AI Model Effectively Identifies People with Probable FH

AI Model Effectively Identifies People with Probable FH
The FH Foundation’s FIND FH® initiative makes precision screening possible to identify the 90% of people with FH who are undiagnosed.

The FH Foundation, a research and advocacy organization focused on familial hypercholesterolemia (FH), published a study today showing that a machine learning model it developed effectively identified individuals with probable FH for the first time at a national scale.

The FH Foundation study was designed to validate and implement the FIND (Flag, Identify, Network, Deliver) FH model, developed as part of the Foundation’s FIND FH initiative, designed to advance precision screening in order to identify the 90% of people with FH who are undiagnosed.

The FIND FH model is designed to scan large, diverse, and disparate healthcare encounter databases to flag individuals with probable FH. According to the study, the model proved to be successful, performing comparably with high precision across two distinct types of healthcare data: a national healthcare encounter database and the Oregon Health & Science University (OHSU) electronic health records (EHR) database.

The study, Precision screening for familial hypercholesterolaemia: a machine learning study applied to electronic health encounter data, was published today online in The Lancet Digital Health and simultaneously presented at the 2019 FH Global Summit in Atlanta.

According to the Foundation, precision screening can help healthcare systems or providers by enabling them to focus on screening individuals most at risk, and thus more likely to benefit from early diagnosis and intervention. Population-based screening for FH in the U.S. has not been widely implemented, the Foundation asserted, even though screening to identify families with FH is recommended by the American Heart Association, American College of Cardiology, Centers for Disease Control and Prevention, and the World Health Organization.

“Since diagnosis of this deadly but treatable condition has stalled in the American medical system, the FH Foundation harnessed artificial intelligence and big data to accelerate identification of those most likely to have FH,” Katherine Wilemon, a co-author of the study and the founder and CEO of the FH Foundation, said in a statement.

The FIND FH model is designed to leverage a healthcare database that has been developed over the past six years by the FH Foundation and includes national healthcare encounter data on over 272 million individuals in the U.S. being treated or evaluated for cardiovascular disease. The initiative also includes a HIPAA-compliant outreach program targeting healthcare providers so they can receive identification of the individuals with probable FH in their practice.

In the study, researchers flagged 1,331,759 of 170,416,201 patients in the national database and 866 of 173,733 individuals in the health-care delivery system dataset as likely to have FH. Experts specializing in FH reviewed a sample of the flagged individuals (45 from the national database and 103 from the health-care delivery system dataset), applying clinical FH diagnostic criteria.

The FIND FH screening algorithm correctly identified individuals with probable FH—and thus having a high enough clinical suspicion of FH to warrant guideline-based clinical evaluation and treatment—87% of the time in the national database and 77% of the time in the OHSU HER, the Foundation said.

“The FIND FH model performed equally well on data from integrated healthcare systems as well as a large national healthcare encounter database, suggesting that it will be applicable to implementation at a variety of healthcare institutions and organizations,” added Kelly Myers, the FH Foundation’s chief technology officer and the study’s corresponding author. “The FIND FH model carries the promise of efficiently identifying many of the over 1 million undiagnosed individuals with FH in the U.S.”

The model significantly reduced the number of people needed to be screened, according to the Foundation. Yet, by using clinical data such as medication use and medical history, the FIND FH model also identified individuals who would be overlooked by relying just on LDL-cholesterol cut-offs often used for identifying people with the condition, such as greater than 190 mg/dL.

“Precision screening for FH is now a reality in any healthcare system with electronic health records,” Daniel J. Rader, M.D., chair of the department of Genetics in the Perelman School of Medicine at the University of Pennsylvania, and a co-author of the study, said in a statement.

“We no longer need to screen everyone to find individuals who are at genetic risk for heart attacks and strokes. After further clinical evaluation, if an FH diagnosis is made, it will trigger screening of relatives as well,” added Rader, who is also the FH Foundation’s chief scientific advisor and a member of its Board of Directors. “While FH is manageable, the greatest benefit is from treatment earlier in life.”