Abstract 16599: Validation Of Flag, Identify, Network, Deliver: Find FH Using The Electronic Medical Record To Identify Familial Hypercholesterolemia Within A Single Healthcare System

IntroductionFamilial hypercholesterolemia (FH) is a common underdiagnosed and undertreated condition that leads to premature cardiovascular disease. A machine learning algorithm (MLA) uses artificial intelligence technology to screen for FH. We validated the use of an MLA ‘FIND FH,’ developed by the FH Foundation, by determining the relationship between the FIND FH score (the output of the algorithm) and either an FH clinical diagnosis or FH-causing mutation in the University of Pennsylvania Healthcare System (UPHS).MethodsFIND FH was trained to detect FH using clinically and genetically diagnosed FH patients from four health systems. Diagnostic performance for FH was evaluated using patients with a cardiovascular co-morbidity at UPHS. Of 700,701 individuals, 181,107 had a FIND FH score above 0.0. The patients were assigned to five pre-defined strata based on FIND FH score‘A’ (≥0.35), ‘B’ (0.20-0.35), ‘C’ (0.16-0.19), ‘D’ (0.06-0.15), and ‘E’ (0.0-0.05). While blinded to genetic results, two lipidologists reviewed medical charts on a sample of patients per strata to establish a clinical diagnosis of FH. Genetic testing was independently performed on these patients by Grifols. A chi-squared analysis and regression model was used to determine the relationship between FIND FH score (strata, continuous) and FH clinical or genetic diagnosis.ResultsIn the validation dataset (n = 414 patients; mean [SD] age, 58.2 [14.6] years; 54% male; 79% white), the prevalence of FH was 33% in strata A (n=109), 25% in strata B (n=109), 19% strata C (n=98), 10% in strata D (n=52), and 2% in strata E (n=46). The relationship between FIND FH score and an FH clinical diagnosis was significant per strata (p-value