A statistical learning framework for predicting left ventricular ejection fraction based on glutathione peroxidase-3 level in ischemic heart disease. (October 2022)

• Record Type: Journal Article
• Authors: Choi, Mun-Ho; Oh, Seok; Choi, Jang Yeol; Kim, Ju Han; Lee, Seung-won
• Is Part Of: Computers in biology and medicine. Volume 149(2022)
• Publication Date: 2022-10
• Subjects: Ischemic heart disease -- GPx3 -- Statistical learning -- Small imbalanced dataset -- LOOCV -- Youden index; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2022.105929
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Ischemic heart disease (IHD) is the most prevalent cardiovascular disease. The left ventricular ejection fraction (LVEF) is a well-validated index of the systolic function of the left ventricle and it gradually decreases in IHD. We aimed to develop a new strategy for examining the relationship between serum glutathione peroxidase 3 (GPx3; a possible antioxidant protector against IHD) and the LVEF of IHD patients. To overcome the problem of small, imbalanced, and multicollinear datasets, we adopted leave-one-out cross-validation to maximize the size of the training set, the Youden index to reflect the biased distribution of events, and regularization or dimension transform techniques to reduce the effect of multicollinearity. For the outcome variable of LVEF, five classification methods were tested for six previously selected features with and without GPx3. High GPx3 levels (≥5.314 μg/mL) were closely related to a reduced LVEF (<50%). The presented statistical learning framework is effective for small and imbalanced data with multicollinearity such as clinical data. Highlights: Performance of multivariate logistic regression and ML analyses for IHD patients. A new classification algorithm, based on LOOCV and Youden index. GPx3 can serve as a predictor explaining LVEF reduction. Our statistical learning strategy proved effective for the small imbalanced dataset.