Seizure recurrence after epilepsy surgery has been classified as either early or late depending on the recurrence time after operation. However, time of recurrence is variable and has been arbitrarily defined in the literature. We established a mathematical model for discriminating patients with early or late seizure recurrence, and examined differences between these two groups.
A historical cohort of 247 consecutive patients treated surgically for temporal lobe epilepsy was identified. In patients who recurred, postoperative time until seizure recurrence was examined using an receiver-operating characteristic (ROC) curve to determine the best cutoff for predicting long-term prognosis, dividing patients in those with early and those with late seizure recurrence. We then compared the groups in terms of a number of clinical, electrophysiologic, and radiologic variables.
Seizures recurred in 107 patients (48.9%). The ROC curve demonstrated that 6 months was the ideal time for predicting long-term surgical outcome with best accuracy, (area under the curve [AUC] = 0.761; sensitivity = 78.8%; specificity = 72.1%). We observed that patients with seizure recurrence during the first 6 months started having seizures at younger age (odds ratio [OR] = 6.03; 95% confidence interval [CI] = 1.06–11.01; p = 0.018), had a worse outcome (OR = 6.85; 95% CI = 2.54–18.52; p = 0.001), needed a higher number of antiepileptic medications (OR = 2.07; 95% CI = 1.16–9.34; p = 0.013), and more frequently had repeat surgery (OR = 9.59; 95% CI = 1.18–77.88; p = 0.021). Patients with late relapse more frequently had seizures associated with trigger events (OR = 9.61; 95% CI = 3.52–26.31; p < 0.01).
Patients with early or late recurrence of seizures have different characteristics that might reflect diversity in the epileptogenic zone and epileptogenicity itself. These disparities might help explain variable patterns of seizure recurrence after epilepsy surgery.
Some patients with medically intractable epilepsy require surgery to control their seizures. The localization of the epileptogenic area is the cornerstone to guide treatment (Rosenow & Luders, 2001). The majority of refractory focal seizures originate from the temporal lobe, and its resection is a well-established procedure (Wiebe et al., 2001). It is estimated that approximately 50–62% of patients will remain seizure-free 5 years after surgery for temporal lobe epilepsy, depending on the pathogenic substrates associated with refractory seizures (Wieser et al., 2003; McIntosh et al., 2004; de Tisi et al., 2011).
However, the elapsed time for seizure recurrence after surgery is not uniform, and patients who continue to have seizures diverge into two different categories. First are the patients who derive no benefit from surgery, with either immediate postoperative seizures or only a brief period of seizure control. Others may have a longer period of seizure control after surgery but then subsequently recur (Schwartz et al., 2006). The differences between these two classes of patients perhaps reflect the appropriateness of resection, intrinsic tissue epileptogenicity, and long-term prognosis (Jehi et al., 2010). For example, patients with late recurrence of seizures usually have fewer attacks and a better quality of life when compared to individuals with early relapse (Lee et al., 2006; Buckingham et al., 2010). To date, each author has defined early or late seizure recurrence arbitrarily, creating obvious difficulties when interpreting data, and making it difficult for investigators to perform comparisons and understand the significance of findings. Mathematical or statistical tools have not, to our knowledge, been used previously for this purpose.
Early or late recurrence of seizures after surgery might be better seen as two distinct events. Although early recurrence might reflect an incomplete resection of the epileptogenic zone, late recurrence might reflect the development of a new epileptogenic process, possibly reflecting an underlying epileptogenic tendency. A better distinction and comprehension of these two different situations could lead to a more appropriate understanding of reasons for surgical failure. This is important because in the short run it would help to establish a more accurate long-term prognosis for patients earlier after surgery, and in the long run it might have an impact on planning better treatment protocols. Therefore, the main objective of this study was to establish a statistically oriented model for better discriminating patients with early or late seizure recurrence and to study the clinical, electrophysiologic, and neuroradiologic differences between these two groups of patients. It is our hope that this might help to better inform patients regarding their prognosis as well as to delineate research objectives for future treatments.
We performed a retrospective cohort study of all consecutive patients diagnosed with medically refractory temporal lobe epilepsy who underwent resective temporal lobe surgery at our center between January 1994 and February 2007. We compared each patient's demographics, seizure semiology, preoperative investigations, surgical technique, pathology results, and outcomes, taking into consideration the time for the first seizure after surgery.
All patients had surgery for temporal lobe epilepsy after being treated and classified as medically refractory by an experienced epileptologist. All patients had a complete investigation with continuous scalp video electroencephalography (EEG), magnetic resonance imaging (MRI) of the brain, and neuropsychological testing. Patients with extra-temporal lobe epilepsy, those with brain lesions that required surgery mainly for tumor resection and not for the intractability of the seizures, and patients with lesions that extended outside of the temporal lobe were excluded.
Patients were initially classified according to the seizure semiology, the presence of an aura, generalized seizures, and postictal state. A history of febrile seizures, handedness, age at onset of symptoms, duration of epilepsy, etiology, frequency of attacks, and number of antiepileptic drugs (AED) was also recorded. EEG and video-EEG examinations were acquired using the 10-20 International System. For interictal and ictal video-EEG, we classified all the epileptiform activity according to their lobar location. For statistical analyses we divided patients as having exclusively temporal concordant abnormalities or any alteration outside temporal lobe. MRI of the brain was obtained according to our institutional protocol, and the presence of any abnormality mentioned in the radiology report was noted. We recorded the presence of any abnormality, temporal lobe involvement, and the radiologic diagnosis. For statistical purposes we also divided the abnormal examinations as single or dual pathology (Kim et al., 2010).
Neuropsychological assessments were performed by trained neuropsychologists according to a standard protocol used at our center. We classified patients as normal, having a concordant temporal alteration, or having any other abnormality outside the epileptic temporal lobe (Keary et al., 2007). When necessary, invasive intracranial monitoring for accurate localization of the ictal-onset zone was performed. In our institution we most frequently use subdural electrodes, usually placed through posterior temporal burr holes (Steven et al., 2007).
Based on the results of the presurgical evaluation, patients were submitted to standard temporal lobectomy, selective amygdalohippocampectomy, lesionectomy, or a tailored temporal neocortical resection. The limits for resection during standard lobectomies were 6–6.5 cm from the tip of the temporal lobe in the nondominant hemisphere following the longitudinal line from the temporal pole at the middle temporal gyrus (Wiebe et al., 2001). The posterior limit in the dominant hemisphere was most often determined by cortical stimulation and intraoperative mapping of temporal speech areas, although a 4–5 cm limit was often observed. As mentioned earlier, patients with any resection outside of the temporal lobe were excluded from the study. Resected areas were submitted for pathologic examination and classified according to histologic abnormalities.
Patients who underwent temporal lobe surgery were classified initially as seizure-free or recurrence. A patient was considered seizure-free if they had no seizures at any point in the postoperative period up until the date of last follow-up. The only exception was patients who had seizures that occurred during the first 2 weeks after surgery. These immediate postoperative seizures were not used for the initial classification because of their uncertain importance for late prognosis (Malla et al., 1998). In addition, if a patient had only nondisabling auras, they were considered to be seizure-free. Patients considered “seizure-free” were discarded and the remainder of the analyses were performed on those patients with recurrent seizures after surgery. The time of the first postsurgical epileptic event was the reference for comparative analysis. We recorded the frequency of the seizures, the semiology, and the presence of trigger events. On the last follow-up, patients were scored according to the International League Against Epilepsy (ILAE) and Engel classifications for outcome (ILAE, 1981; Engel et al., 1993). We divided patients into two main groups for statistical analysis: significant improvement for ILAE 1, 2, or 3; or Engel class IC, ID, or II, and no significant improvement for the others. By definition, there were no Engel class IA or IB patients in this analysis. Patients who still had seizures postintervention but became seizure-free after a period of time (running down phenomenon; Rasmussen, 1970), and those who underwent reoperations, were also analyzed.
Receiver operating characteristic (ROC) curves were calculated initially with the intention to divide the two groups of patients with respect to long-term prognosis. We observed which period of time had the highest accuracy, sensitivity, and specificity to predict long-term outcome rates and used it as an indicator to separate the groups between early and late seizure recurrence. Categorical variables were compared using the two-tailed chi-square test or the two-tailed Fisher's exact test, when the requirements for the first one were not met. Quantitative variables were compared using the t-test for independent samples or nonparametric tests for the non-Gaussian distribution. The existence of a statistically significant difference was defined as probability value p < 0.05, and the confidence interval (CI) accepted was 95%. To verify the independent factors, a logistic regression was used. This approach allowed us to compare both groups, searching for differences that could eventually determine why some patients have earlier seizure recurrences and others have a seizure only after a period of seizure freedom. Kaplan-Meier survival curves with a logrank (Mantel-Cox) test were used to establish differences between good or bad prognosis in time regarding the first seizure after surgery.
Of the 247 patients who underwent surgery for temporal lobe epilepsy during the period under study, 219 satisfied our inclusion criteria. Of these, 107 (48.9%) had relapse of seizures. The median follow-up was 36 months (range 12–60). Based on the first event, 58 (54.2%) of these 107 patients experienced seizure recurrence before 6 months, 18 (16.8%) between 6 months and 1 year, 17 (15.9%) between 1 and 2 years, 5 (4.7%) between 2 and 3 years, 7 (6.5%) between 3 and 4 years, and 2 (1.9%) between 4 and 4 years (Fig. 1). Table 1 demonstrates the clinical characteristics of the included patients. Age at epilepsy onset ranged from 1 to 55 years of age (mean 16 years). Age at surgery ranged from 12 to 65 years (mean 34 years), and the time of epilepsy duration from 1 to 54 years of age (mean 20 years). Seventy-four patients (69.2%) had history of generalized seizures. A structural abnormality of any sort on MRI was identified in 87 (81.3%), with 57 (53.2%) having ipsilateral mesial temporal sclerosis.
Table 1. Characteristics of patients with recurrence of seizures
Using ROC curves, we found that a recurrence time of six postoperative months predicted long-term surgical outcome with the best sensitivity and specificity possible. The result was similar regardless of whether the Engel or ILAE classification was used. Using the Engel classification, patients could be divided into good long-term surgical outcome with an accuracy of 76.1% (95% CI = 0.665–0.867; p < 0.001). According to the ILAE criteria, accuracy was 72.9% (95% CI = 0.622–0.837, p < 0.001). This time frame was also useful for predicting the presence or absence of trigger events precipitating seizure recurrence, where the accuracy for predicting surgery outcome was 79.8% (95% CI = 0.707–0.890; p < 0.001; Fig. 2). Of interest, in all three curves, the highest sensitivity and specificity point to predict long-term seizure outcome was observed at 6 months after surgery (sensitivity 78.8%, specificity 72.1%). We therefore used this time to separate patients in two groups: the early recurrence group in which seizures returned within 6 months of surgery and the late recurrence group in which seizures returned after 6 months of surgery. We studied differences between these two groups in order to better understand factors associated with early or late seizure recurrence.
Differences between early and late seizure recurrence
After dividing the patients into late and early recurrence based on the ROC curves, a univariate analysis was performed on preoperative and postoperative variables (Tables 2 and 3). Age of epilepsy onset was the only presurgical variable found to be significantly different between the two groups (Table 2). Patients with early seizure recurrence after surgery had an earlier age of epilepsy onset (13.4 years) than those with a late seizure recurrence (19.5 years; OR = 6.034; 95% CI = 1.056–11.013; p = 0.018). Neither the types of surgical procedures performed nor the pathologic diagnoses differed between those with late and early recurrence. Patients with early recurrence had a poorer long-term prognosis when compared to the late seizure recurrence group, as classified using the ILAE (OR = 4.545; 95% CI = 1.785–11.111; p = 0.001) or Engel (OR = 7.142; 95% CI = 2.564–20; p = 0.001) outcome scores. The Kaplan-Meier survival curves demonstrated a statistically significant difference when comparing time to recurrence between those with higher and lower Engel or ILAE scores when analyzing outcome at last follow-up (p < 0.001; Fig. 3). These showed that patients with a better Engel or ILAE score tended to recur later than those with a poorer score. Patients with late recurrence were 7.4 times more likely to experience a >50% reduction in seizures than patients with earlier recurrence (95% CI = 1.55–35.4; p = 0.005), considering the Engel classification, and 5.9 times more considering the ILAE outcome score (95% CI = 1.78–19.25; p = 0.002). In addition, the frequency of seizures was higher in the group of patients with early recurrence (p = 0.027). The mean number of attacks was 3.29 (standard deviation [SD] ± 5.83) per month for early recurrence and 1.13 (SD ± 2.23) for late recurrence. Seizures that recurred after 6 months were more often associated with discrete trigger events when compared with seizures that relapsed earlier (OR = 2.82; 95% CI = 1.81–4.39; p < 0.001). Patients with early recurrence required a higher number of AEDs after surgery (p = 0.013). After logistic regression, only age of epilepsy onset (p = 0.05), the presence of a trigger factor (p = 0.002), and severity of seizures (p = 0.032) remained significantly different between patients with early and late seizure recurrence (Table 4).
Table 2. Univariate analysis of the preoperative variables comparing early versus late recurrence of seizures
Table 3. Univariate analysis of the postoperative variables comparing early versus late recurrence of seizures
Early (58) n (%)
Late (49) n (%)
Trigger of seizure
Significant improvement ILAE (1, 2, 3)
Significant improvement Engels (I, II)
ILAE, International League Against Epilepsy; AED, antiepileptic drug; OR, odds ratio; CI, confidence interval.
Frequency (per month)
Current number of AED
Table 4. Logistic regression of variables comparing early and late recurrence
OR (95% CI)
OR (95% CI)
The variables with p-value <0.02 after univariate analysis were taken for logistic regression. The variable current number of AED was excluded from the analysis because of the clear relation with outcome. The overall percentage of correctly predicted cases was 77.9%.
Age of seizure onset
Trigger of seizures
Engel (I, II)
Running-down phenomenon was observed in six patients (5.6%). The time until seizure freedom was achieved varied from 7 to 15 months (mean 11.3 months). Two of these patients had recurrent seizures that had a different semiology when compared to the preoperative events.
In our cohort, patients with early recurrence were significantly more often submitted to an additional surgical procedure for seizure control. Fifteen patients with recurrent seizures (14%) were investigated with subdural electrodes, and 11 underwent additional resective surgery. Of those who had subdural recordings, ictal EEG alterations were lateralized to the side ipsilateral to the original surgery in 13 patients, contralateral in one patient, and in one patient showed a more diffuse epileptogenic area (Table 5). Of the 58 patients with early recurrence, 10 (16.9%) underwent an additional resection, whereas only one of the 49 patients (2%) with late recurrence underwent further surgery (OR = 9.59; 95% CI = 1.18–77.87; p = 0.021). The timing of the second surgery varied from 2 to 11 years after the original operation (mean 6 years). In one case, the repeat resection was abandoned following cortical stimulation, as the epileptogenic area overlapped with the language area. Of the remaining 10 patients, all reoperations were performed in the original hemisphere. The surgical plans accomplished were the removal of the temporal neocortex in four patients with previous selective amygdalohippocampectomies, the resection of the remaining mesial structures in three patients with previous tailored neocortical resections, and the additional resection of the temporal neocortex in three patients who had standard temporal lobectomies. At last follow-up, five of the reoperated patients were Engel class I, two were Engel II, two were Engel III, and one was Engel class IV. In all, 70% of the patients with reoperations had a significant (Engel class I or II) improvement, and 90% had more than a 50% reduction of seizures frequency following the second operation.
Table 5. Characteristics of reoperations
Time of recurrence (months)
Time for reoperation (years)
Side of surgery
Outcome on last follow-up (ILAE/Engel)
Time of recurrence after 2nd surgery (months)
ILAE, International League Against Epilepsy; N/A not applicable.
Surgery was abandoned because seizure onset was found to overlapping with the temporal speech area.
Several studies have investigated the risk for seizure recurrence after surgery by comparing patients who were seizure-free versus those who were not (Foldvary et al., 2000; McIntosh et al., 2004; Janszky et al., 2005; de Tisi et al., 2011). However, it is possible that not all surgical failures are equal and that patients who recur earlier are distinct from those who recur later, possibly reflecting a different mechanism for seizure recurrence as well as a different prognosis. In a few studies, seizure recurrence has been separated into early and late groupings; however, very little comparison of the possible differences between these two groups of patients has been undertaken. Furthermore, there is no agreement about the time cutoff for classifying seizure relapse as early or late. Although some authors advocate for 1 year (Schwartz et al., 2006) and others for 5 years (Sperling et al., 2008), others consider 2 years as the best cutoff (McIntosh et al., 2004; Kelemen et al., 2006). However, these classifications have been mostly arbitrary and do not reflect any possible statistical or neurobiologic mechanism for seizure recurrence. What is unique about the current study is that we had no preconceived notion as to what time cutoff would be used to separate early from late recurrence. Rather than arbitrarily choosing a cutoff, the data were examined and a statistically relevant cutoff was selected. A division at 6 months was best found to fit the data. Patients in whom seizures recurred within the first 6 months of surgery had an earlier age of onset, a worse surgical outcome, and a higher postoperative seizure frequency.
We found that early recidivists had a younger age of epilepsy onset than those with seizure recurrence after 6 months. This is an interesting finding that might reflect a more active epileptogenic process and it is in line with findings in the literature, suggesting lower chances for good seizure control in the early epilepsy onset group (Cendes, 2011). This is also in keeping with evidence that suggests that age of epilepsy onset or duration of epilepsy might be directly related with surgical prognosis (Aull-Watschinger et al., 2008).
Once a patient has the first seizure after surgery, it is useful to have some way to predict the long-term outcome. Our results suggest that patients who have a recurrence within 6 months after surgery have a worse prognosis, with higher seizure frequency and a more frequent need for subsequent intracranial recordings or additional resective surgery than those who have a recurrence after 6 months. This in line with the results of Radhakrishnan and Kelemen who showed that seizures that returned before 1 year after surgery have a worse prognosis (Radhakrishnan et al., 2003; Kelemen et al., 2006). Late relapse seems to be a more benign condition, with less frequent seizures, which is in keeping with Buckingham et al. (2010) who stated that seizures that return after a longer period have better long-term outcomes, with higher chances of remission.
In our study, seizures that returned later than 6 months after surgery were more commonly associated with discreet trigger events. Tapering or withdrawal of AEDs and psychological stress were the main factors associated with late recurrences. It is possible that certain patients without precipitating factors might remain seizure-free for longer periods and may experience a recurrence when faced with one of these trigger events (Schmidt et al., 2004). This may explain why these patients were less frequently considered for reoperation. It is also possible that these patients have a lower threshold for seizures. In the late recurrence group the reduced frequency of seizures, the minor severity of the symptoms, and the lower number of AEDs prescribed for seizure control is perhaps a reflection of a new epileptogenic process (Jehi et al., 2012).
The pathophysiologic difference between early recurrence with incomplete resection versus late recurrence with perhaps a brain with lower threshold for seizures or epileptogenicity is under discussion (Fong et al., 2011). We did not note any histopathologic or radiologic differences between early and late recurrences. It could be argued that 6 months is a precocious period for development of a new focus; however, it is very important to note that for a patient who used to have several seizures a week or a month, being free of attacks for this length of time very likely has some relationship to removal of some or all of an epileptogenic area. In addition, the time for a new area of cortex to become epileptogenic is unknown. It is obvious that the longer the time the greater the chances for relapse, but it is possible that 6 months is sufficient to clinically define this process.
We found that patients who were selected for a second surgery were more often those with an early seizure recurrence. From the 11 patients submitted to another surgical procedure for seizure control, 10 (90.9%) were included in the early seizure relapse group, and all of them had their prior surgical area extended. It is unclear why reoperations were not performed as frequently in the late recurrence group. Most of the postoperative patients with recurrent seizures are reevaluated at our center regardless of the time of recurrence, and time to recurrence is not specifically used as a deciding factor when considering additional surgery. As mentioned earlier, it is possible that due to the milder nature of the epilepsy in those with late recurrence that surgery was not felt to be necessary. It is also possible that the seizures were felt to be either multifocal, generalized, or of contralateral origin, supporting the hypothesis that these patients have either a lower seizure threshold or an underlying propensity to develop seizure foci; this is deserving of further study. Of those patients who underwent repeat surgery, five patients were rendered seizure-free, a finding that strongly implies that the reason for early recurrence in these patients was an incomplete resection of the epileptogenic zone, rather than a placebo effect or some unknown physiologic cause related to surgery in general. These findings are in keeping with Germano et al. (1994), who reported that the initial recurrence of seizures generally occurred during the first 6 months in a series of 40 patients needing reoperation for TLE seizures. However, the partial resection of the epileptogenic area may not be the main physiologic explanation for all early recurrences. Although 16.9% of early recurrences had repeat surgery and many did well, the majority (83.1%) did not have repeat surgery. It is possible that these patients had another epileptogenic zone that was not identified. It is interesting to observe that even when the seizures returned just after few months, the mean time for the second resection was 6 years after the original operation. Considering that the latest observed running-down phenomenon was 15 months postoperatively, and that 90% of the reoperated patients had some benefit from the second surgical procedure, it would be a reasonable strategy to consider investigation for another surgery by 2 years after the first surgical attempt.
The limitations of this study reside in its retrospective nature and the fact that some statistics are exploratory. It is also possible that there might be differences between the late and early recurrence groups that would be detected only with a group of patients larger than the 107 presented here.
Nonetheless, this study presents some important information regarding the timing of recurrence after surgery, and suggests that 6 months might be the most useful time cut-off for defining early versus late recurrence after surgery for TLE. Although preoperative characteristics are good predictors for remission or recurrence of seizures after surgery (McIntosh et al., 2001), once the primary goal is not achieved, the time of the first seizure might be an important predictor for long-term seizure outcome. In this venue our study might help regarding prognosis definition and planning future treatments. We propose that recurrence of seizures after surgery for temporal lobe epilepsy should be separated into early or late recurrences based on the time frame of 6 months following surgery. Patients with seizure recurrence within 6 months have worse outcomes, higher frequency of attacks, tend to use a higher number of AEDs, and carry a higher probability for reoperations when compared with those patients with late seizure recurrence.
None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.
Dr. Goellner is an attending neurosurgeon at Hospital Mãe de Deus in Porto Alegre, Brazil.