Integrated Clinical, Molecular, and Machine Learning Assessment of Familial Hypercholesterolemia

Area of Science:

• Genetics and genomics
• Cardiovascular medicine
• Artificial intelligence in healthcare

Background:

• Familial hypercholesterolemia (FH) diagnosis can be challenging due to phenotypic overlap with other dyslipidemias.
• Current rule-based criteria like Dutch Lipid Clinic Network (DLCN) and Simon Broome (SB) show inconsistent concordance with molecular confirmation.
• Limited genetic testing availability in some regions necessitates reliable clinical diagnostic tools.

Purpose of the Study:

• To investigate phenotype-related discordance between clinical FH criteria and molecular genetic data in a Turkish cohort.
• To evaluate the efficacy of machine learning (ML) models in predicting pathogenic/likely pathogenic variant positivity in suspected FH.
• To compare the performance of ML models against established DLCN and SB criteria for FH diagnosis.

Main Methods:

• Targeted next-generation sequencing of a 9-gene panel was performed on patients with suspected familial hyperlipidemia.
• Machine learning models (Elastic-net, Random Forest, XGBoost) were trained on clinical variables using a stratified split of FH cases with definitive molecular status.
• Performance was compared between ML models and dichotomized SB and DLCN criteria using AUC values.

Main Results:

• Simon Broome (SB) criteria showed higher false-positive rates in mixed dyslipidemia compared to FH, with lower molecular positivity.
• ML models demonstrated superior discrimination for pathogenic/likely pathogenic variant positivity compared to SB and DLCN criteria (XGBoost AUC: 0.808).
• Thirteen novel variants, predominantly in LDLR, were identified, highlighting potential new genetic findings.

Conclusions:

• Machine learning models offer improved prediction of pathogenic/likely pathogenic variants in clinically defined FH cases compared to DLCN and SB criteria in this Turkish cohort.
• ML-based risk stratification can potentially enhance the prioritization of patients for genetic testing.
• External validation of ML models is recommended for broader clinical applicability.