Abstract 17809: Advanced Computational Modeling for Therapeutic Decision-Making in Patients Undergoing Percutaneous Coronary Intervention: A Focus on Gastrointestinal Bleeding
Background: We explored the feasibility of a support vector machine (SVM) model to predict gastrointestinal (GI) bleeding in patients undergoing percutaneous coronary intervention (PCI) and to potentially guide therapeutic decision-making.
Methods: Blue Cross Blue Shield of Michigan Cardiovascular Consortium registry data with 69 pre-procedural and angiographic variables from 68,022 PCI procedures in 2004-2007 were used to develop an SVM model to predict GI bleeding (n=786). The model was optimized for multivariate non-linear performance and tested on 42,310 PCI procedures performed in 2008-2009 (GI bleeding n=389). Model discrimination was assessed using the area under the receiver operating characteristic curve (AUROC). We further compared the observed outcomes of patients at variable predicted risks of GI bleeding, and assessed the complication rate in those treated with bivalirudin versus those treated with platelet glycoprotein IIbIIIa inhibtors.
Results: The SVM model provided high discrimination (AUROC 0.84) and improved reclassification relative to both L1-regularized (NRI 0.18, p<0.001) and L2-regularized (NRI 0.14, p=0.005) logistic regression models trained and tested on the same data. The risk of GI bleeding was lower in patients treated with bivalirudin although clinically relevant differences were mainly seen in patients in the highest risk quartile (figure).
Conclusion: A computational model based on SVM classification can accurately help identify patients at risk for GI bleeding following PCI. Use of bivalirudin instead of platelet GP IIbIIIa inhibitors in the highest risk patients may help reduce the occurrence of this complication.
- © 2012 by American Heart Association, Inc.