Address correspondence and reprint requests to Dr. S.K. Lee at Department of Neurology, Seoul National University Hospital, 28 Yongon-dong, Chongno-gu, Seoul, 110–744, Korea. E-mail: firstname.lastname@example.org
Summary: Purpose: Determining long-term prognostic factors of surgery for mesial temporal lobe epilepsy (MTLE) is important for identifying ideal candidates and predicting the prognosis for individual patients. We tried to identify the prognostic factors of anterior temporal lobectomy (ATL) for MTLE with longitudinal multivariate analysis.
Methods: Two hundred twenty-seven patients with MTLE were included in this study. The primary outcome variable was patient status 1–5 years after surgery: seizure free, or not. Clinical characteristics and recent diagnostic modalities were considered as prognostic factors. Univariate and standard multiple logistic-regression analysis for outcome at 1 and 5 years after surgery and the generalized estimation equation (GEE) model for longitudinal multiple logistic regression of the 5-year follow-up period were used.
Results: The seizure-free rate at 1 year was 81.1% and decreased to 75.2% at 5 years after surgery. By the univariate or standard multiple logistic-regression analysis, age at surgery or hippocampal sclerosis on magnetic resonance imaging (MRI) ipsilateral to surgery was significant for the postsurgical outcome. However, the longitudinal analysis by the GEE model revealed that younger age at surgery [odds ratio (OR), 0.59; 95% confidence interval (CI), 0.43–0.81], absence of secondarily generalized tonic–clonic seizure (2°GTCS; OR, 0.45; 95% CI, 0.26–0.79), and hippocampal sclerosis on MRI (OR, 2.44; 95% CI, 1.11–5.26) were significant predictors of a good surgical outcome.
Conclusions: Age at surgery, presence of 2°GTCS, and hippocampal sclerosis on MRI are independent prognostic factors for ATL in MTLE. These findings suggest that MTLE is a progressive disorder, and surgical outcome is better when early ATL is performed.
Mesial temporal lobe epilepsy (MTLE) is a well-recognized epilepsy syndrome with hippocampal sclerosis (HS) and a major target for epilepsy surgery. Anterior temporal lobectomy (ATL) performed for the treatment of medically intractable MTLE yielded worthwhile improvement (75–95%), including complete remission of seizure (68%–85%) in most patients (1–3). However, ≤30% of patients continue to experience seizures after surgery. Therefore determining prognostic factors is very important in identifying ideal candidates for surgery and predicting the prognosis of individual patients.
Even though some authors have performed multivariate analyses for the prognosis of epilepsy surgery (4–12), most of these studies comprised a mixture of patients who had temporal and extratemporal epilepsy surgery. Few studies include the results of modern presurgical evaluation techniques such as magnetic resonance imaging (MRI), positron emission tomography (PET), and ictal single-photon emission computed tomography (SPECT). Another problem is that the calculation of the proportion of who are seizure free for some years after operation provides only a snapshot of cohort outcomes at that period (13). Only a few studies performed seizure-outcome analysis by using a longitudinal method (10,11). The cross-sectional method of analysis may be problematic, as the correlation of variables with outcome may yield different results depending on the stage of the postoperative course at which outcome was assessed (13,14).
The analytic approach presented for modeling outcome variables that have dichotomous correlated responses is the generalized estimating equation (GEE) model (15). GEE handles continuous and categoric variables with missing values, adjusting for possible correlations within the same patients. If within-subject correlations are ignored as in a typical logistic-regression analysis, then incorrect inferences may result.
We performed multivariate analyses on a large homogeneous series of MTLE patients to determine the predictors of longitudinal prognosis after ATL. We included various clinical profiles and the results of presurgical evaluations such as MRI, PET, ictal SPECT, and Wada test as possible prognostic factors. We applied the GEE model to the multivariate analyses to examine the prognostic value of a number of variables based on the longitudinal responses (e.g., repeated measures of outcome ≤5 years after surgery).
We evaluated a group of 227 consecutive MTLE patients who had undergone standard ATL with amygdalohippocampectomy from October 1994 to October 2000 at the Seoul National University Comprehensive Epilepsy Center. The diagnosis of MTLE was determined by using the guidelines set by Engel (16) and with even stricter criteria for patient selection in this study; patients were diagnosed as having MTLE if (a) HS was seen on MRI and anterior temporal ictal onset was identified during video-EEG monitoring, or (b) normal MRI or other definite lesion was found on MRI in addition to HS, but invasive study with intracranial electrodes confirmed massive and exclusive mesial temporal ictal onset (9). Patients who had only ipsilateral HS on MRI without another definite lesion were classified in the "pure ipsilateral HS" group. Other definite lesions included diffuse hemiatrophy, focal cerebromalacia, and bilateral HS. Mild ipsilateral temporal atrophy associated with HS was not included in "other lesion" group.
Clinical parameters, presurgical evaluation, and surgery
Clinical characteristics registered for each subject included age at surgery, the duration of epilepsy, age at afebrile seizure onset, monthly seizure frequency, presence of febrile convulsion, and existence of aura and secondarily generalized tonic–clonic seizures (2°GTCSs).
Multidisciplinary presurgical evaluations, including a complete neurologic examination, brain MRI with TLE protocol, long-term video-EEG monitoring, fluorodeoxyglucose (FDG)-PET, intracarotid amobarbital test, and ictal and interictal SPECT if possible, were performed as previously described (9,17). The surgical side was localized by the clinical, neuroimaging, and electrophysiologic results. Our standard procedure for treating MTLE was ATL with removal of the medial structures, including amygdalar nucleus, hippocampus, and parahippocampal gyrus. The temporal resection was anterior to the precentral sulcus, including the superior temporal gyrus, irrespective of the language dominance. Temporal lobe was resected en bloc with opening of the temporal horn (9,17). We performed standard anterior temporal resection in all patients. We resected 3.5 cm of lateral temporal lobe from the anterior temporal tip for the left TLE, and 4.5 cm for the right TLE.
Outcome of surgery was defined by the Engel classification (18). We included only patients with a follow-up period of ≥1 year. Outcome was determined at each year up to 5 years after surgery. Outcome status was determined by outpatient clinical interview or telephone interview. Two of authors (S.K.L. and S.W.J.) performed clinical interviews at outpatient clinics. Some of patients' status were confirmed by telephone interview. Patients were categorized as seizure free (Engel class 1) or not (Engel classes 2, 3, and 4). Nondisabling auras were not considered as seizures for the purpose of this study, although we recognize their potential importance for psychological outcomes. Three patients who could not be contacted with outpatient or telephone interview were excluded.
The patients whose outcomes changed during the follow-up period were also identified. In these patients, we attempted to isolate clinical characteristics related to specific patterns of change in outcome (e.g., the running-down phenomenon).
Student's t test was used for continuous variables: age at operation, age at onset, epilepsy duration before surgery (defined by a difference between age at operation and age at onset), and seizure frequency per month before surgery. A χ2 or Fisher's exact test of independence was used for univariate analysis of the following categoric variables: sex, history of febrile convulsion, positive aura, side of surgery, presence of 2°GTCSs, interictal epileptiform discharge mostly from the side of surgery (>70%), ictal epileptiform discharge from the side of surgery (>75%), ipsilateral HS to the side of surgery on MRI, ipsilateral temporal hypometabolism on PET, ipsilateral hyperperfusion on ictal SPECT, and ipsilateral failed memory to surgery on the intracarotid amobarbital test (9).
Univariate analysis was performed on these variables related to outcomes at 1 and 5 years after surgery. Standard stepwise multiple logistic regression using outcomes at 1 and 5 years after surgery also was performed. Standard forward stepwise multiple logistic regression (significance level, ≤0.05) using outcomes at 1 and 5 years after surgery also was performed. In this analysis, we included independent variables with p ≤ 0.25 in the univariate analysis. For multivariate analysis using repeated measures of outcome ≤5 years after surgery, we applied the GEE model (15). Compound symmetry structure was used to model the possible dependency among repeated measurements in a patient. The GEE method was applied because of the possible intrapatient correlations. Some studies indicate that ordinary logistic regression may result in a bias in p values for repeated measures data (15).
Stepwise variable selection also was performed to identify more important prognostic factors for good outcome. For the missing values of the three variables (PET, Wada test, ictal SPECT), binary results for each were created. The first variable was coded with the observed values (1 or 0) for the subjects without missing values and zero for those with missing values. The second variable was coded as one for the subject with missing values and zero for those without missing values. This method treats the missing value as a new category and gives the effects of being missing. By using this method, we were able to include all subjects in the analysis. Data storage and univariate analysis were performed by using SPSS (SPSS Inc., Chicago, IL, U.S.A.). Multivariate analyses were performed by using SAS Proc Genmod (SAS Inc., Cary, NC, U.S.A.).
The mean age at surgery was 29.6 ± 8.2 years (range, 9–55 years). The patient group comprised 137 male and 90 female subjects. The mean duration of follow-up after surgery was 4.6 ± 2.0 years. The proportion of seizure-free patients was 81.1% after 1 year and decreased slightly to 75.2% at 5 years after surgery (Table 1). All patients had interictal and ictal EEG and MRI, whereas PET was performed in 210 patients. Ictal SPECT with interictal SPECT was performed in 144 patients. Twenty-seven patients had surgical resection after intracranial invasive studies.
Table 1. Surgical outcome (Engel class) of mesial temporal lobe epilepsy in each year
1 yr n = 227
2 yr n = 193
3 yr n = 174
4 yr n = 150
5 yr n = 121
The proportion of patients with Engel class I (E class I) is 81.1% at 1 year, and decreases slightly to 75.2% at 5 years.
With univariate analysis, age at surgery (p < 0.001), duration of epilepsy (p = 0.01), and pure ipsilateral HS on MRI (p = 0.02) were significant prognostic factors for postsurgical outcome at 1 year (Table 2). Seven patients had normal brain MRI. Four patients were seizure free at 1 year after surgery. Six of the seven patients were followed up for more than 5 years after surgery. Three patients were seizure free at 5 years (Table 3). Eight patients with seizure-free outcome at 1 year after surgery had another definite lesion in addition to HS, whereas seven patients with persistent seizures had this finding. Among 10 patients who had other lesions in addition to HS, six patients were seizure free at 5 years after surgery.
Table 2. Surgical outcome in 227 mesial temporal lobe epilepsy patients 1 year after surgery
Seizure free (n = 184)
Not seizure free (n = 43)
Univariate analysis on prognostic factors for outcome at 1 year after surgery. Numbers in parentheses indicate percentages.
HS, hippocampal sclerosis
aNot performed, n = 17.
bNot performed, n = 83.
cNot performed, n = 5
Age at surgery (yr; mean ± SD)
28.5 ± 8.0
34.1 ± 9.4
Age at onset (yr; mean ± SD)
12.7 ± 8.0
14.7 ± 9.5
Duration of epilepsy (yr; mean ± SD)
15.7 ± 8.3
19.3 ± 8.7
Seizure frequency per month (mean ± SD)
9.5 ± 20.5
16.4 ± 31.5
Secondarily generalized tonic–clonic seizure
History of febrile convulsion
Side of surgery (right)
Interictal EEG, discharges mostly (>70%) from side of surgery
Ictal EEG, discharges from side of surgery
MRI, pure ipsilateral HS
PET,a abnormal findings in surgical area
Ictal SPECT with interictal SPECT,b Correct localization
For the outcome at 5 years after surgery, only age at surgery was a significant prognostic factor (121 patients; p = 0.005; see Table 3). Duration of epilepsy and pure ipsilateral HS on MRI were marginally significant for 5-year follow-up patients (p = 0.06 and 0.07, respectively).
Standard stepwise multiple logistic regression using outcomes at 1 year after surgery also showed that age at surgery (OR, 1.06; 95% CI, 1.02–1.10) and pure ipsilateral HS on MRI (OR, 4.13; 95% CI, 1.57–10.87) were the significant prognostic factors (p = 0.005 and p = 0.004, respectively). At 5 years after surgery, age at surgery (OR, 1.06; 95% CI, 1.01–1.12) and pure ipsilateral HS on MRI (OR, 3.32; 95% CI, 1.04–10.59) were the significant prognostic factors (p = 0.043 and p = 0.026, respectively). However, by longitudinal analysis using GEE for repeated measures of outcome ≤5 years after surgery, a younger age at the time of surgery (p < 0.001), absence of 2°GTCSs (p = 0.01), and pure ipsilateral HS on MRI (p = 0.03) were the significant predictors for good surgical outcome (Table 4). The younger age group (24 years old or younger) shows 3.9 times greater chance of being seizure free than did the older age group (36 years or older) after adjusting other factors (Table 5).
Table 4. Repeated measures data analysis using the generalized estimation equation (GEE) model showed that age, hippocampal sclerosis on MRI, and history of partial seizures are significant favorable prognostic factors for seizure-free status after temporal lobectomy
95% Confidence interval
aChange per 10 years of age.
Age at surgerya
Partial seizure only
Ipsilateral hippocampal sclerosis on MRI
Table 5. Relation between seizure-free odds ratio and age at surgery
Number of patients
95% Confidence interval
95% Confidence interval
The younger age group (0–23 years old) shows a 3.9 times greater chance of being seizure free than does the older age group (36 years or older) after adjusting for other variables.
Changes of outcome during the follow-up period were observed in 48 of 121 patients who were followed up for ≥5 years after surgery (Fig. 1). The running-down phenomenon was observed in eight patients. Reappearance of seizures after an initial seizure-free state occurred in 20 patients. Transient improvement (three patients), worsening (12 patients), or fluctuation of outcome (five patients) also was found. We could not find a specific clinical profile for a specific changing pattern.
By univariate analysis for surgical outcome, younger age at surgery and only HS without another definite lesion on MRI were found to be favorable factors for outcome at 1 year after surgery. When using multiple logistic-regression analysis, age at surgery and hippocampal atrophy on MRI were significant prognostic factors at 1 or 5 years after surgery. Longitudinal analyses by the GEE model using repeated measures of outcome ≤5 years after surgery also confirmed that early age at surgery and hippocampal sclerosis on MRI were favorable prognostic indicators. In addition to these two factors, this analysis identified the presence of 2°GTCSs as a poor prognostic factor.
It has been argued that MTLE is a progressive disorder and that aggressive treatment should be instituted early in its course (2,19). Surgical outcome with respect to seizures has been reported to be better when surgical resection is performed early (20). However, the majority of studies on temporal lobectomy using multivariate analysis did not report the significance of age at the time of surgery for prognosis (13). Of the two studies that did (7,9), one analyzed yearly prognoses of 125 patients with temporal lobectomy by using a logistic-regression model and reported that febrile seizures, lack of 2°GTCSs, highly localized spikes, and younger age at surgery were all independent favorable prognostic factors. However, this study had set a high significance level of p < 0.1 and did not include MRI and PET data. In addition, it did not discriminate between MTLE and neocortical TLE. Many studies in which multivariate analyses were performed to find prognostic factors of epilepsy surgery included a mixture of patients with mesial and lateral temporal lobe epilepsy (13). Because selection of patients with the broad criteria of temporal lobectomy will cause the inclusion of a mixture of different epileptic syndromes, we applied strict criteria for the diagnosis of MTLE to resolve this problem.
The results of various animal studies have indicated that MTLE is a progressive disorder (21). Progressive epileptogenesis can be examined with a variety of techniques, and secondary epileptogenesis can be found in the kindling models (21–23). Human studies also suggested the formation of secondary epileptogenesis based on the observation that seizures originated in the hemisphere contralateral to a tumor (24,25). Other clinical work demonstrated that the incidence of bilateral spikes increased in proportion to the duration of follow-up, indicating the progressive nature of epilepsy (26).
A recent report showed that patients with chronic epilepsy were significantly more likely to develop neocortical atrophy than were control subjects (27). The increased risk of cerebral atrophy in epilepsy was not related to a history of documented seizures. Risk factors for neocortical atrophy were age and multiple antiepileptic drug exposure (27). These results imply that epilepsy is potentially progressive, in terms of anatomic dissemination of epileptogenic brain tissue. Our results demonstrate that old age at the time of surgery and the presence of 2°GTCSs are poor prognostic factors and also indicate the progressive nature of MTLE and augmentation of epileptogenesis by 2°GTCSs. Age at surgery, not the duration of epilepsy, has affected the surgical outcome, which may mean the presence of the inherited progressive nature of MTLE. Hippocampal epilepsy resulting from early damage followed by synaptic reorganization may be progressive even before the development of epilepsy. Furthermore, repeated 2°GTCSs can enhance epileptogenic properties. Our results show that early surgery is better for MTLE with a dose–response effect (see Table 5).
One previous study focused on the long-term seizure outcome demonstrated that the lack of an obvious abnormality or the presence of diffuse pathology, and preoperative 2°GTCSs were risk factors for recurrence after surgery (11). This finding is consistent with our results identifying presence of pure ipsilateral HS or two GTCSs as a poor prognostic factor.
Three different methods of analysis (univariate analysis, standard multiple logistic regression, and the GEE model) showed slight but important differences in results. To the best of our knowledge, nearly all studies for finding prognostic factors for epileptic surgery are cross-sectional analyses. That is to say, they analyzed the outcome only at a specific time point, such as postoperative 2 years. This approach may be problematic, as the correlation of variables with outcome may yield different results depending on the stage of the postoperative course at which the outcome was assessed (13). A problem exists in repeated univariate analyses, because it can increase type I error or present an inaccurate picture of the significance of variables if the independence of the factors has not been assessed (13). To solve this problem, evaluation of many factors by using a longitudinal and multivariate method is needed. The GEE model we used for this analysis has the advantage of taking into account the correlated nature of the responses within the same subjects over time. Our data are an example of a longitudinal study in which individual outcomes in patients were measured repeatedly over time (yearly up to 5 years). In the analysis of correlated data, like repeated measures data, the correlations within a subject are often ignored in a modeling process like standard logistic regression, which leads to incorrect inferences (15). A longitudinal data analysis in the consideration of the interdependence of factors, like the GEE model, consider this problem into account. This practice reduces bias that might have arisen from a differential follow-up times. Estimates from the GEE can be interpreted as weighted effects over years, taking within subject correlation into account. So we can draw overall conclusions rather than cross-sectional features of the data.
A time-to-event approach would have been applicable (10), if we had observed time to event. We also performed survival analysis with life-table method (Table 6) to calculate cumulative 5-year seizure-free rate. The terminal event was defined when a patient experienced a first seizure after surgery. The observation of seizure recurrence was made with every 1-year interval. This analysis helped us to estimate cumulative 5-year seizure-free rate (59.0%) after surgery. However, unlike survival analysis using cancer patients, changes of outcome during the follow-up period could be observed in epilepsy patients (Fig. 2). It is not appropriate to assign “event” to a single seizure. We thought Engel classification during the last 1 year was the best parameter for deciding surgical outcome. Using only survival analysis could miss the important running-down phenomenon of epilepsy surgery.
Table 6. Survival analysis with the life-table method for cumulative seizure-free rate
Number entering this interval
Number with end of follow-up
Number with recurred seizure
Proportion of seizure-free rate
Cumulative proportion of seizure-free rate
An important issue is the relapse of seizures after achieving a certain seizure-free period in some patients. A phenomenon of running-down of seizures was seen over months or years until the complete cessation of seizures (3). Very few systematic studies have been performed of a patient's natural history and prognostic factors predicting the continued remission of seizure or relapse after surgery (13). Table 1 shows the stability of seizure outcome. One previous report also demonstrated that seizure outcome remained stable after ATL (28). However, although the general proportion of seizure-free patient remained stable in our series, some patients showed the running-down phenomenon or the reversed pattern. However, we could not find a specific clinical profile for prediction of a specific pattern. Future studies that include more patients would be needed for this analysis.
In summary, this study shows that younger age at surgery and only HS on MRI in temporal lobectomy for MTLE are favorable prognostic indicators for seizure remission with the GEE model by using repeated measures of outcome ≤5 years after surgery. We suggest that these results provide supporting evidence that MTLE may be a progressive disorder. However, only a randomized controlled trial can give reliable information about whether early surgery is associated with a better seizure outcome.