Application of Machine Learning to Determine Top Predictors of Non-calcified Coronary Plaque Burden in Psoriasis
Harry Choi1, Eric Munger1, 2, Amit K. Dey1, Youssef Elnabawi1, Jacob Groenendyk1, Noor Khalil1, Justin Rodante1, Milena Aksentijevich1, Aarthi Reddy1, Jenis Argueta-Amaya1, Martin Playford1, Ahmed Hasan1, Moshin S. Jafri2, Veit Sandfort1, Marcus Chen1, David Bluemke3, Joel Gelfand4, Nehal N. Mehta1. 1National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland, United States, 2George Mason University, Fairfax, Virginia, United States, 3University of Wisconsin, Madison, Wisconsin, United States, 4University of Pennsylvania, Philadelphia, Pennsylvania, United States
Purpose of Study Psoriasis, a chronic systemic inflammatory disease, is associated with elevated non-calcified coronary plaque burden (NCB) and increased cardiovascular (CV) events. In this study, we used machine learning (ML) as a tool to determine the top predictors of NCB, which is prone to rupture and cause subsequent MI.
Methods Used The analysis included 82 original biomarkers in 226 consecutive psoriasis patients that were pruned and ranked to 14 variables through the random forest algorithm. These top predictors were then evaluated by linear and logistic regressions to confirm our results.
Summary of Results At baseline, patients with psoriasis were middle-aged, predominantly male, low cardiovascular risk by Framingham risk score and mild-to-moderate skin disease (Table 1). Using the random forest algorithm, the top ten predictors of NCB, in order of importance, were: body mass index, visceral adiposity, psoriasis severity, apolipoprotein A1, small LDL particle, high sensitivity CRP, LDL, white blood cell count, large-medium VLDL, and subcutaneous adiposity. Pearson's and point-biserial correlations of these top variables with NCB yielded similar results in line with our ML outputs.
Conclusions In this study, we applied ML to identify the top predictors of NCB in patients with psoriasis, with top predictors being markers of obesity, dyslipidemia, and inflammation. However, larger studies are needed to validate our findings.
Back to 2019 Abstracts