Abstract 14067: A Novel Tool for Highly Reliable and Accurate Prediction of Multiple Complications in Patients Undergoing Percutaneous Coronary Intervention

Introduction: Precise risk-prediction of post-procedural adverse events and outcomes after percutaneous coronary intervention (PCI) is critical to weighing treatment options and shared decision-making. Post-PCI complications are becoming less coming with quality improvement initiatives, however, there remains a significant risk for acute kidney injury (AKI), bleeding, and death. We hypothesized that with the use of random forest algorithms, we would be able to create accurate models to predict post-PCI complications. Methods: Our study cohort included 71,963 PCI procedures from the BMC2 registry performed at 48 hospitals in Michigan between 4/1/2018 and 9/30/2020. Random forest models were created using 74 pre-procedural clinical and laboratory variables to estimate the risk of each in-hospital outcome including mortality, AKI, major bleeding, and the need for a blood transfusion. A reduced model was created with the 20 most influential variables. The entire cohort was randomly separated into two groups for training (n=35,981) and validation (n=35,982). Model performance was evaluated in an independent validation data set using area under the receiver-operating characteristic curve (AUC). Results: The overall mortality was 1.76% (n = 1,264), there were 1,743 (2,42%) AKI events, 665 (0.92%) major bleeds, and 1,745 (2.42%) transfusions. The reduced models demonstrated excellent discrimination for all measured outcomes (Fig. 1) including mortality (AUC: 0.936 [95%CI 0.931-0.941]), AKI (AUC: 0.857 [95%CI 0.848-0.865]), major bleeding (AUC: 0.859 [95%CI 0.846-0.872]), and need for transfusion (AUC: 0.881 [95%CI 0.874-0.888]). Conclusions: The risk of post-PCI complications including mortality, AKI, major bleeding, and need for transfusion can be accurately calculated using a random forest algorithm of common pre-procedural factors to help better inform clinicians of a patient’s risk for better treatment selection and shared decision-making.