Starting and Stopping Treatment for Seizures and Epilepsy


Address correspondence and reprint requests to David Chadwick at Division of Neuroscience (University of Liverpool), The Walton Centre for Neurology and Neurosurgery, Lower Lane, Liverpool L9 7LJ, U.K. E-mail:


Summary:  Decisions about when to start or to stop antiepileptic drug (AED) treatment must be informed by reliable information of the natural history of epilepsy, the effect of treatment, and the social context of the individual. Ultimately the patient will be the decision maker, the clinician or health-care professional, the provider and interpreter of information. While observational studies will provide information on natural history, the most reliable information on the effect of intervention will come from randomised controlled trials in relevant populations of patients. However, these need to be large enough to allow interpretation not just of the average effect across the recruited patients, but also some estimate of the effects for an individual based on the prognostic effects that most effect outcomes. When trials are of sufficient size they can allow the development of predictive models that assist decision making. The Medical Research Council studies of AED withdrawal and early epilepsy and single seizures provide examples of such trials.

 Decisions about when to start and when to stop therapy vary in their difficulty. For the great majority of people presenting at the point of diagnosis of epilepsy there is little doubt that the benefit of antiseizure drugs (ASDs) will far outweigh the risks. This review does not seek to deal with these patients. There are, however, some individuals with single, few, or infrequent seizures and those with seizures of minor symptomatology for whom there will be doubt and uncertainty. After a period of remission, the decision to withdraw treatment again falls into a grey area of uncertainty for most adults, although the decision will be easy in many children.

It is axiomatic in both areas that decision making should be undertaken jointly between clinician and a well-informed patient. Here, we are fortunate that the area probably provides the best available evidence base in epilepsy for everyday decision making. There are literally hundreds of observational studies that document risks of seizures in both situations. At first sight the results are confusing and contradictory, due to methodological variation, but we are doubly fortunate that for each area we have high-quality systematic review and meta-analysis articles that deal with these observational studies (1,2). These are exemplars in illustrating the value of time-consuming and detailed secondary research.

We do, however, need to inform our patients not solely about the risk of seizures in, what was for the most part in observational studies, the absence of treatment. We also need to quantify the benefits of treatment in reducing the risk of seizures and the potential risk for harm. Here, we can turn to a number of well-conducted randomized controlled trials (RCTs), two of which represent largest yet undertaken in the therapeutic area of epilepsy and which may have sufficient external validity to allow direct patient counseling (3–5).

Before dealing with the results it is worthwhile considering some of the issues that still lead to confusion about RCTs, and how different designs may deal with the problems in different ways.

The great majority of RCTs in epilepsy are sponsored by drug companies for regulatory purposes. For these studies, internal validity is paramount and external validity may be sacrificed. A good example of this is that these studies often use placebo control groups with observations for short periods, making any application of everyday practice to a chronic condition such as epilepsy difficult. If we want to use a study to counsel individual patients about the risk of seizures we need to ensure that it has external validity (6). RCTs can use a number of means to achieve this end which include

  • – Reduce any exclusion criteria to a minimum
  • – Collect data on nonrandomized patients to show that the trial sample is truly representative
  • – Ensure recruited patients are well described
  • – Have sufficient numbers recruited to examine interactions between treatment effects and prognostic factors.

The two UK Medical Research Council (MRC) sponsored RCTs are examples of pragmatic clinical trials of sufficient size and external validity to allow counselling of the individual patient (3,5).


Reported seizure recurrence rates following a first tonic–clonic seizures vary between 23% and 71% over 2 or 3 years (1). This meta-analysis identified several characteristics that influenced recurrence rates including previous brain disease or insults, and abnormal EEG. On average, 50% of people who have had a single seizure do not experience a recurrence, but the risk of future seizures rises with increasing number of seizures; consequently, uncertainty about the need to start ASD treatment diminishes with increasing numbers of seizures (7).

RCTs have been undertaken to compare a policy of immediate versus deferred treatment with ASDs following a single tonic–clonic seizure (4,5,8–11). Most have examined short-term outcomes, with two exceptions looking at longer term remission (4,5). This later study (5) showed that while times to first and second seizures were reduced by immediate treatment any effect on the probability of 2-year remission was lost by 4 years in patients with single seizures and by 6 years in those with multiple seizures. By 6 years after randomization, similar proportions (43–48%) of subjects were taking ASDs in treated and untreated groups, as a result of the crossover that occurred as patients initially treated withdrew from treatment after seizure freedom, and those who deferred treatment were prescribed ASDs after seizure occurrence. Finally, the proportions of patients in each group in a terminal remission of 2 or more years, 5 years after entry to the study, were identical (68%). Indeed immediate treatment resulted in 17% more patients receiving ASDs 5 years into the study in order to achieve the same outcome as a policy of deferred treatment.

Does this mean that all patients, similar to those who entered the study with single, infrequent, or minor seizures did not warrant therapy? A number of factors influenced the risk of seizures, including the number of seizures at randomization, an abnormal EEG, and any indication of a remote symptomatic epilepsy. The higher the risk of seizure recurrence the greater the treatment effect, so that high-risk patients would warrant serious consideration of early treatment to reduce the risk of early seizure recurrence. It was possible to construct a simple predictive model that should help improve patient counseling in this situation (Table 1) (12).

Table 1. Prognostic model for prediction of seizure recurrence for first seizures and early epilepsy(12)
Seizure numberScore 
 One seizure at presentation0 
 Two seizures at presentation1 
 Three or more seizures at2 
Add if present
 Neurological disorder/deficit,+1 
   learning disability, 
   or developmental delay 
 Abnormal EEG+1 
Risk classification groupFinal 
   for seizure recurrencescore 
 Low risk0 
 Medium risk1 
 High risk2–4 
TreatmentProbability ofProbability of
   allocation Medium riskseizure by 1 yrseizure by 3 yr
 High risk

To use, first calculate the score from the left-hand table. For medium- and high-risk groups the probability of seizures for the policy of immediate ASD treatment and deferred treatment can then be read off from the right-hand table.


Patients (up to 70%) diagnosed and treated for epilepsy will attain long-term remission of 2 or more years (13,14). Most of those who enter remission do so shortly after beginning therapy on low doses of ASDs. It is for this group of subjects that decision making about whether or not to continue with ASDs is critical and potentially most difficult.

The difficulty of the decision arises firstly from a lack of understanding of the way in which ASD treatment may or may not interact with the natural history of epilepsy. Are people with a significant period without seizures now cured (i.e., their seizure freedom is no longer dependent on treatment) of the condition, or is it simply controlled by ongoing treatment? Secondly, there is little information available to quantify the risk of unwanted chronic toxic effects of ASDs. Particular concerns may arise for women in the child-bearing years where reduction in the risk of foetal malformation and other adverse effects may be of importance.

The process of providing adequate information ideally involves assessing individual risk and considering how to reduce it, recognizing that some of the factors that effect outcome may be under the control of patients and their advising physician.

How long seizure-free?

Most commonly, periods of 2 years or more are generally considered necessary before consideration of withdrawal, but the longer the period seizure-free the lower the risk of recurrence.

Factors influencing risk of relapse

Studies that include a broad mix of patients and that require 2-year seizure remission before stopping treatment on average show a risk of relapse of 25% in the first year and 29% after 2 years (2). Eighty percent of all recurrences occur within the first year and 90% within the first 2 years. Many factors have been identified that appear to influence the degree of risk. These include the following.

Electroclinical syndrome

Some syndromes are very clearly associated with particular risk of relapse after stopping treatment. Benign rolandic epilepsy (BRE) has been shown by a number of investigators to have an excellent response to treatment, and relapse is almost unknown when medications are stopped (15). Juvenile myoclonic epilepsy (JME) has an excellent response to treatment; however, relapses occur in almost all patients when medications are stopped (16).

Age at onset

Most studies find a favorable prognosis in epilepsy with onset in childhood. Studies including both childhood and adolescent onset epilepsy usually find a substantially increased risk of relapse in those with adolescent onset (20% and 35–40%, respectively) (2).

Underlying etiology

Individuals with an identifiable etiology associated with their epilepsy (remote symptomatic epilepsy) are less likely to enter remission than those with idiopathic or cryptogenic epilepsy. Once in remission, they are about 50% more likely to relapse if medication is stopped.

The electroencephalogram

There is considerable controversy over the value of the electroencephalogram (EEG) in predicting the prognosis for relapse after stopping treatment. Overall, data suggest that the EEG is of greater prognostic help in children than in adults.

Severity of epilepsy

A number of different clinical features that reflect the severity of epilepsy have been studied and reported in the literature. They include a history of status epilepticus, the duration of epilepsy, the number of seizures before remission, the duration of treatment, the requirement for two or more AEDs for remission, and previous failed attempts to stop medication. Most studies indicate that these surrogate measures of severity all adversely affect the risk of recurrence (2).

The influence of individual drugs

It is often suggested that the risk of seizure recurrence may differ depending on the drug that is to be withdrawn. Withdrawal seizures are particularly said to occur with the discontinuation of phenobarbitone. The subject has rarely been exposed to systematic study. There were large subgroups of patients receiving monotherapy with carbamazepine, valproate, phenytoin, and barbiturate drugs (phenobarbitone and primidone) in the MRC Withdrawal Study. The temporal pattern of seizure recurrence was similar in the barbiturate group and the other groups (17).

Prediction of relapse

The MRC AED Withdrawal Study was sufficiently large to develop and test a predictive model for relapse in patients continuing or stopping their medication (18). The model gives decreasing weight to the following factors: whether or not treatment is withdrawn, period of time seizure-free, taking two or more ASDs, being 16 or older at the time of withdrawal, having myoclonic seizures, and having tonic–clonic seizures of any type. The final factor was an abnormal EEG. While the model does not include the presence of remote symptomatic epilepsy, factors retained in the model may, singly or in combination, provide surrogate measures for symptomatic epilepsy and capture those aspects of remote symptomatic epilepsy that are most associated with an increased risk of relapse. The resulting predictive equation was well calibrated for risks between 10% and 80% and correctly identifies JME as having a high risk of relapse. Table 2 outlines the use of this model which can be downloaded from

Table 2. Factors for the calculation of a prognostic index for seizure recurrence by 1 and 2 years following continued treatment or slow withdrawal of antiepileptic drugs, in patients with a minimum remission of seizures lasting for 2 years while on treatment
FactorValue to be added to score
1. Starting score for all patients−175
 Age > 16 yr45
 Taking more than 1 ASD50
 Seizures occurring after35
  the start of treatment 
 History of any tonic–clonic seizure35
  (generalized or partial in onset) 
 History of myoclonic seizures50
 Electroencephalogram while 
  in remission 
  Not done15
 Duration of seizure-free period (years) = D200/D
2. Total scoreT
3. Exponentiate T/100 (Z = eT/100)Z
 Probability of seizure recurrence 
 By 1 yrBy 2 yr
On continued treatment1 − 0.89Z1 − 0.79Z
On slow withdrawal of treatment1 − 0.69Z1 − 0.60Z

Two groups of investigators have developed simple models to predict relapse in children. Dooley et al. (19) developed a model derived from a study of 97 children in whom drugs were withdrawn after 12 months of remission. A point scoring system was devised in which subjects were allocated 1 point for being female, 1 for seizure onset after 10 years, the presence of neurological abnormality and generalized seizures, and 2 points for partial seizures (other than those of BRE). No subjects with 0 points relapsed (they had BRE by definition). Ninety-five percent of patients with 1 point, 80% with 2 points, 45% with 3 points and 5% with 4 points remained seizure-free. Shinnar et al. (20) determined risk of relapse 2 years after stopping ASDs in 264 children as a function of the number of risk factors for relapse a child had, 0–3. This was done separately for those with cryptogenic/idiopathic versus remote symptomatic epilepsy. In the idiopathic group, predictors of relapse were age at onset >12 years, family history of epilepsy, slowing on the EEG, and atypical febrile seizures. The risk of relapse after 2 years was 12%, 46%, and 71% in children with 0, 1, and 2 of these factors. In remote symptomatic cases, predictors were age of onset >12 years, mental retardation, absence seizures, and atypical febrile seizures. The risk of relapse after 2 years was 11%, 35%, 51%, and 78% in children with 0, 1, 2, and 3 risk factors.


The decision to start or stop the drug treatment of epilepsy requires a careful assessment of individual risk of both seizures and treatment. There is abundant evidence about the absolute levels of risk for seizure occurrence in the two situations, and the differences in risk between treatment and no treatment. Evidence about the risks of harm from ASD is more sparse. The main challenge is to offer patients an accurate estimate of their individual risk. Large pragmatic clinical trials with strong external validity offer the best means of doing this.