P1555Using a decision tree analysis to risk stratify patients with coronary artery disease

Abstract

Background and purpose: The development of coronary artery disease (CAD) is a multifactorial process involving traditional, non-traditional and biochemical risk factors that increase the risk of adverse outcomes. Within a population of patients with CAD, different combinations of these risk factors contribute to its morbidity and mortality. Identification of subgroups at varying levels of risk determined by different combinations of risk factors may help in a targeted treatment approach. Decision tree analysis is a useful method to identify subgroups at varying levels of cardiovascular (CV) disease risk based on these risk factors and is well suited for clinical application. Methods: We used a classification tree model to group patients with CAD into different subsets (nodes) for the outcome of death and death/myocardial infarction (MI). We used 26 variables for tree building including clinical risk factors (age, sex, race, body mass index (BMI), smoking, diabetes, hypertension, chronic kidney disease (CKD)), lab values (eGFR, white blood cell count, hemoglobin, platelet, blood urea nitrogen (BUN), sodium, low density lipoprotein, high density lipoprotein, triglycerides, total cholesterol), circulating biomarkers (high sensitivity C-reactive protein, fibrin degradation products (FDP), heat shock protein-70 (HSP-70), soluble urokinase activator receptor (suPAR), high sensitivity troponin (Hs-TnI)) and socioeconomic status (income, education, marital status and living in food deserts). Results: 1897 patients without acute myocardial infarction undergoing coronary angiography were included. There were a total of 326 deaths, 84 MI events (44 patients had both events) over a median follow up of 6.2 years (interquartile range 3.5-8.0 years). Decision tree for death (figure 1A) identified 6 distinct risk groups (nodes 3, 4, 7, 8, 10 and 11) based on 26 variables and demonstrated that those with elevated suPAR, hs-TnI and FDP above specific cutoffs determined by the algorithm (3183.5 pg/mL, 9.95 pg/mL and 0.855 ug/mL respectively) had the worst outcomes and comprised the highest risk group (node 11; HR 16.0, 95% CI 11.6- 22.1 compared to reference group (node 3). Separate decision tree for combined outcome of death/MI (figure 1B) showed similar results (highest risk group (node 9) HR 16.9, 95% CI 11.7-24.5 compared to the reference group (node 3). Conclusion: Decision tree analysis is a unique method to identify subgroups at varying levels of risk for adverse outcomes. Our findings suggest that of the 26 variables tested, circulating biomarkers are the strongest predictors of outcome and identify those at highest risk in patients with CAD. This approach may help identify CAD patients who would benefit from aggressive intervention to reduce morbidity and mortality. (Figure Presented).