Antiepileptic drug withdrawal in medically and surgically treated patients: a meta-analysis of seizure recurrence and systematic review of its predictors

Aim. Many seizure-free patients consider withdrawal of antiepileptic drugs, both when seizure control is achieved by medication alone, or once they became seizure-free following epilepsy surgery. The risk of recurrence is consequently of very important prognostic value. However, estimations of recurrence risks are outdated for both populations. In addition, although many publications have reported predictors of seizure relapse, no comprehensive overview of prognostic factors is available. Methods. A systematic review of the databases of PubMed and EMBASE was conducted, identifying articles on antiepileptic drug withdrawal in patient cohorts. Recurrence risk meta-analyses were performed for both populations at one, two, three to four, and five or more years of follow-up. Within the selected articles, studies presenting multivariable analysis of predictors were identified; all studied predictors were listed, as well as all significant independent predictors. The quality of separate analyses of predictors was assessed. Results. There was no significant difference of long-term cumulative recurrence risk between surgical and medication-only populations, with respectively 29% and 34% recurrences. In medication-only treated patients, 25 factors have been reported as significant independent predictors; 12 have been reported in surgical cohorts. The quality of most analyses of predictors was low to moderate. No predictor was consistently found among all analyses, and for most predictors, study results were contradictory. Conclusion. No consistent set of predictors could be identified because a large number of variables have been identified in the literature, many studies reported contradicting results, study populations varied considerably, and the quality of the original studies was often low. Meta-analysis of individual participant data is necessary, because it allows for (1) correction for differences in follow-up duration between subjects and studies, (2) a study of interaction effects, (3) calculation of more accurate estimates valid across several populations, and (4) the assessment of each predictor's effect size.