to clearly display risks to patients and providers (Figure, shiny.bmc2.org/pci-prediction/). Enhanced risk prediction prior to PCI could help inform treatment selection and shared decision-making discussions.