OUP user menu

Accelerated forgetting in patients with epilepsy
Evidence for an impairment in memory consolidation

R. V. Blake, S. J. Wroe, E. K. Breen, R. A. McCarthy
DOI: http://dx.doi.org/10.1093/brain/123.3.472 472-483 First published online: 1 March 2000


Patients with epilepsy frequently complain of memory difficulties yet perform normally on standard neuropsychological tests of memory. It has been suggested that this may be due to an impairment of very long-term memory consolidation processes, beyond those normally assessed in the neuropsychological clinic. We carried out a prospective study of verbal memory over a long-term retention interval of 8 weeks in patients with epilepsy and in controls. Results were compared with performance on conventional tests of memory. Despite normal learning and retention over 30 min, patients with epileptic foci in the left temporal lobe performed disproportionately poorly on the long-term test compared with both patients with epileptic foci in the right temporal lobe and controls. Our findings provide evidence for an extended period of memory consolidation and point to the critical region for this process, at least for verbal material, in the left temporal lobe. The implications of our findings for clinical assessment and therapeutic management of patients with epilepsy are discussed.

  • epilepsy
  • memory
  • consolidation
  • long-term
  • verbal
  • HAD = Hospital Anxiety and Depression Scale
  • NART = National Adult Reading Test
  • TEA = transient epileptic amnesia
  • WAIS-R = Wechsler Adult Intelligence Scale—Revised
  • WCST = Wisconsin Card Sort Test
  • WMS-R = Wechsler Memory Scale—Revised


Patients with epilepsy frequently complain of memory problems that are often undetected by conventional memory tests. In this paper we have investigated beyond normal retention intervals to prospectively evaluate long-term memory consolidation processes in patients with epilepsy.

Traditionally, theories of human memory have focused on a limited range of time intervals. Information that is assessed after a few seconds is thought to be held within short-term or working memory systems; information that has been stored for longer than a few minutes has usually been assumed to be represented within long-term memory. The establishment of memory within long-term storage is thought to be mediated via some form of consolidation process. While such consolidation may not be complete after a few minutes (and indeed may continue for weeks, months or years) it is often assumed that its general efficacy can be evaluated after relatively brief delays. Thus, clinical tests of memory have tended to focus on the short-term/long-term distinction, measuring memory either after seconds or minutes (e.g. Wechsler, 1987). Comparatively little is known about long-term memory beyond this relatively narrow interval (in the case of standard neuropsychological assessments extending to a mere 30 min), and it is often assumed that there is little to know. This may be an unwarranted assumption.

The most usual means of investigating longer-term declarative memory has been to use a retrospective research strategy: comparing the veracity of material that has been acquired at various times in the past. The aim is one of determining whether there is a difference in the relative vulnerability of memories of different ages, which may illuminate the dynamics of a consolidation process. The relative sparing of older memories in retrograde amnesia [Ribot's law (Ribot, 1882)] is often cited in this regard. On clinical examination, public or personal events occurring in close proximity to the onset of amnesia may seem to be disproportionately impaired and older memories may seem to be stronger and spared. Comparable findings have been reported from patients with transient retrograde memory problems following from ECT (e.g. Squire and Cohen, 1982). The fact that this gradient is directly opposed to normal patterns of forgetting—memory being better at the longest intervals—has been interpreted in favour of an extended period of memory consolidation (Squire et al., 1984). According to Squire and colleagues, while memories are in a transitory consolidation phase, they are relatively vulnerable to disruption. Loss of these more vulnerable memory traces may thus account for a pattern of temporally graded retrograde amnesia in which memories for recent events are more impaired than those for more remote events. However, when put to the scientific test, this evidence of an inverse temporal gradient has been controversial. Memories belonging to different epochs in an individual's life may be stored differently and may even have been acquired in a different way.

More compelling evidence consistent with of a period of extended consolidation can be found from the study of patients with transient epileptic amnesia (TEA). In this syndrome, transient episodes of anterograde amnesia are reported during which the patient has great difficulty in acquiring new information. In the interictal phase, anterograde memory function may appear entirely normal (Zeman et al., 1998)—at least on tasks involving retention intervals of 30 min. However, the simple event of TEA may also lead to permanent gaps in retrograde memory, sometimes affecting events which are personally salient and (most importantly) predate the onset of the attacks of anterograde TEA (Zeman et al., 1998). Indeed, there is evidence that the retrograde amnesia associated with TEA can extend back over a period of 30 years (Zeman et al., 1998).

Retrospective testing of the status of declarative memories acquired at an earlier phase of life is a difficult methodological strategy (McCarthy and Warrington, 1990). At the very least, allowance has to be made for the simple fact that everyone forgets, resulting in an evaluation of changes in function across a shifting (but indeterminate) baseline. Added to this complication we have to allow for the fact that memories of different ages often have different degrees of contextual embedding (e.g. semantic memories versus episodic memories) and are therefore possibly correlated with different neural substrates. Older memories that are recovered in the clinical setting may also be easier to retrieve because they are favourite tales and have been retrieved more frequently. Finally, memories belonging to different time periods may not be of equivalent salience, either at the time of their acquisition or with respect to the individual's current life. Retrospective techniques are therefore not ideal for assessing the status of memories over long periods of time.

The optimum way of studying the changes in memory that take place over an extended period of time is to conduct a prospective study of memory in which material is acquired and tested under controlled conditions. This approach has been adopted with some success in the study of people with organic amnesic syndromes (e.g. Huppert and Piercy, 1979). However, the major problem with amnesic populations has been in establishing adequate levels of initial learning: declarative anterograde memory is globally impaired in the populations of interest.

People with temporal lobe epilepsy provide a natural laboratory for the study of human memory (Snyder, 1997). Complaints of memory impairment are common in epilepsy, and verbal memory deficits have been reported in association with left hippocampal sclerosis (e.g. Baxendale et al., 1998; for a review, see Baxendale, 1995). However, the paradox remains that their problems are often inconsistent with (or are disproportionate to) their functioning on standard anterograde memory tests (Gallassi et al., 1988; Corcoran and Thompson, 1992), which, as we have seen, sample only a limited range of retention intervals. It has been shown that, far from exaggerating their difficulties due to neurosis (Vermulen et al., 1993), people with epilepsy actually underestimate the frequency of their everyday memory failures (Thompson and Corcoran, 1992). As in patients with TEA, the presence of an epileptic focus is likely to disrupt extended processes of memory consolidation but may spare performance when the retention interval is short.

There is some direct evidence consistent with an epileptic consolidation deficit in the single case-study literature. O'Connor and colleagues reported a patient (with temporal lobe epilepsy arising in the context of a paraneoplastic encephalitis) who demonstrated an abnormally fast forgetting rate for a word list over a period of a week despite normal learning and retention after 2 h (O'Connor et al., 1997). A fuller evaluation of memory function after a long retention interval has been reported by Kapur and colleagues (Kapur et al., 1997). They assessed memory prospectively in a patient (PA) whose presenting complaint was of amnesia for events from the preceding 3–24 months. Kapur and colleagues found that PA's memory was similar to controls following a 30 min delay but was impaired following a 6 week delay. While both of these studies are consistent with a consolidation deficit, they do not allow us to determine whether such problems are frequent in epilepsy or whether they are associated with a specific type or location of epileptic activity. There is one group study which suggests that consolidation may be a particular problem for people with temporal lobe epilepsy (Martin et al., 1991). Martin and colleagues found no difference between patients and controls in recall of a word list after a delay of 30 min. However, the patients with temporal lobe epilepsy showed a disproportionate impairment following a delay of 24 h. Although it is uncontroversial that memory disorders are associated with gross pathology in the temporal lobe, and many studies have investigated severe, lesional epilepsy or even, in the case of Martin and colleagues, have included postoperative patients, the situation with regard to milder cases is unclear.

We planned a prospective group study of long-term verbal memory consolidation in patients with epilepsy and healthy controls. By the use of a verbal measure we were able to equate initial learning levels between individual cases. Moreover, we hoped that this approach would enable us to evaluate any interaction between the locus of epileptic activity and very long-term retention. Specifically, we hypothesized that epileptic activity during the retention interval would compromise memory consolidation and that these effects should be strongest for verbal material in patients with a primary focus in the left temporal lobe. A further aim was to examine the relationship of very long-term memory to performance on standard neuropsychological memory tests and psychological factors.


The study was approved by Addenbrooke's and Ipswich Hospitals' ethics committees.


Twenty-three consecutive patients attending epilepsy clinics in Addenbrooke's and Ipswich hospitals with partial epilepsies were recruited (16 females and seven males). The patients were right-handed (22) or, in one case, left-handed, with demonstrated left language dominance in the intracarotid sodium amytal (WADA) procedure. No patient had a history of head injury or neurological illness other than epilepsy. Nineteen matching healthy controls (12 females and seven males) were also recruited (partners, relatives or friends of patients and five neurologically normal volunteers). All subjects gave their informed consent to participation.

Patients were classified according to seizure type and seizure localization by a consultant neurologist who was blind to the neuropsychological and experimental results. Classification was based on one or more of the following: clinical manifestations during seizure; localizing EEG abnormalities (with video-EEG telemetry where appropriate); and MRI abnormalities.

Assessment procedures: standard

Patients and control subjects were assessed on a battery of standardized neuropsychological measurements on two separate occasions ~8 weeks apart. At the time of testing no patient was postictal. Tests administered were subtests from the following: the Wechsler Adult Intelligence Scale—Revised (WAIS-R; vocabulary, similarities, digit span, arithmetic and block design) (Wechsler, 1981) for patients; the Ravens Advanced Matrices (Raven, 1985) for controls; the National Adult Reading Test (NART; Nelson, 1991); the Logical Memory and Visual Reproduction subtests from the Wechsler Memory Scale—Revised (WMS-R; Wechsler, 1987); the Recognition Memory Test (words) (Warrington, 1984); the Camden Topographical Memory Test (Warrington, 1996); Graded Naming Test (McKenna and Warrington, 1983); the Incomplete Letters and Cube Analysis subtests from the Visual Object and Space Perception Test (Warrington and James, 1991); the modified Wisconsin Card Sort Test (WCST; Nelson, 1976); and the Hospital Anxiety and Depression Scale (HAD; Zigmond and Snaith, 1983). In addition, subjects were asked to rate the degree to which their memory was a nuisance in everyday life using a four-category rating scale (ranging from none to severe, as in Corcoran and Thompson, 1992).

Assessment procedures: experimental (assessment of very long-term memory)

Participants were required to learn one of two stories (version 1 or 2) from the Adult Memory and Information Processing Battery (Coughlan and Hollows, 1985) to 90–100% correct over two consecutive trials up to a maximum of 10 trials. They were then tested on recall of the story at a standard interval of 30 min after presentation of the last trial. Participants were randomly assigned to either version of the stories, although within families from which two members were participating the assignment was not entirely random, as the second participant in a family was always given a different version from the first participant. This procedure was designed to exclude the possibility of relearning the stimuli within families during the retention interval. In addition, participants were specifically requested not to discuss details of the tests with other participants. Very long-term recall and recognition was subsequently tested ~8 weeks later. This interval was chosen to be similar to the retention interval of 6 weeks used by Kapur and colleagues (Kapur et al., 1997), and for the practical reason that many patients received routine follow-up appointments in the neurology clinic at 8 week intervals, thus saving participants from making unnecessary journeys.

Story recall was scored strictly according to the manual (i.e. 2 points for each correctly recalled idea unit, 1 point for each partially recalled unit). Recall of the story was converted to `proportion recalled' scores. Recognition was tested for the most salient points of the story only in a forced-choice procedure using 12 questions, for each of which three plausible, closely related answers were given. The questions were designed to assess one idea unit only and to be comparable across story versions.

Statistical analysis

All statistical analyses were carried out using SPSS for Windows. Results were analysed using mixed ANOVA (analysis of variance), independent t tests that did not assume equal variances, Mann–Whitney U tests and Spearman or Pearson correlations as appropriate. Significance levels were set at alpha ≤ 0.05.


Twenty-one patients and 16 controls were included in the final analyses. Two patients and three controls were lost to follow-up. (Participants lost to follow-up did not differ significantly from the remaining sample.)

Patients had a mean age of seizure onset of 12.95 years (SD = 8.26, range 1–37), their mean duration of illness was 20.81 years (SD = 12.70, range 1–45), and they were taking a mean of 2.05 anti-epileptic medications (SD = 0.86, range 0–3). Monthly seizure frequency means were as follows: simple partial seizures, 6 (SD = 20, range 0–87); complex partial seizures, 13 (SD = 21; range 0–91); secondary generalized seizures, <1 (SD = 1, range 0–4).

As has been observed in previous studies, more than half of the patients complained of memory difficulties in everyday life but there was no evidence of significant amnesia on clinical testing. Subjective nuisance ratings of memory problems by patients were quantitatively similar to those reported by Corcoran and Thompson (Corcoran and Thompson, 1992). For comparison, the present data are accompanied by Corcoran and Thompson's findings in parentheses: not a nuisance, 14.3% (12%); mild 38.1% (34%); moderate, 38.1% (35%); severe, 9.5% (19%).

Table 1 describes the patient and control groups in terms of their performance on neuropsychological tests. Independent t tests revealed no significant differences between the patient and control groups with respect to any of the standard neuropsychological variables. Specifically, there were no significant differences between patients and controls on standard memory tests. The patients' mean score on the anxiety scale of the Hospital Anxiety Depression Scale was in the borderline range. The mean score of the controls on the anxiety scale was at the upper end of the normal range. On the depression scale both groups scored within the normal range. There were no significant differences on either scale between groups. Controls were significantly older than patients (P = 0.004), but excluding the four oldest controls to equate groups for age did not have any significant effect on the results.

View this table:
Table 1

Characteristics and neuropsychological data for patients and controls

Mean (SD)RangeMean (SD)Range
* Mean based on 20 patients; mean based on 19 patients; mean based on 18 patients; § mean based on 15 controls.
Males : females 7 : 14 6 : 10
Age (years) 33.76 (9.72)20–49 46.25 (14.54)18–65
Education (years) 12.14 (1.96)10–19 11.75 (2.44) 9–18
General intellectual function
NART IQ103.65* (12.72)80–122101.88 (13.20)71–117
WAIS-R Verbal IQ 93.94 (11.84)74–129
Ravens Advanced Matrices 6.73§ (2.96) 0–11
Recognition Memory Test (words) 47.76 (3.03)40–50 46.00 (4.43)39–50
Camden topographical 26.23 (2.94)18–30 26.44 (2.34)22–29
WMS-R (raw scores)
Logical memory I 28.43 (6.31)14–37 27.50 (5.83)19–37
Logical memory II 22.29 (8.95) 4–36 22.00 (7.89) 9–33
Visual reproduction I 35.57 (4.17)24–41 34.44 (4.57)25–41
Visual reproduction II 32.19 (6.88)15–40 30.56 (6.39)19–40
Naming skills
Graded Naming Test 19.26 (4.50) 9–26 22.25 (4.22)15–29
Perception: Visual Object and Spatial Perception Test
Incomplete letters 19.52 (0.51)19–20 19.25 (0.77)18–20
Cube analysis 9.86 (0.36) 9–10 9.44 (0.89) 7–10
Modified WCST (categories) 5.52 (1.21) 2–6 5.63 (0.81) 3–6
HAD anxiety 8.90 (4.65) 1–17 6.93§ (2.81) 1–12
HAD depression 2.95 (2.13) 1–11 3.67§ (2.53) 1–10

Experimental very long-term memory test

As two versions of the story were given, the possibility that one version was harder to learn or was differentially forgotten following the 8 week delay was considered. Comparing story versions within the control group, there was no significant difference between versions for either the proportion of story recalled during learning trials 1–3 or the number of trials to reach criterion. Nor were there any significant differences between the versions for recall at 30 min or recall and recognition at 8 weeks. Additionally, a mixed ANOVA for effect of delay and story version for controls was performed; it did not reveal either main effects for version or interaction effects for version × delay. It was therefore assumed that the two versions could be considered equivalent for the present analyses.

Although the aim was to see patients following an interval of 56 days, in practice it was not possible to adhere strictly to this criterion as the patient's appointments were often arranged to coincide with clinic appointments and some participants travelled long distances in order to attend. Very long-term memory was assessed at mean intervals of 58.86 days (SD = 5.59, range 53–77) for patients and 61.75 days (SD = 10.85, range 47–91) for controls. The mean difference between groups was not significant. For the purpose of reporting the data obtained, very long-term memory measures will be labelled as having been recorded at 8 weeks.

Analysis 1: patients versus controls

A Mann-Whitney test revealed no significant differences between patient and control groups on the numbers of learning trials. Seventeen patients and 14 controls reached the learning criterion. Three patients and one control failed to reach this criterion and one patient achieved the 90% criterion only on the tenth trial. (Taking learning to criterion was not entirely straightforward. Six cases received between –2 and +2 trials. These scoring discrepancies had no significant effect on the overall pattern of results.)

Each patient received at least three learning trials (Table 2) and independent t tests revealed no significant differences between the patient and control groups on learning trials 1–3.

View this table:
Table 2

Patients and controls: mean, standard deviation and range of proportion of story recalled for learning trials, 30 min and 8 week recall and score for 8 week recognition

Mean (SD)RangeMean (SD)Range
* Means based on 19 patients and 15 controls.
Learning trial 10.44 (0.22)0.07–0.84 0.50 (0.21)0.12–0.93
Learning trial 20.65 (0.25)0.11–0.96 0.74 (0.22)0.17–0.96
Learning trial 30.77 (0.23)0.25–0.98 0.85 (0.16)0.47–0.98
30 min recall0.89* (0.09)0.63–0.98 0.93* (0.06)0.78–1.00
8 week recall0.28 (0.28)0.00–0.71 0.52 (0.27)0.00–0.86
8 week recognition score9.95 (1.96)6–1211.00 (1.10)9–12

Means and standard deviations for recall of the story at 30 min and 8 weeks and recognition at 8 weeks are shown in Table 2. Patients and controls were similar in the proportion of the story recalled following a delay of 30 min. Both groups showed forgetting over the 8 weeks: the mean proportion of the story recalled at 8 weeks was lower than the proportion recalled at 30 min. The mean proportion recalled at 8 weeks was lower in the patient group than in the control group.

A mixed ANOVA with factors of participant group and delay (30 min versus 8 weeks) was performed using recall score as the dependent variable. The analysis revealed main effects for delay (i.e. 8 week recall was significantly poorer than 30 min recall, as expected), a main effect for group [F(1,32) = 6.615, P = 0.015], showing the patient group to be significantly poorer overall and, most critically, an interaction effect of delay with group [F(1,32) = 5.243, P = 0.029], indicating that the patients' recall was differentially poorer than that of the controls following an 8 week delay. Planned independent t tests revealed no significant group differences at 30 min, but confirmed that the patient group was significantly poorer on recall of the story at 8 weeks (P = 0.014) and recognition of the story at 8 weeks (P = 0.047).

Analysis 2: left- versus right-hemisphere patients

The above comparison was repeated between left- and right-hemisphere focus patients. (See Table 3 for clinical and standard neuropsychological data.) Patients did not differ significantly on any neuropsychological or clinical variables. In particular, there were no significant differences between the groups on either standardized verbal or non-verbal memory measures.

View this table:
Table 3

Characteristics and neuropsychological data for left and right hemisphere patients

Left hemisphere (n = 11)Right hemisphere (n = 10)
Mean (SD)RangeMean (SD)Range
* Mean based on n = 10 (left hemisphere); mean based on n = 9 (right hemisphere); mean based on n = 8 (right hemisphere). AED = anti-epileptic drug.
Age 36.45 (8.66)22–49 30.80 (10.38)20–48
Education (years) 12.55 (2.42)10–19 11.70 (1.25)10–14
Interval (days) 59.00 (6.25)55–77 58.70 (5.10)53–70
Age of onset (years) 13.73 (9.79) 1–37 12.10 (6.59) 1–23
Duration of epilepsy (years) 22.73 (13.32) 1–45 18.70 (12.32) 2–34
Number of AEDs 2.00 (0.77) 1–3 2.10 (0.99) 0–3
Monthly seizure frequency
Simple partial 4 (9) 0–30 9 (27)0–87
Complex partial 6 (7) 0–26 21 (28)0–91
Secondary generalized <1 0–1 <10–4
General intellectual functioning
NART IQ106.36 (13.02)80–122100.33 (12.26)81–120
WAIS-R verbal IQ 96.80* (13.64)82–129 90.38 (8.67)74–97
Recognition Memory Test (words) 48.73 (1.56)46–50 46.70 (3.92)40–50
Camden topographical 26.27 (3.44)18–30 26.20 (2.49)23–29
WMS-R (raw scores)
Logical memory I 29.00 (5.95)19–36 27.80 (6.96)14–37
Logical memory II 23.36 (8.71) 4–36 21.10 (9.52) 7–33
Visual reproduction I 36.27 (3.00)31–40 34.80 (5.22)24–41
Visual reproduction II 32.90 (4.74)22–39 31.40 (8.88)15–40
Naming skills
Graded Naming Test 19.64 (4.92) 9–26 18.75 (4.10)13–25
Perception: Visual Object Space Perception Test
Incomplete letters 19.36 (0.50)19–20 19.70 (0.48)19–20
Cube analysis 10.00 (0.00)10 9.70 (0.48) 9–10
Modified WCST (categories) 5.64 (1.21) 2–6 5.40 (1.26) 2–6
HAD anxiety 8.64 (5.75) 1–17 9.20 (3.33) 4–15
HAD depression 2.64 (1.29) 1–5 3.30 (2.83) 1–11

Left- and right-hemisphere patients did not differ significantly in the number of learning trials received or the proportion of the story recalled over learning trials 1–3. Data are shown in Table 4.

View this table:
Table 4

Left- and right-hemisphere patients: mean proportion of story recalled over learning trials 1–3, 30 min and 8 week delayed recall and score for 8 week recognition

Left hemisphereRight hemisphere
Mean (SD)RangeMean (SD)Range
* Mean based on 10.
Learning trial 10.41 (0.21)0.17–0.80 0.47 (0.23)0.07–0.84
Learning trial 20.70 (0.19)0.37–0.93 0.61 (0.31)0.11–0.96
Learning trial 30.83 (0.16)0.47–0.98 0.70 (0.28)0.25–0.96
30 min recall0.88* (0.10)0.63–0.97 0.91 (0.08)0.77–0.98
8 week recall0.15 (0.21)0.00–0.63 0.42 (0.29)0.00–0.71
8 week recognition score9.27 (2.10)6–1210.70 (1.57)7–12

Means and standard deviations of the proportion of the story recalled at 30 min and 8 weeks and recognition score at 8 weeks are shown in Table 4. The mean proportion of story recalled at 30 min was similar between the groups, but the proportion recalled at 8 weeks was lower in the left-hemisphere group than in the right-hemisphere group and the control group.

A mixed ANOVA was carried out with factors of participant group (left- versus right-hemisphere patients versus controls) and delay (30 min versus 8 week), with recall score as the dependent variable. There was a significant main effect for delay and a significant main effect for group [F(2,31) = 6.031, P = 0.006]. There was also a significant interaction of delay by group [F(2,31) = 5.018, P = 0.013]. A planned comparison between left-hemisphere patients and controls was significant, left-hemisphere patients performing significantly poorer at 8 week delayed recall (P = 0.001). In addition, planned comparisons between left- and right-hemisphere patients for 8 week delayed recall revealed that the performance of left-hemisphere patients was significantly poorer (P = 0.028). There were no significant differences between right-hemisphere patients and controls. There were no group differences on recall following a 30 min delay.

Analysis of recognition scores following the 8 week delay also showed that left-hemisphere patients were significantly poorer than controls (P = 0.026). Left-hemisphere patients were also poorer than right-hemisphere patients at 8 week recognition, but this did not quite reach statistical significance (P = 0.096). As with 8 week recall measures, there was no significant difference between right-hemisphere patients and controls on 8 week recognition. There was a significant negative relationship between patients' subjective ratings of memory nuisance and scores on the 8 week recognition test (P = 0.036).

Analysis 3: left- versus right-temporal patients

The analysis was repeated comparing left and right-temporal patients only. Of the left-temporal group, MRI data were available for eight of the nine patients. In five cases the MRI was normal, in one case there was hippocampal sclerosis, in one case there was an arachnoid cyst and in one case there was asymmetry of the temporal horns. In the right-temporal group, MRI revealed evidence of hippocampal sclerosis in four patients and in one case the MRI was normal. The groups did not differ significantly on clinical or standardized neuropsychological variables (Table 5). In particular, there were no significant group differences on standardized verbal and non-verbal memory measures.

View this table:
Table 5

Characteristics and neuropsychological data for left- and right-temporal patients

Left temporal (n = 9)Right temporal (n = 5)
Mean (SD)RangeMean (SD)Range
* Mean based on n = 8 (left temporal); mean based on n = 3 (right temporal). AED = anti-epileptic drug.
Age 36.67 (9.41)22–49 33.20 (10.94)21–48
Education (years) 12.78 (2.64)10–19 11.20 (1.10)10–13
Interval (days) 59.44 (6.88)55–77 58.80 (6.26)56–70
Age of onset (years) 14.33 (10.65) 1–37 9.80 (5.26) 1–14
Duration of epilepsy (years) 22.33 (14.87) 1–45 23.40 (9.53) 8–34
Number of AEDs 2.00 (0.87) 1–3 2.40 (0.55) 2–3
Monthly seizure frequency
Simple partial 5 (10) 0–30 17 (39) 0–87
Complex partial 6 (8) 0–26 15 (17) 1–43
Secondary generalized <1 0–1 <1 0 to 1
General intellectual function
NART IQ105.22 (13.27)80–122101.00 (13.23)86–120
WAIS-R Verbal IQ 98.13* (15.02)82–129 91.60 (6.95)81–97
Recognition Memory Test (words) 48.89 (1.36)46–50 48.40 (2.07)45–50
Camden topographical 26.11 (3.69)18–30 26.40 (2.70)23–29
WMS-R (raw scores)
Logical memory I 28.44 (6.50)19–36 27.80 (5.76)19–33
Logical memory II 21.56 (8.34) 4–29 17.80 (11.39) 7–33
Visual reproduction I 36.00 (3.20)31–40 33.60 (6.95)24–41
Visual reproduction II 32.44 (5.17)22–39 26.80 (10.73)15–39
Naming skills
Graded Naming Test 19.33 (5.36) 9–26 18.67 (2.31)16–20
Perception Visual Object and Space Perception Test
Incomplete letters 19.44 (0.53)19–20 19.80 (0.45)19–20
Cube analysis 10.00 (0.00)10 9.80 (0.45) 9–10
Modified WCST (categories) 5.56 (1.33) 2–6 5.00 (1.73) 2–6
HAD anxiety 8.00 (5.59) 1–15 10.20 (3.19) 6–15
HAD depression 2.67 (1.22) 1–5 3.80 (4.15) 1–11

The standard memory variables were correlated with each other and the story-learning task, but very long-term delayed memory was not predicted by any of these indices (Table 6). In addition, there was no relationship between naming performance, anxiety or depression levels and very long-term memory variables.

View this table:
Table 6

Relationships between standard and very long-term memory tests in left and right-temporal patients

RMT (words)CamdenLM ILM IIVR IVR IITrial 1 30 min recall8 week recall8 week recognition
RMT = Recognition Memory Test; LM = logical memory; VR = visual reproduction.
Standard memory tests
Recognition Memory Test (words)+ve+ve+ve
Camden memory test+ve
Logical memory I+ve+ve
Logical memory II+ve+ve+ve–ve
Visual reproduction I+ve+ve+ve
Visual reproduction II+ve+ve+ve
Experimental memory tests
Trial 1+ve
30 min recall
8 week recall+ve
8 week recognition–ve+ve

Right-temporal patients tended to learn the story more slowly than the left-temporal patients, but there were no significant group differences in either the proportion of story recalled over learning trials 1–3 (Table 7) or the number of trials to reach criterion.

View this table:
Table 7

Left and right-temporal patients: mean (SD) proportion of story recalled over learning trials 1–3, 30 min recall and 8 week recall and score for 8 week recognition

Left temporalRight temporal
Mean (SD)RangeMean (SD)Range
* Mean based on n = 8.
Learning trial 10.38 (0.23)0.17–0.80 0.45 (0.25) 0.07–0.67
Learning trial 20.67 (0.20)0.37–0.93 0.60 (0.31) 0.14–0.86
Learning trial 30.80 (0.17)0.47–0.98 0.68 (0.33) 0.25–0.95
30 min recall0.86* (0.11)0.63–0.97 0.89 (0.08) 0.77–0.97
8 week recall0.09 (0.12)0.00–0.63 0.46 (0.34) 0.00–0.71
8 week recognition9.11 (2.09)6–1211.40 (0.55)11–12

Means and standard deviations for long-term memory measures are shown in Table 7. As with the left- versus right-hemisphere analysis, the left and right-temporal groups were similar to each other and controls on recall of the story at 30 min. There was a difference between the groups in the proportion of story recalled following the 8 week delay. On this measure, the left-temporal patients scored below the level of right-temporal patients and controls, whose performance was similar.

A mixed ANOVA with factors of participant group (left versus right-temporal versus control) and delay (30 min versus 8 week) was performed, with recall score as the dependent variable. There were main effects of delay and group [F(2,25) = 7.451, P = 0.003]. There was also a significant interaction effect of delay with group [F(2,25) = 6.725, P = 0.005]. Planned comparisons between left-temporal patients and controls revealed left-temporal patients to be significantly poorer on recall of the story following an 8 week delay (P < 0.001). Left-temporal patients also tended to be poorer than the right-temporal patients on this measure, but the differences did not quite reach statistical significance (P = 0.070). There were no significant group differences for recall after a 30 min delay. On the 8 week recognition measurement, left-temporal patients performed significantly poorer than both right-temporal patients (P = 0.011) and control patients (P = 0.029). Again, there were no differences between right-temporal patients and controls on either 8 week recall or recognition.

Five of the left-temporal group scored at floor level on 8 week delayed recall, compared with only one right-temporal patient. Three right-temporal patients scored in the 0.6–0.8 range for the proportion of story recalled at 8 weeks. The only right-temporal patient who scored at floor on the 8 week recall test reported particularly severe recurrent generalized seizures over a 3 day period during the retention interval.

Of the left-temporal patients who rated their memory difficulties as a moderate nuisance, 75% scored at floor level on recall of the story at 8 weeks compared with only 40% of patients who rated their memory difficulties as a mild nuisance. Although there appeared to be some relationship between memory nuisance ratings and performance on very long-term delayed tests, there were some patients who complained of significant memory problems yet performed satisfactorily on long-term retention tests.


In this study we have found evidence of accelerated long-term forgetting of verbal material in people with left temporal lobe epilepsy. We have therefore demonstrated the unique contribution of using very long-term retention intervals to evaluate memory functions in patients with epilepsy. These findings indicate that tests sensitive to amnesia are not, by themselves, necessarily the most sensitive or appropriate for people with epilepsy. Our patients' deficits were apparent on recall and recognition of a story following a delay of 8 weeks. The left-temporal epilepsy group were indistinguishable from the right-temporal lobe group and controls on measures of initial learning and recall following 30 min delay, as well as on other standardized neuropsychological memory tests of short-term and episodic memory. Our findings therefore indicate a selective memory deficit in the left-temporal lobe group that is only apparent after a substantial retention interval. In this discussion we will consider the relevance of our findings for theories of memory organization and process and subsequently in terms of their implications for diagnosis and management of memory deficits in epilepsy.

First, however, it is necessary to consider the role of any nuisance variables in influencing our results. One possibility is that accelerated forgetting of verbal material is a non-specific consequence of different learning abilities, different ages of epilepsy onset, seizure frequency, seizure severity, medication, or structural damage arising from repeated epileptic activity. These confounding variables would have been difficult (if not impossible) to evaluate had our memory task not been one which was specifically designed to tap verbal memory. The use of a verbal task allowed us to establish that all of our cases were appropriately matched on initial acquisition and allowed us to specify in advance our hypothesis that people with left-temporal lobe seizure activity would be specifically impaired whereas those with right-sided foci would be largely unaffected. This hypothesis was supported by our data. Patients in the left-sided group were indistinguishable from those in the right-sided group on all of these potentially confounding variables, yet they were the only group to show accelerated forgetting.

We are also able to rule out any significant contribution from comorbidity on other neuropsychological dimensions. One key potential source of confounding is that of language function. Hermann and colleagues (Hermann et al., 1992) and Mayeux and colleagues (Mayeux et al., 1980) suggested that memory deficits in temporal lobe epilepsy might be secondary to anomia. (Mayeux and colleagues demonstrated that dysnomia was prominent in left-temporal lobe patients.) However, as a group, our cases showed normal performance on a stringent test of naming ability (Graded Naming Test). Moreover, performance on the Graded Naming Test was not correlated with performance on very long-term delayed memory tests, indicating that vocabulary size and ease of verbal retrieval in a naming task was not associated with very long-term episodic memory functioning. It has also been suggested that negative mood can exacerbate memory failure in people with epilepsy (Thompson and Corcoran, 1992). While it should be noted that the mean anxiety score in our epilepsy sample was in the borderline range, it did not differ between right- and left-sided cases or from controls. Scores on the measure of depression were comfortably within the normal range. Therefore, as a group the people with epilepsy were not significantly more anxious or depressed than the general population or the control participants in this study. Furthermore, at the individual level, scores on the Hospital Anxiety and Depression Scale bore no relationship to performance on memory tests. Negative mood is therefore insufficient to explain our results. We think that the simplest explanation of our findings is that an active temporal lobe epileptic focus is sufficient for accelerated long-term forgetting of verbal material. We wish to argue that our data provide secure evidence for an extended period of memory vulnerability in the human brain.

Our evidence for an extended period of memory vulnerability has direct implications for the theoretical analysis of memory storage and representation. Critically, our data showed normal learning and normal memory performance after a short delay in all of our patient subgroups, which may be a consequence of the low incidence of hippocampal sclerosis. This replicates many other findings in this field (e.g. Martin et al., 1991; Kapur et al., 1997), but is the first group study of non-surgical patients to show such an effect. We suggest that these findings effectively rule out the possibility of a deficit in encoding or retrieval as a source of the accelerated forgetting result (e.g. Martin et al., 1991; Hermann et al., 1992). The left-temporal group demonstrated intact retrieval processes during learning and following short delays. The integrity of memory processes at shorter delays renders any simple retrieval hypothesis inadequate. Some variant of the retrieval hypothesis might be invoked by proposing that different processes underlie retrieval at different time periods. However, such an interpretation requires an independently specified theory of the interaction between memory age and memory retrieval: without such an account this is hardly a compelling explanation.

We have shown that memories have an extended period of vulnerability; the most straightforward interpretation of these findings is in terms of an extended period of memory consolidation (Squire et al., 1984). A similar interpretation has been advanced in order to account for data from single case studies of accelerated forgetting (e.g. Kapur et al., 1997) as well as for the deficits observed in patients with TEA (e.g. Zeman et al., 1998). However, our study is the first prospective group study to provide secure evidence for such an extended process. Our findings mean that we are able to generalize beyond the data of a single case and to claim that such problems are potentially a significant consideration in any case of temporal lobe epilepsy; moreover, we are able to draw some conclusions regarding the anatomical substrate of the hypothetical consolidation process itself. It seems likely that the disruption of memory consolidation in epilepsy lies on a continuum, with long-term global amnesia lying at one end of the continuum (affecting all types of material), TEA patients occupying an intermediate position (with some autobiographical and event knowledge impairment) and patients with left temporal lobe epilepsy manifesting a material-specific profile of focal memory dysfunction.

The notion of a `consolidation deficit' has a long and respected history in the study of memory disorders. While it is rarely considered to be a complete explanation of the global amnesic syndrome, there is some evidence to suggest that medial temporal lobe injury may disrupt some crucial aspects of the consolidation process (Huppert and Piercy, 1979). Squire and Alvarez have developed these ideas further and have argued for the existence of two long-term memory consolidation processes, characterized by fast and slow rates of storage (Squire and Alvarez, 1995). They have proposed that the `fast consolidation' process is mediated by medial temporal lobe structures (including the hippocampus); the slower neocortical consolidation process arises as a consequence of the repeated and synchronous firing of hippocampal–neocortical connections. The relevant activated neocortical sites gradually form connections between each other: a process that constitutes `slow consolidation'.

Our findings are compatible with such a theory. The results from the left-temporal group suggest that the fast hippocampally based learning system is intact, presumably due to the very low incidence of hippocampal sclerosis in the group, but that the slow-learning process dependent on left-hemisphere hippocampal–cortical interactions is impaired. One possibility is that epileptic activity commencing in the temporal lobe and hippocampus disrupts or degrades the processes of synchronous activation of neocortical cell assemblies. Alternatively, there may be a functional disconnection between the cortical and hippocampal systems arising as an adaptation to temporal lobe epilepsy. Such a deficit might be partial rather than complete.

Our results suggest that very long-term consolidation of verbal material is disrupted by epilepsy with a focus in the left temporal lobe. We cannot assume that activity during the retention interval was confined to the temporal areas or even to the ipsilateral hemisphere, since epileptic activity typically spreads outwards to involve wider regions of the brain, but it should be noted that patients suffered mainly from partial seizures and the incidence of generalized seizures was rare in most of the patients. Thus, seizures spreading to involve contralateral regions was likely to be very infrequent in most of the patients. Our findings simply point towards the importance of a stable environment in the left temporal lobe for slow memory consolidation processes to occur. In this context it is interesting to note that the only right-temporal case who scored poorly on the story task (i.e. at floor level) reported a particularly severe bout of secondary generalized epilepsy during the retention interval.

At present we are unable to determine the exact mechanism by which slow consolidation processes are disrupted in epilepsy, other than to note that overt seizure frequency (as reported by the patients) was not reliably related to memory performance. These findings are similar to those of Bergin and colleagues, who reported that the frequency of overt seizures over a 48 h period bore no relationship to memory scores (Bergin et al., 1995). On first consideration, these observations might be construed as evidence against the interference explanation of consolidation failure. They are quite compatible with the hypothesis of a functional disconnection. However, the relationship between asymptomatic epileptiform EEG abnormalities and memory retention has not been established. It is possible that the total frequency of ictal and interictal epileptic discharges may correspond more closely to the disruption of consolidation processes indicated by accelerated forgetting. In the literature there are some suggestions that a reduction in seizures is associated with improved performance on memory tests (O'Connor et al., 1997). Moreover, memory difficulties in TEA are also reported to improve in response to anticonvulsant medication (Zeman et al., 1998). At the very least, such findings indicate that any disconnection syndrome is potentially reversible.

The inference of a dissociation between fast and slow consolidation processes is not affected by the difficulty involved in determining the precise mechanism by which they are dissociated. The evidence for dissociation rests first on the differential effects of retention intervals over which forgetting occurred and secondly on the laterality effects that we have documented in our patient group. At shorter delays (i.e. ≤30 min) our subject groups were indistinguishable on tests of memory, suggesting that the fast system was functioning in controls and left- and right-sided cases (together with the adjunct systems involved in memory organization, encoding and retrieval). This level of performance would be quite beyond the capability of patients with a classical amnesic syndrome. However, as we have already noted, the left temporal lobe group showed substantial forgetting of the story as the retention interval was prolonged. The fact that this effect was specifically observed in the left-sided cases indicates that it is not a by-product of task difficulty or of any other nuisance variable. While we are unable to claim strong evidence for material specificity (since we did not have a suitable non-verbal test for right-sided cases), we can note that our results mirror previous work on verbal memory impairment in cases of unilateral cerebral dysfunction (e.g. Milner, 1971). Our findings are consistent with a dissociation in function between the fast and slow memory consolidation systems and may even suggest that comparable material-specific constraints may apply across the two domains.

Further research is needed to address the question of whether consolidation of non-verbal material is disrupted by an epileptic focus in the right or left hemisphere. There are anecdotal suggestions that a focus in the right temporal lobe may adversely affect long-term retention of spatial information (Lisyak, 1997). Moreover, bilateral involvement is commonly cited in cases where global failure of consolidation is suspected (Kapur et al., 1997; O'Connor et al., 1997; Zeman et al., 1998). Future work will be necessary in order to fully establish the relationship between the severity of epileptic abnormalities, the locus of epileptic activity and the magnitude of any consolidation failure. Moreover, these variables will need to be systematically related to the cognitive variables of memory type and memory task in order for a clearer picture to be established.

While much work remains to be done in order to clarify the neurobiology of memory in epilepsy, our findings provide a partial explanation for the findings that people with epilepsy may complain of poor memory while continuing to perform normally on clinical tests. In our left-hemisphere group, complaints of poor memory were significantly associated with performance on the very long-term delayed recognition memory test. A problem in consolidation may be directly mirrored by problems experienced in everyday life, at least for one subgroup of patients. However, we only have a partial explanation: in the group as a whole, complaints of poor memory were actually more common than was failure on the very long-term verbal memory test. This almost certainly occurred because we were able to use only a single long-term measure (i.e. story recall) in this investigation. We speculate that it should be possible explain a substantial part of the remaining variance by using a wider sample of very long-term memory tests. Thus, it is possible that a task using non-verbal material might provide an objective index of subjective memory difficulties in the right hemisphere group. The use of very long-term delayed measures can potentially provide us with the means of differentiation between patients who consider themselves to have a poor memory as a result of long-term memory consolidation and patients who consider their memory to be poor for other reasons. This would have direct implications for therapy and patient management. Clinicians might then be in a position to address possible compensatory strategies or confidence-building techniques depending on the precise nature of their clients' memory problem.

By establishing that epilepsy disrupts longer-term consolidation, we hope we have clarified a segment of a particularly confusing literature. We also hope that we have highlighted the unique contribution to clinical assessment that can be made by very long-term testing. Rather than asking the question `why do they complain', we are now in the position of being able to ask the more appropriate question of `how do we make it better'. Patients may benefit from the rehearsal and relearning of material, particularly for critical information, to increase the likelihood of intact storage and resilience to disruption. By taking into account the role of accelerated forgetting, clinicians may be better placed to address the critical issues of remediation in epilepsy.


We wish to thank the patients and volunteers for their participation and Dr Roger Johnson for his support during the study. This study was initially conceived and planned by the late Dr Kristin Breen and is dedicated to her memory.


View Abstract