Screening for familial hypercholesterolemia in the era of artificial intelligence: A contemporary narrative review

Familial hypercholesterolemia (FH) is a genetic condition that significantly increases the risk of coronary artery disease if left unmanaged. Although FH remains underdiagnosed globally, early detection enables timely therapy and reduces cardiovascular risk. Traditional FH criteria, such as the Dutch Lipid Clinic Network and Simon Broome diagnostic criteria, rely on clinical data often missing in real-world practice. Artificial intelligence (AI) offers potential to improve FH case identification.

To review recent advances in simplified FH screening criteria, summarize applications of AI models in FH detection, compare model performance, and evaluate limitations and advantages.

A literature search was conducted to identify peer-reviewed studies on AI in FH screening, diagnosis, or management. The search used Medical Subject Headings and free-text terms for FH and AI technologies. Titles and abstracts were reviewed, and full texts were assessed if potentially eligible.

Machine learning models generally outperform simplified criteria. Logistic regression models remain the most commonly used and are valued for interpretability. Performance comparisons between models are limited by differences in cohort characteristics. Predictive variables commonly used by models include established examples such as family history, lipid profiles, demographics, and prescription records.

AI models that leverage electronic health record (EHR) data can outperform traditional criteria, uncover missed cases, and complement cascade screening. Challenges include external validation, workflow integration, and model transparency. To ensure clinical utility, further work is needed to standardize comparison metrics, broaden applicability, increase interpretability, and address EHR integration complexities.