Decision tree models as a classifier of endothelial function based on strength, pulmonary and cardiac function in COPD: Preliminary analysis

Abstract

Introduction: Decision Trees (DT) models are able to represent knowledge in an interpretable way and an efficient mean of building predictors based on a set of data. Cardiovascular disease (CVD) is pronounced in chronic obstructive pulmonary disease (COPD) patients and endothelial function (EF) is a valuable risk predictor for CVD. Aim: To develop a DT as a classifier of EF in COPD patients. Methods: DT model enrolled 39 patients after COPD exacerbation. Inputs: 1) number of exacerbations (NE) 2) peripheral muscle strength (PMS) 3) predicted forced expiratory volume in 1s (FEV1) and 4) left ventricular ejection fraction (LVEF). EF was determined by brachial artery ultrasonography, flow-mediated dilation and a binary classification was considered worse (0) or better (1). DT models were built, trained (70%) and validated (30%) using Matlab framework. The correctness of the DT model to predict EF was assessed using sensitivity (SE), specificity (SP) and geometric mean (GM) measures. Results: The implementation of the DT models considered three alternatives: 1) NE, PMS, FEV1; 2) NE, PMS, LVEF; 3) NE, PMS, FEV1, LVEF. The performance of the models is depicted in Table I. Conclusion: Results suggest that DT can be used as a predictor of EF after COPD exacerbation discharge, using as input a reduce set of clinical factors. Research Funding Source: FAPESP – process n°2015/12763-4; FAPESP – process n° 2015/26501-1; and CAPES – Brazil.