Address correspondence and reprint requests to Morten Ingvar Lossius National Centre for Epilepsy, P.O. Box 53, 1306 Baerum postterminal, Norway. E-mail: Morten.email@example.com
Objective: Despite side effects associated with the use of antiepileptic drugs (AEDs), withdrawal of AEDs remains controversial, even after prolonged seizure freedom. The main objective of this study was to assess the effects of AED withdrawal on cognitive functions, seizure relapse, health-related quality of life (HRQOL), and EEG results. Additionally, potential predictors for freedom from seizures after AED withdrawal were studied.
Methods: Patients, seizure-free for more than 2 years on AED monotherapy, were recruited for a controlled, prospective, randomized, double-blinded withdrawal study lasting for 12 months, or until seizure relapse. Patients were randomized to AED withdrawal (n = 79) and nonwithdrawal (n = 81) groups. The examination program included clinical neurological examinations, neuropsychological testing, EEG-recordings, cerebral MRI, and assessments of HRQOL.
Follow-up data on seizure relapse were also collected beyond the 12-month study period (median 47 months).
Results: Seizure relapse at 12 months occurred in 15% of the withdrawal group and 7% of the nonwithdrawal group (RR 2.46; 95% CI: 0.85–7.08; p = 0.095). After withdrawal, seizure relapse rates were 27% after a median of 41 months off medication. A normal result to all 15 neuropsychological tests increased significantly from 11% to 28% postwithdrawal. We found no significant effects of withdrawal on quality of life and EEG. Predictors for remaining seizure-free after AED-withdrawal over 1 year were normal neurological examination and use of carbamazepine prior to withdrawal.
Conclusion: Seizure-free epilepsy patients on AED monotherapy who taper their medication may improve neuropsychological performance with a relative risk of seizure relapse of 2.46, compared to those continuing therapy.
Approximately 70% of patients with recent onset epilepsy become seizure-free on antiepileptic drugs (AEDs) (Kwan and Brodie, 2000), although many of them experience different drug side effects, some of which are serious (Specchio and Beghi, 2004). Reduced attention, mental speed, psychomotor slowing, and impaired memory (Vermeulen and Aldenkamp, 1995) are subtle side effects, with the potential to create problems during activities at work and in daily life. Despite the favorable reaction of side effects to AED withdrawal, this procedure remains controversial, even in persons with prolonged seizure freedom, because of the fear of provoking seizure relapse.
The risk-benefit aspect of AED withdrawal, in particularly the neuropschychological effects, has not been fully explored as this information can best be obtained through prospective, randomized, double-blind trials among seizure-free patients. Our primary aim was to conduct such a study, assessing effects of AED withdrawal on seizure relapse rate and possible changes in cognitive functions. Secondarily, we wanted to study possible effects on health related quality of life (HRQOL), and EEG-findings. In brief: Which beneficial effects can be offered by AED-withdrawal and at what risk?
The anticipation of a difference in seizure relapse rates between the withdrawal and nonwithdrawal arms of the study enabled us to minimize the study population for ethical reasons. With this in mind, and in the absence of guidelines on what constitutes a “significant change” in neuropsychological scores, it was decided that a 0.5 standard deviation change in neuropsychological scores would probably be of clinical interest. Based on a Type I error of 0.05 and a Type II error of 0.90, 170 patients would be needed for the study, while a Type II error of 0.80 would require 138 patients. After discussions with our Ethical Committee, it was decided to include between 160 and 180 patients, anticipating a dropout rate of about 10%.
Our legal adviser formulated a legal document referring to possible claims arising from injuries connected to an epileptic seizure, and also clarifying the responsibilities of the Institute.
To increase patient safety further, two sets of coded envelopes were made, one of which was kept in the patient record and the other given to each patient, for use in cases of emergency when it might be important to know whether a patient was on drug therapy or placebo. Envelopes were to be kept sealed and unopened until study termination (to be opened if seizures occurred) when they were to be returned. Given the adverse effects of AEDs we cannot rule out that some patients may have been aware of the treatment policy but this was not systematically assessed.
The Regional Committee for Medical Research Ethics approved the study protocol (Approval S-127/99–99044).
Patients were selected from the epilepsy registries of our hospital and six neurological outpatient clinics in the Oslo area. Inclusion was stopped when 168 had given their informed, written consent to participate. Table 1 lists the inclusion criteria. Figure 1 indicates how the 241 patients, assessed for eligibility, were reduced to 160 who were finally included for randomization (79 to withdrawal and 81 to continued medication). None of the patients included had had epilepsy surgery. In the period between randomization and the start of the intervention, six patients withdrew themselves, two had generalized epileptic discharges on EEG, one experienced seizures while still on his AED, and one suffered a subarachnoid hemorrhage. Thus, 150 patients were included in the intervention part of the study (78 in the nonwithdrawal and 72 in the withdrawal group). Due to protocol violation, one patient was excluded from the analyses.
Table 1. Inclusion and exclusion criteria
Epilepsy (at least two unprovoked seizures)
Two years seizure freedom or longer
Age: 18–67 years
If prior withdrawal attempted and unsuccessful, five years seizure freedom or longer is needed
Juvenile myoclonic epilepsy (JME)
Paroxysmal epileptiform activity in patients with primarily generalized epilepsy
Two prior withdrawal attempts
Pregnant or seeking pregnancy
Progressive neurological disease
Other serious disease which may influence the health status of the patient in the study period
Inclusion started in October 1999 and ended in March 2004.
The study followed a prospective, randomized controlled, double-blind design. Patients came for five visits over a 12-month period, or until seizure relapse. Those experiencing seizure relapse or serious diseases, mandating code breaking, received appropriate treatment. All those randomized to the nonwithdrawal group were offered AED-withdrawal, controlled by the primary investigator, after study termination.
The patients were randomized in blocks of 10, between visit one and visit two, to receive blindly active medication or placebo in prepacked dispensers (dosett, Item Development AB, Stocksund, Sweden), one for each of the 12 withdrawal weeks. Those randomized to withdrawal had AED dose reduction by 20% of the initial dose in the first 6 weeks and 20% of the initial dose every second week over the following weeks. The reduced medication was substituted with placebo to keep the study double-blinded.
2(At one month); intervention started (tablet supply).
3(At three months); clinical control visit (tablet supply).
4(At seven months); repeat of all investigations performed at visit 1, except MRI.
5(At 12 months); study terminated, code-breaking.
In September 2006, we collected additional follow-up data on medication and seizures for all but three patients, who could not be reached despite extensive efforts. Since the study period, four patients had died; two of serious somatic disease not connected to epilepsy and two were apparently sudden unexpected death in epilepsy. Of these, one died only a few weeks after withdrawal and one died 4 years after withdrawal.
Median follow-up was 47 months, and 41 months for patients off medication.
10–13: Simple reaction time with both hands and choice reaction time with lexical and figurative stimuli (Miller, 1990).
14–15: Motor speed and complex coordination with both hands (Kløve, 1963).
All tests have previously been validated thoroughly, and for all tests normative data are defined (Kløve, 1963; Miller, 1990; Heaton et al., 1991). When differentiating between normal and abnormal test scores, a T-score >39 was used as normal (Heaton et al., 1991). A sum score was created, based on the proportion of patients scoring normal on all 15 tests before and after intervention.
EEGs were recorded according to international 10–20 electrode placement system (16 channels). The EEG variables registered were: epileptiform activity, focal epileptiform activity, generalized epileptiform activity, abnormal activity others than epileptiform.
The primary outcomes were seizure relapse and changes in cognitive function. Secondary outcomes were changes in HRQOL and EEG.
Seizure relapse was assed continually and cognitive function, HRQOL and EEG were assessed at inclusion and 7 months after intervention started, that is, 8 months after inclusion and 4 months after end of withdrawal.
Student's t-test and chi-square test were used for testing group differences for continuous and categorical variables, respectively. For occurrence of seizures, a Kaplan–Meier curve with log-rank test was used. Intention to treat analyses was performed. Odds ratios for seizure relapse were estimated with logistic regression analyses. To assess possible predictors for seizure relapse in the withdrawal group, we tested variables of importance in logistic bivariate models. The independent variables were carbamazepine (CBZ) versus other medications, normal neurological examination versus abnormal, partial versus generalized epilepsy, age, gender, duration of seizure freedom, MRI pathology versus no pathology, and age at epilepsy onset.
Since only 11 events (seizure relapse) occurred among those who had withdrawn AEDs, the number of variables in the multivariate logistic model was limited to two (neurological examination that is no sensory, motor, or cranial nerve deficits, and type of medication).
We calculated the overall risk for seizure relapse after withdrawal based on a total of patients that had ended medication both within the study period (12 months) and at study termination (median follow-up 47 months). All reported p-values were 2-sided and not adjusted for multiple testing. No interim analysis was scheduled or performed. Statistical analyses were performed using SPSS, version 14.
Among those who were found to fulfill the inclusion/exclusion criteria, but chose not participate (n = 28), 13 were women (46%), and mean age was 38 years (range: 19–64 years).
There were no significant differences between the withdrawal and nonwithdrawal groups for the following baseline variables: age, gender, epilepsy onset, duration of freedom from seizures, seizure type, MRI pathology, etiology, epileptic activity on the EEG, serum concentration of AED within therapeutic range, and medication (Table 2).
Table 2. Baseline characteristics of study patients
No withdrawal (n = 78)
Withdrawal (n = 72)
NA = not applicable.
Mean age (range)
Epilepsy onset; no. of patients (percent)
Seizure-free; no. of patients (percent)
Epilepsy type; no. of patients (percent)
Seizure type; no. of patients (percent)
Secondarily generalized tonic–clonic seizures
Complex partial seizures
Simple partial seizures
Primarily generalized tonic–clonic seizures
No. of patients (percent)
Normal neurological status
Epileptic activity on the EEG
Serum concentration in therapeutic range
Medication; no. of patients (percent)
Within 12 months, five of the 77 nonwithdrawers (7%) and 11 of the 72 withdrawers (15%) had experienced seizure relapse (RR 2.46; 95% CI: 0.85–7.08; p = 0.095) (Figure 2). The seizures caused no serious harm to the patients in the study period (12 months). One of the 11 experiencing seizure relapse in the withdrawal group, had a seizure provoked by extreme physical exertion after discontinuing CBZ-therapy. He refused to restart medication. The intention-to-treat analyses showed no significant effects on the estimates.
At termination of the double-blind period (12 months), 64 of the 72 patients in the nonwithdrawal group (those that had not experienced seizure relapse during the study) decided to taper their AEDs (i.e., eight chose to remain on medication). Four of 61 in the intervention group (11 had seizure relapse in the randomized controlled trial (RCT) and one patient was not applicable for follow-up) chose, despite no seizure relapse, to recommence their AEDs.
The 1-year risk of seizure relapse among the 60 patients in the nonwithdrawal group, who then stopped their AED-medication subsequent to the double-blind period, was similar to that observed in the intervention group during the first year (Figure 2). Eleven patients experienced seizure relapse after withdrawal in the nonwithdrawal group 1 year after withdrawal started after study termination, that is, 9 months after end of withdrawal.
Among the combined group of patients who tapered their AEDs with a median follow-up of 41 months off medication, we found a declining monthly risk of seizures from 0.010 (95% CI: 0.003–0.017) immediately after tapering to 0.009 (95% CI: 0.002–0.016) at 6 months, 0.009 (95% CI: 0.002–0.015) at 12 months, 0.006 (95% CI: 0.002–0.010) at 24 months, and 0.003 (95% CI: 0.000–0.007) at 36 months.
Improvement in neuropsychological function from first to second assessment (4 months after end of withdrawal) was evident for all tests in both groups. Apart from for two tests, the improvement was greater in the withdrawal group than in the nonwithdrawal group. Table 3 shows the five tests with significant differences between the groups; verbal fluency under time pressure, complex motor coordination with nondominant hand, response inhibition under time pressure, choice reaction time with lexical stimuli, and form discrimination. Further details on the two latter have now been published (Hessen et al., 2006).
Table 3. Changes in main outcome measures recorded before and seven months after intervention
Mean difference in changes between the two groups (CI)
at-Testbchi-square testcpublished elsewhere; NA, not applicable.
No. of patients with all normal neuropsychological tests (percent)
Raw scores of:
FAS-Word fluency (words) mean (SD)
Stroop Color–Word (words) mean (SD)
Motor test, nondominant hand (secs) mean (SD)
−3.66 (−6.64 to −0.68)
Choice reaction time-lexical (ms) mean (SD)c
−24.42 (−40.11 to −8.74)
Form Discrimination (ms) mean (SD)c
−43.05 (−76.97 to −9.14)
Quality of life parameters
QOLIE-89 overall score mean (SD)
0.3 (−1.55 to 2.07)
15D utility score mean (SD)
−0.011 (−0.027 to 0.005)
EQ visual analog scale mean (SD)
3.03 (−0.99 to 7.06)
EEG-findings; no. of
Epileptiform activity on EEG
A normal result to all 15 tests increased from 11% to 28% postwithdrawal (Table 3) contrasting with a decrease in the proportion of normal results for all tests, from 11% to 9%, in the nonwithdrawal group (Table 3). The beneficial effect was similar in all age groups, but significantly better in patients with serum CBZ-values above median compared to those below median (p = 0.03). A more detailed presentation on neuropsychological effects of AEDs has been published (Hessen et al., 2007).
Quality of life
No significant group differences were observed (Table 3).
Abnormal EEG-findings (including epileptiform activity) at baseline and at test 2 remained insignificantly different in the two groups (Table 3).
Predictors of freedom from seizures
Bivariate logistic analysis demonstrated that neither patient age, gender, age of epilepsy onset, partial versus generalized epilepsy, MRI-findings, nor duration of seizure freedom predicted seizure freedom after AED withdrawal. Three of those who had seizure relapse in the withdrawal group had used CBZ, compared with eight on other medications. In a multivariate logistic analysis, this gave an odds ratio of 6.33 (95% CI: 1.23–32.25) of remaining seizure-free for 1 year, when prior use of CBZ was compared with prior use of any other AED. At a median of 41 months, patients on CBZ had an almost threefold chance of remaining seizure-free after withdrawal, compared with patients on any other AED (OR 2.86; 95% CI: 1.31–6.26; p = 0.01). A normal neurological examination was also a significant predictor (multivariate analysis) for seizure freedom at 12 months (OR 2.77; 95% CI: 1.18–142.86; p = 0.036).
The study described here used a double-blind, randomized comparison of consequences of AED-withdrawal versus nonwithdrawal, to enable an assessment of the advantages/disadvantages of AED-withdrawal.
We observed an increase in seizure relapse rate among withdrawers at 12 months, and if all the data are considered, the relative risk may be more than doubled. At two years, 81% of the withdrawers from the RCT remained seizure-free, i.e. a relapse rate of 19%. Three more patients had seizure relapse within 12 months after study termination in the withdrawal group. The declining monthly risk of seizures in the combined group further underlines that most patients will experience seizures after withdrawal within the first six to nine months after ended discontinuation.
Although one of the inclusion criterions was a minimum of 2 years free of seizures, 67% of the study patients had bee seizure-free for more than five years. This may have increased the likelihood of remaining seizure-free.
The MRC Antiepileptic Drug Withdrawal Study Group (1991) reported a two-year relapse rate of 22% among nonwithdrawers, compared to 41% among withdrawers. The marked difference in relapse rate between the two studies is probably due to different inclusion criteria in the two studies. The MRC study included a broad spectrum of patients, also including some on AED polytherapy. Our inclusion criteria may have been strict, but a large proportion of epilepsy patients are seizure-free and on monotherapy (Kwan and Brodie, 2000). We observed a declining risk of seizure relapse over time after withdrawal. This is probably due to continuous selection of those least likely to have seizures. A modest, but general, improvement in neuropsychological function was demonstrated among those who had withdrawn AEDs. The favorable results among those with higher, versus lower, CBZ serum values support this observation. The findings indicate that AEDs have a modest, but reversible, depressing effect on some cognitive functions and complex motor coordination. The practice effect might have meant that greater cognitive improvement due to AED withdrawal could not be observed. The effects on patients who abstain from AEDs for a more prolonged period remain to be seen.
Our results differ from those reported in two previous studies on neuropsychological consequences of CBZ in healthy volunteers and epilepsy patients (Thompson et al., 1980; Aldenkamp et al., 1993), and from those in a comparative study of CBZ and valproate (Prevey et al., 1996). However, our data are in close agreement with those of Meador et al. (1991 and 1993), who showed improved concentration and memory following AED withdrawal. The implications that these changes in neuropsychological functions may have on daily life remain unknown, however it seems probable that the ability to perform activities that demand rapid cognitive performance and complex motor coordination may be improved after AED withdrawal.
Despite improvement in neuropsychological performance, no improvement in HRQOL was observed after AED withdrawal. Adaptation to a new steady state of health (or ill-health) may be reported as no change, if measurement is made after a prolonged drug-free period. The double-blinded design excludes one known positive effect of being off medication; namely not having to take drugs regularly. Fear of seizure relapse due to patients being blinded would be expected to have a negative impact on HRQOL. Both the neuropsychological and HRQOL results may be influenced by the relative short period (four months after the end of the withdrawal period (three months) to the second investigation.
The Medical Research Council (MRC) Drug Withdrawal Group suggested that open AED-withdrawal was associated with an improved (although statistically nonsignificant) feeling of well-being (Jacoby et al., 1992). The overall effects on quality of life associated with AED-withdrawal remain unresolved.
Both at baseline and during follow-up, EEG-findings were without prognostic information on seizure relapse among both those who had withdrawn and those who had not withdrawn AEDs. Since as many as 31% of those who had withdrawn AEDs had epileptic discharges at baseline, and 30% at test 2, the prognostic role of abnormal and epileptiform discharges in EEG of adults with known epilepsy is not fully understood. Although the prognostic significance of EEG pathology at drug withdrawal in children seems undisputed (Specchio and Beghi, 2004), the value of EEG in predicting outcome after withdrawal in adults is controversial (Specchio and Beghi, 2004). Some studies in adults (Callaghan et al., 1988; Specchio et al., 2002) showed an increased risk of relapse in patients with EEG abnormalities in bivariate analysis, but not in multivariate analysis. We are not aware of studies in adults, using multivariate analysis, which report any significant predictive value of EEG abnormalities prior to withdrawal. Reliable predictors for remaining seizure-free after discontinuation of AEDs in adults have been difficult to find. However, multiple AEDs (Medical Research Council Antiepileptic Drug Withdrawal Study Group, 1991 and 1993), adolescent or adult onset epilepsy (Callaghan et al., 1988), an underlying cerebral substrate (Specchio et al., 2002), partial seizures (Specchio et al., 2002), specific AEDs used prior to withdrawal (Medical Research Council Antiepileptic Drug Withdrawal Study Group, 1993), and epilepsy syndromes with high potential for seizure relapse (e.g., juvenile myoclonic epilepsy) (Specchio and Beghi, 2004) are claimed to be negative predictors for remaining seizure-free after withdrawal.
We found normal neurological examination and CBZ use prior to withdrawal to be significant positive predictors for remaining seizure-free after AED-withdrawal at 12 months, and the latter also at 41 months (median). The positive effect of CBZ use prior to withdrawal was also reported in the MRC study (Chadwick, 1999) who found that patients on CBZ-monotherapy would be less likely to have a relapse after withdrawal, than patients stopping treatment with any other AEDs. Since CBZ is the first-line treatment in partial epilepsy, patients demonstrating a good response to CBZ probably have “easy-to-treat” epilepsy although we were unable to find significant data in our study to confirm that the CBZ treated patients had a more benign epilepsy than those treated with other AEDs (data not shown). We still have no evidence to suggest that the observation should be attributed to particular, intrinsic properties of CBZ.
The withdrawal of AED therapy must always be discussed thoroughly with the patient. Most seizures will occur during the first 6 months after withdrawal, and if a seizure does not occur in this period, then the risk after 1 year is almost the same as if AED use had continued (i.e., nonwithdrawal). If CBZ is the AED in use, or if the patient considering withdrawal has normal neurological status, it should be noted that these factors are associated with lower risk of seizure relapse following AED withdrawal. Furthermore, the likely effects of AED withdrawal on neuropsychological functions should be provided in as much detail as possible, whilst ensuring that the patient understands that changes in the quality of life may be modest, at least initially. However, the study described here demonstrates that for the majority of patients with significant predictors for remaining seizure-free after AED withdrawal, controlled AED withdrawal may have beneficial effects.
We would like to thank all the patients who, without any compensation, visited the hospital seven times and spent many hours there due undergoing extensive testing and examinations, despite being busy at work or in education. We also wish to thank the following colleagues: Kari-Anne Huuser, Gunhild Hammarstrøm, Bjørn Guldvog, Dag Hofoss, Ole Morten Rønning, Saga Høgheim, Karl Otto Nakken, Leiv Sandvik, and Kari Mette Lillestølen.
The study received financial support from: The Norwegian Foundation for Health and Rehabilitation (EXTRA FUND), The Norwegian Epilepsy Association, and The Norwegian Chapter of the International League Against Epilepsy.
The various active drugs and placebo tablets were provided from The Pharmacy in Krageroe (Norway), GlaxoSmithKline, Desitin, and Novartis.