Abstracts

Rationale: Administrative ICD-9 and ICD-10 data have been used in Canada for surveillance of chronic conditions such as diabetes. We have previously validated epilepsy ICD-9 and ICD-10 coding from inpatient and emergency visit databases in a large Canadian region. However, a majority of epilepsy patients are managed in outpatient clinics. We therefore conducted this study to: (1) Develop and validate coding algorithms for epilepsy using inpatient and physician claims data (captures both inpatient and outpatient visits); and (2) Assess whether adding an emergency room (ER) database to the inpatient and physician claims databases enhances the epilepsy case validity. Methods: 720/2049 charts (35% of all visits) from 2003 and 1533/3252 visits (47% of all visits) from 2006 were randomly selected for review from 13 neurologists practices as the gold standard for diagnosis. Epilepsy status in each chart was determined by 2 trained physicians with epilepsy management expertise. The optimal algorithm to identify epilepsy cases was then developed by linking the reviewed charts with the following administrative databases (ICD-9 and ICD-10 data from 2000-2008): a provincial health care insurance plan registry, a hospital discharge abstract database, an ER visits database, and a physician claims database in a large Canadian health region (Calgary). Accepting chart review data as the gold standard, we calculated sensitivity (Sn), specificity (Sp), positive predictive value (PPV) and negative predictive value (NPV) for each ICD-9 and ICD-10 administrative data algorithm (case definitions). Results: Of 2253 charts reviewed, 52% and 48% of charts were from epilepsy and neurology clinics respectively. 44% of charts reviewed represented epilepsy cases, 1% convulsion and 55% other diagnoses. The most commonly coded ICD-9 diagnoses based on chart review were 345.4 (localization-related epilepsy and epilepsy syndromes with complex partial seizures = 22%) and 345.1 (generalized convulsive epilepsy = 14%). Of 18 algorithms assessed, the best coding algorithm to identify epilepsy cases was 2 physician claims in 2 years or 1 hospitalization coded with an ICD-9 or ICD-10 epilepsy code (345 and G40/G41, respectively): (1) algorithm tested in 2002-2004: Sn 88.9%, Sp 92.4%, PPV 89.2%, NPV 92.2%; (2) algorithm tested in 2005-2007: Sn 93.1%, Sp 93.0%, PPV 91.9%, NPV 94.0%. Adding the ER database resulted in improved Sn and NPV but with lower Sp and PPV: (1) 2002-2004: Sn 95.6%, Sp 87.4%, PPV 84.3%, NPV 96.6%; (2) 2005-2007: Sn 99.3%, Sp 84.2%, PPV 84.3%, NPV 99.3%. Conclusions: A majority of epilepsy cases can be accurately identified in administrative data using the following case definition: 2 physician claims within 2 years or 1 hospitalization (discharge abstract record) coded with the epilepsy ICD-9 code 345 or ICD-10 codes G40 or G41. Validity of administrative data in recording epilepsy improved over time.