Abstracts

Rationale: Algorithms to identify incident cases of epilepsy using claims data have never been investigated. This study was designed to determine optimal enrollment duration to accurately identify incident cases of epilepsy in administrative data. Methods: We performed a retrospective dynamic cohort study using Ohio Medicaid data from 1992-2006. Individuals were identified as having epilepsy if they had ≥2 claims of epilepsy (ICD-9-CM: 345.xx) or ≥3 claims of convulsion (ICD-9-CM: 780.3 or 780.39), and ≥2 claims of antiepileptic drugs. Each of the diagnosis or pharmacy claims had to be >30 days apart (Figure 1). Individuals were considered incident cases if they met the epilepsy case definition and had at least 1 year of follow-up prior to epilepsy diagnosis (epilepsy-free (EF) interval). Epilepsy incidence was then examined, extending the EF interval to a maximum of 8 years, the greatest interval that we could obtain stable estimates. To determine the optimal duration of EF period that minimizes misclassifying prevalent cases as incident cases, we computed an overestimation of incidence by comparing the incidence in each interval to the incidence for the 5-year interval. We used incidence rate of 5-year EF interval as a reference because this incidence rate among people without pre-existing disability was comparable to that obtained from a recent meta-analysis of population-based studies. The incidence overestimation was then plotted against the EF intervals to identify the optimal EF duration. We also stratified the analysis on pre-existing disability status. Results: Of the 318,123 Medicaid enrollees, 9,661 people met the epilepsy case definition during the study period. Of those, 5,037 individuals were incident cases when there was 1 year of EF interval. As the length of EF interval increased, the incidence rates decreased. Incidence rates in subjects with 1- and 8-year EF interval were 5.50/1,000 and 2.65/1,000 person-years, respectively. Within each interval, the incidence among people without pre-existing disability was lower than that of people with pre-existing disability. Figure 2 shows the plot of the incidence overestimation by EF intervals ranging from 1 to 4 years. As compared to the 5-year EF interval, the 1- and 2-year interval resulted in an overestimation of the incidence by 53.6% and 26.8%, respectively. The incidence overestimation declined to 13.1% and 7.4% when using 3- and 4-year EF interval, respectively. A sharper initial decline in incidence overestimation as EF interval increased was also noted among people with and without pre-existing disability. Conclusions: The 5-year EF interval appears to be the most appropriate to differentiate incident from prevalent cases in administrative data. If this is not possible, 3- or 4-year EF interval might be adequate. Shorter or longer EF intervals could result in over- or under-estimation of epilepsy incidence. This study was approved by the ODJFS and supported by the Epilepsy Foundation.