Abstract 14423: Agreement Among Four Methods for Identifying Patients With Familial Hypercholesterolemia in a Large Healthcare System

IntroductionThe diagnosis of familial hypercholesterolemia (FH) can be difficult even with the availability of genetic testing. Several practical methods exist to phenotype the disease within the electronic health records (EHR), but no gold standard for identification has been agreed upon.HypothesisThere are methods of identifying FH that may be in strong agreement with each other.MethodsA retrospective cohort study at the University of Michigan (UM) was performed. Patients with FH were identified using 4 different approaches, including LDL > 190mg/dL, ICD-10 diagnosis code E78.01, a computable Dutch Lipid Criteria (cDLC), and free-text clinical note search with expert review (TS+ER). The cDLC was the PheKB FH 2.0 algorithm using text search instead of natural language processing. The TS+ER used terms equal to “familial hypercholesterolemia” and review was done by a physician to determine FH status. Agreement analysis was performed between the 4 methods using Cohen’s kappa.ResultsWe identified 70,811 patients that had at least 1 LDL-C level measured between June 2017 and June 2018. Of these patients, 7,094 (10%) were identified as FH by at least 1 method (Figure). N=6,858 (9.7%) were identified via LDL-C>190 mg/dL, 560 (0.8%) via cDLC, 345 (0.5%) via ICD-10 diagnosis code, and 224 (0.3%) via text search. Of those identified via at least 1 method, 56 (0.8%) were identified via all 4. ICD-10 diagnosis code and TS+ER had moderate level of agreement (kappa 0.41), followed by TS+ER and the cDLC with a slight level of agreement (kappa 0.16). LDL >190 mg/dL showed poor agreement with any of the other three methods for identifying FH patients.ConclusionsThere is mixed agreement among the 4 methods. TS+ER of the EHR has the most agreement with actual ICD-10 diagnosis of FH. LDL >190 cutoff may be too sensitive to reliably identify patients with FH. cDLC has the promise to identify patients that may have FH, but more work is needed to address FH under diagnosis.