An algorithm for drug-resistant epilepsy in Danish national registers

Abstract

Patients with drug-resistant epilepsy (DRE) have increased risks of premature death, injuries, psychosocial dysfunction, and a reduced quality of life. Identification of persons with DRE in administrative data can allow for effective large-scale research, and we therefore aimed to construct an algorithm for identification of DRE in Danish nation-wide health registers. We used a previously generated sample of 525 persons with medical record-validated incident epilepsy between 2010-2019, of which 80 (15%) fulfilled International League Against Epilepsy (ILAE) criteria of DRE at the time of the latest contact – this cohort was considered the gold standard. We linked information in the validated cohort to Danish national health registers and constructed register-based algorithms for identification of DRE-cases. The accuracy of each algorithm was validated against the medical record-validated gold standard. We applied the best performing algorithm according to test accuracy (F1 score) to a large cohort with incident epilepsy identified in the Danish National Patient Registry between 1995 and 2013 and performed descriptive and logistic regression analyses to characterize the cohort with DRE as identified by the algorithm. The best performing algorithm in terms of F1 score was defined as ‘fillings of prescriptions for ≥ 3 distinct antiseizure medications (ASMs) within 3 years or acute hospital visit with epilepsy/convulsions following fillings of prescriptions for two distinct ASMs’ (sensitivity 0.59, specificity 0.93, positive predictive value 0.59, negative predictive value 0.92, area under the receiver operating characteristic curve 0.77, and F1 score 0.595). Applying the algorithm to a register-based cohort of 83,682 individuals with incident epilepsy yielded 8,650 cases (10.3 %) with DRE. In multivariable logistic regression analysis, early onset of epilepsy, focal or generalized epilepsy, somatic co-morbidity, and substance abuse, were independently associated with risk of being classified with DRE. We developed an algorithm for the identification of DRE in Danish national registers, which can be applied for a variety of research questions. We identified early onset of epilepsy, focal or generalized epilepsy, somatic co-morbidity, and substance abuse as risk factors for DRE.