Abstracts

Rationale: Epilepsy is a serious neurological condition prone to misdiagnosis, particularly in older patients who may display non-specific symptoms such as altered awareness or motor disturbances. Epilepsy is also prone to over-treatment because of both patient and clinician concerns about future seizures. Assessing the epidemiology of epilepsy and the quality of its diagnosis and care, however, requires accurate information from large populations. We examined the validity of coding used for Medicare claims. Methods: We identified all Medicare beneficiaries between 2006-13 who had an outpatient diagnosis potentially suggestive of seizures (e.g., epilepsy, convulsions, syncope, or collapse), were older than 59 years in 2006, and had at least one outpatient visit at the MGH Neurology Department (n=225). We abstracted the entire electronic health record (EHR) from this sample and we classified seizures based on the examination of physician’s conclusions in the note and the epilepsy definition and classification proposed in 2014 by the International League Against Epilepsy (ILAE). Using the EHR-based classification as the gold standard, we assessed the “test” performance of individual ICD-9 codes for identifying patients with non-epileptic seizures, e.g., sensitivity, positive predictive value, and C-statistic. Results: Among the 225 patients, we had complete EHR and ICD-9 information on 96% of the patients. Among all patients, 106 (47%) had epileptic seizures, 90 (40%) had non-epileptic seizures, and 21 (9%) had neither. Among the 38 patients with epilepsy and recurrent seizures (ICD-9 diagnosis 345.xx), 73% had the diagnosis confirmed in the EHR. Among the 187 patients with other ICD-9 diagnosis, including convulsions, syncope, or collapse (780.1-3x, 780.9), 44% had the diagnosis of non-epileptic seizures confirmed in the EHR. The most sensitive and specific ICD-9 code for differentiating between patients with epileptic vs. non-epileptic seizures was 780.2, named “syncope or collapse” (Sens. 73%, CI 63-82%, Spec. 90%, CI 82-95%, ROC 0.8 CI 0.8-0.9, PPV 86%, NPV 80%). Conclusions: A substantial proportion of Medicare beneficiaries with epileptic or non-epileptic seizures are misrepresented in claims-based studies. Differentiating between non-epileptic and epileptic seizures based on a single code is challenging, and a diagnostic algorithm weighing codes by their accuracy may be necessary to enhance the performance of existing claims-based algorithms. Funding: None