Comparative performance of machine learning vs classical formulas for LDL-cholesterol calculation.

Area of Science:

• Biochemistry and Clinical Chemistry
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Risk Assessment

Background:

• Low-density lipoprotein cholesterol (LDL-C) is a critical cardiovascular risk factor.
• Traditional LDL-C estimation formulas, like Friedewald, have limitations, particularly with high triglycerides.
• Machine learning (ML) presents a promising alternative for accurate LDL-C estimation.

Purpose of the Study:

• To compare the accuracy of various machine learning models against traditional formulas for LDL-C estimation.
• To evaluate the performance of ML models across different triglyceride levels.

Main Methods:

• Retrospective analysis of 34,678 lipid profiles.
• Development and evaluation of 22 machine learning models using Python's PyCaret library.
• Performance metrics included R-squared, MAE, and RMSE, analyzed across four triglyceride subgroups.

Main Results:

• ML models, particularly LightGBM, Gradient Boosting, and XGBoost, demonstrated superior performance (R² > 0.95) compared to traditional formulas.
• Traditional formulas, especially Friedewald (R² = 0.926), showed significantly lower accuracy.
• ML models maintained high accuracy (R² > 0.92) even at triglyceride levels ≥250 mg/dL, where traditional formulas faltered.

Conclusions:

• Machine learning algorithms significantly outperform conventional methods for LDL-C calculation.
• Boosting algorithms (LightGBM, Gradient Boosting, XGBoost) are highly effective for accurate LDL-C estimation.
• Implementing ML models in clinical settings can enhance cardiovascular risk stratification and patient management.