Towards a Better Understanding of Cognitive Deficits in Absence Epilepsy: a Systematic Review and Meta-Analysis

Cognition in absence epilepsy (AE) is generally considered undisturbed. However, reports on cognitive deficits in AE in recent years have suggested otherwise. This review systematically assesses current literature on cognitive performance in children with AE. A systematic literature search was performed in Pubmed, Embase, Cochrane and Web of Science. All studies reporting on cognitive performance in children with AE were considered. In total 33 studies were eligible for inclusion. Neuropsychological tests were classified into the following domains: intelligence; executive function; attention; language; motor & sensory-perceptual examinations; visuoperceptual/visuospatial/visuoconstructional function; memory and learning; achievement. Random-effect meta-analyses were conducted by estimating the pooled mean and/or pooling the mean difference in case-control studies. Full-scale IQ in children with AE was estimated at 96.78 (95%CI:94.46–99.10) across all available studies and in case-control studies IQ was on average 8.03 (95%CI:-10.45- -5.61) lower. Verbal IQ was estimated at 97.98 (95%CI:95.80–100.16) for all studies and 9.01 (95%CI:12.11- -5.90) points lower in case-control studies. Performance IQ was estimated at 97.23 (93.24–101.22) for all available studies and 5.32 (95%CI:-8.27–2.36) points lower in case-control studies. Lower performance was most often reported in executive function (cognitive flexibility, planning, and verbal fluency) and attention (sustained, selective and divided attention). Reports on school difficulties, neurodevelopmental problems, and attentional problems were high. In conclusion, in contrast to common beliefs, lower than average neurocognitive performance was noted in multiple cognitive domains, which may influence academic and psychosocial development. Electronic supplementary material The online version of this article (10.1007/s11065-019-09419-2) contains supplementary material, which is available to authorized users.


Introduction
Typical absence seizures as occurring in childhood absence epilepsy (CAE) and juvenile absence epilepsy (JAE) are characterized by demarcated brief episodes of unconsciousness with generalized~3-Hz spike-and-wave complexes, visible on an electroencephalogram (EEG), in otherwise healthy child r e n ( G u e r r i n i , 2 0 0 6 ; M y e r s & F e c s k e , 2 0 1 6 ; Panayiotopoulos, 2001;Tenney & Glauser, 2013). A clear delineation of the clinical spectrum between CAE and JAE Electronic supplementary material The online version of this article (https://doi.org/10.1007/s11065-019-09419-2) contains supplementary material, which is available to authorized users. is challenging (Hughes, 2009;Tenney & Glauser, 2013;Trinka et al., 2004). CAE occurs mostly between 4 and 10 years of age whereas JAE occurs between 10 and 17 years of age. Furthermore, JAE is characterized by less frequent absence seizures, a higher incidence of tonic-clonic seizures and a higher drug dependency during adulthood.
Starting from early descriptions, cognitive performance in AE has been considered to be normal (Adie, 1924;Currier, Kooi, & Saidman, 1963). However, in a large randomized clinical trial intelligence was lower than normal values, but still within normal range . Nevertheless, 36% of drug-naïve patients presented with attentional deficits. Moreover, subsequent allocation to monotherapy with Valproate was associated with more attentional dysfunction, than allocation to monotherapy with either Ethosuximide or Lamotrigine Glauser et al., 2013;Glauser et al., 2010). Nevertheless, attentional deficits persisted independent of the allocated anti-epileptic drug treatment or seizure control after 1-year of follow-up Glauser et al., 2013;Masur et al., 2013). These are intriguing findings, as apart from therapeutic side effects and (inter)ictal activity other underlying mechanisms may also affect cognition in AE (Aldenkamp & Arends, 2004;Jafarian et al., 2015;Lenck-Santini & Scott, 2015;Nicolai et al., 2012).

Methods
This review adheres to the Preferred Reporting Items for Systematic Reviews and Meta-analysis guidelines (PRISMA) (Moher, Liberati, Tetzlaff, & Altman, 2009). The methods and procedures for this review are available in the paper with additional information provided in the supplemental materials. The review was not registered prior to conducting this review.

Selection Eligibility
Selection criteria were defined according to PICOS: Participants = Children with AE, either defined as 3-4 Hz spike-wave complexes or a syndromic classification of CAE and/or JAE; Intervention/diagnostic = neuropsychological tests (batteries) or reports on school performance; Comparison = normative values (for example reports using standardized scores as these represents scores relative to the normative sample of the test) or a control group; Outcome = neuropsychological function based reported as a median/ average score or prevalence of impairment based on cut-off values or a direct comparison of test scores with a control group; Design = Observational studies (cohort studies/casecontrol studies) or clinical trials (depending of the design of the clinical trial these data are also regarded as observational data, for example baseline neuropsychological results).

Systematic Literature Search
The search strategy consisted of indexed terms and free text words on absence epilepsy in combination with terms on observational research and clinical trials (the search in Pubmed is provided as a Supplemental file). The following electronic databases were searched: Pubmed (until 29-09-2017), EMBASE (until 23-03-2017), Cochrane (until 05-04-2017) and Web of Science (all databases) (until 06-04-2017). Furthermore, references of included articles were hand searched to find additional relevant articles. In case multiple publications were available on the same study sample or if reports reported overlapping results, only the most recent data were used. Only studies written in English, Dutch or Spanish were considered. Firstly, titles and/or abstracts identified by the search were first screened by EFW to remove any unrelated hits. Secondly, the remaining abstracts were screened by EFW and another author (GSD or SK or MD or JH) based on the predefined selection criteria. The second researcher was blinded for journal, authors, title, date of publication and publication language. Any discrepancies were resolved by consensus or by screening the full-text subsequently. Eligibility assessment of full-text articles was performed by EFW and one of the neurologists (MD or SK). Any discrepancies were resolved by consensus or by consulting a third author.

Data Extraction
The following data were extracted: year of publication, country, study design, inclusion/exclusion criteria, number of included patients, number of patients per cognitive assessment, age at onset, age at the time of the study, anti-epileptic drug use and neuropsychological tests results.
Neuropsychological test results were classified to cognitive domains according to Baron and secondly according to the authors of the study or consensus within our team (Baron, 2004). We distinguished the following cognitive domains: intelligence; executive function; attention, language; motor and sensory-perceptual examinations; visuoperceptual/visuospatial/visuoconstructional function; learning and memory. Additionally, we included results on: achievement tests, parent/teacher (by proxy) reports on attention or attentional deficiency disorders; reported prevalence's of school difficulties; neuropsychological and/or neurodevelopmental problems.

Risk of Bias in Individual Studies
Risk of bias was evaluated using a modified version of the Newcastle Ottawa Scale made suitable for this review (available in the Supplementary Materials). Two authors (EFW & SK) independently rated each study, and any disagreement was resolved by consensus. For the interpretation of the total scores we used the following cut-off values as used previously in a systematic review (Marengoni et al., 2018). Scores >7 were considered a low risk of bias; 5 to 7, a moderate risk; and < 5, a high risk. For the meta-analysis sensitivity analyses were carried out by excluding studies with a score lower than seven.

Meta-Analyses
Statistical analyses were conducted in R using the "meta" package (version 4.8-2) and "metaphor" package (version 1.9-9) (Schwarzer, 2007;Viechtbauer, 2010). Single-arm meta-analyses (a weighted pooling of the reported means including studies without a control group) were performed by estimating the weighted mean using a random effect model. For this, we used the reported mean and calculated the standard error (SE) for each study. In case a z-score with a 95% confidence interval was reported we calculated the standard score and its standard deviation. The estimated mean was considered significantly different from normal if the 95% confidence interval did not include the normative mean of the neuropsychological test.
In addition, a random-effects meta-analysis on the mean difference between cases and controls was performed. Studies that did not report usable data for pooling of results were discarded for the meta-analyses but were included in this systematic review. Pooling of results was conducted per neuropsychological test if methodology and reporting of the results allowed a direct comparison between studies. P values of ≤0.05 were considered to infer statistical significance.
The presence of small study effect and/or publication bias was assessed by visually inspecting funnel plots for asymmetry. In case of potential outliers, a sensitivity analysis was performed by recalculating the effect size after removal of these studies. Tau-squared (T 2 ) was used to estimate the true variance of the true effect sizes (Borenstein, Higgins, Hedges, & Rothstein, 2017). In addition, I 2 is reported for descriptive purposes. Due to the limited amount of studies in the metaanalyses we were not able to perform subsequent metaregressions to examine the impact of moderator variables (Shuster, 2011;Thompson & Higgins, 2002).

Selection of Studies
The study selection process is depicted in Fig. 1. A total of 3833 individual articles across all electronic databases were screened. Additionally, nine references from selected articles were screened for eligibility. A total of 506 abstracts were selected for further reading, after which 351 were selected for full-text screening. In the end, 33 articles were included in this systematic review.
Based on our assessment for the risk of bias eighteen studies had a low risk of bias and eleven studies had a moderate risk of bias. No studies were classified to have a high risk of bias. The highest risk of bias was due to inadequate selection and/or description of the non-exposed cohort and a small sample size. Full-text arƟcles excluded: (n = 281) -No report on cogniƟon in AE:(n = 220) -Other language: (n = 50) -Unavailable full-text: (n = 6) -Other publicaƟon type: (n = 5)
The estimated mean for the Verbal Comprehension Index was 94.43 (95% CI: 91.11-97.75; T 2 = 7.39; I 2 = 70.2%), the estimated mean for the Processing Speed Index was 96.30 (95% CI: 93.74-98.86; T 2 = 2.77; I 2 = 41.9%) and the estimated mean for the Perceptual Organization Index was 97.09 (95% CI: 95.50-98.69; T 2 = 0; I 2 = 0%). One study, which used the WISC-IV, reported a mean Working Memory Index of 94.8 (SD 14.2). Sensitivity analyses by excluding outliers based on the funnel plots did not yield different conclusions. Results on the subtests of the Wechsler Intelligence tests are available in Table S1. Non-verbal intelligence tests were used in three studies. In the study with by far the largest sample of Masur et al. (n = 316) non-verbal intelligence assessed with the TONI-3 test was significantly higher compared to the normative mean . Raven's Progressive Matrices was used by Oostrom et al. which

The study of Levav et al. used the Rosvold Continuous
Performance test and also reported diminished scores in visual sustained attention (2002). However, in this study, the mean age of the control group used was 12 years older than the group with AE. On the Trail Making Test A, the pooled estimated mean difference compared to controls was 12.37 s (95% CI: 7.02-17.73; T 2 = 0; I 2 = 0%) longer for children with AE and on the Trail Making Test B an estimated mean difference was found of 51.48 s (95% CI: 12.49-90.46; T 2 = 1061.59; I 2 = 89.9%) (Fig. 2) (Conant et al., 2010;D'Agati et al., 2012;Levav et al., 2002;Mostafa et al., 2014).
Furthermore, Cerminara et al. found significantly lower scores in some measures of alertness, divided attention, impulsivity, and selective attention in CAE compared to controls (2013). The divided and selective attention tasks were characterized by more errors of omission, whereas the impulsivity task was characterized by more commission errors. Reaction times had significantly more variability during the tonic arousal, phasic arousal and impulsivity task, but not in the divided or selective attention task. Focused attention did not differ from controls; however, Cheng et al. (2017) did find a significantly longer choice reaction time, which is similar to the focused attention task in the study performed by Cerminara et al. (2013). Mostafa et al. reported significantly lower mean scores compared to controls in an expressive attention and receptive attention task; however, it was not clear whether this was still statistically significant when corrected for the 6 years age difference with controls (2014). Siren et al. used FEPSY auditory/visual reaction times and STIM tasks to assess attention and did not find a significant difference in this small sample (n = 10) (2007).
The study of Conde-Guzon and Cancho-Candela in typical AE patients reported significantly lower performance in phonemic hearing, articulation/repetition, denominating/narration and phonetic analysis (2012). Comprehension and understanding of simple grammar did not differ significantly.
In a recent study by Cheng et al. semantic comprehension and word rhyming was not significantly worse compared to controls (2017). Furthermore, Henkin et al. reported on auditory event-related potentials and found significant increased N2 amplitudes for phonetic and semantic processing, as well as, a significantly increased latency for semantic stimuli for P3 compared to controls (2003).

Random effects model
Heterogeneity: I 2 = 64%, τ 2 = 7.5718, p < 0.01 Lopes et al., 2013Masur et al., 2013Masur et al., 2013Gencpinar et al., 2016D'Agati et al., 2012Caplan et al., 2008Kernan et al., 2012Nolan et al., 2004Pavone et al., 2001Conde−Guzon et al., 2012 Fig. 2 Forest plots of the single-arm meta-analyses (weighted average) and meta-analyses of the mean difference (difference in performance in case-control studies). For each study the mean is represented by a square (size is proportional to the study's weight) and the 95% confidence interval (CI) is represented by a horizontal line. The overall weighted mean is represented by a diamond shape Furthermore, Guerrini et al. reported a higher prevalence of dysgraphia in AE (21% vs. 8% for controls) and reported diminished overall performance in a handwriting fluency test compared to controls (2015). Conde-Guzon and Cancho-Candela used the LURIA-battery to assess motor and sensory functions compared to control subjects (2012). They reported significantly lower performance in the manual subtest, but verbal regulation did not differ from normative values. Sensory functions such as rhythmic hearing and tactile subtests were significantly worse compared to controls, while kinesthesia and stereognosis did not differ from normative values.

Visuoperceptual, Visuospatial, and Visuoconstructional Function
Two studies assessed visuomotor planning and integration. Of the studies using the Beery-VMI, the largest study from Masur et al. (n = 106) reported a mean of 98.4 (SD 16.5), which is well within normal limits in children with CAE younger than 6 years of age (2013). Conant et al. included a smaller subset of patients (n = 16) and reported a significantly lower score of 89.3 (SD 6.3) compared to controls (2010). The single-arm random effect meta-analyses for the Beery-VMI was estimated at a mean of 93.85 (95% CI: 84.93-102.76; T 2 = 38.88; I 2 = 93.9%), which is not significantly different from a normative mean of 100 (Fig. 2).
On the contrary, Conant et al. found significant lower performance using the KABC-HM (imitation of hand movements) a test for visuomotor planning and integration (Conant et al., 2010).
Five studies have investigated visuospatial skills. In total three studies used the Rey-Complex Test (RCFT). Nolan et al. found significant worse performance in the RCFT compared to normative data but did not include the average scores (2004). Pavone et al. found lower performance in the RCFT compared to the control group, although still within normal clinical range (2001). Henkin, et al. did not find a significant difference in the RCFT with the control group (2005). Cheng et al. (2017) reported that children with CAE did not perform worse in a 3D mental rotation test, however, Conde-Guzon and Cancho-Candela did find significant worse performance in the visuospatial subtest of the LURIA-DNI neuropsychological battery (2012).
Two studies assessed visual search ability. Levav et al. used a Letter Cancellation test and reported a large difference in completing the test compared to controls, however, the age difference between patients with CAE and controls was 12 years (2002). In a study by Mostafa et al. total duration of a Visual Search Test was significantly worse compared to controls, however, controls were 6 years older on average (2014).
Schraegle et al. reported intact verbal memory and list learning compared to normative data using the California Verbal Learning Test (auditory memory) (2016). However, Kernan et al. (2012) and Henkin et al. (2005) did find significant memory impairment using this test compared to control subjects. Kernan et al. reported a significant difference in the total mean score, whereas Henkin et al. reported significant differences in immediate recall, delayed recall, and retrieval.
Findings on the recall of the Rey-Osterrieth Complex Figure Test (visual memory) did not differ compared to controls, although Nolan et al. reported significant worse performance compared to normative data (Henkin et al., 2005;Lopes et al., 2014;Nolan et al., 2004).
In  (Gencpinar et al., 2016). Number recall also seems diminished in the study of Mostafa et al., however controls were on average 6 years older (Mostafa et al., 2014).
Lopes et al. reported normal performance on LIST learning for words (2014). Masur et al. reported a Sentence Repetition mean standard score of 9.2 which is significantly lower compared to a normative mean of 10 in children with CAE younger than 6 years of age (2013). Conde-Guzon and Cancho-Candela reported impairments in immediate (short-term) memory and logical memory in typical absence epilepsy compared to controls (2012). The performance on several other memory tests (Corsi Block Tapping Test (D'Agati et al., 2012;Lopes et al., 2014), Visual Aural Digit Span Test (Gencpinar et al., 2016), Rey Auditory Verbal Learning Test (Levav et al., 2002), Spatial Memory Test (Mostafa et al., 2014), Incidental Verbal Memory Test (Mostafa et al., 2014), STIM-tasks (Siren et al., 2007) and Doors and People (Kernan et al., 2012) did not differ in performance from control subjects.
In children with CAE, 40-50% had more than 1-year delay in school grade levels compared to 10% of the control group. Conde-Guzon and Cancho-Candela found significant worse performance in writing, reading, numerical structure and arithmetic abilities compared to controls using the LURIA-DNI battery (2012).

Reports on Attention or Attentional Deficiencies (by Proxy)
Vega et al. reported attentional problems, especially forgetfulness and distractibility to be more prevalent in children with CAE compared to controls using the Behavior Assessment System for Children (2010).
The prevalence of attentional problems in 38% of children with AE differed significantly compared to 16% in control subjects assessed with the Child Behavior Checklist (CBCL) in the study by Caplan et al. (2008). However, Shinnar et al. (2017) reported a lower percentage of 15% clinically significant attentional problems in CAE compared to the study of Caplan et al. (2008) but used a higher cut-off value. Conant et al. (2010)

Discussion
The aim of this review was to systematically assess the literature on cognitive performance in AE. Children with AE are regarded to have cognitive functioning within normal range (Adie, 1924;Currier et al., 1963). Nevertheless, we found multiple studies reporting lower cognitive performance across a wide spread of cognitive domains. However, the exact degree of impaired cognitive functioning is difficult to estimate as the methodologies across studies vary and multiple neuropsychological tests have been used, which hampers comparisons between studies. Moreover, it is currently difficult to distinguish momentary effects on cognitive performance during the active stage of AE from long-lasting effects on cognitive functioning, as most studies reported on cognitive performance at different time-points after seizure onset. Some studies tested prior to introduction of anti-epileptic drug treatment, while others reported on a mixed population of children on or off anti-epileptic drug treatment and with or without ongoing seizures.

Intelligence
Full-scale IQ was estimated to be approximately three points lower on average compared to normative values. It is important to note that, although average performance in intelligence measures are statistically lower, the pooled averages still fall well within normal values. However, in case-control studies, the mean difference is larger with a difference of~8 points in full-scale IQ,~9 points in verbal IQ and~5 points in performance IQ. There may be several explanations for these results. The mean IQ of the control subjects was significantly higher than normative values, which may simply resemble a higher average IQ in the studied population or geographical area. However, it may be due to exclusion of patients with a low IQ in several case-control studies (Caplan et al., 2008;D'Agati et al., 2012;Gencpinar et al., 2016;Henkin et al., 2005;Kernan et al., 2012;Lopes et al., 2013). Another explanation for a higher IQ in controls may be bias due to convenience sampling (e.g. children from academics), although this was not evident based on the methods used for the recruitment of control subjects in these studies. However, the estimated mean IQ in the single-arm meta-analysis in patients with AE is less subject to bias due to sampling error as a far larger proportion of the total population is being tested. Moreover, the estimated true variance (T 2 ) for the single-arm meta-analyses was totally dependent on the three studies (Caplan et al., 2008;Conde-Guzon & Cancho-Candela, 2012;Kernan et al., 2012) with the highest mean IQ scores in AE and controls. The estimated true variance (T 2 ) was small for the pooled difference in case-control studies.

Executive Function
Lower than average performance in executive functioning was noted in cognitive flexibility, planning and verbal fluency (Cheng et al., 2017;Conant et al., 2010;D'Agati et al., 2012;Gencpinar et al., 2016;Henkin et al., 2005;Kernan et al., 2012;Levav et al., 2002;Masur et al., 2013). However, not all tests were indicative of lower executive functioning as results on the STROOP test did not differ.

Attention
There are clear indications for a lower performance in attention, such as sustained attention, selective attention and divided attention. From our meta-analysis, we can conclude that trail making A and B scores take significantly more time to perform by children with AE compared to controls, especially when attentional shifts (divided attention) are necessary. The estimated true variance (T 2 ) for the trail making test B was high, however, this may be explained by differences in study design, as the study by Levav et al. reported in children ≥13 years of age, which would probably require less seconds to finish the test than younger counterparts in the other two studies. Furthermore, in the study by Masur et al. sustained attention was mostly affected due to attentional lapses (errors of omission) rather than reflecting disinhibition (errors of commission) (2013).

Language
Results on specific language tests are of particular interest, as verbal IQ was estimated to be lower in case-control studies, and a relatively low mean verbal comprehension index (in children >6 years of age) was reported in the study by Masur et al. (2013). However, receptive vocabulary was not affected with the Peabody Picture Vocabulary Test in the study by Masur et al. (Masur et al., 2013). Two studies with a decent sample size raise concerns regarding expressive language as found by Caplan et al. (although average functioning is still within normal clinical range) and Conde-Guzon & Cancho-Candela (Caplan et al., 2008;Conde-Guzon & Cancho-Candela, 2012). However, these two studies were characterized by relatively high verbal IQ's in controls. Therefore, data on language tests in AE remains inconclusive, but warrants further research.

Motor Function
Studies on simple motor tasks were inconclusive (Conant et al., 2010;Henkin et al., 2005;Siren et al., 2007). Complex motor tasks may be impaired but were only assessed in one study (Conant et al., 2010) and the control group was characterized by higher scores than the normative mean. Furthermore, one study found a higher prevalence of dysgraphia and diminished performance in handwriting fluency. These findings were associated with abnormal neurophysiological findings, which led the authors to conclude that these patients had a form of dystonic dysgraphia. Interestingly, 12 out of 17 patients with initial dysgraphia were re-tested 5-years later and showed resolution of dysgraphia and improved handwriting skills. In addition, Conde-Guzon and Cancho-Candela also reported worse performance in a writing test (2012).

Visuoperceptual, Visuospatial, and Visuoconstructional Function
Visual-motor integration from the Beery-VMI test did not differ from the normative mean in our meta-analysis, although only two studies were available. The lower mean score in the Beery-VMI in the study by Conant et al. may be due to a small study effect. Studies on visual search tests and visuospatial skills remain inconclusive (Cheng et al., 2017;Conde-Guzon & Cancho-Candela, 2012;Henkin et al., 2005;Levav et al., 2002;Mostafa et al., 2014;Nolan et al., 2004;Pavone et al., 2001). Indeed, visual-spatial thinking ability in the meta-analysis of Loughman et al. in idiopathic generalized epilepsies did also not differ significantly in case-control studies (2014).

Learning and Memory
Results on memory have yielded inconclusive results. The largest study on memory by Masur et al. suggests lower visual memory function, although still within normal clinical range (2013). However, other studies on non-verbal memory tests have overall found average performance (Conant et al., 2010;D'Agati et al., 2012;Gencpinar et al., 2016;Henkin et al., 2005;Kernan et al., 2012;Lopes et al., 2014;Nolan et al., 2004;Pavone et al., 2001;Siren et al., 2007). Furthermore, studies on verbal memory mostly showed normal performance (Levav et al., 2002;Lopes et al., 2014;Schraegle et al., 2016). Nevertheless, some studies using subtests for memory for stories and with the California Verbal Learning Test did find lower performance compared to controls (Conant et al., 2010;Henkin et al., 2005;Kernan et al., 2012). However, the observed differences with controls in these studies might suggest a sample bias related to a higher functional level of the controls. Only, one study found an overall lower than average performance in memory functioning, especially in short-term auditory, visual memory and logical memory, although this study was also potentially characterized by a better than average control group as reflected by the performance of the control group on the WISC-R (Conde-Guzon & Cancho-Candela, 2012). Sentence repetition was lower compared to the normative mean in the study by Masur et al. (2013), although scores still fall within normal clinical range and may be related to attentional deficits.

Achievement
Arithmetic ability may be vulnerable based on our metaanalysis of the WRAT-3 test, although test scores fall within the normal clinical range. Nevertheless, arithmetic ability was also worse compared to controls in children with AE in the study by Conde-Guzon and Cancho-Candela (2012). Evidence regarding reading ability, is contradictory, as the meta-analysis on the WRAT-3 test was not significantly lower, but other reading tests by Vanasse et al. (2005) and Conde-Guzon & Cancho-Candela (2012) did report worse performance compared to controls.

By Proxy (Parent-Reported Functioning)
Studies using parental questionnaires also point towards attentional problems and a higher prevalence of attentional deficit hyperactivity disorder.

Prevalence of School Difficulties
School difficulties seem more prevalent in patients with AE than in the normal population, as the three largest studies reported school difficulties in 23% to 52% (Berg et al., 2014;Covanis et al., 1992;Urena-Hornos et al., 2004).

Prevalence of Neuropsychological and/or Neurodevelopmental Problems
A p r e v a l e n c e o f n e u r o p s y c h o l o g i c a l a n d / o r neurodevelopmental problems were found in approximately 22% to 32% of the patients with CAE (Berg et al., 2014;Fastenau et al., 2009;Sinclair & Unwala, 2007;Wirrell et al., 1996).

General Discussion
Cognitive deficits in one area may be related to the performance in another cognitive domain. Masur et al. reported a direct sequential effect among attention, memory, executive function, and academic achievement (2013). Therefore, the emergent pattern of clear attentional deficits in a proportion of children with absence epilepsy may influence performance in other cognitive domains. Nevertheless, the overall pattern suggests vulnerabilities in intelligence, attention and executive function. Less conclusive results were found for (expressive) language, motor function, visuo-perceptual functioning and learning & memory. Ultimately, vulnerabilities in cognitive domains may impact neurocognitive development and lead to more academic difficulties.
How children with AE mature into adulthood is far less researched. The few studies that have investigated cognitive function during adulthood in patients with AE suggest that lower performance on neuropsychological tests may persist, but current studies are small in sample size. A study in (n = 10) adults with CAE in remission reported a full-scale IQ of 92 (69-99), a performance IQ of 85 (66-117), with particularly low scores in the picture arrangement, block design and object assembly subtests of the WAIS-R (Hommet et al., 2001). Language and executive function were also tested using specific tests, but did not differ with a control group. Recently Loughman et al. reported on cognitive function using the Woodcock-Johnson-III test of cognitive abilities in an adult population of genetic generalized epilepsy syndromes, including CAE (n = 10) and JAE (n = 21) patients (2017). This study reported lower scores on brief intellectual ability, crystallized intelligence, new learning/memory and speed of processing (Loughman, Bowden, & D'Souza, 2017). In conclusion, neurodevelopment may ultimately differ in AE, as a study also found abnormal cortical thickness connectivity in various regions after in young adults with a history of CAE (Curwood et al., 2015).
What the impact may be on their working careers and psychosocial well-being is another important question. In a telephone follow-up study, 25% of 52 retrospectively identified patients of 11 to 36 years old with CAE had a history of psycho-pedagogical help (Martinez-Ferrandez et al., 2017). Another study in patients focusing on long-term psychosocial outcome in juvenile myoclonic epilepsy (JME) with a mean age over 60 years used AE patients as an epilepsy matched-control group (Holtkamp, Senf, Kirschbaum, & Janz, 2014). To the surprise of the authors, a significantly lower amount of patients with a history of AE accomplished a university degree compared to controls. Nevertheless, the overall psychosocial outcome of this study was favorable. However, another study compared a group of patients with rheumatoid arthritis to patients with CAE and similar impact of "chronic disease" (Camfield & Camfield, 2007). Young adults with a history of CAE were more prone to working in jobs requiring minimum education, having more behavioral problems and psychiatric consultations. Therefore, the potential impact later on in life cannot be underestimated and deserves further scientific attention.
Cognitive findings may also be related to (inter)ictal activity, pharmacological effects or another underlying (genetic) vulnerability, but disentangling all factors may not be feasible. The use of valproate was associated with attentional dysfunction in the study by Masur et al. (Cnaan et al., 2017;Dlugos et al., 2013;Glauser et al., 2010;Masur et al., 2013). In addition, seizure duration of more than 20 s correlated well with attentional dysfunction and errors of omission . In the prospective cohort study of Caplan et al. verbal IQ was associated with duration of illness and AED treatment. Moreover, ADHD was associated with the duration of illness and seizure frequency (2008). However, these associations have not been consistent across the literature.

Limitations
We opted to use single-arm meta-analyses and/or a metaanalysis of the difference between cases and controls. This was chosen as sampling error may be of greater influence in a relatively small sample of the normal population, than it would be to estimate the true mean in AE, as a larger proportion of the group is likely to be included in a given geographical area. In addition, some studies did not match a control group specifically to children with AE as other types of epilepsy were also included. Therefore, residual confounding may still be present in those studies where there is a large age difference between patients with AE and controls. Some studies may have included patients with AE, which may have concomitant features, such as myoclonic jerks or atypical features. Also, some of the included studies focused on multiple epilepsy groups and therefore the findings of sub-analyses in patients with AE were often underpowered. Furthermore, some studies were relatively small in sample size and may be subject to a small-study effect. Importantly, multiple comparisons may have yielded false-positive findings (type 1 errors) in the included studies. Furthermore, the Flynn effect, which is a tendency of IQ to rise by~3 points per decade on average, may have a small but significant bearing on our results (Trahan, Stuebing, Fletcher, & Hiscock, 2014). Unfortunately, inconsistent reporting of neuropsychological test results across studies hindered comparison between studies and pooling of results. Granting that cognitive domains may overlap, assigning neuropsychological tests to a specific cognitive domain may lead to an oversimplified view.

Future Work
Future work should focus on large (multi-center) long-term observational studies to assess cognitive development and to identify children with AE most at risk for neuropsychological co-morbidity. Importantly, this may clarify whether repeated neuropsychological tests could be helpful and how cognitive development differs from healthy peers. Emphasis should be made to characterize children with AE according to semiology, EEG characteristics, and epilepsy syndrome. Especially separate reporting in JAE is lacking. Early identification of children with concomitant cognitive dysfunctions may help health care workers to initiate additional supportive programs, as well as communicate vulnerabilities and discuss teaching strategies with their school. A broad neuropsychological test battery, focusing on executive functioning, attention and verbal abilities should be used. Ideally, this should ideally be done before the start of anti-epileptic drugs and repeated at least once to track cognitive development. Given that cognitive dysfunctions may be case-specific, personalized supportive programs may be most useful, which may be discussed in a multidisciplinary team, consisting of at least a neurologist, neuropsychologists and a school expert.