Abstract 9822: Yield of a Familial Hypercholesterolemia Genetic Diagnosis After Electronic Health Record and Genomic Data Screening

Abstract

Background: Data mining of electronic health records (EHR) has been used as a strategy to identify patients with undiagnosed familial hypercholesterolemia (FH). Most studies have been limited by the absence of both phenotypic and genotypic data in the same cohort. Methods: Using a subset of the Geisinger MyCode Community Health Initiative (MyCode) cohort with both exome sequencing and EHR data (n=130,257), we ran two FH screening algorithms to determine genetic and phenotypic diagnostic yields: the Mayo Clinic algorithm (Mayo), which identifies those with LDL-C levels > 190 mg/dL, and FIND FH®, the Family Heart Foundation’s machine learning model, to identify individuals with phenotypes suggestive of FH. With 29,243 excluded by Mayo (for secondary causes of hypercholesterolemia, no lipid value in EHR), 52,034 excluded by FIND-FH (insufficient data to run the model), and 187 excluded for prior FH diagnosis, a final cohort of 59,729 participants screened by both algorithms was created. Genetic diagnosis was based on the presence of a pathogenic or likely pathogenic (P/LP) variant in 3 genes implicated in FH via genomic screening. Charts from 180 variant negative participants (60 controls; 120 identified by FIND FH and/or Mayo) were reviewed to calculate Dutch Lipid Clinic Network scores; a score > 5 defined probable or definite FH. Results: Mayo flagged 10,415 subjects; 164 (1.6%) had an FH P/LP variant. FIND-FH flagged 573; 28 (4.9%) had an FH P/LP variant giving a net yield from both algorithms of 167/240 (70%). Confirmation of a phenotypic diagnosis was constrained by lack of EHR data on physical findings or family history (high cholesterol, premature atherosclerotic disease) required for score calculation. Phenotypic FH by chart review was present by Mayo and/or FIND-FH in 13/120 vs 2/60 not flagged by either (p< 0.09). Conclusion: After excluding those with a prior FH diagnosis, applying two recognized phenotypic FH screening algorithms to the eligible MyCode cohort identified 70% of those with a P/LP FH variant. Limitations to this approach include participant exclusions for each algorithm, a low yield of positive genomic screening for Mayo, and a low yield of participants for FIND FH. Phenotypic diagnosis was rarely achievable due to missing data.