Patients and methods

The present cohort consists of eighty consecutive patients with medically intractable seizures who underwent en bloc temporal lobe resection in whom pathological examination of the excised material disclosed a focal lesion. All patients were operated on by the same neurosurgeon (CEP) between 1975 and 1995 who also obtained a detailed and standardised history from all patients. Preoperative clinical and investigative data were gathered retrospectively from the case notes. Details on medical and developmental history, seizure characteristics, duration of epilepsy, age at surgery, results of interictal EEG recording, neuropsychological testing, and neuroimaging were entered into a computerised database.

In the great majority of cases the decision to operate was made on the basis of congruent data from the clinical history and examination, neuropsychological assessment, routine and sleep EEG recordings, and the findings of structural neuroimaging, most commonly CT because high resolution MRI became available only in 1993. For patients in whom a structural lesion was present on imaging an operation was performed when the principal EEG focus was congruent and the lesion could be removed without risk of deficit. The presence of generalised spikes or other discrete extratemporal foci were not themselves a contraindication. For cases with normal CT the decision to operate was made on the basis of repeated interictal EEG recordings together with considerations of the seizure characteristics and the neuropsychological profile of the patient. Most patients proceeded to surgery without ictal recordings but if these were performed foramen ovale electrode recording was usually employed. Neuropsychological assessment was carried out using the Wechsler intelligence scales for adults and children, obtaining verbal (VIQ) and performance (PIQ) intelligence quotients. Verbal and visuospatial memory were assessed using a battery of investigations including the Wechsler logical memory subtest and the Benton and the Ray-Osterreith tests. Specific verbal or performance IQ (VIQ, PIQ) deficits and deficits in verbal or visuospatial memory were regarded in the assessment process as suggesting dominant and non-dominant hemispheric or temporal dysfunction respectively. An intracarotid amytal test was performed in patients with uncertain language dominance and in those considered at risk of a postoperative amnesic syndrome. Cerebral dominance was presumed to be left sided if the patient was right handed and the neuropsychological data from patients who were left handed or ambidextrous and who did not have a Wada test were omitted from the analysis of pyschological variables.

A standardised anatomical resection was performed which consisted of removal of the anterior part of the temporal lobe, extending backwards 5.5 cm from the temporal pole in the dominant hemisphere, sparing the posterior half of the superior temporal gyrus, and 6.5 cm backwards in the non-dominant hemisphere.17 In every case the amygdala and hippocampus were included in the resection. Acute electrocorticography was routinely performed but rarely dictated the extent of resection, which primarily depended on anatomical constraints and the feasibility of complete removal of a mass lesion.

Postoperative follow up was conducted personally by the senior author (CEP) at least on an annual basis for the first 5 years and information on seizure status was recorded on a computerised database. For the purposes of actuarial analysis patients were considered “seizure free” if they attained 12 consecutive months of absolute seizure freedom. Patients with persistent auras were classified as “seizure free” but all other postoperative seizures such as isolated tonic-clonic seizures on attempted antiepileptic drug withdrawal were classified as not “seizure free”.

Where possible all original neuropathological material was re-examined for classification of tumour histology and determination of completeness of resection (in the tumour cases) by examination of the resection margins. In the tumour cases a qualitative assessment was made of the presence or absence of neuronal loss and gliosis in the hippocampus. In addition, the presence of coexistent cortical dysplasia denoted by the presence of dysplastic neurons at the periphery of the tumour and abnormalities in cortical lamination in the adjacent neocortex was noted in the tumorous cases. Pathological reclassification was carried out by one neuropathologist, who was blinded to the postoperative outcome in all cases. Detailed diagnostic criteria for the diagnosis of dysembryoplastic neuroepithelial tumour have been published previously.18

STATISTICAL ANALYSIS

Survival analysis using Kaplein-Meier curves and actuarial statistics19 was used to establish the variables associated with eventual seizure remission after surgery. Although we were primarily interested in the chances of patients eventually achieving seizure freedom, survival analysis was used to take into account the different lengths of follow up as some patients took more time to achieve remission than others. The log rank test was used to test whether the Kaplein-Meier curves differed between subgroups defined by certain clinical and investigative variables. To describe remission in the subgroups we also computed the percentage of patients in remission after 5 years (the median follow up) in each subgroup. The significance value given below in the tables is for the log rank test, which does not necessarily reflect the subgroup differences in the percentage remission at 5 years, reflecting instead the difference in the entire survival function. The different variables tested using the log rank test are not independent of each other; for example, duration of epilepsy and age at surgery were found to be highly correlated. Therefore, to determine whether an apparent effect of a single variable is direct or due to confounding with an another variable, Cox regression analysis20 was used to assess the direct effect of each of a set of variables after controlling for the other variables in the group. An automatic forward selection procedure based on the likelihood ratio was used with p values of 0.1 and 0.2 for entry and removal respectively. Statistical analysis was performed using SSPS (SSPS v 8.0, SSPS Inc, Chicago, IL, 1997).

Results

Of the 80 patients, 39 were male and 41 were female. The median age at surgery was 19 years (range 4–48 years) and the mean duration of postoperative follow up was 60 months. The clinical characteristics, investigative results, and pathological features, and their relation to outcome are set out in tables 1 and 2. Overall, the probability of achieving a 1 year seizure remission was 58% by 1 year, 65% by 2 years, and 71% by 5 years. Thereafter, no further patients remitted.

CLINICAL VARIABLES AND OUTCOME

The incidence and influence on seizure outcome of the clinical and investigative features are given in tables 1 and 2. Age at surgery and duration of epilepsy were both significantly associated with outcome. If operated on before the age of 15 years, 84% achieved a good outcome, compared with a remission rate of 44% if operated on after the age of 30 years. Similarly, a duration of epilepsy of less than 10 years was associated with a remission rate of 88%, compared with 57% if the duration had exceeded 20 years. Forty nine patients had onset of epilepsy before the age of 10 but this variable did not influence seizure remission rates. There was also a significantly worse outcome if patients had required special schooling. A non-significant trend (p=0.10) towards a poor outcome was noted for patients with a prior psychiatric history but the presence of pregnancy or perinatal complications, delayed developmental milestones, family history of epilepsy, significant brain injury, complicated febrile convulsions or seizure characteristics, particularly the presence of generalised tonic-clonic seizures, were not associated with outcome.

INVESTIGATIVE VARIABLES

EEG

Forty five patients had interictal epileptiform EEG abnormalities confined to the resected temporal lobe. A further 21 patients had generalised or multifocal discharges and nine had multifocal discharges confined to the operated hemisphere. The presence of multifocal or generalised discharges, however, conferred no negative effect on outcome. Fourteen patients were operated on on the basis of preoperative seizure localisation with foramen-ovale electrode ictal recordings. The seizure remission rate in this group was 55% which was not significantly different from the remission rate of patients without preoperative seizure recording (74%, p=0.23).

Neuroimaging

Overall the focal lesion was recognised preoperatively in 54 patients and in 47 of the tumorous cases (78%). In two patients where dysembryoplastic neuroepithelial tumours were subsequently discovered at surgery preoperative scanning had suggested atrophy of the temporal lobe. In another case where focal cortical dysplasia was diagnosed the preoperative CT had shown ipsilateral hemiatrophy and, in a case of gangliglioma, the CT had shown contralateral hemiatrophy. The preoperative CT was regarded as normal in 12 patients. However, cases in which surgery was performed on the basis of the electroclinical features in the face of normal or discordant imaging did as well as the cases in which the lesion had been recognised preoperatively.

Neuropsychology

The influence of VIQ-PIQ difference and material specific verbal and non-verbal memory deficits were analysed separately for left and right sided resections. Absolute levels of either verbal or performance IQ were not correlated with seizure outcome except for right sided operations in which all 11 patients with a performance IQ>100 achieved remission (p=0.05) whereas in the same group none of a small number of four patients who had performance IQs<80 achieved remission. Formal neuropsychometry was possible only in 66 of the patients because some patients were too young to test and the presence of a severe learning or behavioural disturbance in others meant cooperation with formal testing did not prove possible. Of the 10 patents who were classified as learning disabled (not simply because of low IQ but also those who had severe behavioural and attentional difficulties) only 37% remitted compared with 74% of patients who were cognitively normal (p=0.10).

An increasing and appropriate VIQ-PIQ difference was associated with an increasing percentage remission in left sided operations with a corresponding poor outcome for an inappropriate difference of any magnitude. The opposite applied to right sided resections in which inappropriate VIQ-PIQ differences were associated with remissions ranging from 84% to 100% as the magnitude of the difference increased compared to 50%-75% remission rates if the IQ difference was appropriate. Impaired verbal memory in left sided resections was associated with a significantly better outcome than if the deficit was inappropriate or bilaterally abnormal. By contrast, the reverse applied to right side resections in which an appropriate (non-verbal) memory deficit was associated with a remission rate of only 50%.

COX REGRESSION ANALYSIS

In the Cox regression analysis of the clinical and investigative variables duration of epilepsy and the need for special schooling were selected in the forward selection procedure. An increase in duration of epilepsy from one category to the next was associated with a decrease in the probability of remission by 25% (hazard ratio 0.75, 95% confidence interval (95% CI) 0.59–0.97, p=0.03). Furthermore, the need for special schooling reduced the probability of remission to about a half (hazard ratio 0.49, 95% CI 0.23–1.04, p=0.06). There was no significant interaction between duration of epilepsy and special schooling.

PATHOLOGICAL CONSIDERATIONS

The pathological classification of the focal lesions is presented in table 3. Two major subdivisions were recognised: (1) neoplasms and (2) developmental and other lesions. The influence of pathological variables on outcome is presented in table 4. Apart from dysembryoplastic neuroepithelial tumour the numbers of other neoplasms and other abnormalities were comparatively small but a significantly worse outcome was noted for the nine patients with dysplastic lesions when compared with other developmental lesions such as hamartomas and vascular malformations and to neoplasms.

A considerable revision of pathological diagnosis was made when the original histological material was re-examined. Thirty two patients (operated on after 1988) had originally been classified as having dysembryoplastic neuroepithelial tumours, one of which was reclassified as a nodular heterotopia. A further 20 patients were reclassified as dysembryoplastic neuroepithelial tumour. Previous diagnoses included hamartoma (n=5), hamartoma with cortical dysplasia (n=1), pleomorphic hamartoma (n=2), fibrillary astrocytoma (n=2), low grade oligodendroglioma (n=3), low grade glial tumour (n=2), oligoastrocytoma (n=2), ganglioastrocytoma (n=2), and ganglioglioma with areas of malformation (n=1). On reclassification only one case was regarded as an astrocytoma and no case of oligodendroglioma was found. Three of the vascular lesions were classified as arteriovenous malformations. There was one case of cavernous haemangioma and the final vascular lesion was of uncertain classification consisting of large vascular channels with a thin endothelium, islands of fibrillary glia, well formed trabecular bone, and patches of dark brown pigment consistent with melanin. This lesion was considered as a hamartoma with a prominent vascular component. Two cases were classified as glioneuronal hamartomas; these were easily visualised well demarcated masses with glial, neuronal, and fibrovascular components.

It was possible by examination of the resection margins of the lesion to verify completeness of resection in 53 of the tumour cases and in 70 cases overall. In six cases the specimen was fragmented and in the remainder the original material had been lost. Of the tumorous patients only 15 (28%) had histological confirmation of complete removal and, overall, 74% of patients had incomplete removal of the lesion. However, although greater numbers achieved remission if the lesion was completely resected, this was not statistically significant.

The mesial temporal structures were examined in 60 of the 80 patients (75%). The presence of neuronal loss and sclerosis was qualitatively documented in 33% of cases overall and in 38% of tumours. The mesial temporal structures were replaced by tumour in five patients but were not clearly identified in nine other patients. The presence or absence of mesial temporal sclerosis in the specimen did not influence subsequent seizure remission.

Forty nine of the patients with dysembryoplastic neuroepithelial tumour were examined for the presence of cortical dysplasia. Changes suggestive of associated dysplasia were seen in 12 (25%) but this feature did not influence the seizure remission rate.

Discussion

We evaluated the influence of the clinical, neuropsychological, investigative, and pathological features on seizure remission after the en bloc removal of temporal lobe foreign tissue lesions. Overall, 71% of patients achieved a good outcome. Factors predictive of seizure remission included a shorter duration of seizures and younger age at surgery, the absence of significant cognitive impairment requiring special schooling, learning disability, and appropriate verbal performance IQ differences and verbal memory impairment for the left sided resections. Generalised tonic-clonic seizures, interictal EEG discharges confined to the resected lobe, demonstration of the lesion preoperatively on CT, complete histological resection of the lesion, the presence of mesial temporal sclerosis in the specimen and appropriately lateralised VIQ-PIQ deficits and non-verbal memory for the right sided resections were not significant predictors of seizure outcome. Of the clinical and investigative findings, duration of epilepsy, and the need for special schooling were independently predictive of outcome on logistic regression analysis.

We specifically chose to use actuarial statistics and Cox regression as these are the methods of choice in analysing prognostic factors in epilepsy with prolonged periods of follow up. Although not widely used in epilepsy surgery, which usually gives figures on outcome after 1 or 2 years, actuarial statistics allow the ready evaluation of the relative importance and possible interaction of prognostic variables. Therefore, actuarial statistics have become the method of choice for analysing the prognosis for seizure control in epilepsy, recurrence rates after a first seizure, relapse rates after drug withdrawal, and response to drug treatment in anticonvulsant trials involving prolonged follow up.21-24 These methods, however, measuring probability of achieving 1 year remission at any time will imply a slightly better outcome and the issue of subsequent relapse is clearly important in interpreting the relevance of the identified prognostic factors. We have computed the Engel classification grading1 of each patient at a mean of 6 years of follow up. Twenty patients (25%) were classified as Engel grade 3 or 4 (no worthwhile seizure improvement) at this point of follow up. Importantly, all of these patients had been in these groups since the time of surgery—that is, no patient relapsed to having frequent seizures after an early remission. Similar findings have been reported by the UCLA group who noted only two relapses in a similar patient group after 10 years of follow up.25 Within the lower Engel grades (1 and 2, seizure free or greatly improved) 40 patients (50%) remained with a IA classification (seizure and aura free) at 6 years of follow up. The remaining 20 patients (25%) transferred between the remaining Engel 1 and 2 groups, including eight patients who reverted to an Engel IA outcome. Three patients remained seizure free but continued to have occasional auras. Thus, the actuarial analysis of time to 1 year continuous seizure freedom seems justified in discriminating factors prognostic of a good or bad outcome.

The present results are derived from a consecutive cohort of patients, most assessed before the routine use of MRI and preoperative ictal EEG. Whether this means that the results are applicable to modern surgical practice is uncertain. There are, however, few data in the literature that suggest preoperative ictal-EEG improves outcome, however intuitive this may seem. Indeed, contrary data exist26 and seizure outcome in this series was not statistically superior when preoperative monitoring was employed. Similarly, it is nor clear whether the use of MRI would have substantially altered the present findings. Whether this too detracts from the relevance of the current findings to modern surgical practice is a matter for debate. This is further emphasised by the finding that patients with negative or discordant CT did as well as those with a preoperative CT identification of the resected lesion. It is probable (but not yet proved) that preoperative identification by MRI of associated mesial temporal atrophy may allow a more conservative resection and sparing of mesial temporal structures. It is also probable that MRI allows better anatomical localisation particularly with regard to suprasylvian or posterior temporal extension and so allows better preoperative prognostication. On a practical level all patients had a maximum/standardised resection based on anatomical constraints and incomplete histological removal was not associated with a worse seizure outcome. A detailed re-examination27 of patients with no worthwhile seizure improvement after resection of temporal dysembryoplastic neuroepithelial tumour (n=10) has not related the postoperative epileptic syndrome to residual tumour on MRI (n=3) but rather implicates extratemporal/multifocal epileptogenesis as a cause of persistent seizures.

Table 5 highlights the major published series of surgically treated temporal lobe lesions associated with intractable epilepsy. It is apparent that considerable heterogeneity exists between series, particularly in tumour diagnosis. This may be related in part to differences in the referral populations, the present series having been referred for management solely of intractable epilepsy. Furthermore, most of the previous series have not reanalysed tumour histology, collected in most over many years. Indeed the first clinical and pathological classification of dysembryoplastic neuroepithelial tumour16 includes 19 cases previously classified as low grade astrocytoma, oligodendroglioma, and oligoastrocytoma in the Mayo Clinic Tissue Registry. The underestimated incidence of dysembryoplastic neuroepithelial tumour has also been conceded in another large series of temporal lesions associated with chronic epilepsy.25 This may not simply reflect the recent appreciation of this entity but also reflects the histological heterogeneity of dysembryoplastic neuroepithelial tumours18 which are characteristically associated with small round oligodendroglia-like cells, astrocytes, and mature neurons in varying proportions with multinodular, solitary nodular, and diffuse arrangement. In a recent series of 40 cases28dysembryoplastic neuroepithelial tumours resembled pilocytic astrocytomas (four cases), astrocytomas (16 cases), anaplastic astrocytoma (one case), oligodendrogliomas (10 cases), oligoastrocytomas (eight cases) or anaplastic oligoastrocytoma (one case). Furthermore, proliferating cell nuclear antigen labelling studies are absent or low in most and rarely above 40%.18Thus, despite some heterogeneity of histology, the clinical, pathological, and indolent biological features indicate that these tumours constitute a single entity. Only one of the 52 cases of dysembryoplastic neuroepithelial tumour in this series (most lacking histological evidence of complete resection) has demonstrated radiological progression 10 years after temporal lobectomy (CE Polkey, unpublished data) emphasising the need to recognise the morphological range of dysembryoplastic neuroepithelial tumour and thus avoid adjuvant radiotherapy.

The striking finding from the pathological data is that only 25% of tumours were completely removed as defined by postoperative analysis of the resection margins. Outcome, however, was not statistically different for most patients who had incomplete resections. Daumas-Duport16 reported similar findings in a dysembryoplastic neuroepithelial tumour series in which 44% of patients had incomplete or subtotal removal, most remaining seizure free after a mean follow up of 9 years. The other series (table 5) report less consistent findings although Boon et al 29 reported an 83% seizure free outcome with resection to tumour free margins as determined by intraoperative frozen section analysis and Awad et al 30 found that complete resection of the lesion resulted in 90% seizure free outcome, regardless of the additional epileptogenic tissue resected. Others advocate additional resection of electrically active surrounding cortex using intraoperative electrocorticography.31-33 It is uncertain if any definitive conclusions can be drawn from these approaches because of the mixture of temporal and extratemporal cases in some and by the relatively few patients diluting the strength of observations and statistical associations. It would seem that lesionectomy alone in temporal lobe cases has a less favourable outcome34 35suggesting that the epileptogenic zone is more extensive, involving both the hippocampus and the tumorous area. Electrophysiological involvement of the hippocampus (in acute electrocorticography) has been well documented in association with extrahippocampal mass lesions31 32 and Malthern et al,36 using depth recordings, confirmed that the hippocampus was the site of origin or first propagation in six of seven cases of seizures associated with temporal lobe tumours. We found an incidence of qualitatively described mesial temporal sclerosis in 38% of 47 tumour specimens in the present series but because all patients had a standardised resection including the mesial temporal structures we are unable to say whether those patients with normal hippocampi would have done as well with lesion removal alone. Joomaet al 34 reported a residual seizure rate of 53% in patients who had severe hippocampal cell loss in association with temporal lobe lesions compared to 11% if the cell loss was mild.The ability of high resolution MRI to diagnose mesial temporal structures in vivo may allow preoperative decision making on whether to perform a mesial temporal removal and thus avoid the deficits in material specific memory associated with removal of a normal hippocampus.37 38

The seizure outcome after surgical treatment of dysembryoplastic neuroepithelial tumour is similar to that of the other two smaller series16 39 and emphasises the favourable prognosis for seizure remission after resection of this lesion. A similar satisfactory outcome occurred after temporal lobectomy for hamartomas and vascular lesions. This is by contrast with the results of surgical treatment of temporal lobe developmental lesions such as focal cortical dysplasia and nodular heterotopia in which only a third of a small group of nine patients achieved seizure remission. The poor outcome in these patients is paralleled in other series4 40-42 and probably reflects the diffuse pathological nature of these entities.

It is clear, however, that a substantial number of patients with temporal lobe dysembryoplastic neuroepithelial tumours fail to become seizure free after surgery and the present report has highlighted two important clinical variables which influence prognosis. The first is the duration of the preoperative seizure disorder which is the most important statistical variable in predicting outcome. In the other large series of temporal lobe lesions25 using actuarial analysis of time to seizure remission after surgery, duration of epilepsy was also found to be a statistically significant predictor of outcome. Similarly in a series of 38 gangliogliomas43 a statistically significant better outcome with earlier age at surgery was found. The mechanisms whereby chronicity of seizures imparts a poor outcome on postoperative seizure control is uncertain but may reflect the process of secondary epileptogenesis. Apart from dysplastic lesions44 the great majority of temporal foreign tissue lesions are not in themselves intrinsically epileptogenic and presumably give rise to seizures by some remote effect. In non-surgical populations of patients with epilepsy those with a younger age at onset and longer duration of epilepsy generally have more severe epilepsy.45 It is probable that there is a continuing process of secondary epileptogenesis, which is the probable explanation for the poor outcome with longer duration of epilepsy. In patients with temporal lobe epilepsy due to glial tumours, lesions that are unlikely to occur bilaterally, Morrell46 reported evidence of contralateral seizure generation in 15%. The failure of contralateral seizures to disappear after surgical resection of the primary epileptogenic region was correlated with both frequency and duration of seizures preoperatively. We did not find evidence, however, of contralateral seizure generation as a mechanism of postoperative epileptogenesis in 10 patients with dysembryoplastic neuroepithelial tumours who had no worthwhile improvement in seizure control after temporal lobe surgery27 Morrel et al,47 however, also suggested that adjacent cortex could be recruited into the epileptogenic zone with increasing duration of epilepsy and this would seem to be a more important mechanism in the present patients.

The requirement for special schooling was an independent predictor of seizure outcome after surgery. Most patients with epilepsy attend regular school48 and in the United Kingdom the need for special schooling usually denotes significant cognitive or behavioural disturbance. It is intriguing that this in itself should carry a negative prognosis for seizure freedom after resection of temporal lobe structural lesions. Similarly, the presence of significant psychiatric disturbance and learning disability were associated with a poor outcome. Severe deficits in cognitive and behavioural functioning may reflect more widespread cerebral abnormalities in patients with dysembryoplastic neuroepithelial tumour, cortical dysplasia, and ganglioglioma, which comprised most of the pathological lesions. Cortical dysplasia often occurs as part of the pathological range of dysembryoplastic neuroepithelial tumours16 18 39 49 and both dysembryoplastic neuroepithelial tumour and ganglioglioma may occur together with focal cortical dysplasia in the same patient,50 51 suggesting a close histogenetic relation between these entities and implying a common origin as a neuronal migrational abnormality. The association with cortical dysplasia, and the known extensive disturbances in neuronal circuitry39 52 in this setting, may provide an explanation for the poor outcome in cognitively and behaviourally abnormal patients. We have, however, been unable to show a negative effect on seizure outcome of coexistent cortical dysplasia in the temporal specimen in patients with dysembryoplastic neuroepithelial tumours. This does not exclude the presence of subtle extratemporal cortical developmental abnormalities accounting for learning disability and other deficits and, presumably, epileptogenesis.

Lateralised neuropsychological deficits predicted outcome well in patients with left sided lesions with increasingly appropriate verbal performance IQ discrepancy and impaired verbal memory being associated with a better outcome. Discordant findings, however, in right sided resections, were associated with a better postoperative seizure outcome. There are few data in the epilepsy surgery literature on the utility of neuropsychological tests in either lateralising temporal lobe structural lesions or predicting outcome after their removal. Boonet al 29 noted that neuropsychological assessment was more sensitive to the presence of left hemisphere lesions than to right sided pathology but that, overall, these tests provided congruent lateralisation in only 56% of patients. It was concluded that cortical language deficits reflect a greater anatomical specificity than non-verbal spatial function and that subtle non-dominant deficits could easily be missed. Hermannet al 53 evaluated the ability of the Wechsler adult intelligence scale to discriminate left and right sided temporal lobe epilepsy in 215 patients with invasively verified epilepsy of unilateral temporal lobe origin without lesions demonstrated by MRI. Only the vocabulary subtest reliably separated the groups, with the left temporal lobe epilepsy group scoring significantly lower. None of the other indices reliably distinguished left and right temporal lobe epilepsy. Thus, it seems that the non-specificity of tests designed to test right hemispheric and temporal function may account for the present results. However, if the tests were truly non-specific then no effect on outcome would be predicted, which is not the case. It would seem from the present results that gross impairment of non-dominant hemisphere tests in right sided resections may actually reflect a more pervasive neuropsychological deficit than simply lateralised non-dominant hemispheric dysfunction. It is important to note that the results, although internally consistent within the groups, are derived from multiple tests on few patients and we acknowledge that the results may be fortuitous.

The failure of the preoperative interictal EEG to predict outcome is surprising in view of the clear predictive role of this investigation in previous series5-7 9 54 55 of outcome predictors after temporal resection (most cases had either mesial temporal structures or non-lesional pathology). In particular, a predominantly unilateral anterior temporal focus has been consistently demonstrated to predict a favourable seizure outcome. Yet, the occurrence of diffuse bilateral and generalised EEG discharges are commonly found in childhood brain tumours56 as well as in adults with temporal lobe tumours in general57 and, specifically, in dysembryoplastic neuroepithelial tumour39 and ganglioglioma.43 The mechanisms involved are uncertain and may reflect secondary epileptogenesis58 or a diffuse disorder of neuronal circuitry in the context of cortical dysgenesis.39 The present data cannot substantiate these hypotheses since extratemporal or bilateral EEG changes were not predictive of an unfavourable outcome unlike duration of epilepsy and cognitive impairment. Whatever the mechanism it is important to be aware that the diffuse EEG abnormalities associated with temporal lobe lesions should not preclude further assessment for epilepsy surgery or, in themselves, dictate the need for invasive EEG studies.

In conclusion, the present findings confirm the overall favourable results of epilepsy surgery for temporal lobe foreign tissue lesions and highlight the importance of dysembryoplastic neuroepithelial tumour as the commonest pathological entity in this setting. The presence of cognitive and behavioural disturbance not only has implication for postoperative psychological and socioeconomic rehabilitation but actually influences seizure outcome as well. Finally, the major contribution of duration of preoperative epilepsy highlights the need for early recognition and surgical treatment of lesional temporal lobe epilepsy.