Cognitive Decline in Severe Intractable Epilepsy


  • Pamela J. Thompson,

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology UCL, London; and National Society for Epilepsy, Gerrards Cross, United Kingdom
    Search for more papers by this author
  • John S. Duncan

    1. Department of Clinical and Experimental Epilepsy, Institute of Neurology UCL, London; and National Society for Epilepsy, Gerrards Cross, United Kingdom
    Search for more papers by this author

Address correspondence and reprint requests to Dr. P. Thompson at National Society for Epilepsy, Chesham Lane, Gerrards Cross, Bucks SL9 0RJ, U.K. E-mail:


Summary: Purpose: To explore the relation between seizure-related variables and cognitive change in patients with severe intractable epilepsy.

Methods: A retrospective analysis of data from 136 patients who had undergone a cognitive assessment on two occasions at an interval of ≥10 years. Cognitive measures included tests of memory and executive skills in addition to intelligence quotients (IQ). Details were available regarding seizure type and frequency in the intertest interval.

Results: Cognitive decline was severe and occurred across a wide range of cognitive functions. The frequency of generalised tonic–clonic seizures was the strongest predictor of decline. Complex partial seizure frequency was associated with a decline in memory and executive skills but not in IQ. Seizure-related head injuries and advancing age carried a poor cognitive prognosis, whereas periods of remission were associated with a better cognitive outcome. Early age at onset was not implicated, and duration of epilepsy was a much less potent predictor of cognitive decline than has been reported in cross-sectional studies. No evidence indicated that a higher level of cognitive function protected against cognitive decline.

Conclusions: Our findings, together with those from animal studies and surgically treated patients, suggest that seizures can have a direct adverse effect on cognition and that good seizure control even after years of intractability can have a beneficial impact on cognitive prognosis. This study was based on individuals who merited two cognitive assessments ≥10 years apart and hence is biased in favor of those with the most severe forms of refractory epilepsy and those with decline.

Cognitive decline has long been recognized as a sequel of intractable epilepsy (1). A number of factors have been identified as having a role, including underlying pathology, seizures, and medication (2,3). Of the main factors identified, the role of seizures has been less well studied, and the available evidence does not indicate as strong a relation as might be anticipated from clinical experience and animal studies. Research involving animal models of epilepsy has demonstrated that status epilepticus and frequent recurrent chemically and electrically induced seizures can result in cognitive impairment (4–7).

Cross-sectional studies of humans have provided some evidence of a relation between seizure frequency and cognitive impairment (8–10). Longer duration of epilepsy has been reported to be associated with greater cognitive impairments (11–13). The negative impact of duration may be in part due to the cumulative impact of seizures, but also to other factors including antiepileptic drug (AED) treatment and pathologic interictal brain activity. Two studies have estimated cognitive decline on the basis of language abilities. One found a relation between cognitive decline and seizure severity but not frequency (14), and the other reported that duration of epilepsy was the best predictor of cognitive decline, and other epilepsy-related variables did not make any additional contribution to the variance (15).

In 2004 Dodrill (16) reviewed nine longitudinal studies that measured intellectual functioning in children followed up for a maximum of 4 years. Several reported intellectual decline, and Dodrill concluded that poorly controlled seizures were likely to have some causal role. In 1968 Rodin (17) reported intellectual decline in adults with poorly controlled seizures in contrast to improved intellectual functioning in patients who had experienced periods of remission ≥2 years in the intertest interval. Similarly, Seidenberg et al. (18) reported intellectual gains in association with improved seizure control. Other investigators reported stable cognitive functioning in individuals with good seizure control (19–21) and, conversely, a poorer cognitive outcome in patients with continuing seizures (22,23). No cognitive change, however, has been noted in some studies in association with continuing seizures (24,25). Cognitive improvement or arrest of cognitive decline has been reported in patients rendered seizure free by temporal lobe surgery (22), although one study failed to find an association between longer-term memory outcome and postoperative seizure control (26).

Dodrill (27) assessed cognitive change in 35 patients with active focal epilepsy and 35 healthy controls at an intertest interval of 10 years (27). The epilepsy group had experienced on average >1,000 partial seizures and >50 generalised convulsions. Significantly improved scores were noted on three cognitive measures in the controls compared with the epilepsy group, and one improved score was noted on a measure of motor dexterity in the patient group. No association was found between the frequency of partial seizures and changes in test scores, and only two modest associations between the frequency of generalized convulsions and decline in full-scale IQ (FSIQ) and a measure of speed of mental flexibility. Four patients had experienced an episode of generalized status epilepticus, and these showed a significant decline on two measures of memory.

Higher mental reserve capacity has been suggested as affording protection from cognitive decline. Patients who performed at higher levels during baseline testing retained better long-term performance than did those who started from a lower level (22). Individuals with lower mental reserve capacity, as assessed by years of education, showed the highest correlations between cognitive impairment and duration of epilepsy (13). It has been suggested that an early age at onset is a risk factor for cognitive decline, as brains are less able to develop a functional reserve capacity to cope with subsequent loss (28). Advancing age, conversely, has been proposed as increasing the risk of cognitive decline, because having epilepsy accelerates the cognitive aging process (29).

A number of deficiencies with existing studies should be addressed if our understanding of the mechanisms underlying cognitive decline in epilepsy and the role of seizures is to be advanced. The main problems are first, the short intertest intervals, and follow-up periods of ≥25 years may be required (12). Second, lack of adequate data is found on seizure types and frequencies. Third is the lack of control groups matched for age and education, and fourth, measurement of cognitive abilities is often limited to IQ measures. A prospective longitudinal study addressing these deficiencies is required. Even if such studies are currently under way, it will be many years before findings emerge. In an attempt to bridge the gap between past studies and future prospective projects, we carried out a retrospective study of patients with intractable seizures with a long follow-up, detailed data regarding seizure type and frequency, and using measures likely to be sensitive to change.


Patient population

Adults with active epilepsy were included if they had undergone two neuropsychological assessments at a minimum intertest interval of 10 years, with adequate documentation of their seizure frequencies between the testing sessions. Individuals also had to be functioning above the mentally impaired range of intellectual ability at the time of the first assessment. Surgically treated patients were included if the operation had taken place before the first assessment session.

The sample was drawn from patients living at a residential centre for individuals with complex epilepsy and from the inpatient and outpatient services of a tertiary referral centre. In general, assessments were prompted in the outpatient and residential group because of concern over cognitive decline. In the inpatient group, 51 reassessments were undertaken as part of a preplanned multidisciplinary assessment of individuals undergoing medical and presurgical review. Accordingly, the group is biased toward patients with more severe forms of epilepsy. Seizures were recorded prospectively by care staff in the residential setting and in seizure diaries in inpatients and outpatients. All assessments were undertaken by the same psychology service, ensuring consistency in approach.

Neuropsychological measures

Neuropsychological test data were recorded if cognitive measures were common to both assessments and were available for the majority of the sample.

Intellectual level

All cases had been assessed on the WAIS-R (30).

Intellectual potential

All but two cases had been administered the Nelson Adult Reading Test (NART) on the first session. This is a measure of reading competence that has been validated as a measure of premorbid intellectual functioning (31).


Scores were included from the List Learning subtest of the Adult Memory and Information Processing Battery (AMIPB) (32). In this test, the subject is read aloud a list of 15 unrelated words and asked to recall as many as possible over five trials. The total recalled provides a measure of verbal learning. The subject is then required to recall the list again after a brief period of distraction, and the total number of words retained provides a measure of verbal recall.


The McKenna Naming Test was used as a measure of expressive language functioning (33). The subject has to name 30 line drawings of decreasing frequency.

Executive skills

Data were available from a measure of mental flexibility and two tests of verbal fluency. In the Trail Making Test of mental flexibility, the subject is first required to join a series of consecutive numbers together (Part A), and the time to complete is recorded. Subsequently the subject is required to join alternate letters and numbers (Part B), and again, the time to complete is recorded. The verbal fluency measure requires the subject to recall as many words as possible in a minute beginning with the letter “s” (phonemic fluency). Subsequently the subject is required to state as many animals as possible in 60 s (semantic fluency) (34).

Demographic and clinical characteristics of the sample

Neuropsychological test data were available for 136 adults (Table 1). Sixty-seven were living in a residential centre. The median age of onset at recurrent seizures for the group was 8 years, and the median duration of epilepsy was 35 years; 125 had focal epilepsy, of whom 95 had secondarily generalized seizures. The remaining 11 patients had generalized epilepsy.

Table 1. Demographic and clinical characteristics of the sample
  1. SE, Status epilepticus.

Gender: male88 (65%)   
Handedness: right115 (85%)    
 Residential care67 (49%)   
 Tertiary referrals69 (51%)   
Education (yr; median, range)11 (9–16)  
Age at seizure onset; median, range8 (1–39) 
Duration of epilepsy; median, range35 (10–64) 
Epilepsy type
 Generalized11 (8%)    
 Focal125 (92%)    
Annual seizure frequency; median, range
 Complex partial seizures52 (0–420) 
 Generalized tonic–clonic6 (0–240)
 Drop attacks (tonic and atonic)0 (0–300)
History of SE (intertest interval)27 (20%)   
History of remission (intertest interval)34 (25%)   
Seizure related head injuries (intertest interval)50 (37%)   
 Acquired insults31 (23%)   
 Hippocampal sclerosis27 (20%)   
 Malformation of cortical development11 (8%)    
 Tumor5 (4%)   
 Idiopathic generalized5 (4%)   
 Cryptogenic57 (41%)   
MRI scans
 Focal atrophy40 (29%)   
 Generalized atrophy23 (17%)   
 Cerebellar atrophy only11 (8%)    
 No atrophy62 (46%)   
Repeated scans (n = 45); progressive atrophy23 (17%)   

The median intertest interval was 13 years (range, 10–27 years). The median age of the group at the first assessment was 31 years and 44 years at the second assessment. The group as a whole had frequent seizures, although a few had relatively good seizure control. Thirty-four patients had experienced a period of remission during the intertest interval of at least a year, with the maximum seizure-free period being 15 years.

MRI scans

Of the sample, 98% had undergone at least one MRI scan, but only 35 (26%) were scanned before or at the time of the first assessment. Forty (29%) scans showed focal atrophy, 23 (17%), generalized atrophy, and 11 (8%), cerebellar atrophy only. The remaining 62 (46%) scans were judged to show no evidence of atrophy. Forty-five (33%) had repeated MRI scans; 23 of these were reported as showing progressive atrophy.


Details of AED treatment are given in Table 2. All the patients were taking AEDs at the time of the first assessment; 7% were taking a single drug, 60% were taking two drugs, and 33% were taking three drugs. Eighteen different drugs were being used, with carbamazepine (CBZ), phenytoin (PHT), and sodium valproate (VPA) being the most frequently prescribed, and VPA plus CBZ the most popular combination.

Table 2. AED treatment on the two testing sessions
 Session 1Session 2
Sodium valproate49363324
Primidone282012 9
Phenobarbitone1612 3 2
Clobazam12 93828
Lamotrigine 8 63123
Acetazolamide 5 4 6 4
Vigabatrin 6 4 4 3
Clonazepam 4 310 7
Ethosuximide 2 1 1<1 
Levetiracetam 0 02216
Topiramate 0 01511
Oxcarbazepine 0 011 8
Sulthiame 0 0 5 4
Tiagabine 0 0 4 3
Other 1  6 
Diazepam Pregabalin 1
Allopurinol 1
Zonisamide 1
Diazepam 1
Felbamate 1
Piracetam 1

At the time of the second assessment, 23 different drugs were used. The number receiving monotherapy was 15%, and the number taking two had decreased to 40%. The number taking three drugs had increased to 39%, and 9% were prescribed four different AEDs. CBZ remained the most popular drug used. The use of PHT had decreased to 17%, and VPA, to 24%. Of the newer AEDs, topiramate and levetiracetam were the most widely used, being taken by 11% and 16% of the sample, respectively.

Statistical analysis

Analysis was undertaken with parametric tests, and data that were not normally distributed were subject to logarithmic transformations (SPSS version 9).

The significance of cognitive change between the two testing sessions was analysed by Student's t tests. Percentage change in scores was calculated for each of the nine cognitive measures and used as the dependent variable. A number of independent seizure-related variables were identified as of interest (age at onset; duration of epilepsy; annual frequency of generalized tonic–clonic seizures, complex partial seizures, drop attacks; years of remission from seizures; history of seizure-related head injuries; history of status epilepticus; aetiology and degree of cerebral atrophy; and intertest interval). In addition, NART IQ and years of education were used to explore the higher-mental-reserve hypothesis. The association between these variables and the measures of cognitive change was explored first in a series of analyses to select key variables for entry into subsequent multivariate analyses. The relation between cognitive change and the independent variables was assessed by using two-tailed Pearson's correlation test for continuous data, Student's t tests for history of head injury and status, and analysis of variance (ANOVA) for degree of atrophy (none, focal, generalized, and cerebellar). Because of the number of analyses performed, the level of significance was set at p < 0.005.

To explore the relative contributions of the clinical variables, a series of standard multiple regression equations were then calculated, with the dependent variables being the measures of cognitive change and the independent variables being the significant variables from the previous analyses. Chronologic age and baseline cognitive performance were included as additional independent variables. Entry criterion for the analyses was set at p < 0.05.


Neuropsychological test results

The results from the two testing sessions are given in Table 3. On all tests, scores were lower for the second assessment session, and the decline recorded was statistically significant for all tests (paired Student's t test, p < 0.0005).

Table 3. Neuropsychological test data on the first and second assessment sessions
 No.Assessment oneAssessment two
Mean (SD)CentileMean (SD)Centile
  1. Centile scores are derived from normative/standardization samples of comparative ages.

NART13497.6 (10.7)50th 
VIQ13690.3 (11.0)25th82.3 (13.4)9th
PIQ13691.0 (14.4)25th84.5 (19.4)9th
FSIQ13690.7 (10.9)25th83.1 (17.6)9th
Verbal Learning13444.7 (8.2)10th34.5 (2.9)<1st
Verbal Recall134 9.1 (2.3)10th 6.1 (3.8)<1st
Naming13414.1 (5.7)10th11.5 (7.0)5th
Verbal Fluency
 Phonemic “s”13512.9 (5.3)10th 8.2 (5.9)5th
 Animals13515.9 (5.7)10th11.3 (6.8)5th
Mental Flexibility Median Range Median Range 
Trails A11431.5 (10–229)30th49.0 (15–370)<1st
Trails B11453.0 (21–234)50th120.0 (25–430)<1st
Age (yr)13630.5 (16–54) 43.9 (26–69) 

An estimate of intellectual potential based on the NART places the potential of the group toward the middle of the average range of ability. On the first assessment, the intellectual level derived decreases toward the lower end of the average range on both verbal and performance subtests. This suggests that some intellectual decline had occurred before the first testing session.

On the first assessment of memory, the group functioned well below average, with scores falling at the 10th centile for the learning trials and after a delay. Similarly on tests of naming, the scores were at the 10th centile. In both the straightforward and alternating condition of the Trail Making Test, the performance of the group was below average, at the 10th and 20th centiles. On the measures of verbal fluency, the group functioned below average but not in the impaired range.

At the second assessment, the group functioned in the low-average range of intellectual ability on both verbal and performance subtests of the WAIS-R. On tests of learning, the scores of the group decreased to an impaired level. Naming performance had declined to the 5th centile. On the tests of executive skills, the performance deteriorated, and on all measures, scores decreased to the first centile or below for normative samples for similar age groups.

Given the retrospective nature of this study, no test–retest data are available for controls. However, as is demonstrated in the table when comparisons are made with normative data derived from samples of comparable ages, it is clear that the gross deterioration in scores obtained cannot be attributed to the passage of time.

Associations with cognitive decline

Results are summarised in Table 4. Significant correlations were found between the average annual frequency of generalized tonic–clonic seizures during the intertest interval and declines in VIQ and PIQ (both at p < 0.0005; Fig. 1). Significant associations also were found between generalized tonic–clonic seizure frequency and verbal learning, delayed verbal recall (p < 0.0005), naming (p < 0.001), and semantic fluency (p < 0.005).

Table 4. Significant correlations between the percentage cognitive change on the neuropsychological measures and seizure-related variables


Duration of

  1. Correlation coefficients are not applicable to status and head injury.

  2. GTCS, generalized tonic–clonic seizure; CPS, complex partial seizure.

  3. ap < 0.005.

  4. bp < 0.001.

  5. cp < 0.0005.

VIQ0.42c c 
PIQ0.34c c 
Verbal learning0.33c 0.26ab  0.25a−0.36c
Verbal recall0.33c 0.26aab −0.39c
TMT Part A a−0.34c 0.42c
TMT Part B −0.30b −0.37c 0.35c
Semantic fluency0.27b 0.30b a −0.31c
Phonemic fluency c 0.34c−0.25a
Figure 1.

Average annual frequency of generalized tonic–clonic seizures (log scale) versus percentage decrease in VIQ (p < 0.0005).

The frequency of complex partial seizures was associated with a decline in verbal learning (p < 0.005), delayed recall (p < 0.005), Trial Making part B (p < 0.001), and semantic fluency (p < 0.001). Patients with a history of status epilepticus showed greater declines on measures of verbal learning (p < 0.001) and verbal recall (p < 0.005). No significant associations were found between frequency of drop attacks and any of the cognitive-change scores.

A history of recurrent head injuries was associated with greater declines in VIQ and PIQ (p < 0.0005), verbal recall (p < 0.001), semantic and phonemic fluency (p < 0.005 and p < 0.001, respectively), and a decrease in speed on part A of the Trail Making Test (p < 0.005). A number of significant results were obtained for years of remission, with cognitive decline being less marked with increasing numbers of years of seizure freedom for verbal learning, verbal recall, Trail Making Parts A and B, semantic fluency (all at p < 0.0005), and phonemic fluency (p < 0.005).

Age at onset of epilepsy was not significantly associated with cognitive decline on any of the measures. Longer duration of epilepsy was associated with greater declines in verbal learning (p < 0.005), an increase in mental speed on both parts of the Trail Making Test (p < 0.0005), and poorer phonemic fluency (p < 0.0001). Cognitive decline was greater at longer intertest intervals for VIQ (p < 0.005), verbal learning (p < 0.0005), Trail Making Part A (p < 0.001), and Naming (p < 0.005).

No significant difference was noted in the rate of cognitive change between the etiologic categories for all cognitive variables. Individuals with generalised atrophy on MRI showed greater declines in scores than did individuals with no atrophy or only focal or cerebellar atrophy for all cognitive variables (p < 0.001 to p < 0.0005). In the 45 patients who had repeated MRIs, no differences were seen in the rates of decline, at the significance level set, between patients showing progressive atrophy on repeated MRI and those assessed to have stable imaging.

No significant correlations were recorded between the measures of cognitive decline and NART IQ and years of education.

Multiple regression analyses

The results from the analyses are summarized in Table 5. Frequency of tonic–clonic seizures was the most significant predictor of a decline in both VIQ and PIQ, with a history of head injuries making an additional contribution, and the length of the intertest interval, a more modest contribution for decline inVIQ. PIQ at the first testing session also was a predictor of decline, with higher initial PIQ being associated with greater declines.

Table 5. Results from the multiple regression analyses
GTCS 2.020.313.70.0005
Head injuries 
Interval 0.580.192.40.018 
Head injuries 
Interval 0.810.202.40.017 
Verbal Learning0.57 
Age 0.420.172.20.03  
Status 11.1  
Remission −1.5   −0.16−2.00.05  
Verbal Recall0.49 
Age 0.830.212.70.008 
Complex partial seizures  
Trail Making A0.50 
Remission 13.480.192.20.03  
Head injuries −77.66−0.19−2.10.04  
Trail Making B0.47 
Complex partial seizures −24.22−0.25−2.70.009 
GTCS 6.480.263.10.003 
Age 0.920.242.80.006 
Phonemic fluency0.46 
Head injuries 22.3 
Semantic fluency0.55 
GTCS 6.390.253.20.002 
Complex partial seizures 

For verbal learning, the intertest interval was the major predictor of decline, with greatest decreases occurring with longer intertest intervals. Frequency of generalized tonic–clonic seizures, age, a history of status, and years of remission made additional but smaller contributions. For delayed verbal recall, having frequent generalized tonic–clonic seizures was the most significant predictor, but age made a more substantial contribution than did verbal learning. Frequency of complex partial seizures also contributed to the variance. A history of head injuries and years of remission were not significant. Decline in naming was predicted by age and frequency of generalized tonic–clonic seizures.

For the Trail Making Test, a history of head injuries and remission were significant predictors of performance declines on Part A, and for Part B, frequency of complex partial seizures was the sole significant predictor. For phonemic fluency, a history of head injury was the significant predictor, with duration of epilepsy, years of remission, and age not being significant. For semantic fluency, age and frequency of generalized tonic–clonic and complex partial seizures were the significant predictors of decreased output.


Cognitive decline was severe in this group of patients with epilepsy. This was to be expected, as the sample represents a high-risk population, includes several individuals with the severest forms of epilepsy, and was biased toward those with cognitive decline, who merited repeated evaluations. On the first assessment session, scores ranged from the 10th to the 50th centile. On the second session, a minimum of 10 years later, the majority of test scores were impaired, and the best group performance fell at the 9th centile for measures of intellectual capacity. The study was retrospective, and it was not possible to include a control group, but data could be compared with age-related normative values. The extent of the cognitive deterioration recorded, however, was clearly much greater than could be explained by changes resulting from the passage of time that would be unrelated to epilepsy. Indeed the cognitive decline was of such a magnitude that a number of individuals would be clinically classified as having a presenile dementia and as a consequence currently require high levels of staff support and care. Furthermore, the extent of the decline recorded is likely to have been underestimated, as the average age at the first testing session was 30 years, and performance on the NART, a measure of intellectual potential, suggested that moderate decline had already occurred before this.

One of the strongest relations found was between cognitive decline and generalized tonic–clonic seizure frequency. Longitudinal research to date has indicated, at most, modest levels of significance for this variable (27). A likely reason contributing to the significance of our findings is the wide range of seizure frequencies. The total number of convulsive seizures between testing sessions for our group ranged from zero to >3,000. Our intertest interval was ≥10 years, and the length of the intertest interval was a significant predictor of cognitive decline; this supports proposals made by others of the importance of substantial follow-up periods (11,12,16). Status epilepticus was not as strongly associated with cognitive decline as were generalized tonic–clonic seizures and was significant only for memory decline, as noted by Dodrill (27).

Frequent complex partial seizures was identified as a factor associated with cognitive decline. This has seldom been the case in longitudinal investigations. The relation of complex partial seizures occurred with measures of memory and mental flexibility but not for measures of intellect. This endorses the view that measures of intelligence are less sensitive to cognitive change than are measures of memory and executive skills (16,27), and that studies and clinical assessments using only intelligence measures may underestimate the extent of cognitive decline. Further, measures of memory and executive functions reflect functions associated with frontal and temporal brain regions that are commonly involved in epileptic foci and networks.

The frequency of drop attacks did not have any significant associations, and the severity of drop attacks may have been a more relevant variable. A history of seizure-related head injuries, which most commonly arise from atonic and tonic seizures, was implicated as a contributory factor. This finding supports an earlier study in which seizure-related head injuries and skull fractures were significantly more prevalent in a cognitively deteriorated group (14).

No significant relation was found between cognitive change and age at onset, and thus no evidence supported the hypothesis that early age at onset is a risk factor for decline, as some have suggested (28). This does not mean that age at onset is not an important variable, but this may be related more to cognitive impairment and cognitive development in children and have a more-limited role in cognitive decline in adults. Furthermore, a more prominent role for age at onset might have been found if we had data from cognitive assessments undertaken earlier in the course of our patients' seizure histories.

Duration of epilepsy was a much less potent predictor of cognitive decline than has been highlighted in cross-sectional studies (11–13), and no relation was observed for intellectual decline. The strongest relations were observed on tests of executive functions. Duration of epilepsy was of significance in a number of the bivariate analyses, but the impact was reduced by the inclusion of other variables in the multivariate analyses, most notably, frequency of generalized tonic–clonic seizures. In keeping with Rodin (17) in an earlier longitudinal study, we found seizure remission to be a pertinent factor, and this would reduce the role of duration of epilepsy (17). Periods of remission therefore may have a protective role and at the least may result in an arrest of ongoing decline. This is in keeping with reports that the cognitive outcome is good in temporal lobe surgical patients who are rendered seizure free after temporal lobe resection (22). This suggests that achieving complete seizure control, even after years of intractability, can have a beneficial impact on cognition as well as on psychological well-being and quality of life (35).

Our study does not suggest that a higher level of initial function protects against decline. Those individuals with higher abilities assessed by intellectual potential (NART) or longer educational histories were not less likely to decline cognitively. Indeed, higher PIQ was a predictor of intellectual decline. It is possible, however, that higher mental reserve capacity is of importance over shorter epochs of <10 years. In addition, this study was of those with the most severe forms of generally refractory epilepsy, and the results cannot be extrapolated to those with less-severe forms of epilepsy. Older age was found to be a significant risk factor associated with more-marked declines on measures of object naming, memory, and executive functions providing support that age-related cognitive decline occurs earlier in intractable epilepsy (29).

MRI data were available for all but three cases. Classification of the MRI-scan findings was qualitative, and no quantitative MRI measurements were used. However, patients reported as having more widespread atrophy did have higher rates of cognitive decline than did those with milder forms of atrophy or no recorded abnormalities. A small subgroup had repeated imaging. No differences were recorded between the groups rated as showing progressive MRI changes and those with more stable imaging at the level of significance set. For several cases, however, the interval between scans was <5 years, and MR imaging did not coincide with the two cognitive assessments. The absence of relations between progressive cognitive decline and progression of atrophy therefore is not surprising. Prospective studies including repeated quantitative MR imaging and cognitive assessments will be needed to explore whether cognitive decline is reflected in changes in brain structure.

Although this study provides the strongest evidence to date of a relation between seizure frequency and cognitive decline, this can only be part of the mechanism, as the amount of the variance explained by the various regression analyses was relatively modest. Other factors must clearly have a role. AEDs are likely to make some contribution to cognitive change (36–38).The impact of drugs has been followed up only over relatively short periods, and no evidence exists about the effect of use over several decades. Data were available on the numbers and types of medications taken on the two assessment sessions. The variety of changes in medication that had taken place and the varying combinations of drugs did not lend itself to any meaningful analysis, with the further complexity of the possible effects of the AEDs taken between the two assessments.

It remains unclear whether the association of cognitive decline with seizure frequency is due to a direct adverse effect of seizures, or that a higher seizure frequency is a marker of a more-degenerative cerebral process that gives rise to cognitive decline and a propensity to seizures. Although the latter may occur, particularly in some defined syndromes, the inference from animal models of induced seizures (6), our current findings suggesting the positive influence of years of remission and patients treated surgically (22) imply that the former is generally a more probable explanation.