Long-term outcomes of reoperations in epilepsy surgery

Abstract

Objective:

To analyze longitudinal seizure outcomes following epilepsy surgery, including reoperations, in patients with intractable focal epilepsy.

Methods:

Clinicoradiological characteristics of patients who underwent epilepsy surgery from 1995 to 2016 with follow-up of ≥1 year were reviewed. In patients undergoing reoperations, the latest resection was considered the index surgery. The primary outcome was complete seizure freedom (Engel I) at last follow-up. Potentially significant outcome variables were first identified using univariate analyses and then fit in multivariate Cox proportional hazards models.

Results:

Of 898 patients fulfilling study criteria, 110 had reoperations; 92 had one resection prior to the index surgery and 18 patients had two or more prior resective surgeries. Two years after the index surgery, 69% of patients with no prior surgeries had an Engel score of I, as opposed to only 42% of those with one prior surgery, and 33% of those with two or more prior resections (P < .001). Among surgical outcome predictors, the number of prior epilepsy surgeries, female sex, lesional initial magnetic resonance imaging, no prior history of generalization, and pathology correlated with better seizure outcomes on univariate analysis. However, only sex (P = .011), history of generalization (P = .016), and number of prior surgeries (P = .002) remained statistically significant in the multivariate model.

Significance:

Although long-term seizure control is possible in patients with failed prior epilepsy surgery, the chances of success diminish with every subsequent resection. Outcome is additionally determined by inherent biological markers (sex and secondary generalization tendency), rather than traditional outcome predictors, supporting a hypothesis of “surgical refractoriness.”

1 |. INTRODUCTION

Epilepsy surgery is the most effective and widely accepted treatment option for patients with drug-resistant focal epilepsy, leading to seizure freedom in 60%–80% of surgical patients with intractable temporal lobe epilepsy and in 50%–60% after extratemporal resections.^1–3 Although most patients benefit from epilepsy surgery, some might not have a worthwhile reduction in seizure burden.⁴ Ongoing or recurrent postoperative seizures continue to impair quality of life and add to the morbidity and mortality. With ever-improving imaging and electrophysiological localization tools, there is a growing urge to trigger a consideration of further surgery, and optimism to relocalize the epileptogenic zone, and reoperate after a surgical failure. The underlying assumption is that a reoperation would address any residual epileptogenic tissue, and therefore should be successful if the epilepsy is accurately localized and completely resected. Recent data suggest, however, that inherent, potentially genetic, and patient characteristics may increase the risk of surgical failure regardless of localization accuracy and completeness of resection.⁵

From a pragmatic perspective, it can be daunting for patients and physicians to opt for a reoperation while balancing it against the possibility of functional deficits, operative risks, and cost of repeated presurgical evaluations.^6,7 This is a particular challenge considering that data on long-term outcomes and their predictors after reoperations are limited. As highlighted in a recent comprehensive meta-analysis of the heterogeneous available literature,⁸ most studies were under-powered, not controlled, and some were restricted to a specific age group,⁶ a specific pathology,⁹ or a specific lobe.^10–13 In addition, all published experience is limited to a single reoperation, limiting its applicability in patients requiring multiple reoperations. A long-term view assessing the longitudinal benefits of an initial epilepsy surgery and subsequent reoperations would be helpful to guide decision-making, particularly given the emergence of multiple neuromodulatory treatment options.

In this study, we analyze the evolution and predictors of long-term outcomes after consecutive epilepsy surgeries in a large cohort of comprehensively evaluated drug-resistant focal epilepsy patients. Our extensive longitudinal follow-up allows us to additionally focus on the subgroup with multiple reoperations.

2 |. MATERIALS AND METHODS

2.1 |. Patient selection

With approval from the institutional review board, all consecutive patients who underwent epilepsy surgery at the Cleveland Clinic from January 1995 to December 2016 were screened (991 patients). We excluded patients whose first surgery was a hemispherectomy, corpus callosotomy, or tumor debulking (35 patients), as well as high-grade tumor patients. Patients who underwent the first surgery at another institution for whom insufficient information was available were also excluded (58 patients). For patients who underwent temporal lobe resections for mesial temporal sclerosis, a standard temporal lobectomy was done. Patients considered to have drug-resistant epilepsy according to the International League Against Epilepsy definition¹⁴ were subjected to a standard presurgical assessment, collecting data points of history, physical examination, video-electroencephalographic (vEEG) monitoring, brain magnetic resonance imaging (MRI), and positron emission tomography (PET) scan. At a multidisciplinary patient management conference, the decision was made to proceed with surgery or obtain further testing including invasive EEG. Routine postoperative follow-up was scheduled at 6 weeks, 3 months, 6 months, and 1 year and yearly thereafter for seizure-free patients. Patients whose seizures recurred were seen more frequently. When a reoperation was considered, repeat neuroimaging (typically including MRI, PET, and ictal single photon emission computed tomography) was obtained in addition to a repeat vEEG and often invasive EEG evaluations.

To evaluate the independent contribution of the number of previous epilepsy surgeries on outcomes, we considered the most recent operation as the index surgery and evaluated the prognostic significance of having prior resections in addition to the traditionally evaluated outcome predictors. We also considered the timing of the initial surgery to account for any potential effect of the evolution in surgical workups and techniques on outcomes.

2.2 |. Data collection

Because our goal was to investigate outcomes after reoperations, our study variables characterize the most recent epilepsy surgery (epilepsy duration, age at surgery, history of generalization, preoperative monthly seizure frequency, invasive evaluation, lobe and side of resection, and postoperative pathology). For the analysis, MRI findings prior to the initial surgical intervention were included, because those were most reflective of the underlying epilepsy etiology.

In patients who had prior resections, relevant data were also collected for each one of the surgeries (prior invasive evaluation, lobe and side of resection, postoperative pathology, and timing of seizure recurrence). Focusing on the patients who failed the first surgery and were considered for a reoperation, a detailed comparison of EEG, MRI, and overall surgical hypothesis pre- and postsurgery was done to identify reasons for surgical failure.

Seizure outcomes were ascertained from the outpatient follow-up visits or telephone contact (at 6 weeks, 3 months, 6 months, and 1 year and yearly thereafter). For patients who did not have an exact date of recurrence, the recurrence date was estimated to be halfway between the two dates of ascertained outcome neighboring it. For seizure recurrence after surgery, the time to recurrence was grouped as early (within 6 months) or late (>6 months) recurrence.¹⁵ For patients with acute postoperative seizures, the first seizure occurring after 1 week of the operative date was considered the date of recurrence.

2.3 |. Outcome definitions

We investigated two outcomes: first, freedom from any seizure recurrence after surgery (isolated auras were not classified as seizures); and second, probability of obtaining an Engel score of I (eventual favorable seizure outcome allowing for some initial postoperative seizures, or seizures occurring only with physiological stress such as drug withdrawal).

2.4 |. Statistical analysis

Categorical variables were described using frequencies and percentages, whereas continuous variables were described using means and standard deviations (for normally distributed data) or medians and quartiles (for nonnormal distributions). The relationship between number of previous surgeries and variables of interest was assessed using t tests or analysis of variance models (normally distributed continuous variables), Wilcoxon rank sum tests or Kruskal-Wallis tests (ordered data or nonnormally distributed continuous variables), or Pearson chi-square tests (nonordered categorical data). Ad hoc comparisons were run for significant three-group comparisons between patients who had one surgery, two surgeries, or three or more surgeries. In patients with a reoperation, to evaluate relationships between time of recurrence of previous surgery and current outcome based on Engel score, mixed effect models that accounted for multiple surgeries were performed. For continuous measures, the recurrence time in months was log-transformed prior to analysis and then back-transformed for presentation. Models were run using all patients with valid recurrence dates, and then as a sensitivity analysis including those with missing recurrence date and imputing either the date of next surgery (max) or midpoint between the two surgeries (midpoint). Hazard ratios from Cox proportional hazards models were presented for all measures univariably. Condition indices and variance inflation factors were used to evaluate multicollinearity among the predictors. Condition indices and variance inflation factors > 10 indicate multicollinearity. Offending variables were identified and removed, and this process was repeated until maximum levels of both measures were <10. As a result, the multivariate model was fit without including invasive surgery status, age at onset, and seizure frequency. High multicollinearity indicates that predictors are highly correlated, and that including the variables in the model will affect the interpretability and significance of other factors in the model. These factors may hold prognostic value, but it is not possible to evaluate their impact concurrently with other predictors. Multivariate Cox proportional hazards models were fit with the remaining variables. Analyses were performed using SAS Software (v9.4).

3 |. RESULTS

3.1 |. Patient characteristics

3.1.1 |. Overall patient characteristics

A total of 898 patients fulfilled study criteria. Half were female, one-third were pediatric patients, and 68% had temporal lobe resections. Median age of the pediatric cohort was 9 years (interquartile range [IQR] = 3.0–13.0), whereas the median age for the adult group was 36 years (IQR = 24.0–48.0). Fifty-two percent of resections were on the left side. Median age at onset of epilepsy was 9 years (IQR = 3.0–19.0), with a median duration of epilepsy of 13 years (IQR = 5.0–24.0). The median preoperative seizure frequency was 8 seizures/mo (IQR = 3.0–40.0), and 70% of the cohort had secondary generalization. Mean follow-up was 4.2 years (SD = 3.3 years; Table 1).

TABLE 1.

The relationship between patient characteristics and number of previous surgeries from the dataset is shown below

	Overall, N = 898
Factor	n	Statistics	First surgery, n = 788	Second surgery, n = 92	Third or more surgery, n = 18	P
Gender
F	898	448 (49.9)	393 (49.9)	45 (48.9)	10 (55.6)	.88a
M		450 (50.1)	395 (50.1)	47 (51.1)	8 (44.4)
Age at onset	896	9.0 [3.0–19.0]	9.0 [3.0–19.0]	10.0 [3.5–16.0]	6.5 [2.0–13.0]	.58b
Age at surgery	898	29.0 ± 16.2	29.3 ± 16.3	26.6 ± 15.5	28.3 ± 15.5	.33C
Pediatric vs adult
Adult	898	251 (28.0)	214 (27.2)	32 (34.8)	5 (27.8)	.30a
Pediatric		647 (72.0)	574 (72.8)	60 (65.2)	13 (72.2)
Epilepsy duration	894	13.0 [5.0–24.0]	13.0 [5.0–24.0]	11.5 [5.0–19.5]	16.5 [11.0–26.0]	.28b
Epilepsy duration < 5 y
3 or more years of epilepsy	894	795 (88.9)	694 (88.5)	84 (91.3)	17 (94.4)	.54a
<3 y		99 (11.1)	90 (11.5)	8 (8.7)	1 (5.6)
Preop seizure frequency, per mo	892	8.0 [3.0–40.0]	8.0 [3.0–30.0]f	16.0 [3.0–60.0]e	12.0 [3.0–90.0]	.027b,f
GTC
No	893	273 (30.6)	231 (29.5)	35 (38.0)	7 (38.9)	.18a
Yes		620 (69.4)	552 (70.5)	57 (62.0)	11 (61.1)
Initial brain MRI
Abnormal	895	773 (86.4)	677 (86.1)	80 (87.9)	16 (88.9)	.85a
Normal		122 (13.6)	109 (13.9)	11 (12.1)	2 (11.1)
PET abnormality
Localized	768	551 (71.7)	497 (72.4)	47 (66.2)	7 (63.6)	.31a
Nonlocalized		91 (11.8)	80 (11.7)	8 (11.3)	3 (27.3)
Surgical side
L	898	469 (52.2)	411 (52.2)	47 (51.1)	11 (61.1)	.053a
R		428 (47.7)	377 (47.8)	44 (47.8)	7 (38.9)
Lobe
Extratemporal	865	276 (31.9)	244 (31.0)	25 (38.5)	7 (58.3)	.065a
Temporal		589 (68.1)	544 (69.0)	40 (61.5)	5 (41.7)
Simplified pathology
Gliosis	893	223 (24.9)	199 (25.3)d	18 (20.0)e	6 (33.3)	<.001a,f
MCD		246 (27.5)	205 (26.0)	35 (38.9)	6 (33.3)
MTS		262 (29.3)	248 (31.5)	12 (13.3)	2 (11.1)
Stroke/vascular		43 (4.8)	39 (5.0)	4 (4.4)	0 (0.0)
TUM		121 (13.5)	96 (12.2)	21 (23.3)	4 (22.2)
Acute postop Sz
No	275	217 (78.9)	196 (80.7)	17 (68.0)	4 (57.1)	.12a
Yes		58 (21.1)	47 (19.3)	8 (32.0)	3 (42.9)
Engel
I	894	589 (65.9)	545 (69.3)d,g	38 (42.2)e	6 (33.3)e	<.001a,f
II+		305 (34.1)	241 (30.7)	52 (57.8)	12 (66.7)
Seizure freedom at 6 mo
No	878	277 (31.5)	229 (29.6)d	39 (44.8)e	9 (50.0)	.004a,f
Yes		601 (68.5)	544 (70.4)	48 (55.2)	9 (50.0)
Seizure freedom at 12 mo						.007a,f
No	836	300 (35.9)	250 (34.0)d	42 (50.0)e	8 (50.0)	.007a,f
Yes		536 (64.1)	486 (66.0)	42 (50.0)	8 (50.0)	.007a,f

Note: Statistics are presented as mean ± SD, median [interquartile range], median (min, max) or n (column %). A significance level of .017 was used for pairwise ad hoc comparisons. Bold values indicate significant findings.

Abbreviations: F, female; GTC, generalized tonic-clonic seizures; L, left; M, male; MCD, malformation of cortical development; MRI, magnetic resonance imaging; MTS, mesiotemporal sclerosis; PET, positron emission tomography; R, right; TUM, tumor; Sz, seizures.

^aPearson chi-square test.

^bKruskal-Wallis test.

^cAnalysis of variance.

^dSignificantly different from 1.

^eSignificantly different from 0.

^fStatistically significant.

^gSignificantly different from 2+.

3.1.2 |. Patient characteristics relative to number of previous surgeries

Of the 898 patients, 92 patients had one prior resective surgery, and an additional 18 patients had a total of two or more prior resective surgeries. A more detailed description of clinical, imaging, and pathological findings of these different subgroups can be found in Table 1. Patients with one or more prior surgeries were more likely to have resections of the left side (P = .053) and a higher frequency of preoperative seizures per month (P = .027). There was no significant association between demographics like sex, age at surgery (pediatric or adult), epilepsy duration, or initial MRI or PET findings with the number of epilepsy surgeries a patient underwent.

3.2 |. Overall postoperative seizure outcome

3.2.1 |. Complete postoperative seizure freedom

At 2 years, 58% (95% confidence interval [CI] = 55–62) were seizure-free among those with no prior surgery. Seizure freedom was significantly lower (49%, 95% CI = 39–60; hazard ratio [HR] = 1.52, 95% CI = 1.16–2.00; P = .003) among those with one prior surgery, and also in those with two or more prior surgeries (39%, 95% CI = 14–64; HR = 1.90, 95% CI = 1.07–3.38; P = .028; Figure 1).

FIGURE 1.

Kaplan-Meier curves showing seizure freedom in relation to: A, number of surgeries; B, sex of the patient; and C, history of generalization. GTC, generalized tonic-clonic seizures

3.2.2 |. Engel outcomes

Among patients who underwent no prior surgery, 69% had an Engel score of I at 2 postoperative years, whereas the majority (60%) of the patients with multiple surgeries had an Engel score greater than I (P < .001). Table 1 details the Engel outcomes relative to the number of prior resections.

3.3 |. Predictors of recurrence

3.3.1 |. Univariate analysis

Variables of interest were used to identify predictors of recurrence (Table 2). If the initial brain MRI was normal, 61% recurred, whereas 52% failed to achieve seizure freedom with abnormalities on the MRI (HR = 0.75, 95% CI = 0.59–0.96; P = .023). When studying the effect of the decade in which the surgery took place, higher recurrence rates were observed in patients who had their index surgery in more recent years (HR = 1.03, 95% CI = 1.01–1.06; P = .002). In addition, gliosis on pathology of prior surgery and increased preoperative seizure frequency were significantly associated with an increased risk of recurrence but did not remain significant in the multivariate model. Fifty eight percent of patients with no prior invasive evaluation had a recurrence as compared to 73% with an invasive evaluation done before the resective surgery (HR = 1.44, 95% CI = 0.90–2.29; P = .13.

TABLE 2.

Predictors of recurrence: univariate

Variable	n	Events, n (%)	1-y survival, % (95% CI)	2-y survival, % (95% CI)	Cox univariate hazard ratio (95% CI)	Cox univariate Wald P	Cox univariate overall P
Previous surgery
0	788	406 (52%)	67.3 (64.0–70.7)	58.2 (54.6–61.8)	1.00 (REF)		001a
1	92	59 (64%)	52.1 (41.6–62.6)	49.4 (38.8–60.0)	1.52 (1.16–2.00)	.003
2+	18	12 (67%)	54.3 (30.7–77.9)	38.8 (14.0–63.6)	1.90 (1.07–3.38)	.028
IS
No IS	62	36 (58%)	56.4 (43.7–69.0)	50.2 (37.2–63.3)	1.00 (REF)		.13
Previous IS	48	35 (73%)	47.5 (32.9–62.1)	44.7 (30.0–59.5)	1.44 (0.90–2.29)	.093
Gender
M	450	254 (56%)	62.9 (58.4–67.4)	52.8 (48.0–57.6)	1.00 (REF)	<.001	.044a
F	448	223 (50%)	68.3 (63.9–72.7)	61.1 (56.4–65.8)	0.831 (0.693–0.995)
Abnormal initial MRI
No	122	75 (61%)	62.9 (54.2–71.6)	48.3 (39.0–57.7)	1.00 (REF)	.044	.023a
Yes	773	400 (52%)	66.0 (62.6–69.4)	58.2 (54.6–61.8)	0.75 (0.59–0.96)
GTC
No	273	130 (48%)	69.4 (63.9–75.0)	61.3 (55.3–67.4)	1.00 (REF)	.023	.030a
Yes	620	344 (55%)	63.7 (59.9–67.6)	54.8 (50.7–58.9)	1.25 (1.02–1.53)
Index surgery pathology						.030
MCD	246	131 (53%)	65.9 (59.9–71.9)	53.1 (46.5–59.8)	1.00 (REF)		.042a
MTS	262	139 (53%)	69.3 (63.7–74.9)	62.9 (57.0–68.9)	0.81 (0.64–1.03)
Gliosis	220	126 (57%)	60.9 (54.3–67.4)	52.0 (45.0–58.9)	1.10 (0.86–1.41)	.093
Other lesion	167	79 (47%)	66.2 (58.9–73.4)	59.6 (51.8–67.3)	0.81 (0.61–1.08)	.43
Proximity to Eloq cortex
No	606	309 (51%)		58.8 (54.7–62.8)	1.00 (REF)		.16
Yes	292	168 (58%)		53.0 (47.0–59.0)	1.15 (0.95–1.38)	.16
First surgery in OSH
No	86	58 (67%)		48.7 (37.8–59.7)	1.00 (REF)		.56
Yes	24	13 (54%)		46.4 (25.3–67.4)	1.20 (0.65–2.22)	.56
Index surgery year	898	477 (53%)		56.9 (53.6–60.3)	1.038 (1.014–1.062)	.002	.002a
Continuous measures						.15
Onset age, per decade	896	476 (53%)			1.03 (0.97–1.11)
Surgery age, per decade	898	477 (53%)			1.01 (0.96–1.07)	.35	.35
Epilepsy duration, per decade	894	476 (53%)			0.98 (0.92–1.05)	.62	.62
Preop seizure frequency, per 30 d	892	475 (53%)			1.01 (1.00, 1.02)	.64	.64
						.016	.016a

Abbreviations: IS, invasive surgery; Eloq, eloquent; CI, confidence interval; REF, reference; M, male; F, female; MRI, magnetic resonance imaging; GTC, generalized tonic-clonic seizures; MCD, malformation of cortical development; MTS, mesiotemporal sclerosis; OSH, outside hospital.

^aStatistically significant.

3.3.2 |. Multivariate analysis

After multivariate Cox proportional hazard modeling, three variables retained their significance as independent recurrence predictors (Table 3):

1. As detailed above, a higher number of previous surgeries is a predictor of recurrence (one surgery: HR = 1.47, 95% CI = 1.11–1.96; two or more surgeries: HR = 2.08, 95% CI = 1.16–3.73; P = .002; Figure 1).

2. Male patients had a lower (44%) chance of seizure freedom after surgery as compared to women (50%; HR = 0.79, 95% CI = 0.66–0.95; P = .011; Figure 1).

3. A positive history of secondary generalization was significantly associated with increased seizure recurrence (HR = 1.29, 95% CI = 1.05–1.59; P = .016; Figure 1).

TABLE 3.

Predictors of recurrence: multivariate

Variable	n	Events, n (%)	Cox multivariate hazard ratio (95% CI)	Cox multivariate Wald P, n = 885	Cox multivariate likelihood ratio P, n = 885
Previous surgery
0	788	406 (52%)	1.00 (REF)		.002
1	92	59 (64%)	1.47 (1.11–1.96)	.008
2+	18	12 (67%)	2.08 (1.16–3.73)	.014
Gender
M	450	254 (56%)	1.00 (REF)		.011
F	448	223 (50%)	0.79 (0.66–0.95)	.011
Abnormal initial MRI
No	122	75 (61%)	1.00 (REF)		.10
Yes	773	400 (52%)	0.81 (0.63–1.04)	.10
GTC
No	273	130 (48%)	1.00 (REF)		.016
Yes	620	344 (55%)	1.29 (1.05–1.59)	.016
Index surgery pathology
MCD	246	131 (53%)	1.00 (REF)		.12
MTS	262	139 (53%)	0.82 (0.63–1.08)	.16
Gliosis	220	126 (57%)	1.05 (0.81–1.35)	.72
Other lesion	167	79 (47%)	0.79 (0.59–1.06)	.12
Surgical side
R	428	218 (51%)	1.00 (REF)		.43
L	469	258 (55%)	1.08 (0.90–1.29)	.43
Surgery age, per decade	898	477 (53%)	1.071 (0.992–1.156)	.079	.079
Epilepsy duration, per decade	894	476 (53%)	0.95 (0.86–1.04)	.25	.25

Abbreviations: CI, confidence interval; F, female; GTC, generalized tonic-clonic seizures; L, left; M, male; MCD, malformation of cortical development; MRI, magnetic resonance imaging; MTS, mesiotemporal sclerosis; R, right; REF, reference.

3.4 |. Possible reasons for failure of irst surgery

Of the total 898 patients studied, 58% (516) had a recurrence after the first surgery. Of these, 209 were evaluated for a reoperation and 110 underwent subsequent surgery (Figure 2). Upon review of postoperative MRI, EEG, and the surgical hypothesis and comparing them with the corresponding preoperative data, the probable reasons for failure were divided into three categories:

1. Mislocalization, when postoperative EEG or MRI abnormalities were in a different lobe/noncontiguous; this was seen in 29 patients (only three patients with true dual pathology were identified);

2. Incomplete resection, when there was residual lesion on postoperative MRI (n = 30) or the resection was in eloquent cortices (n = 7); and

3. Unexplained by above two categories (n = 44), when there was no clear reason for surgical failure.

FIGURE 2.

Flowchart delineating possible reasons for recurrence after a failed first surgery. EEG, electroencephalographic

Among the three groups, no significant difference was found between seizure outcome after the next surgery (P = .8). However, we also compared the time to recurrence after first surgery in patients in these groups. Patients with an incomplete resection were the fastest to recur (mean = 5.1 months), followed by patients where the focus was mislocalized (mean = 7.5 months), and finally the unexplained group had the longest mean time to recurrence of 10 months (P = .2).

Further details regarding surgical failure in patients that underwent two or more surgeries are enumerated in Table 4.

TABLE 4.

Patients who underwent multiple surgeries grouped by initial MRI findings

	Surgery 1				Surgery 2				Surgery 3				Further surgeries: invasive, pathology, reason for recurrence, resection	Seizure recurrence
ID	Invasive	Pathology	Resection	Initial MRI	Invasive	Pathology	Reason for recurrence	Resection	Invasive	Pathology	Reason for recurrence	Resection
1	No	Unknown	TL	Cystic lesion	No	Gliosis	Unexplainable by prior findings	TL	SDE + depth	Gliosis	EEG finding elsewhere	TL		Seizure-free
2	SDE	MCD	FL	MCD	SDE	Gliosis	Unexplainable by prior findings	FL	SEEG	MCD	EEG finding elsewhere	FL + insula		Seizure-free
3	No	MCD	FL	MCD	No	MCD	MRI residual lesion	FL	No	NA	Unexplainable by prior findings	Callosotomy		Recurred
4	SDE + depth	MCD	FL	MCD	No	MCD	Unexplainable by prior findings	FL	SDE + depth	MCD	Unexplainable by prior findings	FL		Recurred
5	No	NA	TPO	MCD	Unknown	NA	MRI residual lesion	TPO	No	MCD	MRI residual lesion	TPO		Recurred
6	No	Gliosis	TL	MTS	No	NA	MRI residual lesion	TL	SDE + depth	MTS	MRI residual lesion	TL + FL	Last surgery (total 6 surgeries): no invasive, MCD, EEG findings elsewhere, FL + PL	Recurred
7	No	MTS	TL	MTS	No	MTS	MRI residual lesion	TL	SDE	Gliosis	Unexplainable by prior findings	TL		Recurred
8	No	MTS	TL	MTS	SDE	Gliosis	EEG finding elsewhere	FL + TL	SEEG	Gliosis	EEG finding elsewhere	TPO		Recurred
9	No	MTS	TL	MTS	SDE	Gliosis	Unexplainable by prior findings	TL	SEEG	Gliosis	Unexplainable by prior findings	TL	4th surgery: no invasive, pathology NA, EEG finding elsewhere, insula	Seizure-free
10	SEEG	MTS	TL	MTS	No	MTS	MRI residual lesion	TL	SEEG	MCD	Dual pathology	TL		Seizure-free
11	Depth	NA	TL	NA	SDE	NA	unexplainable by prior findings	PL	SDE + depth	NA	EZ close to eloquent areas	PL	4th surgery: SEEG, gliosis, unexplainable by prior findings, PO	Recurred
12	SDE	MCD	FL	Normal	SDE + depth	MCD	Unexplainable by prior findings	FL	SEEG	Gliosis	EEG finding elsewhere	TL + FL	4th surgery: no invasive, nonspecific, unexplainable by prior findings, FL	Seizure-free
13	SDE	Nonspecific	FL + TL	Normal	No	Gliosis	EEG finding elsewhere	FL + TL	SEEG	Gliosis	EEG finding elsewhere	PO		Recurred
14	No	Tumor	TL	Tumor	No	Tumor	MRI residual lesion	TL	No	Tumor	MRI residual lesion	TL	4th surgery: no invasive, tumor, residual lesion, TL	Recurred
15	No	Tumor	TL	Tumor	No	Tumor	MRI residual lesion	TL	SDE	Tumor	MRI residual lesion	TL	4th surgery: no invasive, gliosis, unexplainable by prior findings, TL	Recurred
16	No	Tumor	TL	Tumor	SDE	Gliosis	EEG finding elsewhere	TL + FL	SEEG	None	Unexplainable by prior findings	TL + FL		Recurred
17	No	Tumor	TL	Tumor	SDE	Gliosis	unexplainable by prior findings	TL	SDE + depth	Gliosis	EEG finding elsewhere	TL + insula	4th surgery: SEEG, gliosis, eloquent areas, TL	Recurred
18	No	Tumor	TPO	Tumor	No	Tumor	MRI residual lesion	TPO	No	Gliosis	Unexplainable by prior findings	TPO		Seizure-free

Abbreviations: depth, depth electrodes; EEG, electroencephalographic; EZ, epileptogenic zone; FL, frontal lobe; MCD, malformation of cortical development; MRI, magnetic resonance imaging; MTS, mesiotemporal sclerosis; NA, not available; PL, parietal lobe; PO, parietal-occipital; SDE, subdural grids; SEEG, stereo-EGG; TL, temporal lobe; TPO, temporal-parietal-occipital.

3.5 |. Timing of recurrence

A subgroup analysis was done of the 110 subjects who had a total of 134 reoperations. The patients with a poor outcome after reoperation (Engel outcome of II or more) had a median time to recurrence after the previous surgery of 3.3 months, as compared to 5.3 months in the group with an Engel outcome of I after their reoperation. Of those who had a poor outcome after their reoperation (Engel II or more), 63.3% recurred within the 6 months of their previous surgery, as compared to 51.3% among those who had a good outcome after the reoperation (Engel I).

4 |. DISCUSSION

We report the first single-center study of longitudinal outcomes, and their predictors after epilepsy reoperations in a large cohort of drug-resistant focal epilepsy patients, with a subgroup undergoing multiple repeat surgeries. This is a complex and difficult to treat patient population, and little literature is available to guide the selection of patients for reoperations. The meta-analysis by Krucoff et al⁸ effectively shed light on outcomes in patients with a single reoperation; however, the effect of consecutive reoperations was not addressed. Additionally, it had the inherent disadvantage of heterogeneity that a meta-analysis carries, as the compilation included studies from several countries with variations in diagnostic, explorative, and resective techniques.

A prior attempt to study reoperations in temporal lobe epilepsy surgery failures was made by our center in 2010 that helped characterize the proximity of the new epileptogenic substrate in early versus late recurrences. The current study expands on this concept to include temporal as well as extratemporal surgeries with an overlap of 15 patients between the two study cohorts.¹⁶

4.1 |. Characteristics of patients with multiple surgeries

In this series, we found that patients with multiple reoperations were more likely to have had left-sided surgeries. A poor surgical outcome with left-sided surgeries has been well documented after a single surgery,¹⁷ and we extend this observation to include reoperations. A likely explanation for such an observation is the adoption of a conservative approach leading to smaller tailored resections limited by eloquent cortex with surgeries involving the dominant hemisphere.¹⁸ This may result in an incomplete resection of the epileptogenic zone leaving behind residual epileptogenic tissue, preventing seizure freedom or consequently causing a recurrence of seizures soon after surgery. Moreover, incomplete resections have also been reported to cause a postoperative paradoxical worsening of seizures,¹⁵ making these patients candidates for an evaluation for reoperations. Side, however, had no bearing on eventual seizure outcomes, and neither did a specific analysis of proximity to eloquent cortex (Table 2), suggesting that the higher tendency for seizure recurrence in our cohort cannot be attributed to surgical limitations imposed by epileptogenic zone localization close to eloquent function. The second characteristic that achieved significance in this cohort was a higher preoperative seizure frequency. This is in concordance with previous published evidence,^19–21 and we hypothesize that it might suggest the presence of a highly epileptogenic substrate reflected by its ability not only to seize more frequently but also to generate a new epileptogenic focus after resection, culminating in a surgical failure. Another explanation might be the persistence of this high baseline seizure frequency after seizure recurrence, prompting consideration for a reoperation to improve quality of life.

Put together, these two findings highlight a complex patient population (dominant hemisphere epilepsy with a high seizure burden) that is particularly vulnerable to being chosen for reoperations while searching for a treatment.

4.2 |. Overall outcomes

In our series, although a substantial percentage of patients became seizure-free after reoperations, the chances of achieving complete seizure freedom and an acceptable Engel outcome decreased with every subsequent surgery. There are limited possible explanations for such a trend. In this presumably challenging subgroup of patients, a simple hypothesis could be the mislocalization of the epileptogenic zone on two separate attempts. Although plausible in some cases, this traditional justification of surgical failure may not explain the phenomenon entirely. We will take next a deeper look into the drivers of seizure recurrence in this patient cohort to gain a better understanding of the underlying mechanisms of postoperative seizure recurrence.

4.3 |. Predictors of seizure recurrence

In keeping with available experience and evidence, the univariate analysis showed patients with abnormal MRI were likely to have a better seizure outcome.^20,22 The available explanation is an improved ability to guide the margins for resection. However, this observation lost statistical significance in the multivariate model. In epilepsies where the epileptogenic zone extends beyond the obvious lesion on MRI, even a well-tailored lesionectomy may fail to prevent seizure recurrence. During the repeat surgical evaluation in such patients, we are faced with a wider than initially anticipated epileptogenic network. This may account for the absence of influence of initial lesional MRIs in predicting outcomes in repeat surgical patients. The need for a more suitable tool to identify the area of resection still continues.

The univariate analysis also showed a decrease in the number of surgeries being conducted over time and a worse outcome in index surgeries that have taken place in more recent years. Although surprising, it is becoming a well-documented effect that culminates from a few reasons. First, with more sophisticated invasive EEG techniques and their reduced morbidity, there is a lower threshold for the use of invasive EEG to evaluate the epileptogenic network for surgical failures.²³ Nonetheless, there are an increasing number of invasive evaluations that do not lead to subsequent resections in recent years.²⁴ Second, there has been a shift in the demographic of the presenting patient population from “straight-forward” mesial temporal lobe epilepsies to more complex extratemporal and nonlesional substrates.

In this cohort, after multivariate analysis, a worse seizure outcome was seen in patients with a history of secondary generalization. This effect has been shown previously, especially after temporal lobe surgeries.^25,26 A hypothesized reason to explain this observation is the presence of an “evolved” epileptogenic network that facilitates generalization, which even after the resection of a critical node has a lower threshold for recurrent epileptogenesis.²⁷

The predictive value of sex in our analysis, with male patients having more recurrence, is consistent with published inherent sex-based differences in the underlying epileptogenic substrate. Similar differences have previously been reported by Bianchin et al.²⁸ Male patients with mesial temporal lobe epilepsy (MTLE) were reported to have more secondarily generalized tonic-clonic seizures and suffer more seizure-induced damage than females.^29,30 A recent study of a two-hit rat model of epileptogenesis (a freeze lesion-induced cortical malformation at postnatal day 1 followed by a prolonged hyperthermic seizure at day 10) found that after both insults females did not develop MTLE, whereas all males did,³¹ supporting sex-specific epileptogenic tendencies.

4.4 |. Possible reasons for failure of first surgery

In trying to understand the reasons for surgical failure in the patients who underwent reoperations, we divided these patients into three groups on the basis of their postoperative evaluation: patients who may have failed because their epileptogenic network was mislocalized, those who had an incomplete resection either inadvertently or because the resection was limited by eloquent cortex, and thirdly patients for whom no clear explanation for failure was available. In comparing their time to recurrence after first surgery the “unexplainable” group took the longest time to recur.

Vaugier et al have hypothesized complex epileptogenic networks and abnormal connectivity as culprits for surgical failures in mislocalizations.²³ Incomplete resections may occur due to inability to visualize the extent of the lesion, making appropriate tools to identify the area of resection a dire necessity. However, the third group of patients continues to remain poorly understood, with no obvious explanation for recurrence. Their epilepsy continues to be in the same lobe or contiguous despite their prior lesion being completely resected to the best of our knowledge. We hypothesize that these may be the individuals who have a unique genetic epileptogenic substrate that is capable of producing/activating new foci even after a complete resection. This may be supported by the finding that they differ in their time to recurrence and need longer after a resection to rouse another focus. Jehi et al published differences in gene expression in neuroinflammatory and brain remodeling pathways in patients who have late recurrences after epilepsy surgery.⁵ Furthermore, it remains unclear whether postoperative encephalomalacia itself can beget new epileptogenic foci around the bed of the resection, which needs further review.

Additionally, the detailed table of reoperations (Table 4) highlights two trends:

1. The need for intracranial recording is seen to increase with every failed surgery. This is possibly due to the increasing complexity of the epileptogenic network and inability to completely resect it.

2. Despite there being unique challenges with each patient listed, it appears from the table that most patients have a definitive pathology after the first surgery yet do not achieve seizure freedom. With a second surgery, for the majority, the pathology is nonspecific gliosis, and this trend continues to exaggerate with subsequent surgeries. Again, this indicates a challenging epileptic network, well described in recent years,^32–35 involving not only the lesion but other accessory nodes, perilesional and distant, which have the potential to develop subsequent hyperexcitability.

4.5 |. Timing of recurrence

The subgroup analysis looked at the effect of timing of recurrence on the outcomes in patients with reoperations. The patients with a poor Engel outcome (II or more) at last follow-up had a shorter time to recurrence after their previous surgery, the majority recurring within 6 months, when compared with those with who had a good final Engel outcome. Earlier work on temporal and frontal lobe epilepsies have explained such phenomenon, where an early recurrence was thought to occur from a mislocalized or incompletely resected epileptogenic zone, whereas a late recurrence was due to development of a new epileptogenic zone.^15,17,36 Our findings expand on the above concepts and urge us not only to think in terms of focus mislocalization or restricted surgical margins due to proximity of an area of eloquent cortex as the cause of early surgical failure, but to consider the possible role of an underlying epileptogenic substrate, especially in patients undergoing repeat surgery.

4.6 |. Surgical refractoriness

The correlation we observed between an increasing number of previous surgeries and worsening seizure outcomes remained significant on multivariate analysis even after controlling for all other traditional outcome predictors and surrogates of poor localization. The poor outcome predictors in the cohort with multiple surgeries were more closely correlated with inherent biological characteristics (sex and tendency for secondary generalization—the only variables that retained significance after multivariate analysis) rather than with poorly localized epilepsy. This suggests that at least some of the patients who continue to have seizures despite multiple reoperations may be “surgically refractory.” They have a widespread, “malignant” epileptogenic network with a tendency to develop new epileptogenic zones, on resection of the prior zone, causing a progressive disease state⁸ that may be “surgically refractory.” It may reasonably be argued that such patients are inappropriate surgical candidates at the time of first surgery. However, the concept of a poor surgical candidate is still nebulous, and considering the limitations of our cohort size, future research may help in better delineation of such patients while refining the identification of ideal candidates.

5 |. CONCLUSION

In this most challenging group of focal epilepsy patients, it is possible to achieve long-term seizure control with reoperaions; however, a reoperation should be offered in carefully selected patients. This cohort showed that the chance of seizure freedom drops with each subsequent reoperation, invoking a concept of “surgical refractoriness of the epileptogenic zone.” This further stresses the need for a prudent selection of candidates for reoperations. Significant predictors of recurrence were male sex, secondary generalization, a higher preoperative monthly seizure frequency, an epileptogenic focus in the dominant hemisphere, need for an invasive evaluation, and early recurrence after previous surgery. These characteristics when present should be handled cautiously in the selection process.

Key Points.

• This study is unique in analyzing the long-term surgical outcomes after consecutive epilepsy reoperations

• Repeat surgery correlates with diminishing odds of seizure freedom, irrespective of other clinical characteristics

• Seizure outcomes are predicted by more inherent and innate (gender, history of convulsions) characteristics rather than being surgery-specific

• We introduce a concept of “surgical refractoriness” as a probable conclusion of the diminishing outcomes after repeat surgeries