Abstract Body (Do not enter title and authors here): Introduction: Familial hypercholesterolemia (FH) is a genetic dyslipidemia characterized by increased low-density lipoprotein cholesterol (LDL-C) and greater risk for early atherosclerotic disease and premature mortality. Despite being a fairly common disorder, most individuals with FH are undiagnosed. To combat this disparity, the Family Heart Foundation launched the Flag, Identify, Network Deliver^TM initiative (FIND-FH), a multi-institution collaboration that integrates an artificial intelligence (AI) machine learning algorithm (MLA) into institutional structured electronic medical record (EMR) data to identify patients with high likelihood of having FH. This study describes the methods and progress to date of one academic center that is part of the FIND-FH^TM initiative.

Methods: The FIND- FH^TM MLA identified 491 patients within the EMR of a single academic center between the years of 2017-2022. Of these 491 patients, 70 were excluded. Exclusion criteria were as follows: patient deceased, duplicate medical record number, existing FH diagnosis, and patient under 18 years old. Subsequently, patients were manually risk stratified using Dutch Lipid and Simon-Broome criteria. Outreach was prioritized by risk-stratification tier, with patients being deemed “high risk” for FH the first cohort to be contacted. Patient outreach was conducted by a team of medical students and a research coordinator. Outreach was multimodal and consisted of two phone calls, a message in the EMR, and an email to the patient and a message to the patient’s primary care clinician (PCC).

Results: After manual risk stratification, patients were categorized as “established FH” (n=32), “likely FH” (n=113), or “suspected FH” (n=309). Outreach to the “likely FH” cohort was prioritized first. To date, outreach to all 113 “likely FH” patients has been completed, with 16 patients subsequently evaluated for FH in cardiology clinic. Of these 16 patients, 8 have been diagnosed with FH/probable FH and 8 have been considered possible FH. No patients have had FH ruled out.

Conclusions: MLA-integrated models provide scalable approaches for systems level screening for undiagnosed diseases. Robust outreach systems are needed to complement AI-mediated screening. Future work aims to engage the suspected FH cohort for risk stratification and clinical evaluation, as well as real-time integration and implementation of the MLA for prospective, continuous screening.