Outcomes of resections that spare vs remove an MRI-normal hippocampus

Abstract

Objective:

To characterize seizure and cognitive outcomes of sparing vs removing an magnetic resonance imaging (MRI)–normal hippocampus in patients with temporal lobe epilepsy.

Methods:

In this retrospective cohort study, we reviewed clinical, imaging, surgical, and histopathological data on 152 individuals with temporal lobe epilepsy and non-lesional hippocampi categorized into hippocampus-spared (n = 74) or hippocampus-resected (n = 78). Extra-hippocampal lesions were allowed. Pre- and postoperative cognitive data were available on 86 patients. Predictors of seizure and cognitive outcomes were identified using Cox-proportional hazard modeling followed by treatment-specific model reduction according to Akaike information criterion, and built into an online risk calculator.

Results:

Seizures recurred in 40% within one postoperative year, and in 63% within six postoperative years. Male gender (P = .03), longer epilepsy duration (P < .01), normal MRI (P = .04), invasive evaluation (P = .02), and acute postoperative seizures (P < .01) were associated with a higher risk of recurrence. We found no significant difference in postoperative seizure freedom rates at 5 years between those whose hippocampus was spared and those whose hippocampus was resected (P = .17). Seizure outcome models built with pre- and postoperative data had bootstrap validated concordance indices of 0.65 and 0.72. The dominant hippocampus-spared group had lower rates of decline in verbal memory (39% vs 70%; P = .03) and naming (41% vs 79%; P = .01) compared to the hippocampus-resected group. Partial hippocampus sparing had the same risk of verbal memory decline as for complete removal.

Significance:

Sparing or removing an MRI-normal hippocampus yielded similar long-term seizure outcome. A more conservative approach, sparing the hippocampus, only partially shields patients from postoperative cognitive deficits. Risk calculators are provided to facilitate clinical counseling.

1 ∣. INTRODUCTION

Data on the surgical management of temporal lobe epilepsies (TLEs) in patients with a magnetic resonance imaging (MRI)–normal appearing hippocampus are inadequate. The prevailing “feeling” is that sparing a nonlesional hippocampus should prevent postoperative memory decline but will likely compromise the odds of seizure-freedom. Some studies suggest that overall seizure outcome after TLE surgery is the same regardless of surgical strategy,^1-3 whereas others focusing on specific etiologies demonstrate higher rates of seizure-freedom whenever the mesial structures are resected.^4,5

As to the question of cognitive outcomes, hippocampus resection is associated with neuropsychological (NP) decline, particularly when the hippocampus is nonlesional.^6-11 Resection of an MRI-normal hippocampus is a more difficult proposition than removal of a sclerosed hippocampus. There is limited information as to how well partial resections, with varying degrees of hippocampus sparing, achieve their stated goal of sparing memory decline while maintaining seizure-freedom.

Overall, the existing literature is limited by small sample sizes, conflicting results, and isolated outcome measurements. This is particularly vexing given that quality of life declines most markedly in surgical patients who fail to achieve seizure freedom AND sustain a decline in memory.^8,12

To better inform surgical decision-making and advance current knowledge, we reviewed both seizure and NP outcomes of TL surgery in patients with an MRI-normal hippocampus. In addition, we explored the potential benefits of hippocampal sparing compared to standard TL resection as well as resection of each temporal anatomical sub-region on outcomes. To individualize our results, we developed models to predict seizure outcome.

2 ∣. METHODS

2.1 ∣. Patient selection and clinical data

In this retrospective cohort study, we identified all patients who had TL surgery for epilepsy at the Cleveland Clinic (Ohio, USA) from 2010 to 2018 (total of 418 patients). We then excluded patients with pathological features of hippocampal sclerosis (HS) (n = 154) and patients with MRI signs of HS not confirmed by pathology (n = 14), as reviewed by two epilepsy-imaging specialists (SJ and FC) and a neuropathologist specialized in epilepsy (IB). Lesions outside of the hippocampus (Table S2) or the presence of hippocampal gliosis on pathology were allowed, as our goal was to evaluate the consequences of removing vs sparing a normal-appearing hippocampus on MRI.

We excluded patients with prior neurosurgeries, other hippocampal lesions, postoperative follow-up of less than 6 months, postoperative events of unclear nature, and patients younger than 16 years old. Postoperative events of unclear nature were defined as cases for which we were not able to clearly classify the postoperative seizure outcome based on video electroencephalography (video-EEG) evaluation and seizure semiology. In some cases, psychogenic seizures were recorded, and in others, no events were captured on video-EEG but there was a strong clinical suspicion of psychogenic seizures. The 12 cases excluded had the hippocampus resected. We were left with 152 patients without MRI signs of HS who fulfilled all study criteria (Figure 1).

FIGURE 1.

Flow chart describing the selection process and seizure outcome according to preoperative evaluation. Other: no invasive evaluation with normal magnetic resonance imaging (MRI). Type of surgery decided based on concordant data and presence of comorbidities (high risk for surgical complications)

All patients underwent a comprehensive pre-surgical assessment including clinical history, MRI, video-EEG, magnetoencephalography, and nuclear imaging with positron emission tomography (PET) and/or single-photon emission computed tomography (SPECT) when indicated. Demographic and clinical data were collected from electronic medical records. The preoperative high-resolution MRI was acquired using 1.5T (n = 50) or 3T (n = 102) scanners. The epilepsy protocol included thin-section coronal T1-weighted and T2-weighted fluid-attenuated inversion recovery (FLAIR) images, in addition to a volumetric T1-weighted sequence, two dimensional (2D) sagittal T1-weighted, and axial diffusion. An MRI-normal hippocampus was defined as a hippocampus with no MRI signs of HS defined by visual analysis, specifically assessing signs of volume loss, increased T2/FLAIR signal, and loss of internal architecture. Relaxometry was not performed in this study. Hippocampal volumetric analysis was performed using Neuroquant, an automated volumetric software. We considered the presence of volumetric signs of hippocampal atrophy whenever the hippocampal volume was less than the fifth percentile compared to a gender- and age-matched control group. Patients were classified according to the type of resection (spared-hippocampus [n = 74] or resected-hippocampus [n = 78]) (Figure 1). We classified the pathology and the preoperative extra-hippocampal MRI findings as normal or abnormal. We also reviewed the hippocampal pathology from the hippocampus-resected group.

2.2 ∣. Seizure outcomes

The primary outcome was time to seizure recurrence (eg, months to first postoperative seizure). Acute seizures were defined as seizures occurring in the first month after surgery and were not considered as epilepsy recurrence unless they persisted beyond the acute postoperative phase. Patients classified as Engel Ib (only auras after surgery) at last follow-up (n = 22, spared [n = 8] and resected-hippocampus [n = 14]) were included in the seizure-free group.

2.3 ∣. Cognitive outcomes

Only a subset of patients in the overall sample completed pre- and postoperative NP assessments (n = 86). We compared patients who completed an NP assessment to those who did not (n = 66) with regard to their baseline characteristics to ensure that this patient subset was representative of the larger cohort. Patients who did not complete both pre- and postoperative NP testing were comparable on all but two baseline characteristics: Patients with available NP testing had higher rates of abnormal MRI (68.6% vs 48.5%; P = .02) and surgery on the dominant side (65.1% vs 48.5%; P = .05). We stratified patients according to surgical side while studying cognitive outcomes.

NP outcomes were assessed on measures of delayed verbal memory (Logical Memory II subtest from the Wechsler Memory Scale–Third or Fourth Edition, Rey Auditory Verbal Learning Test–Delayed Free Recall), delayed visual memory (Faces II and Family Pictures II subtests from the Wechsler Memory Scale–Third Edition or Visual Reproduction II subtest from the Wechsler Memory Scale–Fourth Edition), and confrontation naming (Boston Naming Test).^13-16 Preoperative test scores on these measures were subtracted from postoperative test scores, and change scores were classified as “declined” or “not declined using epilepsy-specific Reliable Change Indices.^17,18 For measures without available Reliable Changes Indices (ie, Visual Reproduction), a score reduction of one standard deviation or more compared to healthy control norms was considered a decline. For NP outcome analyses, patients who declined on at least one cognitive measure/subtest within a domain were classified as “declined” for that domain. Similar methods were used for analyses to examine postoperative cognitive improvement.

Given the material specificity of cognitive measures (ie, language and verbal memory functions more strongly associated with dominant temporal lobe function and visual memory function more strongly associated with nondominant temporal function), we stratified all analyses by side of surgery (dominant [N = 56] vs nondominant [N = 30]). To make this stratification, we first relied on results of language lateralization procedures, when available (22/86 [25.6%] had only language functional MRI, 4/86 [4.7%] had only Wada testing, and 7/86 [8.1%] had both). When language was bilaterally represented, the predominant side was considered the dominant side (n = 4). In the absence of language testing, the left side was considered dominant 53/86 (61.6%). All but two of the individuals without language testing were left-handed, and both of these individuals underwent left-sided resections. The analysis of cognitive outcomes was repeated without these two individuals to rule out the influence of potential atypical language dominance in these left-handed individuals.

We also classified the side of surgery of patients without NP testing using the same criteria to better characterize the full cohort.

2.4 ∣. Association of outcomes with anatomical temporal sub-regions

Postoperative MRIs (n = 145) were reviewed with neurosurgery supervision (JB), blinded to outcomes, to define the extent of resection. Eight different areas within the TL were individually analyzed, and the resection of each area was classified as complete, partial, or no resection. The sub-regions studied were amygdala, fusiform, hippocampus, inferior temporal, middle temporal, parahippocampus, superior temporal piriform cortex, and temporal pole.

We analyzed the association between resection of each sub-region and recurrent epilepsy, verbal memory, and naming decline for dominant surgeries and visual memory decline for nondominant surgeries.

2.5 ∣. “Postoperative double loss”

Individuals were classified as presenting a double loss if they experienced clinically meaningful decline in verbal or visual memory and were not seizure-free after surgery. Naming decline was not included in this analysis.

2.6 ∣. Statistical methods

2.6.1 ∣. Demographics and clinical data

To describe the cohort, we used the median (interquartile range) for numeric variables, and counts (%) for categorical data. Kruskal-Wallis and Fisher exact tests were used to test for univariate associations of numeric and categorical variables with the treatment, respectively.

2.6.2 ∣. Analysis of seizure outcomes

Overall survival

We obtained Kaplan-Meier estimates of longitudinal seizure-recurrence rates.

Univariate analysis

of potential seizure outcome predictors: We compared baseline characteristics for seizure-free patients vs those who had a recurrent seizure and obtained univariate hazard ratios and P-values from a Cox-proportional hazard model for each baseline factor. Additional survival curves were created for MRI, acute postoperative seizures, and invasive evaluation.

Multivariate modeling

Cox proportional-hazards models were built stratified by surgery type (ie, hippocampus-spared vs resected). The following potential seizure outcome predictors were considered: age at surgery, duration of epilepsy, history of generalized tonic-clonic seizures, MRI, side of surgery, and gender. Pathology and acute seizures were included for the development of the postoperative model. A backward selection process was then used to reduce each treatment-specific model according to Akaike information criterion (AIC). AIC quantifies the tradeoff between how well the predictors explain the outcome and model complexity.¹⁹ Starting with the full model, variables were removed one at a time until the AIC no longer decreased. To obtain the final models (ie, pre/postoperative), a single, combined model was constructed where common variables retained in the stratified models (with similar effect direction) were entered as independent predictors while uncommon predictors were entered as interactions with the treatment variable. A second backward selection process was then implemented by sequentially removing interaction terms (and keeping main effects) until the AIC no longer decreased.

Online risk calculator

Rather than producing multiple separate risk calculators (one for seizure outcomes with preoperative variables alone, one with additional postop variables, plus separate ones for cognitive outcomes), we built one simple, clinical tool combining both models of seizure outcome prediction (preoperative and postoperative models) with one of our previously published naming outcome prediction nomograms.⁶ We generated a unified online risk calculator simultaneously presenting the estimated individualized chance for any given patient for seizure-freedom at 1 year after surgery AND risk of naming decline with either treatment option: spared-hippocampus vs resected-hippocampus.

2.6.3 ∣. Analysis of cognitive outcomes

Fisher exact test and Wilcoxon rank sum test were conducted to examine differences in cognitive outcome between patients with spared or resected hippocampus. Some patients did not complete cognitive measures in all three domains of interest in this study, and measures not completed were treated as missing data (naming 7/86, verbal memory 1/86, visual memory 11/86).

2.6.4 ∣. Association of outcomes with anatomical temporal sub-regions

Log-rank and Fisher exact tests were performed to evaluate differences in risk of seizure recurrence and cognitive outcomes for each resection region(s), respectively.

2.6.5 ∣. “Postoperative double loss”

Univariate analysis

We compared baseline characteristics for patients classified as presenting “double-loss” (not seizure free and with memory decline) vs those who were not. Fisher exact test and Wilcoxon rank sum test were used to assess associations.

2.7 ∣. Classification of evidence

This study provides class IV evidence that sparing or removing an MRI-normal hippocampus yielded similar long-term seizure outcome, and that sparing the hippocampus only partially shields patients from postoperative cognitive decline.

3 ∣. RESULTS

3.1 ∣. Demographics and clinical data

The median follow-up time in our cohort was 29.5 months (25th/75th, 17.25/54.5). Table 1 displays summary statistics for all 152 patients, and Table 2 displays these statistics for the subset of 86 patients who had available NP data after surgery. Table S2 describes seizure outcome according to type of resection, preoperative, and postoperative evaluation results. No difference was found between these groups. There was no difference in the proportion of patients having invasive EEG (iEEG) between the spared-hippocampus group (50%) and the resected-hippocampus group (47%). Figure 1 provides further information on seizure outcome and preoperative evaluation results.

TABLE 1.

Summary statistics by type of surgery for the full cohort

Variable	Level	Spared (N = 74)	Resected (N = 78)	P-value
Age at surgery (years)		36 (25, 44.75)	36.5 (26, 50.25)	.59
Sex	Female	37 (50%)	44 (56.4%)	.516
	Male	37 (50%)	34 (43.6%)
Duration of epilepsy (years)*		11.5 (4, 17.19)	11 (6, 20.75)	.274
Age at epilepsy onset (years)*		20.5 (12, 44.5)	20.5 (12, 45.1)	.852
GTCS	No	13 (17.6%)	10 (13%)	.501
	Yes	61 (82.4%)	67 (87%)
MRI	Normal	26 (35.1%)	35 (44.9%)	.249
	Abnormal	48 (64.9%)	43 (55.1%)
Pathology	Normal	28 (37.8%)	49 (62.8%)	.003
	Abnormal	46 (62.2%)	29 (37.2%)
Invasive	No	37 (50%)	41 (52.6%)	.871
	Yes	37 (50%)	37 (47.4%)
Surgery side	Dominant	53 (71.6%)	35 (44.9%)	.001
	Nondominant	21 (28.4%)	43 (55.1%)
Acute postoperative seizures	No	58 (78.4%)	66 (84.6%)	.403
	Yes	16 (21.6%)	12 (15.4%)
Distance from the lesion	Close to HC	28 (37.84%)	38 (48.72%)	.003
	Distant from	20 (27.03%)	5 (6.41%)
	No lesion	26 (42.68%)	35 (44.87%)

Abbreviations: GTCS, generalized tonic-clonic seizures; HC, hippocampus; MRI, magnetic resonance imaging of the brain outside the hippocampus.

^*Median scores (interquartile range), distance to hippocampus (HC) was defined by estimating on MRI the smallest distance between the lesion and HC. If distance was below 10 mm, we classified the lesion as being close to the HC.

TABLE 2.

Summary statistics by type and side of surgery for patients with available neuropsychological data

		Dominant (N = 56)			Nondominant (N = 30)
Variable	Level	Spared (n = 29)	Resected (N = 27)	P-value	Spared (N = 7)	Resected (N = 23)	P-value
Age at epilepsy onset (years)		19 (9.5-37.5)	20 (10-29)	.702	35 (33-46)	21 (13-38)	.004
Age at surgery (years)		38 (25, 44)	32 (25, 48)	.724	44 (36.5, 51.5)	31 (26, 49)	.122
Sex	Female	11 (37.9%)	15 (55.6%)	.284	3 (42.9%)	18 (78.3%)	.153
	Male	18 (62.1%)	12 (44.4%)		4 (57.1%)	5 (21.7%)
Duration of epilepsy (years)*		12 (4, 19)	11 (5, 17.91)	.876	5 (2.5, 11.5)	10 (6.5, 12.5)	.154
GTCS	No	7 (24.1%)	3 (11.5%)	.303	1 (14.3%)	5 (21.7%)	1
	Yes	22 (75.9%)	23 (88.5%)		6 (85.7%)	18 (78.3%)
MRI	Normal	11 (37.9%)	10 (37%)	1	0 (0%)	6 (26.1%)	.29
	Abnormal	18 (62.1%)	17 (63%)		7 (100%)	17 (73.9%)
Pathology	Normal	16 (55.2%)	18 (66.7%)	.423	0 (0%)	13 (56.5%)	.01
	Abnormal	13 (44.8%)	9 (33.3%)		7 (100%)	10 (43.5%)
Invasive	No	12 (41.4%)	12 (44.4%)	1	6 (85.7%)	16 (69.6%)	.638
	Yes	17 (58.6%)	15 (55.6%)		1 (14.3%)	7 (30.4%)
BNT baseline*		49 (46.75, 52)	47.5 (44.75, 52.25)	.243	54 (53.5, 56.5)	53.5 (48.25, 56.75)	.474
Delayed Verbal Memory Baseline*		94 (86, 98)	78 (71, 88)	<.001	104 (93.5, 106)	100 (86.5, 108)	.768
Delayed Visual Memory Baseline*		94 (82.5, 98.5)	84 (78, 91)	.012	97 (94, 109)	94 (87.5, 103)	.279
Acute postoperative seizures	No	19 (65.5%)	25 (92.6%)	.021	7 (100%)	22 (95.7%)	1
	Yes	10 (34.5%)	2 (7.4%)		0 (0%)	1 (4.4%)
Naming decline		11/27 (40.7%)	19/24 (79.2%)	<.01	1/7 (14.3%)	1/21 (4.8%)	.44
Verbal memory decline		11/28 (39.3%)	19/27 (70.4%)	.03	0/7 (0%)	3/23 (13%)	1
Visual memory decline		4/22 (18.2%)	7/24 (29.2%)	.49	3/6 (50%)	13/23 (56.5%)	1

Abbreviations: BNT, Boston Naming Test; GTCS, generalized tonic-clonic seizures; MRI, magnetic resonance imaging of the brain outside the hippocampus.

^*Median scores (interquartile range).

We reviewed the hippocampal pathologies of patients whose hippocampus was resected. Because the presence of electrodes in those who had iEEG implantations can lead to pathological changes, we excluded the ones who underwent invasive evaluation (n = 37). The presence of gliosis was associated with higher rates of seizure freedom (24/29 [82.8%] vs 4/12 (33.3%), P = .003).

3.2 ∣. Predictors of seizure freedom

Approximately 61% (confidence interval [CI] 53.3%-69.4%) of patients were seizure-free in the first year, 53% (CI 45.5%-62.3%) in the second year, and 45% (CI 36.1%-56%) within 5 years of surgery. The occurrence of auras in the initial 6 months following surgery did not influence the rates of recurrence (P = 1, Table S2).

3.2.1 ∣. Univariate analysis

We compared baseline characteristics for patients who remained seizure-free vs those who had a recurrent seizure. At a univariate level, these variables were associated with a higher risk of seizure recurrence: male (hazard ratio [HR] = 1.66, P = .03), longer epilepsy duration (HR = 1.03, P < .01), normal MRI (HR = 0.62, P = .04), history of invasive evaluation (HR = 1.77, P = .02), and acute postoperative seizures (HR = 5.75, P < .01). The estimated chance of seizure-freedom at 2 postoperative years was 61.3% (95% CI 50.9-73.8) in the hippocampus-resected group vs 44.4% (95% CI 33.7%-58.5%) in the hippocampus-spared group, but the overall effect of surgery type was not significant (P = .17), as the two curves converged by last follow-up.

Figure 2 shows survival curves stratified individually by treatment group (hippocampus-resected vs spared), presence of acute postoperative seizures, performing an iEEG evaluation, and extra-hippocampal MRI findings. We show the sample size still at risk for each time point.

FIGURE 2.

Survival curves for seizure freedom. Survival curves for seizure freedom separately stratified for presence of acute postoperative seizures (A) A, treatment group (hippocampus: resected vs spared) (B), magnetic resonance imaging (MRI) findings outside of the hippocampus (C), and preforming an invasive electroencephalography (EEG) evaluation (D). Number of patients with available follow-up was 127 at 1 y, 103 at 2 y, and 34 at 5 y

3.2.2 ∣. Multivariate analysis and nomogram development

The final preoperative model contained interaction terms that included surgery side, age at surgery, and epilepsy duration, each interacting with treatment and MRI as an independent factor (Table 3). Note that the HRs for numeric variables (epilepsy duration and age) in Table 3 compare the 75th and 25th percentiles from the full cohort. The bootstrap-validated concordance index (1000 iterations) was 0.65.

TABLE 3.

Hazard ratios and P-values for each characteristic in the final pre- and postoperative models

Preoperative model
	Spared		Resected
Variable	Hazard ratios (95%)	P-value	Hazard ratios (95%)	P-value
MRI (Abnormal vs normal)	0.55 (0.34, 0.89)	.016	0.55 (0.34, 0.89)	.016
Surgery side (Nondominant vs Dominant)	1.59 (0.75, 3.37)	.224	0.49 (0.25, 0.98)	.045
Age (47 vs 26 y)	0.55 (0.31, 0.98)	.043	1.25 (0.76, 2.06)	.38
Epilepsy duration (19 vs 5 y)	2.74 (1.83, 4.11)	<.001	1.08 (0.71, 1.66)	.711
Postoperative model
	Spared		Resected
Contrast	Hazard ratios (95%)	P-value	Hazard ratios (95%)	P-value
Acute seizures (Yes vs No)	5.66 (3.13, 10.23)	<.001	5.66 (3.13, 10.23)	<.001
MRI (Abnormal vs normal)	0.47 (0.28, 0.8)	.005	0.47 (0.28, 0.8)	.005
Surgery side (Nondominant vs dominant)	2.23 (1.05, 4.72)	.036	0.45 (0.22, 0.92)	.028
Pathology (Abnormal vs normal)	0.45 (0.22, 0.91)	.028	1.87 (0.88, 3.95)	.102
Age (47 vs 26 y)	0.67 (0.4, 1.12)	.126	1.25 (0.77, 2.04)	.372
Epilepsy duration (19 vs 5 y)	2 (1.31, 3.05)	.001	1.22 (0.77, 1.93)	.389
Gender (Male vs female)	1.86 (1.12, 3.08)	.016	1.86 (1.12, 3.08)	.016

Note: Hazard ratios, 95% confidence intervals, and P-values (testing against the null effect) for each characteristic in the final pre- and postoperative models by treatment group estimating the risk of seizure recurrence. Note that numeric variables compare the observed 75th percentile to the observed 25th percentile. Bold values are statistically significant.

The final postoperative model contained surgery side, pathology, age at surgery, and duration of epilepsy each interacted with treatment, acute seizures, MRI, and gender as independent factors (Table 3). The bootstrap-validated concordance index (1000 iterations) was 0.72.

3.2.3 ∣. Online risk calculator

The model is available at http://riskcalc.org:3838/SparingHippocampus/. Figure 3 shows an example demonstrating the use and the output from the unified online risk calculator.

FIGURE 3.

Example of how to use the online risk calculator to predict seizure and naming outcome. Case: A 20-year-old, right-handed man with drug-resistant temporal lobe epilepsy is undergoing epilepsy surgical evaluation. He has had seizures since the age of 10 y and completed 13 y of education. The pre-surgical assessment suggested his seizures were coming from the left temporal region. Magnetic resonance imaging (MRI) showed a cavernoma in the left anterior middle temporal gyrus with a normal-appearing hippocampus (abnormal MRI). Wada test confirmed that the left hemisphere was dominant for language. Two treatment options were offered: a temporal lobe resection sparing the hippocampus or a complete resection including the hippocampus. The information was added to the risk calculator (Step 1). The chances of seizure freedom after 1 y were calculated for both procedures along with the risk of clinically meaningful naming decline (Steps 2 and 3). Right after surgery, the patient had an acute seizure and pathology confirmed the presence of a cavernoma (abnormal pathology). The risk calculator was used to recalculate the chance of seizure-freedom adding the postoperative information (Steps 2 and 3)

3.3 ∣. Predictors of cognitive outcomes

On the dominant side, those whose hippocampus was spared had lower rates of clinically meaningful postoperative declines in verbal memory (39% vs 70%; P = .03) and confrontation naming (41% vs 79%; P < .01; Table 2) compared to those whose hippocampus was resected. Improvements in memory and language were relatively rare among patients with dominant-sided surgeries. Five patients (10%) showed postoperative improvements in verbal memory, all of whom had their hippocampus spared. No patients with dominant-sided resections (with or without removal of the hippocampus) showed clinically significant improvements in naming. Ten patients (22%) with dominant resections showed postoperative improvements in visual memory (n = 4 resected, n = 6 spared).

On the nondominant side, the rates of postoperative declines in verbal memory (0% vs 13%; P = 1), visual memory (50% vs 57%; P = 1), and confrontation naming (14% vs 5%; P = .44) were not significantly different between hippocampus-spared and hippocampus-resected groups. Only three patients (11%) who underwent nondominant resections showed postoperative improvements in visual memory (n = 2 resected, n = 1 spared). Eight patients (29%) who underwent nondominant resections showed improvements in verbal memory (n = 6 resected, n = 2 spared), and only one showed an improvement in confrontation naming.

3.4 ∣. Association of outcomes with anatomical temporal sub-regions

Of the 152 study patients, 145 had postoperative MRI studies available for review. Table S1 displays the counts and percentages of patients whose seizures recurred and who had NP decline by the resection status of each anatomical temporal sub-region. Further subgroup analyses stratifying by side of surgery were not possible given the sample size.

3.5 ∣. “Postoperative double loss”

When we compared baseline characteristics for patients who were classified as presenting a “double-loss” vs those who were not, abnormal MRI status (P = .02) and nondominant side surgery (P = .04) were associated with lower risk of “double loss.” The odds of presenting a “double loss” for patients with surgery on the nondominant side are ~70% less (95% CI 2.7%-91.1%) than those with surgery on the dominant side and for patients with an abnormal MRI the odds are ~72% less (95% CI 2.3%-89.9%) than those with a normal MRI. Among the 54 patients with dominant resections, patients with hippocampus resection had similar odds of presenting a “double loss” as those without hippocampus resection (odds ratio [OR] 0.85; 95% CI 0.28-2.6). The same was true for the 20 patients with both dominant resections and normal MRI (OR 0.67; 95% CI 0.11-3.91).

4 ∣. DISCUSSION

4.1 ∣. Seizure outcome

Our seizure outcomes are in agreement with published literature on surgical nonlesional TLE.^20,21 In a recent meta-analysis, 51% of patients with TLE were seizure-free when MRI was nonlesional vs 75% in lesional cases.²² However, beyond such general statements, current understanding of long-term seizure-freedom in relation to various degrees of sparing a normal-appearing hippocampus remains limited. Our article advances the discussion in a number of important ways:

1. Sparing hippocampus and seizure-freedom:

Some studies suggest that resecting the hippocampus is unnecessary to achieve seizure control^1,3,23,24; however, these studies had small sample sizes and/or used control cohorts with HS. Conversely, studies of patients with TL cavernomas⁵ or tumors⁴ found better seizure outcomes with hippocampus resection compared to lesionectomy alone. In our study, the risk of recurrence was similar when the hippocampus was spared vs resected and evaluated at 5 postoperative years. However, our survival curve suggests a divergence in early seizure recurrence between the two groups. Future studies are needed to confirm this finding. Our prior work hypothesizes that early surgical failures are driven by incomplete resections of the current epileptogenic zone (more likely if an epileptic, albeit normal-appearing, hippocampus is left behind), although later outcomes are likely driven by genetic and other inherent biological characteristics (unaffected by whether a hippocampus is removed or spared).^25-27

2. Drivers of seizure outcomes when hippocampus is spared vs not:

In this study, multiple variables correlated with seizure outcomes on a univariate level (Table 3), consistent with available literature.²⁸ At the multivariate level, the final preoperative model included surgery side, MRI, age at surgery, and epilepsy duration. The postoperative model added acute seizures and pathology. Overall, acute postoperative seizures and long epilepsy duration were associated with the poorest seizure outcomes, whereas a defined extra-temporal lesion on MRI correlated with better seizure outcomes, in concordance with published literature.^22,28,29 Mechanistically, acute postoperative seizures likely suggest that the epileptogenic zone has been untouched, long epilepsy duration may increase the risk of hippocampal epileptogenicity, and the lack of a lesion in the TL complicates localization. Male gender was also correlated with a higher risk of seizure recurrence.^30,31 Surgery side appears to have opposite effects in the treatment groups. Worse seizure outcomes were observed following surgery on the dominant hippocampus-resected group, and on the nondominant hippocampus-spared group. This may be attributed to cohort differences in the use of iEEG between the dominant and nondominant resections. More patients on the nondominant hippocampus-spared group had extra-hippocampal MRI lesions (17/21 vs 22/43; P = .03) and underwent a resection without iEEG, so seizures may have recurred due to a false sense of security with a lesionectomy, leaving behind an epileptic hippocampus. Conversely, more patients had iEEG done to explore a normal-appearing dominant hippocampus (51/88 had iEEG on dominant hemisphere vs 23/64 of nondominant patients; P = .008). Yet, an equal proportion of dominant hippocampi was spared whether iEEG was performed or not (63% spared with iEEG and 57% spared without iEEG; P = .66), suggesting that iEEG was attempted to rationally guide the resection, but in reality was a marker of a challenging and poorly localized epilepsy. Without a clearer understanding of EEG patterns observed in hippocampal electrodes at ictal onset, it is difficult to adequately differentiate true hippocampal ictal onset from spread patterns to the hippocampus. Recent data from our group³² highlight the potential significance of some stereo-EEG ictal patterns, emphasizing the need to study the role of iEEG in guiding surgical decision-making. Finally, our postoperative seizure recurrences may be due in part to sparing an actually “sclerosed” hippocampus missed on imaging. This is unlikely: Of the 83 MRI-normal hippocampus patients whose hippocampus was resected, only 5 (6%, excluded from this analysis) had HS on histopathology. This approximates a similarly low proportion in the hippocampus-spared group.

3. Individualizing outcome prediction:

To analyze the combined role of all predictive variables, and translate our findings to clinical care, we developed two nomograms (Figure 2): one to be used preoperatively (concordance-statistic 0.65) and one to be used postoperatively (concordance-statistic 0.72). Nomograms provide an individualized prediction of seizure recurrence with potential clinical utility33.

4. Gliosis on an MRI-normal hippocampus:

The presence of hippocampal gliosis was associated with higher rates of seizure freedom. Potential explanations for this finding are either indication of early stage HS or an indirect marker of the proximity of the hippocampus to the epileptogenic zone.

4.2 ∣. Cognitive outcome

There were 86 patients with available NP data. Not surprisingly, given the high potential for postoperative cognitive morbidity, patients with nonlesional epilepsy in the language-dominant hemisphere were more likely to have completed pre- and postoperative NP testing compared to the remaining cohort. As previously reported, the dominant hippocampus-spared group had significantly lower rates of decline in naming and verbal memory compared to the dominant hippocampus-resected group. Published rates of memory decline after a left (dominant) anterior temporal resection ranged from 22% to 75%,^34,35 with an estimated 44% risk of verbal memory decline and 34% of naming decline.³¹

Although cognitive outcomes were clearly better among patients whose dominant hippocampus was spared, it is essential to highlight that even when the dominant hippocampus was spared, the rates of domain-specific memory and naming decline were rather substantial (39%–41%). By disrupting the functional brain memory network, deficits can occur even when the hippocampus is spared.³⁶ Verbal memory decline was reported following dominant-sided hippocampus sparing resections in the TL¹. Thus sparing the hippocampus does not guarantee an absence of postoperative cognitive morbidity, and patients should be counseled accordingly.

Only seven patients in the nondominant spared-hippocampus group had NP data, limiting conclusions in this group.

4.3 ∣. Association of outcomes with anatomical temporal sub-regions

Sparing the amygdala, middle temporal, or temporal pole was associated with worse seizure outcomes in our study. A multicenter study of laser ablation similarly found that more extensive amygdalar ablations were associated with Engel class I outcomes.³⁷

A similar analysis classifying the resection of the different temporal structures as total, partial, or spared found resection of the left parahippocampus led to a greater decline in verbal learning.¹ We expand on these results showing higher rates of verbal decline when the dominant hippocampus, parahippocampus, and fusiform gyrus were resected.

Rates of naming decline on the dominant side were overall lower in the “spared” sub-region group. Laser ablation studies also offer a better memory outcome than open resection,³⁶ with naming ability less likely to be impacted.³⁸ These data are limited, however, by short-term follow-up. In our series, the rate of verbal memory decline was similar regardless of the extent of hippocampus resected, suggesting limited value of attempting to “spare memory” by partial hippocampus resection.

4.4 ∣. Postoperative double loss

Individuals with persistent postoperative seizures accompanied by memory decline (double loss) were those with normal MRI who underwent dominant-sided resections, supporting the prevailing clinical impression.

4.5 ∣. Study limitations

This was a retrospective study, and baseline characteristics of spared-hippocampus and resected-hippocampus groups were not similar. For example, the number of patients with lesions distant from the hippocampus was higher in the hippocampus-spared group, potentially leading to a selection bias. We also constructed a predictive model pre-selecting variables more likely to influence the surgical outcome, and so acknowledge that other variables may also be important in defining the surgical outcome. We compensated whenever possible for this heterogeneous cohort by adjusting our analyses: We formally adjusted for the seizure outcomes with the multivariable analysis, and stratified the smaller subgroup with cognitive outcome analyses by surgery side.

When analyzing the survival curve comparing spared vs resected-hippocampus groups, the high rate of early seizure recurrences (57 patients recurred by 1 postoperative year) led to a decreasing number of patients at risk beyond 5 years after surgery, which may have contributed to the convergence of the seizure outcome curves beyond that time point. Because the decision to spare the hippocampus is likely significantly influenced by individual needs and expectations, patient recruitment to a prospective randomized trial would be problematic, but larger observational studies will help address some limitations of our seizure outcome analyses. A larger sample size would also help us to better analyze NP outcomes in nondominant TLE.

Not all patients had pre- and postoperative NP testing; therefore the high percentage of patients with NP decline may be influenced because patients experiencing cognitive changes are more motivated to have postoperative testing, thereby skewing our results.

The short interval between surgery and postoperative NP testing (mean 7.9 months) is another limitation. Although postoperative memory deficits improve over time in children,³⁹ postoperative cognitive outcomes in adults were stable when evaluated between 5 and 22 postoperative years.⁴⁰ Longer-term NP outcomes (5 postoperative years minimum) will be helpful to ascertain in our cohort.

A volumetric assessment of the extent of resection, beyond the qualitative assessment of resection of specific gyri done here, may have provided additional information.

4.6 ∣. Discussion: Summary

We recognize when deciding whether or not to spare the hippocampus, individual needs must be taken into account. The more conservative approach may lower, without eliminating, the risk of NP decline. The risk of recurrence over 5 years was not statistically different, but the pattern of our data suggests that future work should investigate earlier recurrence specifically.

When seizures continue after a hippocampus-sparing lesionectomy, the possibility of a more extensive resection could be considered to maximize the likelihood of seizure freedom, but should also involve careful consideration of the potential risks for cognitive decline.

Key Points.

• The risk of seizure recurrence was similar between groups with spared vs resected the hippocampus at 5 postoperative years (P = .17)

• Patients with language-dominant temporal lobe epilepsy (TLE) had lower rates of postoperative decline in naming and verbal memory with hippocampus sparing surgery.

• Thirty-nine percent of patients who underwent surgery sparing the hippocampus on the dominant side still had verbal memory decline, and 41% demonstrated naming decline.

• We developed an online risk calculator to facilitate clinical counseling on seizure and naming outcomes in patient with a normal hippocampus on magnetic resonance imaging (MRI).

DATA AVAILABILITY STATEMENT

Data not provided in the article will be available to any qualified investigator upon request.