Machine learning to predict elevated lipoprotein(a)

Lipoprotein(a), Lp(a), is a genetically determined, lifelong risk factor for atherosclerotic cardiovascular disease (ASCVD). Despite broad guideline support for universal one-time testing, Lp(a) measurement remains rare in clinical practice. This review summarizes recent advances in machine learning-based strategies that can enhance the efficiency, yield, and equity of Lp(a) screening.

To date, three studies have developed and validated machine learning models to identify individuals with elevated Lp(a) using routinely available clinical variables. The ARISE framework, derived from the UK Biobank and validated across multiple US cohorts, reduced the number needed to test by more than 50% while maintaining consistent discrimination across demographic subgroups. Additional studies have confirmed the feasibility of decision-tree and neural network models to improve case finding for elevated Lp(a) in both clinical and population-based settings.

Machine learning-based strategies provide a scalable means of operationalizing universal Lp(a) testing recommendations within health systems. When developed using unbiased data, externally validated, and assessed for fairness and interpretability, these models can support systematic identification of individuals with elevated Lp(a) and integration of Lp(a) measurement into routine cardiovascular risk assessment.