Lateralized differences for verbal learning across trials in temporal lobe epilepsy are not affected by surgical intervention

Abstract

This research aims to broaden understanding of learning verbal material in participants with left- and right-sided mesial temporal lobe epilepsy (MTLE). We modeled word list-learning to determine how anterior temporal lobe resection affects verbal learning. Verbal learning (across trials) was assessed using the first five trials of the Rey Auditory Verbal Learning Test (RAVLT) in 128 participants with MTLE. Mixed-effects modeling was used to determine whether learning curves differed between left- and right-sided MTLE patients pre- and post- anterior temporal lobe resection. Laterality of MTLE had a significant effect on both the model intercept and the linear slope, whereby participants with left-sided MTLE retained fewer words on both the first trial and on each subsequent trial than right-sided MTLE participants; and this held regardless of anterior temporal lobe resection status (t(117)=−3.516, p<.001; t(120.50)=−2.049, p=.042, for intercept and linear slope, respectively). There were no significant differences in the learning curves after anterior temporal lobe resection surgery in either left- or right-sided MTLE. Our findings suggest that acquisition of verbal information may be especially impaired in patients with left-sided MTLE. Further, we show that verbal learning across trials was not affected by surgical intervention. This finding contributes to the broader understanding of the impacts of anterior temporal lobe resection on verbal memory function, and has important implications for the clinical management and surgical planning for patients with temporal lobe epilepsy.

1. Introduction

Mesial temporal lobe epilepsy (MTLE) is characterized by recurring seizures originating from mesial temporal structures [1,2]. While the majority of patients with MTLE respond to antiepileptic medication, approximately one-third are not able to gain adequate control of their seizures with medication alone [2,3]. In these cases, anterior temporal lobe resection (ATLR) is an established and effective treatment, leading to seizure freedom for up to 70% of patients [2,4,5]. While effective at treating medically refractory MTLE, resections are not without consequence. Meta-analyses suggest that approximately 45% of left- and 20% of right-sided ATLR patients demonstrate significant decline in verbal memory ability following resection [6,7].

“Verbal memory” refers to the ability to encode, consolidate, and retrieve verbal information [8–11]. Verbal memory underlies skills and abilities necessary for successful participation in social environments and management of daily living, such as remembering a grocery list, relaying a message to a colleague, or recalling directions to an unfamiliar location. Assessment of verbal memory in populations with epilepsy is commonly done via standardized neuropsychological tests that measure learning across trials, recognition, and recall of verbal information. In broad terms, these measures are believed to correspond roughly to memory processes of encoding (acquisition), consolidation, and retrieval, respectively [9,12–14] (Fig. 1). We would hasten to acknowledge that these memory constructs are complex and likely reflect overlapping neural circuits and psychological processes [15,16]. Nonetheless, they provide useful heuristics for understanding the impact of MTLE on verbal memory.

Figure 1.

Schematic of theoretical mental processes underlying verbal memory and the respective psychological measures used to assess them, adapted from the work of Vakil and Blachstein [9].

Following ATLR, patients with MTLE show decline in both recognition and recall of verbally-presented information, indicating a loss of consolidation and retrieval abilities [17–23]. This post-surgical decline is both more common and more severe for patients with left-sided MTLE (L-MTLE) than for patients with right-sided MTLE (R-MTLE) [6,7]. The effects of ATLR on verbal learning across trials and the underlying encoding process it is believed to reflect, however, have been less studied. While recall and recognition are important components of verbal memory assessment, to fully elucidate the impact of ATLR on verbal memory, it is necessary to consider changes in verbal learning across trials as well.

The existing literature generally suggests that prior to surgery, participants with R- and L-MTLE do not show group differences in tasks of verbal learning across trials [17,20,21,24,25], though participants do show below-expected performance when compared to a healthy sample [20,24,26]. After ATLR, group differences emerge as participants with L-MTLE demonstrate decline [17,20,24,26–28]. However, interpretation of this literature must be approached with caution, as these studies have methodological limitations. Of primary concern in the existing literature is the operationalized definition of verbal learning across trials as the sum of recalled words over all learning trials. This is a limitation because summed scores reflect the aggregate performance over all trials and do not capture the influence of previous trials through primacy, recency, and resistance to interference effects [29,30]. Additionally, there is a significant effect of working memory on the first trial that is independent of verbal memory performance [9,31]. Considering all trials in aggregate mixes together behaviorally and statistically distinct constructs, making the aggregate score hard to interpret in terms of specific memory processes. Secondly, many individual and group characteristics – such as pre-operative performance, language lateralization, age, and seizure freedom – can influence word list-learning performance, which can further obfuscate inferences [20,24,27,32]. To address both issues, a modeling approach that differentiates trials and accounts for individual and group effects is necessary to fully understand the impact of ATLR on verbal memory.

Such modeling approaches have been used by Dulay and colleagues to study verbal learning over trials in MTLE [33]. Their research found that learning curves of list-learning tasks can be accurately estimated using quadratic models, where the intercept corresponds to initial recall after one learning trial, the linear slope indicates rate of learning through subsequent trials, and the quadratic slope corresponds to the degree at which the slope is changing over trials. Their research showed that post-operatively, participants with L-MTLE have learning curves that are flatter than participants with R-MTLE. Dulay and colleagues additionally displayed figures that suggest a decline in verbal learning across trials after ATLR. However, their work did not include analyses that would indicate whether the apparent change in intercept and slope were indeed statistically different. Despite meaningful contributions to the literature, their work did not test for the effect of ATLR on learning curves, leaving this important question unanswered.

The current study was designed to address limitations and knowledge gaps in the literature on verbal memory in MTLE by utilizing a mixed-effects approach to model word list-learning curves in a large cohort of participants prior to and following ATLR. We hypothesize a main effect of pre/post-ATLR status on verbal learning across trials, such that participants with MTLE will show decline in verbal learning slope following ATLR, as well as an interaction between laterality and ATLR, with L-MTLE participants exhibiting greater post-operative decline in verbal learning slope when compared to R-MTLE participants.

2. Methods

2.1. Participants

Participant data were obtained from a database of patients who underwent ATLR as treatment for MTLE at the University of Iowa Hospitals & Clinics. Database data collection was completed with approval by the Institutional Review Board at the University of Iowa and informed consents were obtained from all participants. Data from 128 participants were available for analysis. Sixty-seven participants had left-sided ATLR (L-ATLR) and 61 had right-sided ATLR (R-ATLR). Sample characteristics are included in Table 1. Language lateralization was defined as the result of pre-operative Wada testing [34]. Participants for whom Wada data were unavailable were categorized as likely left language lateralized if they were fully right-handed and belonged to families comprised of entirely right-handed individuals. Seizure freedom was determined via retrospective chart review and defined as no reported seizures since surgery, excluding seizures directly related to missed doses of antiepileptic medication.

Table 1.

Sample characteristics of patients with medial temporal lobe epilepsy who underwent left- and right-sided temporal lobe resection surgery. Non-count values indicate sample means (and standard deviations).

	L-ATLR (n = 67)	R-ATLR (n = 61)
Gender	39 females, 28 males	28 females, 33 males
Handedness	55 right-handed, 9 left-handed, 3 mixed-handed	51 right-handed, 9 left-handed, 1 mixed-handed
Hemispheric language lateralization	50 left-lateralized, 6 right-lateralized, 5 mixed, 1 likely left-lateralized, 5 unknown	45 left-lateralized, 4 right-lateralized, 2 mixed, 7 likely left-lateralized, 3 unknown
Education (years)	13.22 (2.20)	13.55 (2.15)
Age at surgery (years)	40.16 (12.66)	34.44 (10.31)
Age of seizure onset (years)	16.92 (13.00)	18.34 (7.91)
Seizure freedom status after ATLR	43 achieved, 18 not achieved, 6 unknown	44 achieved, 13 not achieved, 4 unknown

L-ALTR = left anterior temporal lobe resection; R-ATLR = right anterior temporal lobe resection

All candidates for ATLR are clinically referred for comprehensive neuropsychological evaluation prior to surgery. Verbal memory was assessed at two epochs: approximately six months prior to surgery and one year after. Data for both timepoints were available for 113 of the 128 participants; 9 participants had pre-operative data only and 6 had post-operative data only. Analyses accounted for missing data.

2.2. Neuropsychological Measure

Verbal learning across trials was assessed using the first five trials of the Rey Auditory Verbal Learning Test (RAVLT) [35]. Briefly, the RAVLT is a word list-learning task in which participants are read a list of 15 unrelated words and asked to recall as many as possible over the span of five trials. Participants are read the entire target list before each trial. Over successive trials, the number of successfully remembered words tends to increase, indicating learning. Alternate forms [36,37] of the test were used during re-testing to minimize practice effects, which are common in tests of memory [8]. The RAVLT and its alternate forms have been shown to have high test-retest reliability [38].

Word list-learning paradigms are among the most sensitive assessments for verbal memory, as they are not bound by the associative context inherent in prose-based tasks [8]. Due to the supraspan word list length and number of trial repetitions, the RAVLT is particularly useful in its ability to measure rate of learning. Performance on the five trials generates a curve, and the slope of this curve provides a measure of verbal learning across trials [39].

2.3. Statistical Analyses

Growth curve analysis [40] was used to analyze word list-learning over five trials. A quadratic mixed-effects model was fit to participants’ learning trials of the RAVLT using the lmer function in the lme4 package [41] in R [42]. Degrees of freedom and p values for main effects and interactions were obtained from the lmerTest package [43]. Figures were produced using the package ggplot2 [44].

The following independent variables were effect coded for analysis: Laterality (Left = 1, Right = −1) and Epoch (Pre-surgery = 1, Post-surgery = −1). For purpose of analysis, trials were coded as 0–4; they are reported in this manuscript in typical 1–5 notation.

3. Results

Participants with L-MTLE and R-MTLE did not differ significantly in gender ratio, handedness, education, age at the onset of seizures, time of testing relative to surgery, or post-operative seizure freedom (all p-values >.10). A Mann-Whitney-Wilcoxon test indicated a significant difference between the two groups in their age at surgery: participants who underwent L-ATLR were on average 5.71 years older than those with R-ATLR (p=.018). For this reason, age at surgery was included as an additional parameter in the full fixed-effects model.

To model verbal learning, the initial full fixed-effects model was:

$$\text{Score}=(\text{Trial}\times \text{Laterality}\times \text{Epoch}\times \text{Age})+({\text{Trial}}^2\times \text{Laterality}\times \text{Epoch}\times \text{Age})$$

The full model was pared down by systematically removing fixed effects and using model comparisons to arrive at the final maximally-supported model. Fixed effect terms were removed based on complexity (i.e., multi-way interactions removed before two-way interactions) and empirical interest (i.e., age and laterality terms removed before slope terms). The maximal random-effects structure supported by the data was derived in the same way. The initial full random-effects model was (Intercept + Trial + Trial² | Participant) and terms were removed based on complexity (i.e., quadratic term removed before linear term). The final random-effects structure included random participant intercept and linear slope as well as random covariance between linear slope and intercept. The following model was used for analysis:

$$\text{Score}=\text{Trial}\times \text{Laterality}\times \text{Epoch}+{\text{Trial}}^2+(\text{Intercept}+\text{Trial}|\text{Participant})+\text{error}$$

Model parameter estimates for fixed effects are presented in Table 2.

Table 2.

Fixed effect parameter estimates for model of word list-learning curve in participants with mesial temporal lobe epilepsy. “Epoch” denotes time in relation to anterior temporal lobe resection (i.e., pre- or post-surgery). “Laterality” indicates left- or right-lateralized epilepsy.

	Beta	Std. Error	df	t	p
Intercept	5.71	.14	157.7	39.34	<.001
Trial (Linear Slope)	2.06	.11	1036	17.96	<.001
Trial2 (Quadratic Slope)	−.20	.02	932.0	−7.62	<.001
Epoch	.12	.07	989.2	1.58	.113
Laterality	−.49	.13	118.1	−3.72	<.001
Epoch × Laterality	−.01	.07	989.2	−.21	.830
Linear Slope × Epoch	.01	.03	992.2	.51	.607
Linear Slope × Laterality	−.09	.04	121.6	−2.04	.043
Linear Slope × Epoch × Laterality	.009	.003	992.2	.304	.761

Categorical variables were effect-coded: Laterality (Left = 1, Right = −1); Epoch (Pre-surgery = 1, Post-surgery = −1)

3.1. Trial Effects

Participants were able to recall on average 5.71 words on the first trial of the RAVLT, t(157.7)=39.34, p<.001. A significant linear slope component of 2.06 words per trial was present, t(1036)=17.96, p<.001 as well as a significant quadratic component of −.201, t(932.0)=−7.62, p<.001. Taken together, intercept, linear slope, and quadratic slope values indicate that, on average, participants recalled 7.57 words on the second trial, 9.03 words on the third trial, 10.09 words on the fourth trial, and reached on average 10.75 words on the fifth trial.

For comparison to normative data, pre-operative trial scores were converted to z-scores using age-matched RAVLT norms [45]. A one-sample t-test indicated that the participant mean z-score (−.789) was significantly lower than the population mean, t(605) = −16.224, p<.001. This analysis was repeated on a trial-by-trial basis, with Mann-Whitney-Wilcoxon tests used in Trials 1 and 5 due to non-normal distributions (Shapiro-Wilk test Trial 1: W=.976, p=.037; Trial 5: W=.972, p =.014). Participants performed below age-matched population expectations in all trials of the RAVLT (Trial 1: V=1828, p<.001; Trial 2: t(117)=−8.828, p<.001, Trial 3: t(117)=−8.783, p<.001; Trial 3: t(117)=−8.0507, p<.001; Trial 5: V=1427, p<.001).

3.2. Laterality Effects

The interactions between laterality of epilepsy and intercept, and between laterality of epilepsy and linear slope, were both significant. Participants with R-MTLE recalled on average .98 more words during the first trial of the RAVLT than participants with L-MTLE, t(118.1)=−3.72, p<.001. On each subsequent trial, participants with R-MTLE retained more words than L-MTLE participants, t(121.6)=−2.04, p=.043, leading to a difference of 1.7 words by the fifth trial (Fig. 2).

Figure 2.

Number of words recalled on the five trials of the Rey Auditory Verbal Learning Test by participants with left- and right-sided medial temporal lobe epilepsy (“Laterality”) pre- and post-anterior temporal lobe resection (“Epoch”). Figure displays growth curve model estimate means with standard error bars overlaid on raw data. Raw data points have been jittered for visualization purposes.

Comparisons to normative data were also executed as above. When stratified by laterality, both R- and L-MTLE were associated with below-median performance on the learning trials of the RAVLT (Right median= −.409, p<.001; Left median= −1.1, p<.001). A Mann-Whitney-Wilcoxon test showed that the difference was significant between the two groups: participants with L-MTLE showed significantly lower z-scores than participants with R-MTLE (p<.001). Additional trial-by-trial analyses were computed, again using non-parametric tests in Trials 1 and 5. Participants with R- and L-MTLE did not significantly differ in their pre-operative Trial 1 z-scores, W=1575, p=0.386. For the remainder of the trials, participants with L-MTLE had significantly worse performance relative to an age-matched healthy population than participants with R-MTLE (Trial 2: t(115.9=−2.901, p=.004, Trial 3: t(112.5)=−4.301, p<.001; Trial 3: t(115.9)=−3.176, p=.001; Trial 5: W=1220.5, p=.003). (Fig. 3).

Figure 3.

Overlayed box and violin plot of trial-by-trial pre-operative verbal learning performance in participants with right- and left-sided mesial temporal lobe epilepsy (MTLE), converted to z-scores for comparison to a healthy age-matched population. Participants with MTLE performed significantly below the expected population mean in all trials of the Rey Auditory Verbal Learning Test. Participants with left-sided MTLE had significantly more negative z-scores than participants with right-sided MTLE on trials 2–5. Asterisks indicate between-group comparison p-values <.01 (**) and <.001 (***).

3.3. Surgery Effects

There was no main effect of ATLR on the number of words recalled during the first trial, t(989.2)=1.58, p=.114, nor on the linear component of the learning slope, t(992.2)=.513, p=.608. Surgery epoch also did not significantly influence the effects of laterality on the number of initial words recalled, t(989.2)=−.214, p=.830, nor the learning slope, t(992.2)=.304, p=.761 (Fig. 3).

3.4. Further Analyses to Address Potential Confounds

To ensure that our results were not influenced by atypical participants, the model was re-run on subsections of the dataset. When including only participants with left-lateralized language, the same model parameters remained significant and their respective beta values did not change by more than one-tenth of a word. This was also observed when the model was run using only data from participants that experienced seizure freedom following ATLR.

To examine whether any datapoints may have had undue influence on our findings, we used the romr.fnc function from the LMERConvenienceFunctions package [46]. Six datapoints, representing 0.5% of the dataset, were identified as having a standardized residual greater than 3 standard deviations from 0. When the model was re-run with these 6 datapoints excluded, the significant parameters remained, and their beta values did not differ by more than .2 words.

3.5. Recall and Recognition

For comparison to previous literature, performance in the delayed recall and recognition trials of the RAVLT were also analyzed. Briefly, delayed recall scores are the number of words correctly recalled after a 20–25 minute delay and are believed to index retrieval abilities. Recognition scores indicate the number of words correctly recognized from a larger list of 50 words and can be used to infer consolidation of verbal information. The following model was used for both:

$$\text{Score}=\text{Laterality}+\text{Epoch}+(\text{Laterality}\times \text{Epoch})+(\text{Intercept}|\text{Participant})$$

Participants recalled an average of 7.74 words in the delayed recall trial of the RAVLT, t(123.3)=29.25, p<.001. There was a significant effect of laterality, with L-MTLE participants recalling on average 2.72 fewer words than participants with R-MTLE, t(123.3)=−5.15, p<.001. The interaction between laterality and surgical epoch was not significant, and there was no main effect of ATLR.

In the recognition trial, participants were able to correctly recognize on average 11.14 of the 15 target words, t(121.6)=34.54, p<.001. Participants with L-MTLE recognized on average 2.20 fewer target words than participants with R-MTLE, t(121.6)=−3.41, p<.001. There was a significant interaction between surgical epoch and laterality: following ALTR, participants with L-MTLE recognized on average 6.26 fewer words than participants than R-MTLE, t(110.5)=2.59, p=.01. The main effect of surgical epoch was not significant but was trending toward ATLR being associated with .62 fewer words recognized on average, t(110.5)=1.73, p=.084.

4. Discussion

In this study, we addressed limitations of previous work examining verbal learning across trials in participants with MTLE. We used mixed-effects modeling to obtain learning curves from a word list-learning task before and after ATLR. Our research indicates that there are significant differences in both initial verbal recall and verbal learning slope between participants with R- and L-MTLE, with L-MTLE being associated with worse performance. We also found that ATLR does not impact Trial 1 recall, verbal learning slope, or laterality effects, indicating that surgical intervention does not affect verbal learning over trials. Our hypotheses that participants with MTLE would demonstrate a decline in verbal learning slope following L-ATLR and that L-MTLE would be associated with greater decline than R-MTLE were not supported.

This work clarified how RAVLT performance may be impaired among participants with L-MTLE. We showed significant differences in performance in a large cohort of participants with lateralized MTLE. Our results indicate L-MTLE participants demonstrate a smaller Trial 1 recall as well as flatter learning slope when compared to R-MTLE participants, yielding fewer learned words on a trial-by-trial basis. This suggests diminished performance in working memory and encoding, as indexed by the Trial 1 recall and learning slope respectively. Using mixed-effects modeling, we were able to disentangle these two processes and show that L-MTLE may adversely affect the number of words recalled in the RAVLT by negatively impacting both verbal working memory and encoding.

The current literature generally suggests that there are no differences in pre-operative verbal learning across trials between R- and L-MTLE participants [17,20,21,24], though some studies suggest otherwise and present evidence of L-MTLE being associated with lower verbal learning scores [47,48]. Our findings support these latter studies, indicating that participants with L-MTLE demonstrate significantly lower Trial 1 recall and flatter learning slope when compared to R-MTLE participants in a word list-learning task. This may be due to our use of mixed effects modeling, which represents a more efficient analysis technique that affords more analytical power [49]. Alternatively, previous studies may have been insufficiently powered to detect differences between groups. Supporting this interpretation, one study with a similarly-sized cohort reported a non-significant trend toward participants with L-MTLE having lower verbal acquisition scores pre-operatively [28].

We found no evidence of decline in verbal learning slope following ATLR. This was unexpected as even when utilizing a more conservative measure of decline, Baxendale and colleagues reported significant declines in verbal learning following ATLR [26,27]. Other literature reports a decline in verbal learning performance following L-ATLR only, with participants undergoing R-ATLR demonstrating either no change or even small improvements [17,20,21,28]. There are several reasons that may explain these inconsistencies and why our results do not replicate previous work. It is possible that differences in surgical approach may lead to disparate findings, although current evidence does not support surgical approach as having a significant effect on verbal memory outcomes [50]. Our sample may have been unusually resilient against verbal memory decline as we also failed to replicate previous studies that show decline in delayed recall following ATLR [6,7], though we did observe significant decline in recognition of verbal information in participants with L-MTLE. It is also possible that any declines in verbal memory due to ATLR may have been mitigated by improvements due to post-surgical seizure freedom, as both seizure frequency and antiepileptic drug load have been shown to affect verbal memory [51–53]. It is important to note that we utilized alternate forms of the RAVLT to account for practice effects [8], whereas previous research used corrected z-scores [24,26,27], or did not account for practice effects [17,20,21]. As outlined above, our operational definition of verbal learning was also different than what was previously used to assess patients with MTLE. For these reasons, it is possible that our measure may be different enough from previous work to make direct comparisons difficult, and the measures may even be indexing separate constructs. The above concerns serve to highlight a need for further meta-analytical work to elucidate the impacts of ATLR on verbal learning across trials, similar to the work that has already been done for recall and recollection of verbal and non-verbal information.

This research has several limitations. First, a lack of detailed mapping of epileptic foci hinders our ability to identify neural correlates of verbal learning across trials more precisely within the anterior temporal lobe. A more detailed mapping of the locus of epileptiform activity would allow for more finely determining the brain regions involved in lateralized differences in verbal learning across trials between participants with L-MTLE and R-MTLE. Second, verbal learning across trials was assessed using a list-learning task of semantically unrelated words. It is unclear whether the additional context of prose or semantic associations, which more accurately reflect daily verbal memory demands, would have influenced our findings. Third, due to the RAVLT consisting of only 5 learning trials, we were not able to determine whether participants with L-MTLE are impaired in their total encoding capacity in addition to rate of learning. Finally, it is important to note a significant selection bias in our sample, as only participants with pharmacoresistant epilepsy undergoing ATLR were assessed. It is unclear if lateralized differences in verbal learning curves are present in the larger population of patients with MTLE whose seizures can be controlled by medication.

5. Conclusion

In conclusion, this research has demonstrated that participants with L-MTLE perform worse that those with R-MTLE in a task of verbal learning across trials, though MTLE was associated with performance below the normative median regardless of lateralization. L-MTLE was associated with fewer words recalled on the first trial of the RAVLT and a significantly flatter learning slope, suggesting deficits in working memory and encoding. A notable finding was that performance in the RAVLT was not affected by surgical intervention. These findings indicate the existence of decreased verbal learning across trials in participants with pharmacoresistant L-MTLE prior to surgery, warranting further study into the underlying causes and potential interventions. The results also underscore the importance of assessing multiple components of memory when evaluating the cognitive profiles of participants with MTLE.

Highlights.

• Left mesial temporal lobe epilepsy is associated with lower verbal learning.

• Deficits in learning are due to poor working memory and verbal encoding.

• Learning across trials is not affected by anterior temporal lobe resection.

Abbreviations

MTLE

mesial temporal lobe epilepsy

L-MTLE

left-sided MTLE

R-MTLE

right-sided MTLE

RAVLT

Rey Auditory Verbal Learning Test

ATLR

anterior temporal lobe resection

L-ATLR

left-sided ATLR

R-ATLR

right-sided ATLR