Abstracts

Rationale: Electronic health records (EHR) are powerful tools to facilitate disease surveillance, conduct population-based outcomes research, and improve the quality of care in people with epilepsy. The development of validated case definitions allows researchers to accurately identify epilepsy patients in EHRs. In this study, we evaluated the accuracy of various claim- and antiseizure medicine (ASM)-based algorithms to identify patients with epilepsy in a large racially diverse multicenter health system in New York City.

Methods: The EHR (Epic Systems Corporation) was queried to identify patients of all ages on an ASM (regardless of indication) or with an encounter coded with an epilepsy- or seizure-related International Classification of Diseases Ninth or Tenth Revisions with Clinical Modification (ICD-9-CM or ICD-10-CM) code between 2012 and 2018 within an urban academic health system containing six tertiary and community hospitals. 800 charts were randomly selected from 476781 and reviewed by trained medical students (n=2) and a neurology resident (n=1). Epilepsy diagnoses were confirmed by two board-certified epileptologists. Previously published ICD-9-CM, ICD-10-CM, and ASM-based algorithms were identified and additional algorithms were developed to enhance the accuracy of existing algorithms. Sensitivity (Sn), specificity (Sp), negative predictive value, positive predictive value, and Youden’s Index (YI) were calculated to evaluate the accuracy of the algorithms. Statistical analyses were performed using SAS 9.4.

Results: Of the 800 patients included, 511 (63.9%) were between the ages of 18-64 years old, 182 (27.7%) were Medicaid recipients, 181 (27.6%) were Medicare recipients, 424 (53%) had been seen in a neurology practice and 435 (54.4%) had epilepsy. We tested 94 algorithms. The performance metrics of the best algorithms are listed in Table 1. The algorithms with highest Sn were those including a combination of ICD-9-CM (usually a 345.x together with 780.x codes) and ICD-10-CM (G40.x, G41.x and R56.x) codes, or ICD-10-CM codes alone. These latter algorithms had no restriction on encounter type or time frame. The algorithms with the highest Sp also combined ICD-9-CM and ICD-10-CM codes, but required that these be associated with a hospitalization or multiple outpatient visits in a set time frame. The algorithm with the highest YI was an ICD-10-CM based algorithm using G40.x in the primary position or in ³2 encounters in any diagnostic position.

Conclusions: We identified EHR-based case definitions that were sensitive and specific for a diagnosis of epilepsy. The application of these claims-based definitions will allow researchers to conduct disease surveillance and assess important health outcomes in people with epilepsy.

Funding: Please list any funding that was received in support of this abstract.: N. Jette is the Bludhorn Professor of International Medicine. CS Kwon is the holder of the Leon Levy Fellowship. LJB received salary support from American Epilepsy Society, Epilepsy Foundation and the Mount Sinai Claude D Pepper Older Americans Independence Center (5P30AG028741-11).