Optimizing drug-resistant epilepsy identification in the Veterans Health Administration

Accurate identification of drug-resistant epilepsy (DRE) is crucial for accurate disease measurement, effective clinical intervention and improved patient outcomes. Prior attempts to define DRE in administrative data using the 2010 International League against Epilepsy (ILAE) criteria have faced complexities.

This retrospective study utilized national administrative data from the Veterans Health Administration (VHA) to identify patients with possible DRE. This was a multicenter national cohort that uses a common, non-commercial medical record system. A panel of six epileptologists conducted chart reviews to identify DRE using the 2010 ILAE criteria. Logistic regression was used to analyze epilepsy-related variables of interest to develop algorithms identifying DRE.

Among 260 included patients, 93 (35.8 %) had DRE, 148 (56.9 %) did not have DRE, and 19 (7.3 %) were undetermined. Out of 96 algorithms assessed, the best-performing algorithm had a high accuracy (F1 score=0.726) and defined DRE as those on ≥ 3 ASMs in addition to those on ≥ 2 ASMs for ≥ 365 days with at least one intractable ICD code. The algorithm demonstrated high sensitivity (0.74), specificity (0.81), and area under the curve (AUC 0.78). Factors such as age, number of ASMs, EEG, and MRI procedures, and intractable epilepsy ICD codes were associated with DRE.

Our optimal algorithm for DRE identification is like previously published algorithms that determined the importance of number and duration of ASMs. However, it differs in the particular combination of factors that best identified DRE. These differences highlight the importance of fine-tuning algorithms for specific care settings. Further validation in a larger, more heterogenous cohort are needed to determine our algorithm's applicability and potential impact.