Abstract 14448: Machine Learning Methodology Identifies Predictors of a Cardiovascular Composite Measure Among Severe Peripheral Artery Disease Patients
Introduction: The aging population has resulted in greater prevalence of peripheral artery disease (PAD), associated with significant morbidity and mortality. A better understanding of the risk factors for morbid and fatal outcomes may lead to improved treatment strategies in PAD patients.
Methods: A retrospective analysis was conducted on 7,249,982 patients, using the Optum+Humedica integrated database (2007-15). Patients age ≥50 with severe PAD (characterized by rest pain, ulceration or gangrene with first diagnosis as index date) who had 6-month and 12-month clinical activity or died during the pre and post-index periods, respectively, were selected. Patients with history of intracranial hemorrhage, stroke and transient ischemic attack were excluded.
A Bayesian machine learning platform, Reverse Engineering and Forward Simulation (REFS TM), built an ensemble of 256 predictive models to examine the association between baseline patient characteristics (demographics, physician visits, diagnoses and treatment history) with a 12-month post-index cardiovascular composite measure (CCM). We defined CCM as ischemic stroke, myocardial infarction (MI), major amputation, or all-cause death. We assessed effect estimates across the ensemble of models using mean odds ratios (OR).
Results: A total of 13,016 severe PAD patients were selected. Overall mean age was 70.2 years, females comprised 44.5%, and 47.3% of patients had comorbid diabetes. Approximately 28.5% of PAD patients had the CCM. Among patients with diabetes, the CCM risk was higher compared to patients without diabetes (33.6% vs 24.0%, p<0.001).
From baseline characteristics, REFSTM models identified factors that were highly predictive of the CCM. These included (OR, SD): acute MI (3.5, 0.2), gangrene (2.3, 0.1), congestive heart failure (1.9, 0.1), acute and unspecified renal failure (1.8, 0.1), and diseases of the heart (1.5, 0.05).
Conclusion: In this large, geographically diverse study, severe PAD patients had an event rate of 28.5% for ischemic stroke, MI, major amputation, or all-cause death. Based on a machine-learning platform, REFSTM, three conditions were identified as the strongest predictors of major vascular events: acute MI, gangrene, and congestive heart failure.
Author Disclosures: W. Ting: Consultant/Advisory Board; Significant; Janssen Pharmaceuticals. L. Haskell: Employment; Significant; Janssen Pharmaceuticals. F. Lurie: Research Grant; Modest; Tactile medical. Consultant/Advisory Board; Modest; Janssen, Cook. J.S. Berger: Consultant/Advisory Board; Significant; Janssen Pharmaceuticals, AstraZeneca. Z. Eapen: Ownership Interest; Significant; Pattern Health Technologies. Consultant/Advisory Board; Modest; Amgen, Cytokinetics, Janssen, Medtronic, SHL Telemedicine. Consultant/Advisory Board; Significant; Novartis. M. Valko: Consultant/Advisory Board; Modest; Janssen Pharmaceuticals. V. Alas: Consultant/Advisory Board; Modest; Janssen Pharmaceuticals. K. Rich: Consultant/Advisory Board; Modest; Janssen Pharmaceuticals. C. Crivera: Employment; Significant; Janssen Pharmaceuticals. J. Schein: Employment; Significant; Janssen Pharmaceuticals.
- © 2016 by American Heart Association, Inc.