Prediction of low-density lipoprotein cholesterol levels using machine learning methods.

Area of Science:

• Biochemistry
• Medical Informatics
• Computational Biology

Background:

• Traditional equations for calculating low-density lipoprotein cholesterol (LDL-C) often lack sufficient accuracy.
• Accurate LDL-C measurement is crucial for cardiovascular risk assessment and dyslipidemia management.

Purpose of the Study:

• To develop and evaluate a more accurate machine learning-based model for predicting LDL-C levels.
• To compare the performance of machine learning models against established LDL-C calculation equations.

Main Methods:

• Individual characteristics, lipid profiles, and laboratory results were used as predictors for a 2-step prediction model.
• Machine learning algorithms including random forest and XGBoost were employed.
• The novel 2-step model and other machine learning methods were benchmarked against Friedewald, Martin, and Sampson equations.

Main Results:

• The proposed 2-step prediction model achieved the lowest root mean squared error (RMSE) of 7.015, indicating superior accuracy.
• The 2-step model demonstrated the highest concordance rate (85.1%) with directly measured LDL-C levels.
• Machine learning models generally outperformed existing LDL-C calculation equations.

Conclusions:

• Machine learning approaches provide a more accurate method for calculating LDL-C than conventional equations.
• The novel 2-step prediction model represents a significant advancement in LDL-C estimation accuracy.