Introduction

According to Milner’s (1966) early description of the dual “material-based” model, if seizure onset originates in the left language dominant temporal lobe, verbal learning and memory are adversely affected (i.e. lateralization of verbal memory; Saghafi et al., 2018; Sherman et al., 2011; Witt, Elger, & Helmstaedter., 2015; Witt et al., 2019). Although the data is somewhat less robust (Vaz, 2004), if onset derives from the right non-dominant temporal lobe, learning and memory for non-verbal materials such as designs, or faces is affected. More than 50 years later, there is still strong pre-operative and post-operative empirical support for the relationship between verbal memory impairment and dysfunction in the left mesial temporal lobe in adults with TLE (Saghafi et al., 2018; Sherman et al., 2011; Witt et al., 2015; Witt et al., 2019), but a lack of consensus with regard to lateralization and localization of nonverbal memory (Kennepohl, Sziklas, Garver, Wagner, & Jones-Gotman, 2007; Vaz, 2004).

Furthermore, although research suggests there is an increased risk of cognitive impairment in children with temporal lobe epilepsy (TLE) and that the memory domain is most likely to be affected (Flint et al., 2017; Menlove & Reilly, 2015), it is unclear whether children with TLE exhibit patterns of material-specific memory lateralization that are similar to those described for adults. Some researchers have suggested that children with temporal lobe epilepsy may have less localization than adults and thus exhibit a broad pattern of cognitive deficits, beyond those considered to subserved by the epileptogenic region (Hermann et al., 2008; Jambaqué, Dellatolas, Dulac, Ponsot, & Signoret, 1993; MacAllister & Schaffer, 2007; Pulsipher et al., 2009). For example, Stefanatos (2015) only found minor differences between the neuropsychological profile of children with frontal and temporal lobe epilepsies. In contrast, Gonzalez, Anderson, Wood, Mitchell, and Harvey (2007) reported higher levels of memory problems in children with mesial TLE (MTLE) than lateral TLE, but no lateralization of verbal or non-verbal memory, suggesting that memory in children with TLE is localized but not lateralized (Stefanatos, 2015). However, these studies typically have small sample sizes (e.g. Pulsipher et al., 2009), with heterogeneous participants and pathologies (e.g. Jambaqué et al., 1993), making it difficult to draw a robust conclusion regarding lateralization and localization of the memory domain in children with epilepsy.

The question of localization and lateralization of memory in children with TLE is particularly critical in clinical practice, in cases of intractable epilepsy, when resection of the epileptogenic zone is considered as a proactive treatment for seizure relief (Loring, 2010). Patients with intractable TLE typically undergo comprehensive monitoring and assessment to help localize the epileptogenic focus. Such assessments include EEG monitoring, neuroimaging and functional mapping (fMRI). The neuropsychological assessment is also part of the pre-operative routine which is used to detect subtle cognitive dysfunctions (Baxendale & Thompson, 2010) and predict, at least with some certainty, the risk of post-operative memory decline, following temporal lobe surgery (Helmstaedter & Witt, 2017). Although post-operative neuropsychological impairments, especially in the memory domain, have been widely documented in adults with TLE (Lee, Yip & Jones‐Gotman, 2002; Witt et al., 2019), the effects of such surgery in childhood have been less extensively studied. Vulnerability of the verbal memory function to left temporal lobe resection in children has been reported in some studies (Dlugos et al., 1999; Jambaqué et al., 2007; Meekes et al., 2013; Szabó et al., 1998), but not in others (Gonzalez, Mahdavi, Anderson, & Harvey, 2012; Mabbott & Smith, 2003; Oitment et al., 2013; Williams Griebel, M. L., & Dykman, 1998). In a longitudinal study, Gleissner, Sassen, Schramm, Elger, and Helmstaedter (2005) reported cognitive decline three months post operation, in both children and adults. While the children showed recovery after nine months, the adults remained impaired. Skirrow et al. (2015) found long-term improvement of verbal memory following resection of the non-dominant right hemisphere in children. In a recent systematic qualitative review, Flint et al. (2017) concluded that after temporal lobe surgery, there was some evidence of material-specific memory deficits based on the resection side, namely verbal memory decline after left resection. Nevertheless, heterogeneity in sample sizes, methods, memory measures and outcomes, make it difficult to draw firm conclusions from a qualitative review. The inconsistent finding in pediatric TLE patients’ memory function can be explained in a developmental perspective (Helmstaedter and Elger, 2009). Several authors have suggested that specific difficulties can manifest throughout childhood (Culhane-Shelburne, Chapieski, Hiscock, & Glaze, 2002; Helmstaedter & Elger, 2009; Gonzalez et al., 2012; Smith, 2016). For example, in Gonzalez et al. (2012) the participants were children at baseline and adolescents or young adults at follow-up. Verbal memory deficits were apparent in left TLE patients, only at follow-up, suggesting that verbal-specific memory deficits emerge over time. Researchers have claimed that there is a critical developmental period for verbal memory lateralization. Helmstaedter and Elger (2009) indicated that memory impairment in children with TLE becomes most pronounced in the decade after puberty, in contrast to Gonzalez et al. (2012) who posited that the critical developmental period for verbal memory lateralization occurs earlier, in mid-childhood when language function is strongly lateralized and the capacity for the reorganization of memory diminishes (Everts et al., 2010).

The purpose of the current study was to conduct a quantitative meta-analysis of studies investigating the effect of lateralization (right childhood TLE versus left childhood TLE), material-type (verbal versus nonverbal) and age (age of onset, duration of epilepsy and age at surgery), on memory functions in children with TLE. The crucial advantage of a quantitative meta-analysis over a qualitative review is that it provides quantitative data on the between-study heterogeneity (Miller et al., 1994), thus enabling more accurate conclusions.

Specifically, the current study investigates (a) whether memory functions are impaired in children with TLE compared to healthy controls, (b) whether memory functions are lateralized in children with TLE compared to healthy controls, (c) whether memory functions decline following temporal lobe surgery, and finally (d) whether earlier surgery leads to better memory outcomes.

Based on the literature, it was hypothesized that children with TLE would show impaired memory function compared to healthy controls and further, they would have material-specific memory deficits involving verbal memory. That is, it was assumed that verbal memory would manifest a lateralization effect. Specifically, verbal memory was predicted to be impaired in children with left TLE (LTLE), but preserved in children with a right focus and it was also expected to be better before than after childhood temporal lobe surgery. Finally, we hypothesized that earlier surgery would lead to better memory outcomes based on the notion that brain plasticity occurs mainly in the developing brain (e.g. Berl, 2014; Cross et al., 2006).

Methods

Data Sources

Review material was drawn from the PsycINFO, Medline, and Google Scholar databases for the years 1990-2019. An existing optimized child search strategy described by Boluyt et al. (2008) was used to select child studies. The key search terms were: *TLE* OR “temporal lobe epilepsy” OR "HS*" OR “hippocam*” OR "MTS*" OR “mesial temporal sclerosis*” AND “Surgical*” OR “Operati*” OR “Resecti*” OR "Temporal lobectomy" AND “child*” OR "pediatric*" OR “Paediatric*” OR “Paediatric*” OR “school age*” OR “schoolchild*” OR “school*” OR “kid*” OR “adolescent*” OR “teen*” OR “boy*” OR “girl*” OR “Junior*” AND "memory*" OR “learning and memory” OR “recall*”.

Inclusion Criteria

Based on the PRISMA systematic review guidelines (Moher et al. 2015), the inclusion criteria for the studies were 1. fll-length, English language studies published between 1990 and October 2019, 2. contained samples of pediatric patients (aged 6-16) with RTLE or LTLE, 3. reported statistics for the comparison of memory data, or when not available, contained images that enable data extraction, and 4. had eight participants or more.

Data Extraction

The statistics extracted included verbal and nonverbal memory performance comparisons between patients and healthy age-matched control groups, or verbal and nonverbal memory performance comparisons pre- and post-surgery.

The quantitative analysis focused on delayed memory outcome, which is considered the gold standard for assessing memory integrity in TLE patients (Witt et al., 2019) and is also the most consistently reported. Other measures may have limited efficacy in differentiating paediatric epilepsy patients from healthy children or pre- to post-surgical patients. For example, children with epilepsy scored lower than the norm on long not short delay memory tasks, which were within the normal range (Hershey, Craft, Glauser, & Hale, 1998). The statistical data included the standardized means, and Standard Deviation (SD) of the delayed recall outcomes per group and memory material type. If the mean or the standard deviation values were not reported, these were calculated based on the reported confidence intervals (CI) or p-values. If the standard error (SE) was reported, this was converted to the standard deviation. Several studies reported more than one result for the same group of participants (e.g. multiple memory results for a single memory paradigm or multiple paradigms in a single memory domain). However, since inclusion of non-independent observations risks underestimating the error variance associated with each effect size (Borenstein et al. 2011) in cases of multiple results of a single measure, the results for delayed recall were used. More than 90% of the studies compared multiple independent experimental groups (RTLE and LTLE) with a single control group or multiple results for different memory domains (i.e., verbal and nonverbal domains) within the same sample. Since calculating an average effect size that collapses over the observations would result in the omission of important moderator data and therefore is not appropriate (see Higgins & Green 2011), effect sizes for each of these non-independent comparisons were included as separate datasets. Nevertheless, to avoid underestimating the error variance associated with each effect size, the sample sizes used to calculate the standard errors for each group were divided by the number of inclusions (see Higgins & Green, 2011; Webb et al., 2012).

Statistical Analyses: Meta-Analyses

The effect size was calculated using the standardized delayed memory scores for patients compared to controls (Higgins & Green, 2011). For before and after studies, the effect size was calculated as the change in memory performance from baseline. As recommended in the Cochrane Handbook for Systematic Review of Interventions and others (Follmann, Elliott, Suh, & Cutler, 1992), imputed correlation coefficients of 0.50 were used to impute the change from the baseline standard deviation required to calculate an effect size. Effect sizes were calculated in a multi-stage process. The first stage involved calculating effect sizes for each test that was included from each individual study. A higher memory test score indicated better performance. Therefore, a negative effect size indicates that patients performed worse than controls, or patients' scores decreased after surgery. To correct for small sample sizes, Hedges’ g effect size computation was used: g = d[1 - (3/4 N) - 9], where N represents the cumulative sample size for both patients and control groups (Hedges & Olkin, 2014). The magnitude of Hedges’ g coefficient is equivalent to Cohen’s d effect sizes. Effect sizes of 0.2 were considered small, effect sizes of 0.5 were considered medium, and effect sizes of 0.8 were considered large (Cohen, 1988). The calculation of effect size for each outcome, the pooling of effect sizes and tests of heterogeneity were conducted using Meta-Essentials version 1.4 (Suurmond, van Rhee & Hak, 2017). Two-tailed significance tests were employed at a p value of 0.05. Given that the studies included in the meta-analyses differed in terms of several variables (e.g., memory test type, participants' characteristics, etc.), a random effects model was used to calculate the between or within group comparison pooled effect. As noted by Borenstein, Hedges, Higgins, and Rothstein (2011), the random effects model is often the most appropriate model to use in a quantitative synthesis of existing literature because the random-effects model allows for the true effect size to vary from study to study unlike the fixed-effect model which is based on the assumption that all studies in the meta-analysis share a common (true) effect size. The effect sizes were calculated to the 95% confidence intervals.

Heterogeneity in effect sizes was tested using the Q statistic (\(C{hi}^{2}\)), \({I}^{2}\) and \(tau-squared ({\uptau }^{2})\), for each comparison. Q provides significance testing for heterogeneity, with p < 0.05 indicating significant heterogeneity. \({I}^{2}\) represents the ratio of the variance in the true effect compared to the variance due to sampling errors and was reported for descriptive purposes (Borenstein, Higgins, Hedges, & Rothstein 2017), where 25%= low, 50%= moderate, and 75%= high heterogeneity (Higgins, Thompson, Deeks, & Altman, 2003). \({\uptau }^{2}\) estimates the between-study variance of the effect sizes, which is an indication of absolute variance (Borenstein, Higgins, Hedges, & Rothstein, 2011).

To explore heterogeneity, the effects of several moderator variables were investigated including the side of the epileptogenic foci (LTLE or RTLE), and the memory test material type (non-verbal or verbal). In addition, meta-regressions were used to investigate the roles of duration of epilepsy and age at onset. In addition, Forest plots were generated to visualize integrated size effects and heterogeneity.

Two quantitative meta-analyses were conducted. The first meta-analysis examined heterogeneity and estimated the effect sizes for differences in memory performance between children with TLE and healthy controls. The second meta-analysis examined the heterogeneity of the results and estimated effect sizes with regard to memory performance differences in children with TLE before and after temporal lobe surgery.

Each analysis first examined whether there were memory outcomes present in both the right and left foci, by assessing the heterogeneity of the findings of both groups as a whole. Then, in order to investigate the source of variation across studies, memory performance and heterogeneity in children with RTLE and LTLE were analyzed separately. Under the assumption that learning, and retention of verbal materials is associated with the left temporal memory system, tests were run on the RTLE and LTLE subgroups separately to determine whether the pooled estimate of memory outcomes and heterogeneity were influenced by specific differences in material. Finally, to investigate the potential influence of age parameters (age at onset, duration of seizures and age at surgery) on memory performance, moderator analyses was conducted to determine whether age moderated the memory performance in children with TLE and whether earlier surgery leads to better memory outcomes.

Results

Retrieval The literature search yielded 158 references of which 33 (21%) were duplicates and 98 (62%) did not meet inclusion criteria. After their removal, 27 studies met the inclusion criteria. Of these 27 studies, four studies were by the same research group and had identical numerical outcomes (Guimarães et al., 2007; Guimarães et al., 2014; Rzezak, Guimarães, Fuentes, Guerreiro, & Valente, 2011; Rzezak et al., 2012) therefore only the earlier version of the studies (Guimarães et al., 2007; Rzezak et al., 2011) were included, resulting in 25 remaining studies. A flow chart of the systematic review phases is presented in Fig 1.

Fig. 1
figure 1

adapted from Moher, Liberati, Tetzlaff, & Altman, 2009)

A four-phase flow diagram of the systematic review (

Full size image

The studies and their characteristics are presented in Tables 1-2.

Table 1 Summary of studies (N = 13) that compared children with TLE to controls
Full size table
Table 2 Summary of studies (N = 14) that compared children with TLE before and after epilepsy surgery on memory paradigms
Full size table

11 studies compared patients to controls, 12 compared preoperative and post-operative results of the patient groups and two had both experimental designs (Beardsworth & Zaidel, 1994; Lendt, Helmstaedter, & Elger, 1999). However, in these two studies, only the between groups comparison data was available for numerical calculations. Two studies with a pre-post design measured post-operative memory twice, at three- and 12-months post-op (Gleissner et al., 2005; Lendt et al., 1999). Only the 12-month assessment outcomes were included in the current analyses. Five studies had a single patient group containing both RTLE and LTLE patients (Bailey, 2013; Guimarães et al., 2007; Mankinen et al., 2014; Rzezak, Guimarães, Guerreiro, & Valente, 2017; Smith, Elliott & Lach, 2006). The rest of the studies (n = 19) were composed of two patient groups of children with either LTLE or RTLE. All of the studies except two (Beardsworth & Zaidel, 1994; Martins et al., 2015), used a verbal memory paradigm, 16 of which (64%) also used a nonverbal paradigm whereas the remainder only used a verbal memory paradigm (Gleissner et al., 2002; Jambaqué et al., 2009; Lah & Smith, 2015;Law, Benifla, Rutka, & Smith, 2017; Mankinen et al., 2014; Szabó et al., 1998). Three studies used a memory of faces paradigm (Beardsworth & Zaidel, 1994; Gonzalez et al., 2012; Mabbott & Smith, 2003; Smith et al., 2006). As shown in Table 3, the children were older in the pre-post studies compared to the between groups studies, with respect to both the age at onset (p < .05) and the age at assessment (p < .01), but no differences were found between the pre-operative and post-operative studies and patient-control studies, with regard to the duration of epilepsy.

Table 3 Summary of cognitive measures and outcomes in studies (N = 13) that compared children with TLE to controls
Full size table

Quality assessment The following criteria were employed to assess study quality: (1) randomization, (2) double blinding and (3) proper treatment of withdrawals or dropouts (Jadad et al., 1996). The first two criteria were not applicable because the groups were divided by disease status and had apparent behavioral differences. The dropout criterion was not reported in any of the studies. Nevertheless, patients’ withdrawal was reported in five studies: Bailey (2013), 31 patients at the post-operative assessment, Gascoigne et al. (2014), two patients and two controls, Gonzalez et al. (2012), 19 patients at the post-operative assessment, Mankinen et al. (2014), 11 patients, Smith et al. (2006), three patients and five controls. Effects of publication bias were examined with Egger’s tests (Egger, Smith, Schneider, & Minder, 1997) and were also inspected visually by a funnel plot.

Quantitative Analysis (Meta-analysis)

Patients Versus Controls Differences

Thirteen studies compared memory performance in children with TLE and healthy children, with age-matched controls using either verbal, nonverbal or both paradigms (Table 4).

Table 4 Summary of cognitive measures and outcomes in studies (N = 12) that compared children with TLE before and after surgery
Full size table

The 13 studies included 39 comparisons of children with either LTLE (18 comparison), RTLE (18 comparisons) or both (three comparisons) to controls. The first stage of the meta-analysis covered 38 effect sizes that were derived from the verbal and nonverbal paradigms for children with LTLE-RTLE groups in total. The magnitude of the differences was relatively moderate (df = 37, Hedges’ g = −.48, p < .0001, 95% CI [−.75, -.22]). A statistical examination of the funnel plot using Egger's regression test resulted in a non-significant model, suggestive of a lack of significant publication bias (t(38) = -.56, p > .05). Q value indicated heterogeneity (Q = 184.6, p < .01, \({\tau }^{2}\) = 0.51, I2= 80.5%) and therefore the presence of potential moderators (Sánchez-Meca & Marín-Martínez, 1997). Accordingly, in the second stage of analysis, the group type was entered as a moderator. For LTLE patients the magnitude of the differences was relatively small (df = 17, Hedges’ g = −.41, p < .01, 95% CI [−.74, -.09]; Q = 56.44, p <0.05, \({\tau }^{2}\) = 0.31, I2 = 71.65%). For RTLE patients the magnitude of the differences was also relatively small (df = 17, Hedges’ g = −.34, p < .01, 95% CI [−.78, .10]; Q = 83.71, p < .01, \({\tau }^{2}\) = 0.64, I2 = 82.08%, see Fig. 2a). For both groups, the Q values indicated effect size heterogeneity and therefore the presence of additional moderator(s). Accordingly, in the third stage, paradigm type was entered as a moderator in each of the two patient groups: LTLE patients, verbal memory paradigm, df = 8, Hedges’ g = −.76, p < .01, 95% CI [−1.9, .47]; Q = 45.60, p <0.01; LTLE patients, nonverbal memory paradigm, df = 8, Hedges’ g = −.51, p < .01, 95% CI [−1.52, .28]; Q = 30.72, p <0.01; RTLE patients, verbal memory paradigm, df = 8, Hedges’ g = −.32, p < .01, 95% CI [−1.6, 1.07]; Q = 54.93, p <0.01; RTLE patients, nonverbal memory paradigm, df = 8, Hedges’ g = −.87, p < .01, 95% CI [−2.3, .61]; Q = 63.87. Therefore, with regard to patients and controls differences, since the Q value still indicated heterogeneity, the mean effect size was not considered a reliable estimation for the data.

Fig. 2
figure 2figure 2

a-b Forest plot of standardized effect sizes and confidence intervals for verbal memory and nonverbal memory in children with TLE compared to controls (2a) and in children TLE pre- compared to post-operative status (2b). Study name are in year of publication order

Full size image

Pre-operative versus post-operative differences Twelve studies compared the memory performance of children with TLE before and after surgery, using either verbal, nonverbal or both paradigms (Table 5).

Table 5 Means in years (SD) of age at onset, age at assessment, and duration with epilepsy in pre-post studies compare to control-patient studies
Full size table

The 12 studies included 39 comparisons of children with either LTLE (18 comparisons), RTLE (16 comparisons) or both (five comparisons). Verbal memory paradigms were used in 21 comparisons and nonverbal paradigms were used in 17 comparisons. In line with the between-groups analyses, the first stage of the meta-analysis included 39 effect sizes that were derived from the verbal and nonverbal paradigms for children with LTLE-RTLE groups in total. The magnitude of the differences was minimal (df = 38, Hedges’ g = −.02, p >.05, 95% CI [−.16, .13]). A statistical examination of the funnel plot using Egger's regression test resulted in a non-significant model, suggestive of a lack of significant publication bias (t(38) = 1.82, p = .07). The Q value indicated heterogeneity (Q = 57.60, df = 38, p < .05, \({\tau }^{2}\) = .06, I2= 32.28%) and therefore the presence of potential moderators (Sánchez-Meca & Marín-Martínez 1997). Accordingly, in the second stage of analysis, the group type was entered as a moderator: LTLE patients, n = 18, Q = 22.70, df = 17, p >.05, \({\tau }^{2}\) = .04, I2 = 25.12%; RTLE patients, n = 16, Q = 20.00, df = 15, p > .05, \({\tau }^{2}\) = .05, I2 = 25% (Fig. 2b). The Q value indicated homogeneity and therefore the mean effect size was considered the best estimation for the data. The overall magnitude of the differences was small. In the LTLE studies Hedges’ g = -.16, SE: .14, CI [-.36, .03] and in RTLE studies Hedges’ g = .21, SE: .09, CI [-.30 , .73].

Finally, paradigm type (verbal, nonverbal) was added as a moderator for each of the two patients groups: LTLE patients, verbal memory paradigm, n = 10, Q = 12.03, df = 9, p > .05; LTLE patients, nonverbal memory paradigm, n = 8, Q = 11.93, df = 7 p > .05; RTLE patients, verbal memory paradigm, n = 9, Q = 18.93, df = 8, p > .01; RTLE patients, nonverbal memory paradigm, n = 7, Q = 22.18, df = 6, p <.01. Homogeneity of the estimated mean effect size was found in LTLE patients for both the verbal (Hedges’ g = -.38, SE: .13, CI [-0.71, -.06]) and nonverbal (Hedges’ g = .09, SE: .16, CI [ -.46, .65]) paradigms and in RTLE patients for the verbal paradigm (Hedges’ g = .06, SE: .16, CI [-.76, .89]) but not for the nonverbal paradigm.

Covariance of age at onset and age at assessment Regression analysis revealed that age at onset, duration of epilepsy and age at assessment did not serve as moderators of the Hedges’ g effect sizes for patients versus controls (age at onset, β = -.29, p = .08; duration of epilepsy; β = .21, p > .05; age at assessment , β = -.06, p > .05). This was also true with regard to pre-operative versus post-operative effect sizes, neither for the age parameters (age at onset, β = -.22 \(,\) p > .05; duration of epilepsy, β = -0.3, p > .05; age at assessment, β = -.12, p > .05 ).

Discussion

The purpose of this quantitative meta-analyses was to examine the role of lateralization, material type and age, on memory functions in children with TLE, both before and after surgery. It was hypothesized that there would be evidence of material-specific memory deficits, but only with regard to verbal memory in LTLE patients, in control-patient comparisons and in pre-post-surgery comparisons. The hypotheses were partly confirmed. In control-patient comparisons, in contrast to our predictions, heterogeneity indicated inconsistent results, It is difficult to draw conclusions based on the studies that reported memory differences between controls and pediatric TLE patients.

Control-Patient Findings

Although most studies that compared patients to controls have revealed a significant verbal or nonverbal memory impairment in children with TLE compared to controls (for example; Cohen, 1992; Jambaqué et al., 1993; Guimarães et al., 2007; Jambaqué et al., 2009; Leunen et al., 2009) the current meta-analysis failed to reveal consistent differences between groups. The main explanation is related to the substantial methodological differences across studies. The small sample sizes, within group variability, different measuring methods and outcome measures all increased the heterogeneity value, and hampered efforts to draw integrative conclusions. Although it is challenging to identify test measures that are common across the world, the establishment of an international uniform neuropsychological assessment protocol would constitute a valuable first step which would increase the homogeneity in results across studies and enable a coordinated international approach to neuropsychological assessment in epilepsy (Baxendale et al., 2019). Additionally, the idea that TLE might be regarded as a wider brain network disorder affecting brain areas beyond the temporal lobes, as opposed to the ‘domain specificity’ model, raises the question of whether memory function alone is the most appropriate cognitive measure to evaluate TLE cognitive dysfunctions (Hermann, Loring, & Wilson, 2017). Further studies should examine whether children with TLE exhibit impairment in other cognitive areas such as attention and language (Mankinen et al., 2014; Hermann et al., 2016; Lah & Smith, 2014).

Memory Decline Following TLE Surgery in Children

The results suggest that the side of the epilepsy foci (RTLE versus LTLE), alongside the material type (verbal versus non-verbal) may serve as a moderator for children with TLE memory changes after epilepsy surgery. These analyses revealed a mild effect size in children with LTLE for decline in post-operative verbal memory as compared to their pre-operative verbal memory performance. In addition, with regard to post-operative compared to pre-operative nonverbal memory performances, there was practically a zero-effect size which indicates stability. Similarly, in children with RTLE, a virtually zero effect size was revealed, indicating stability in verbal memory performance before and after epilepsy surgery.

Memory Decline Following Left TLE Surgery in Children

The main finding indicates that delayed verbal memory becomes impaired in children after left temporal lobe surgery. This result is in line with the literature on adult TLE, that has reported relatively consistent rates of verbal memory decline after left-sided surgery (Lee et al., 2002; Sherman et al., 2011). These results suggest that in children, as in adults, verbal memory is lateralized to the left hemisphere and that verbal memory tasks are sensitive to LTLE surgery (Flint et al., 2017). The data regarding lateralization of the nonverbal memory to the right hemisphere or demonstrating sensitivity of nonverbal tasks to RTLE surgery were not confirmed. Hence, better post-operative cognitive outcomes might be expected with seizure freedom. One possible interpretation of these results is that seizure frequencies and seizure control might not serve as the main contributor with regard to memory function in children with TLE (Smith et al., 2006). Rather, the findings support the notion that an underlying neurological impairment may influence memory, not the seizures themselves. Note that seizure control was not included in the current analysis since not all studies provide data on this. Further research should include seizure control and seizure frequency as covariates when measuring memory function after pediatric TLE surgery.

Age at Onset: Developmental Considerations

Several authors have suggested that specific difficulties emerge throughout childhood (Culhane-Shelburne et al., 2002; Helmstaedter & Elger, 2009; Gonzalez et al., 2012; Smith, 2016). The current meta-analysis failed to show a similar association between age and effect size. Nevertheless, the patients versus controls effect sizes were marginally negatively associated with age at onset (p = .08). These findings may hint that early onset is related to better memory performance in children with TLE compared to controls, and therefore indirectly supports the notion that the vulnerability of memory in children with TLE has a developmental basis. The findings on adults which support a material-specific pattern of impairment (Saling, 2009) is consistent with the notion that the nature of verbal memory impairment changes over time. The most widespread explanation for these results is based on the concept of brain plasticity that occurs mainly in the developing brain (Berl, 2014). Further prospective longitudinal research is needed to fully explore the influence of age on memory performance in children with TLE. Another important issue which requires further investigation is whether earlier surgery leads to better memory outcomes. The current results show no association between patients' age at surgery and memory decline after surgery. Nevertheless, the possibility that memory functions becomes localized over the developmental period could indirectly suggest that earlier surgery might be more beneficial, given that the greater plasticity of the developing brain enables greater reorganization of cognitive functions and minimizes the functional impact of removal brain tissue (Cross et al., 2006). In contrast, the crowding hypothesis argues that lesion (removal of tissue) induced reorganization of cognitive function to the unimpaired hemisphere results in compromises in cognitive processing due to limited computational capacity (Strauss, Satz, & Wada, 1990). Interestingly, the crowding hypothesis has recently received support in a study on pediatric TLE (Danguecan & Smith, 2019).Future longitudinal studies could shed light on the issue of the recovery abilities of the developing brain with regard to memory function after TLE surgery.

Limitations and Future Directions

The results of this quantitative meta-analysis differ partially from the results of the preliminary qualitative analysis, which suggested that verbal memory is not affected after LTLE surgery. The results also partially counter the Menlove and Reilly (2015) systematic review results which suggested, although cautiously, that there was improved memory performance after left surgery in children with TLE.

As presented in Table 5, most studies that compared pre-surgery to post-surgery verbal memory in children with LTLE, did show a trend toward verbal memory decline, even though this was not always significant. The quantitative statistical analysis revealed that this trend was consistent. This difference between analyses highlights the advantages of performing a quantitative meta-analysis over other non-quantitative methods when the goal is to generate conclusions based on a literature review. Most of the pediatric studies that were included in the Menlove and Reilly (2015) qualitative review were excluded here, because they included small sample sizes, combined scores from different tests or test versions or measured change in ways that precluded the calculation of the estimated mean which is required for a quantitative analysis. Nevertheless, larger sample sizes in pre-surgery to post-surgery studies with longer follow-up periods and comprehensive memory batteries are needed to establish more robust conclusions.

Another limitation that needs to be considered is that some of the excluded studies did not match the inclusion criteria but still make an important contribution to the literature. For example, the Skirrow et al. (2015) study was excluded because the pre-post assessment interval was outside the delay frame and certain statistical data, namely SD, was missing. The authors reported that nine years post-operation, patients who underwent surgery in childhood tended to show a hemisphere-dependent material-specific improvement in memory functions in the intact temporal lobe (i.e. verbal memory improvement at the follow-up after right temporal lobe surgery and non-verbal memory improvement after left temporal lobe surgery). These results can be interpreted as the release of reserve capacities which were suppressed or damaged by epilepsy (Helmstaedter et al., 2003), i.e., the release of memory function in the non-operated temporal lobe. As mentioned in the Method section, we limited the scope of the studies in the current meta-analysis to either patients vs. controls or pre-surgery – post-surgery comparisons. Since most of the studies used standardized memory scores, a future meta-analysis might consider the inclusion of studies without a control group, in order to prevent the exclusion of valuable data from the meta-analysis. Furthermore, other factors might also influence memory functions in children with TLE, such as intelligence, medical treatment, seizure control, seizure frequency and presence of comorbid conditions. These could not all be controlled for since the data was not always presented in the studies. Future studies could also address the question of whether other factors influence pediatric epilepsy patients’ memory scores, and in particular the characteristics of memory paradigms, which are shown to have major influence on memory outcomes in the adult literature (Saling, 2009). The choice of method for handling non-independent effect sizes is a topic of ongoing debate (Borenstein et al. 2009) and the applicability of newly developed methods (Cheung, 2019) should be considered in future meta-analyses. Finally, for obvious ethical reasons none of the studies reviewed here conducted randomized clinical trials on the effect of surgery on memory performance and the conclusions cannot be extrapolated to frontal lobe or other extratemporal epilepsies.

Conclusions

The results of the current meta-analysis showed that studies have consistently found evidence for a mild verbal memory decline after LTLE surgery, whereas nonverbal memory performance stayed the same after either LTLE or RTLE surgery. The results also imply a better memory outcome at an earlier age. While further longitudinal investigation is needed, these results clarify to some extent the ambiguous findings on memory performance in children with TLE, before and after surgery. The results highlight the need for a coordinated international approach to neuropsychological studies of childhood TLE utilizing some common data elements. Special Acknowledgement should be referred to the latest ILAE guidelines regarding the role of neuropsychological assessment in epilepsy.