Estimation of Low-Density Lipoprotein Cholesterol Concentration Using Machine Learning.

Area of Science:

• Clinical Chemistry
• Biomedical Informatics
• Machine Learning in Healthcare

Background:

• Accurate estimation of low-density lipoprotein cholesterol (LDL-C) is crucial for cardiovascular disease risk assessment.
• Traditional formulas like Friedewald and Martin-Hopkins have limitations, particularly in specific patient populations.

Purpose of the Study:

• To develop and validate novel LDL-C prediction models using machine learning techniques.
• To compare the performance of these new models against established Friedewald and Martin-Hopkins formulas.

Main Methods:

• Utilized a large dataset (n=59,415) of lipid profiles, partitioning into training and testing sets.
• Developed models using linear regression, gradient-boosted trees, and artificial neural networks (ANN).
• Statistical significance determined by P < .001 and effect size > .2, employing t-tests and Wilcoxon signed-rank tests.

Main Results:

• Friedewald formula underestimated and Martin-Hopkins overestimated LDL-C for triglycerides (TG) ≥177 mg/dL (P < .001).
• Machine learning models (linear regression, gradient-boosted trees, ANN) demonstrated superior accuracy compared to traditional formulas for TG ≥177 mg/dL and LDL-C < 70 mg/dL.
• Performance validation confirmed by comparison with a homogeneous assay and classification accuracy.

Conclusions:

• Machine learning models offer more precise LDL-C estimations than conventional formulas.
• These models are particularly advantageous for individuals with TG levels between 177-399 mg/dL and LDL-C < 70 mg/dL.
• The developed models represent a significant advancement for clinical lipid management.