Individualised prediction model of seizure recurrence and long-term outcomes after withdrawal of antiepileptic drugs in seizure-free patients: a systematic review and individual participant data meta-analysis
Herm J Lamberink,Willem M Otte,Ada T Geerts,Milen Pavlovic,Julio Ramos-Lizana,Anthony G Marson,Jan Overweg,Letícia Sauma,Luigi M Specchio,Michael Tennison,Tania M O Cardoso,Shlomo Shinnar,Dieter Schmidt,Karin Geleijns,Kees P J Braun
DOI: https://doi.org/10.1016/S1474-4422(17)30114-X
IF: 59.935
2017-07-01
The Lancet Neurology
Abstract:BACKGROUND: People with epilepsy who became seizure-free while taking antiepileptic drugs might consider discontinuing their medication, with the possibility of increased quality of life because of the elimination of adverse events. The risk with this action, however, is seizure recurrence. The objectives of our study were to identify predictors of seizure recurrence and long-term seizure outcomes and to produce nomograms for estimation of individualised outcomes.METHODS: We did a systematic review and meta-analysis, and identified eligible articles and candidate predictors, using PubMed and Embase databases with a last update on Nov 6, 2014. Eligible articles had to report on cohorts of patients with epilepsy who were seizure-free and had started withdrawal of antiepileptic drugs; articles also had to contain information regarding seizure recurrences during and after withdrawal. We excluded surgical cohorts, reports with fewer than 30 patients, and reports on acute symptomatic seizures because these topics were beyond the scope of our objective. Risk of bias was assessed using the Quality in Prognosis Studies system. Data analysis was based on individual participant data. Survival curves and proportional hazards were computed. The strongest predictors were selected with backward selection. Models were converted to nomograms and a web-based tool to determine individual risks.FINDINGS: We identified 45 studies with 7082 patients; ten studies (22%) with 1769 patients (25%) were included in the meta-analysis. Median follow-up was 5·3 years (IQR 3·0-10·0, maximum 23 years). Prospective and retrospective studies and randomised controlled trials were included, covering non-selected and selected populations of both children and adults. Relapse occurred in 812 (46%) of 1769 patients; 136 (9%) of 1455 for whom data were available had seizures in their last year of follow-up, suggesting enduring seizure control was not regained by this timepoint. Independent predictors of seizure recurrence were epilepsy duration before remission, seizure-free interval before antiepileptic drug withdrawal, age at onset of epilepsy, history of febrile seizures, number of seizures before remission, absence of a self-limiting epilepsy syndrome, developmental delay, and epileptiform abnormality on electroencephalogram (EEG) before withdrawal. Independent predictors of seizures in the last year of follow-up were epilepsy duration before remission, seizure-free interval before antiepileptic drug withdrawal, number of antiepileptic drugs before withdrawal, female sex, family history of epilepsy, number of seizures before remission, focal seizures, and epileptiform abnormality on EEG before withdrawal. Adjusted concordance statistics were 0·65 (95% CI 0·65-0·66) for predicting seizure recurrence and 0·71 (0·70-0·71) for predicting long-term seizure freedom. Validation was stable across the individual study populations.INTERPRETATION: We present evidence-based nomograms with robust performance across populations of children and adults. The nomograms facilitate prediction of outcomes following drug withdrawal for the individual patient, including both the risk of relapse and the chance of long-term freedom from seizures. The main limitations were the absence of a control group continuing antiepileptic drug treatment and a consistent definition of long-term seizure freedom.FUNDING: Epilepsiefonds.
clinical neurology