Machine Learning Model Spots Heart Disease Patients Most Likely to Carry High Lp(a)

A peer-reviewed study published in JACC: Advances has shown that a machine learning model developed by the Family Heart Foundation can reliably single out patients with atherosclerotic cardiovascular disease who are most likely to have elevated lipoprotein(a), a genetically determined and often overlooked risk factor for heart attack and stroke. The study, titled “FIND Lp(a) MLM: Targeted Screening Enrichment of Elevated Lipoprotein(a) in Atherosclerotic Cardiovascular Disease,” describes both the development and the initial validation of the FIND Lp(a) Machine Learning Model, which was built using the Foundation’s Family Heart Database, a large real-world repository of clinical information. According to the findings, patients flagged by the model were more than 2.2 times as likely to have high Lp(a), defined as a concentration of 125 nmol/L or greater, compared with the overall population of people with atherosclerotic cardiovascular disease in the database. That degree of enrichment, the authors report, demonstrates that the algorithm can help healthcare systems prioritize limited screening resources toward the patients most likely to benefit from knowing their Lp(a) status.

Lipoprotein(a) is a particle similar in structure to low-density lipoprotein cholesterol but carrying an additional protein called apolipoprotein(a), whose size varies from person to person because of genetic variation. Unlike most cardiovascular risk markers, Lp(a) levels are largely locked in at birth by genetics and remain relatively stable throughout life, which means a single measurement in adulthood can, in principle, characterize an individual’s lifelong exposure. High Lp(a) is recognized as an independent risk factor for atherosclerotic cardiovascular disease, and recent estimates cited by the Family Heart Foundation suggest that roughly one in five people carry levels high enough to matter for their long-term cardiovascular risk. Despite that prevalence, awareness of the biomarker remains strikingly low among clinicians, health system administrators, payors, and the general public, and the Foundation reports that about 99 percent of people in the United States have never had their Lp(a) measured at all.

The gap between guideline recommendations and everyday clinical practice is central to why the Foundation pursued a machine learning approach. Recently released United States dyslipidemia guidelines now recommend Lp(a) screening for all adults, a position known as universal screening. Yet translating that recommendation into routine care requires changes to electronic health record workflows, laboratory ordering practices, reimbursement structures, and clinician education, and the Foundation’s researchers argue that this integration will take many years, if not decades, to accomplish. In the interim, a targeted strategy offers a pragmatic bridge: rather than screening everyone at once, health systems can use predictive analytics to identify the patients within their existing records who are statistically most likely to have elevated Lp(a) and offer testing to them first. The new study provides evidence that this enrichment strategy can work at meaningful scale.

“Although, recently released U.S. dyslipidemia guidelines recommend Lp(a) screening for all adults, integrating this into routine clinical practice will take many years, if not decades,” said Diane MacDougall, vice president of Research at the Family Heart Foundation and principal author of the study. “The FIND Lp(a) model supports targeted screening by helping identify people most likely to have high Lp(a), accelerating the adoption of universal screening and creating more opportunities for individuals living with high Lp(a) and their healthcare teams to manage cardiovascular risk.” Her framing captures the dual purpose of the project: the model is presented not as a replacement for universal screening but as a way to accelerate progress toward it by demonstrating value, building institutional experience, and reaching high-risk patients sooner than a purely sequential rollout would allow.

Technically, the FIND Lp(a) Machine Learning Model is a central component of the Foundation’s Flag, Identify, Network, and Deliver quality improvement program, a framework that applies predictive analytics to data already stored in electronic medical records. The model examines patterns in routinely collected clinical information to estimate which adults are most likely to have high Lp(a), and health systems that implement the program then offer Lp(a) screening to the individuals the algorithm flags. Patients confirmed to have elevated levels receive appropriate clinical care and are invited to engage with the Family Heart Foundation for additional education and support, an arrangement designed to close the loop from prediction to diagnosis to management. Because the approach relies on data that health systems already collect, it avoids the need for new data capture infrastructure, which the Foundation identifies as one reason the model could be deployed in real clinical environments rather than remaining a research prototype.

The deployment record distinguishes the project from many machine learning initiatives in medicine, which frequently stall after publication without ever reaching operational use. The FIND Lp(a) model has already been successfully deployed at five large United States healthcare systems participating in the FIND Lp(a) Program, an initiative the Foundation created to catalyze Lp(a) screening nationwide. Through a Collaborative Learning Network, the Foundation and its five health system partners are prospectively validating the model’s real-world performance while sharing implementation best practices and developing sustainable approaches to screening and follow-up care. In addition to this prospective validation, the study team reports that further validation using both retrospective and prospective healthcare system data is currently ongoing, an effort intended to confirm that the enrichment observed in the Family Heart Database holds up across heterogeneous patient populations, record systems, and care settings.

The study’s methodology was classified as a data and statistical analysis investigation, and the research subject was people, reflecting the use of de-identified patient records rather than laboratory or animal models. The initial validation result, a greater than 2.2-fold enrichment of elevated Lp(a) among model-flagged patients, functions as a measure of precision: of the patients the model prioritizes, a substantially higher proportion turn out to have genuinely high Lp(a) than would be found by screening the cardiovascular disease population at random. For health system planners, that precision translates directly into efficiency. Every screening order directed at a model-flagged patient is more likely to return a clinically actionable result, reducing wasted testing, laboratory costs, and clinician time while still moving the institution toward the guideline-recommended goal of universal measurement.

The clinical rationale for finding these patients is grounded in the biology of the lipoprotein itself. Because Lp(a) is genetically determined, individuals with elevated levels cannot lower them substantially through diet or exercise in the way that other lipid measures can sometimes be modified, which makes awareness and risk management through other means especially important. Knowing one’s Lp(a) status can prompt more aggressive control of modifiable risk factors, earlier and more intensive monitoring, and appropriate family counseling, since the trait can be passed across generations. The Family Heart Foundation, founded in 2011 as the FH Foundation, has built its mission around exactly this problem, working to prevent heart attacks and strokes caused by familial hypercholesterolemia and elevated Lp(a), two common genetic disorders whose effects ripple through families and that frequently go undiagnosed until a cardiovascular event occurs.

The Foundation describes itself as a pioneer in applying real-world evidence, patient-driven advocacy, and multi-stakeholder education to inherited lipid disorders, and the FIND Lp(a) study fits that pattern by turning a large clinical database into a practical screening tool. The organization notes that it receives contributions and sponsorships from individuals, foundations, and pharmaceutical companies, and that this phase of the FIND Lp(a) initiative was funded by Novartis, although the company played no role in the study design, conduct, interpretation, or publication plans. For the broader field of cardiovascular prevention, the study offers a concrete demonstration that predictive analytics can be moved from theory into deployed clinical workflow, and that machine learning can serve as a bridge between what guidelines recommend and what health systems can realistically deliver. If ongoing validation confirms the initial results, targeted screening programs built on models like FIND Lp(a) could shorten the long road between a guideline recommendation and the day when the vast majority of Americans finally know their Lp(a) number.

Subject of Research: Machine learning-based targeted screening for elevated lipoprotein(a) in atherosclerotic cardiovascular disease

Article Title: Family Heart Foundation study demonstrates how the FIND Lp(a) machine learning model enables targeted Lp(a) screening

Article References: Family Heart Foundation study demonstrates how the FIND Lp(a) machine learning model enables targeted Lp(a) screening. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: lipoprotein(a), machine learning, cardiovascular disease, screening, Family Heart Foundation, predictive analytics, electronic health records, JACC: Advances, risk factors, guidelines, targeted screening, genetic risk