A statistical learning framework for predicting left ventricular ejection fraction based on glutathione peroxidase-3...

Area of Science:

• Cardiology
• Biochemistry
• Statistical modeling

Background:

• Ischemic heart disease (IHD) is a leading cause of cardiovascular mortality.
• Left ventricular ejection fraction (LVEF) is a key indicator of cardiac systolic function, often declining in IHD.
• Serum glutathione peroxidase 3 (GPx3) is investigated as a potential antioxidant biomarker in IHD.

Purpose of the Study:

• To develop and validate a statistical learning strategy for analyzing the relationship between serum GPx3 levels and LVEF in IHD patients.
• To address challenges posed by small, imbalanced, and multicollinear clinical datasets.
• To determine if GPx3 levels are associated with reduced LVEF in IHD.

Main Methods:

• Employed leave-one-out cross-validation to maximize training data size.
• Utilized the Youden index to account for event distribution bias.
• Applied regularization and dimension transformation techniques to mitigate multicollinearity.
• Tested five classification methods on six features, with and without GPx3, for LVEF prediction.

Main Results:

• High serum GPx3 levels (≥5.314 μg/mL) were significantly associated with reduced LVEF (<50%).
• The statistical framework demonstrated effectiveness in handling small, imbalanced, and multicollinear clinical data.
• GPx3 emerged as a relevant feature in predicting LVEF in the context of IHD.

Conclusions:

• Serum GPx3 levels are closely related to decreased LVEF in patients with ischemic heart disease.
• The developed statistical learning framework provides a robust method for analyzing complex clinical datasets in cardiovascular research.
• GPx3 warrants further investigation as a potential biomarker for assessing cardiac function in IHD.