Machine learning modelling of blood lipid biomarkers in familial hypercholesterolaemia versus polygenic/environmental...

Area of Science:

• Biochemistry and Genetics
• Cardiovascular Disease Research
• Machine Learning in Healthcare

Background:

• Familial hypercholesterolaemia (FH) elevates low-density lipoprotein cholesterol (LDL-C), causing premature cardiovascular disease.
• Current genetic screening for FH has limited yield, as most hyperlipidaemic individuals have polygenic forms.
• Improved identification of monogenic FH is crucial for effective cascade screening and treatment.

Purpose of the Study:

• To identify novel biomarkers for improved FH diagnosis.
• To develop advanced machine-learning models for classifying monogenic FH.
• To enhance the specificity and yield of genetic screening programs for FH.

Main Methods:

• Utilized a machine-learning approach on a pediatric dataset with extended lipid profiles and genetic testing results.
• Developed classification models incorporating traditional and novel lipid parameters.
• Compared model performance against existing clinical diagnostic criteria and simple lipid cut-offs.

Main Results:

• Machine-learning models achieved significantly higher specificity in classifying FH patients compared to current methods.
• Novel parameters like apoB/apoA-I, TG/apoB, and LDL1 improved monogenic FH identification.
• Models using only total cholesterol (TC) and LDL-C showed enhanced specificity over simple cut-offs.

Conclusions:

• Advanced machine learning models offer superior classification accuracy for monogenic FH.
• Incorporating novel lipid parameters significantly improves the identification of individuals with monogenic FH.
• These findings can optimize genetic screening programs, increasing yield and reducing costs.