Machine learning-based prediction of LDL cholesterol: performance evaluation and validation.

Area of Science:

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

Background:

• Traditional formulas for estimating low-density lipoprotein cholesterol (LDL-C) have limitations, particularly in patients with hypertriglyceridemia.
• Accurate LDL-C measurement is crucial for cardiovascular risk assessment and management.

Purpose of the Study:

• To validate and optimize machine learning (ML) algorithms for precise LDL-C prediction.
• To compare the performance of various ML models against established LDL-C calculation methods.

Main Methods:

• Compared multiple ML models (Random Forest, XGB, MLP, etc.) and traditional formulas (Friedewald, Martin, Sampson) using lipid profiles from over 120,000 subjects.
• Evaluated predictive accuracy using R-squared (R²), mean squared error (MSE), and Pearson correlation coefficient (PCC).

Main Results:

• ML models significantly outperformed traditional formulas in LDL-C prediction.
• Random Forest and XGBoost models achieved the highest accuracy (R² = 0.94).
• ML models maintained high accuracy across all triglyceride levels, unlike traditional methods.

Conclusions:

• Machine learning algorithms offer superior accuracy for LDL-C estimation, especially in challenging cases like hypertriglyceridemia.
• Improved LDL-C prediction using ML can enhance cardiovascular risk stratification and personalize treatment strategies.
• These advanced models hold potential for improving patient outcomes through more informed clinical decisions.