Objective: Seizure recurrence in new onset epilepsy (NOE) can result from various factors. Although drug ineffectiveness is frequently investigated, other causes-such as nonadherence, inadequate treatment, nonepileptic events (e.g., functional/dissociative), or acute symptomatic seizures-also impact patient outcomes. This retrospective, longitudinal single-center study evaluated relapse causes over a 5-year period in a cohort of NOE patients.
Methods: We reviewed the NOE registry of Geneva University Hospital, including adult patients diagnosed between 2005 and 2019. Of 740 consecutive patients, 330 with more than 5 years of follow-up were included in the analysis. Relapses after antiseizure medication (ASM) initiation were classified into seven categories: poor adherence, inadequate treatment, ineffective treatment, acute symptomatic seizures, nonepileptic seizures, multiple, and unknown causes. Mixed-effects logistic regression and post hoc analyses were used to assess temporal trends.
Results: A total of 202 of 330 patients (61%) experienced at least one relapse; 181 occurred after ASM initiation. Of these, 32% (57/181) were due to treatment ineffectiveness, whereas 68% were attributable to other causes. Inadequate treatment was the most frequent cause in year 1 (37%) but declined to 10% by year 5 (p < .001). Poor adherence remained stable (27%-38%). The proportion of drug ineffectiveness as a cause of relapse increased over time (22% in year 1 to 41% in year 5). Drug-resistant epilepsy (DRE) was diagnosed in 7% of the entire cohort. Generalized tonic-clonic (GTC) seizures at relapse were associated with poor adherence, whereas non-GTC seizures were linked to ineffectiveness (p < .001).
Significance: In our cohort, most early relapses in NOE were due to modifiable factors. Systematic identification of relapse causes can prevent misclassification of DRE and reduce unnecessary treatment escalation. The high proportion of patients lost to follow-up is a relevant limitation of this study; however, a substantial selection bias is unlikely.
