Validated operational case definitions of drug-resistant epilepsy across health systems: a systematic review

Drug-resistant epilepsy (DRE) affects nearly one-third of people with epilepsy and carries substantial clinical and public health burden. Although a 2010 International League Against Epilepsy (ILAE) consensus definition exists, its implementation and validation vary across health systems resulting in inconsistent patient identification. We conducted a systematic review to synthesize validated operational case definitions of DRE that have been applied in clinical, administrative and research settings, comparing their validation methods, performance and applications.

Following PROSPERO registration (CRD420251074563), we systematically searched MEDLINE, Embase and PsycINFO (through July 21, 2025). Eligible studies formally validated operationalized DRE definitions using clinical outcomes, administrative or electronic health records (EHR) data or expert consensus. Data extraction and risk of bias assessment (QUADAS-2) were performed in duplicate. A narrative synthesis summarized validation approaches, performance metrics, and health system applications.

Of 1,720 records screened, 15 met inclusion criteria. Approaches included (1) reliability/reappraisal of consensus definitions (n = 5), (2) administrative/EHR computable phenotypes (n = 4), (3) predictive models using clinical, EEG, or imaging (n = 4) and (4) syndrome-specific tools (n = 2). Administrative algorithms achieved moderate discrimination (AUC ∼ 0.78-0.80), while multimodal predictive models incorporating EEG/MRI radiomics reported higher performance (AUC ∼ 0.98). Reliability studies supported ILAE framework but highlighted misclassification and variability across epilepsy types. Syndrome-specific reappraisals suggested different ASM thresholds by epilepsy type.

This review demonstrates that DRE is identified in practice through diverse validated operationalizations of existing definitions rather than a single uniform approach. Administrative/EHR algorithms enable population surveillance but with tradeoffs in accuracy, whereas multimodal predictive models show promise but require external validation. Syndrome-specific refinements to consensus definitions may enhance predictive validity. Harmonization of validated case definitions across health systems is critical to ensure consistency in clinical care, research and policy applications.