Long-term characterization of cognitive phenotypes in children with seizures over 36 months

Rationale: Children with new-onset epilepsies often exhibit co-morbidities including cognitive dysfunction, which adversely affects academic performance. Application of unsupervised machine learning techniques has demonstrated the presence of discrete cognitive phenotypes at or near the time of diagnosis, but there is limited knowledge of their longitudinal trajectories. Here we investigate longitudinally the presence and progression of cognitive phenotypes and academic status in youth with new-onset seizures as sibling controls. Methods: 282 subjects (6 – 16 years) were recruited within 6 weeks of their first recognized seizure along with 167 unaffected siblings. Each child underwent a comprehensive neuropsychological assessment at baseline, 18 and 36 months later. Factor analysis of the neuropsychological tests revealed four underlying domains – language, processing speed, executive function, and verbal memory. Latent trajectory analysis of the mean factor scores over 36 months identified clusters with prototypical cognitive trajectories. Results: Three unique phenotypic groups with distinct cognitive trajectories over the 36-month period were identified: Resilient , Average , and Impaired phenotypes. The Resilient phenotype exhibited the highest neuropsychological factor scores and academic performance that were all similar to controls; while the Impaired phenotype showed the polar opposite with the worst performances across all test metrics. These findings remained significant and stable over 36 months. Multivariate logistic regression indicated that age of onset, EEG, neurological examination, and sociodemographic disadvantage were associated with phenotype classification. Conclusions: This study demonstrates the presence of diverse latent cognitive trajectory phenotypes over 36 months in youth with new-onset seizures that are associated with a stable neuropsychological and academic performance longitudinally.


Introduction
Amongst children with new-onset epilepsy, it is well-documented that a myriad of neuropsychiatric and neurobehavioral co-morbidities may be evident, which is also characteristic of the disorder [1][2][3][4][5].Cognitive impairment is one of the more common co-morbidities and often pre-dates the first diagnosed seizure [6][7][8].Executive dysfunction, poor attention, impaired working memory, prolonged processing speed, expressive and receptive language delays, and verbal and visual memory concerns are common cognitive concerns in these children [9][10][11].These impaired cognitive domains can adversely affect learning ability and focus at school, which can be quite disruptive to academic and school performance.Accordingly, diminished academic performance may also predate the first known seizure indicating that academic performance and cognitive impairment cannot fully be explained by the effects of overt seizure activity [12][13][14].This impairment is likely due to a multitude of risk factors including the effect of seizures on brain development, sociodemographic factors, and underlying genetic predispositions to epilepsy [15,16].
Even though children with epilepsy are known to be at greater risk for cognitive impairment and other comorbidities than their healthy counterparts [17], the degree to which each child is affected varies and is yet to be fully characterized [15].While some children may function within the typically developing scope, others exhibit significant impairment.A growing body of literature has indicated consistent evidence of heterogeneity in the extent of cognitive impairment in both children and adults with epilepsy, which can be characterized into cognitive phenotypes [18][19][20][21][22][23].Although classifying children by epilepsy type is helpful in seizure management, cognitive impairment is not uniform across children diagnosed with the same epilepsy syndrome or across different epilepsy syndromes [12].The lack of classification based on cognitive patterns alone rather than epilepsy syndrome/type limits the effectiveness of interventions by physicians, caretakers, and educators.Identifying underlying latent cluster profiles along with potential risk categories can be vital for accurate understanding and treatment of epilepsy.
Characterizing cognitive phenotypic categories is an approach that has been utilized across a variety of fields including Alzheimer's disease [24], 22Q11.2deletion syndrome [25], pediatric asthma [26], and multiple sclerosis [27] as well as in the field of epilepsy.Amongst different epilepsy syndromes including generalized idiopathic epilepsy [22], frontal lobe epilepsy [28] and temporal lobe epilepsy [2], these unique cognitive phenotype patterns often have associated distinct clinical epilepsy characteristics and sociodemographic factors [12].Overall, cognitive phenotyping has primarily focused on the adult epilepsy population with fewer phenotyping analyses among children with new-onset epilepsy, and often focused on quality of life [19,22,23,29,30].In addition, cognitive phenotyping in epilepsy has focused primarily on cluster patterns at a specific time with limited focus on longitudinal patterns [31,32].Latent trajectory analysis is a unique approach that identifies clusters based on similar longitudinal trajectories that remain stable and persistent over time [23,[33][34][35].This approach uses unsupervised machine-learning techniques to demonstrate the presence of distinct long-term patterns.By applying this latent longitudinal approach to children with new-onset seizures, children can potentially be classified into cognitive phenotypes early and receive the appropriate interventions based on their predicted trajectory.
Currently, there is no knowledge of the subsequent course or trajectories of these cognitive phenotypes in children with new-onset epilepsy.Here we investigate the presence of latent trajectories of cognitive and academic status in youth with new-onset seizures over 3 years.Capturing the nature and implications of this variable susceptibility to significant cognitive co-morbidities is an important clinical and research topic as unique predictors and biomarkers of cognitive dysfunction in epilepsy can be a significant advance for the treatment of cognitive problems in this disorder.The goal of this study was to incorporate the unique approach of latent trajectory analysis to characterize the underlying course of these cognitive phenotypes over time.In addition, we investigated the relationship of these phenotypes to baseline clinical epilepsy and baseline sociodemographic characteristics.We also evaluated potential predictors of these phenotypic groupings.We hypothesize that there are discernable groups characterized by varying presence and severity of cognitive problems.In addition, the latent trajectory of each unique group will remain stable and persistent over a prolonged period.We further hypothesize that the cognitive phenotypes will be characterized by differing baseline clinical epilepsy characteristics and sociodemographic characteristics; which will vary across the groups and serve to tease out the inconsistencies that have been noted in the literature.Specifically, we hypothesize that characteristics associated with higher seizure burden and higher levels of sociodemographic disadvantage would be associated with latent trajectory groups with poorer cognitive performance.To our knowledge, this is the first study of its type addressing the long-term trajectory of cognitive phenotyping in children with seizures and its correlates over a 36-month period.

Participants
Study participants included children with newly diagnosed seizures, their siblings as controls, and their primary caregivers in each household [36,37].The core investigation was conducted at Indiana University and Cincinnati Children's Hospital at the University of Cincinnati between 2001 and 2008.Children were recruited through electroencephalogram (EEG) laboratories, emergency departments, and pediatric neurologists in two large children's hospitals (Indianapolis and Cincinnati) and from practices of private pediatric neurologists in Indianapolis.When children met the criteria, refusals were less than 10 %.All children with newly diagnosed seizures in this sample met International League Against Epilepsy criteria for epilepsy [38].
Exclusion criteria for both children with seizures and siblings were: a co-morbid chronic physical disorder, intellectual disability (based on either clinic records or parent report), or seizures precipitated by an acute event (e.g., intracranial infection, metabolic derangement, and recent head injury).Children who had had two or more febrile but no afebrile seizures or who were placed on daily antiseizure medication (ASM) after a febrile seizure were also excluded.In addition, children with infantile spasms (hypsarrhythmia), electrical status epilepticus in sleep and epilepsy with continuous spike-wave during slow wave sleep were excluded from the study.Parental informed consent and child assent were obtained prior to data collection.Siblings did not have epilepsy and were not on medication that could affect mental status.The study was approved by the institutional review boards at Indiana University and Cincinnati Children's Hospital Medical Center.
A total of 349 children with seizures were recruited within 6 weeks of their first recognized seizure (Mean = 35 days).Of the 349 children who agreed to participate in the study, 23 scored below an IQ of 70 on screening, 11 did not provide any data, and 33 were unable to complete testing (e.g., typically scheduling/travel), for a final sample of 289 [6].There were no significant differences between those who completed neuropsychological testing and those who did not on age, sex, race, or socioeconomic status (p > 0.10).For the 36-month longitudinal analysis, 228 of the 289 children completed at least one follow-up visit.Overall, there were a total of 282 children with epilepsy in the baseline group, 245 children at 18 months (18-month attrition rate of 15 %), and 228 children at 36 months (36-month attrition rate of 7 %).
The sibling control sample was a comparison group of 167 healthy siblings of the children with epilepsy.Only one sibling was recruited per family.For each child in the seizure group, we attempted to recruit a healthy sibling age 2-18 years (preferring ages 6+ for cognitive testing).If there were multiple siblings, the sibling that was closest in age to the child with the seizure was included in the study.When the sibling was too young (<6 years), had another chronic condition (e.g., asthma), or was too old (>18 years), he/she was not included in the study.There was minimal difficulty recruiting siblings when siblings were available.Of the 252 eligible siblings, 50 were too young (<6 years) to complete neuropsychological testing and 35 others could not travel to the medical center for testing, resulting in a final sample of 167 sibling controls.Overall, there were 167 sibling controls in the baseline group, 147 siblings at 18 months (18-month attrition rate of 12 %), and 147 siblings at 36 months (36-month attrition rate of 0 %).
Data were first collected within 6 weeks of the first recognized seizure (baseline; B) from both children with newly diagnosed seizures and siblings.All participants were followed prospectively and reassessed 18 months later (M18) and finally, 36 months later (M36).All data were included in the analysis regardless of the number of visits completed.

Measures
Cognitive Evaluation -All participants completed a comprehensive neuropsychological evaluation that included standardized clinical measures of intelligence, language, immediate and delayed verbal and visual memory, executive functions, speeded fine motor dexterity, and academic achievement at baseline, M18 and M36.The specific administered tests included: Clinical Evaluation of Language Fundamentals, 3rd Edition (CELF-3) [39]; Comprehensive Test of Phonological Processing (CTOPP) [40]; Conners' Continuous Performance Test, 2nd Edition, (CPT-II) [41]; Kaufman Brief Intelligence Test (K-BIT) [42]; Coding and Symbol Search subtests of the Wechsler Intelligence Scale for Children, 3rd Edition (WISC-III) [43]; Wide Range Assessment of Memory and Learning (WRAML) Design Copy [44]; and the Wisconsin Card Sorting Test (WCST) [45].Testing was administered by psychometrists who were trained, observed, and certified on the test battery and its scoring by a pediatric neuropsychologist [6].
To assess general intellectual ability, the full-scale K-BIT IQ score was used.All youth had an IQ equal to or greater than 70.In addition, each test was administered according to standardized procedures and scores were converted to age-corrected standardized scores using the best available national norms for all tests except WRAML Design Copy, which was designed by this study's research group [44]; this test was normed internally, using our own sample to generate age-corrected scores.
Factor analysis of this neuropsychological test data revealed four underlying factors: (1) Language, (2) Processing Speed, (3) Executive Function/attention/construction (EF), and (4) Verbal Memory and Learning [46,47].The Language factor consisted of measures of verbal concept formation, phonological awareness, and phonological memory.The Processing Speed factor consisted of measures assessing psychomotor speed and rapid naming.The Executive Function (EF) factor consisted of measures assessing sustained attention, problem solving, and visual-construction.The Verbal Memory and Learning factor consisted of measures of rote verbal learning and story recall.Higher factor scores indicate better neuropsychological performance [6].Briefly, the rotation method used was promax and the factor extraction method used was principal axis factoring [45].
Academic Achievement -All children and sibling controls were assessed using three subtests of the Woodcock-Johnson Revised Tests of Achievement (WJ-R): Letter-Word Identification, a measure of word reading; Calculation (a measure of math computation skills) and Dictation (a measure of spelling, punctuation, and syntax in writing) [48].Standard scores for age were generated from national norms for that test.
Seizure characteristics and sociodemographic data (e.g., caregiver's highest education level, caregiver's household income, child's age of onset of seizures, and child's sex) were collected via structured interviews by trained research coordinators as well as psychometrists.Clinical seizure variables including seizure classification, results of EEG and imaging were collected from the electronic medical record and were coded independently by study physicians blinded to the cognitive data.The sociodemographic assessment was conducted at the baseline (B) visit only.

Statistical analysis
To identify distinct patterns of cognitive performance change over a 36-month period, the four factor scores were included in an analysis of latent group-based trajectory modelling (LGBM) of longitudinal data, which was carried out by SAS® Proc Traj [48,49].LGBM captures the heterogeneity of subgroups among a specific population by simultaneously estimating several trajectories as opposed to fitting an overall population mean.For each participant, the mean of all four cognitive factor scores was collated for each visit (B, M18, & M36) and utilized as the cognitive evaluations over the three-year period.In order to find the optimal number of trajectories, Bayesian Information Criterion (BIC) compares the fitness of models between trajectories with a differing number of groups or between different shapes of a trajectory.We required a minimum of 10% of the overall sample size for each of the trajectory groups.Multivariate regression was utilized to identify the significant risk factors for each of the cognitive phenotypes in the trajectory analysis.The risk factors assessed in the analyses included clinical seizure characteristics (epilepsy syndrome (0 = primarily generalized, 1 = localization-related), EEG results (0 = normal, 1 = abnormal), MRI results (0 = normal, 1 = abnormal), neurologic examination at baseline (0 = normal, 1 = abnormal), age of onset of first recognized seizure, seizure frequency (number of seizures/year), and percent taking first anti-seizure medications (ASM) as well as demographic characteristics (age, sex, education, handedness, and sociodemographic disadvantage score).The Sociodemographic Disadvantage (SD) score is an index based on four sociodemographic variablesmother's education level, race (self-identified), household income, and marital status.Briefly, families classified into the SD-3 category were primarily non-white and showed the lowest levels of income, caregiver education, and married parental status, indicating highest levels of sociodemographic disadvantage; while families who fell into the SD-0 category were all white and showed the highest levels of income, caregiver education and married parental status, indicating lowest levels of sociodemographic disadvantage [16].
The level of significance α=0.05 was used for the multivariate logistics regression.LGBM data analyses were all conducted using SAS® version 9.4.
Additional analyses were conducted using the Statistical Package for Social Sciences (SPSS) software (Version 27.0, IBM, Chicago IL).Using SPSS, one-way analysis of variance (ANOVA) compared cognition in children with new onset seizures with sibling controls and compared cognitive phenotype groups by clinical seizure characteristics and academic performance at each time point (B, M18, M36).When the F statistic was significant, Tukey Honest Significant post-hoc comparisons were conducted among the cognitive phenotypes.Chi-square test was utilized to compare categorical variables across groups.The data was normally distributed so parametric testing was appropriate for analysis.In addition, power calculations were based on a larger investigational study.

Sample clinical Characteristics
Table 1 summarizes demographic characteristics for both groups (children with seizures and siblings) and clinical seizure characteristics in the seizure group.Briefly, a total of 282 children with newly diagnosed seizures aged 6-16 years and 167 sibling controls were included in the analyses.There were no significant differences between the groups.The clinical epilepsy characteristics indicate that the children with seizures in this sample had an average age of onset of seizures of 9.35 years and about 65% of the seizure group was comprised of focal epilepsy syndromes.The five most frequently prescribed ASMs were lamotrigine, oxcarbazepine, carbamazepine, phenytoin and valproic

Table 1
Sample Characteristics for Seizure and Sibling Control Groups.No significant differences between Seizure and Sibling Groups on any demographic variable.Data presented as mean (SD).SD = standard deviation, %=percent, M/F = male/female.

Cognitive performance over 3 years
A comparison of the cognitive factor scores between children with new-onset seizures and sibling controls across the three-year period shows significant differences across all cognitive domains (Fig. 1).Children with new-onset seizures perform significantly poorer across almost all cognitive domains compared to sibling controls.This pattern remains stable and consistent over the three-year period.

Latent trajectory Analysis
Latent trajectory analysis was utilized to evaluate unique longitudinal patterns amongst children with new-onset seizures over 36 months using the mean cognitive factor scores.This analysis approach resulted in 3 distinct phenotype groups -Resilient (N = 121, 41.9 %), Average (N = 129, 44.6 %) and Impaired (N = 39, 13.5 %) (Fig. 2).The Resilient group performed better than the Average and Impaired groups and performed similarly to the sibling control group consistently.The Average group performed only slightly below normal (Z = 0) consistently, while the Impaired group consistently averaged around 1.5 standard deviations below the normal range.Of note, latent trajectory analysis was also conducted to assess for phenotype patterns for individual cognitive domains (Language, Processing Speed, Executive Function and Verbal Memory).Findings showed similar results to the mean cognitive factors scores with language (18.3 %) and processing speed (19.4 %) showing the highest number of patients in the impaired phenotypes (Fig. 3).In addition, latency trajectory analysis was also conducted to assess for distinct phenotype clusters in 4 subtests -CELF Total Language (Expressive and Receptive), CPT Processing Speed (Reaction time), WCST Executive Function (% Errors T scores), and WRAML Verbal Learning/Memory.The results, once again, found three distinct phenotypes for each subtest with language (15.7 %) and processing speed (17.9 %) showing the highest number of patients in the impaired phenotypes (Supplementary Fig. 1).

Cognitive phenotype and factor scores over 36 months.
Using the specific cognitive domain factor scores incorporated in the primary latent trajectory analysis, Table 2 shows the differences between the three phenotype groups.The resilient group consistently performs ~ 0.5 to 1 SD above the average range (zero) while the Impaired group consistently performs ~ 1 to 1.75 SD below the mean in all the categories.This finding is consistent over a 3-year period.
Cognitive phenotype test scores over 36 months.To understand the specifics of cognitive performance amongst each distinct cognitive phenotype, individual cognitive tests performance was examined (Fig. 4).Individual cognitive tests were scaled using standardized scores -Mean = 100, Standard Deviation = 15.The Impaired group is globally affected in all cognitive domains, especially in attention and processing speed.This finding remains significant over the 36-month period.Similar consistency in performance across all measures, neither deteriorating or improving over time, is also noted in the Resilient and Average groups over three years.

Clinical and sociodemographic characteristics of the 3 cognitive phenotypes
Notably, baseline clinical epilepsy characteristics generally show that there are significant differences amongst the cognitive phenotypes in age of onset of seizures, EEG findings and neurological exam findings (Table 3).The Tukey posthoc tests indicate that the children in the Impaired cognitive phenotype are more likely to have a younger age of onset of seizures and exhibit abnormalities on EEG and on neurologic exam compared to the Average and Resilient cognitive phenotypes.There was also a trend towards significance in biological sex at testing.Sociodemographic Disadvantage was also significantly different amongst phenotypes.Specifically, the Resilient cognitive phenotype is Predictors of Phenotype Class Membership: Using multinomial logistical regression analyses, we determined which clinical epilepsy and sociodemographic factors best predict phenotype class membership (Table 4).The Average Cognitive Phenotype was used as the reference group.Compared to the Average phenotype, age of onset, baseline neurological examination, and sociodemographic disadvantage played a significant role in predicting class membership.Specifically, children in the Impaired phenotype had a younger age of onset, were more likely to have an abnormal neurological examination at baseline, and more likely to come from a more disadvantaged sociodemographic background compared to the average phenotype.On the other hand, children in the Resilient phenotype had an older age of onset, were more likely to have a normal neurological examination at baseline, and were more likely to come from a less disadvantaged sociodemographic background compared to the average phenotype.

How do cognitive phenotype classes relate to academic performance over time
To determine how the cognitive phenotype groups translate to school performance, academic achievement tests as well as global intellectual ability were evaluated in relationship to these phenotype groups.During the 36-month period, academic achievement (Woodcock Johnson) and global intellectual ability consistently remains significantly different between the cognitively Resilient, Average, and Impaired phenotypes (Table 5).The large effect size indicates the relevance of these findings and remains within the moderate to large range in all testing categories.

Discussion
Heterogeneity in neuropsychological status among youth with new onset epilepsies was investigated in this large cohort with the goal to characterize latent prospective cognitive phenotypes and determine the stability of associated factors.These findings corroborate prior studies [21,51,52], while extending the literature by depicting long-term outcome trajectories.Much of the prior literature on cognitive phenotypes are cross-sectional in nature, leaving long-term trajectory patterns unknown.In our study, latent trajectory analysis classified children with new onset seizures into three groups based on cognitive ability over a 36-month period: Resilient, Average and Impaired phenotypes.These three distinct groups showed discrete differences in the severity of cognitive impairment and range of cognitive characteristics affected.Amongst the three phenotypes identified-the Resilient phenotype performed best and even comparable to sibling controls; the Average phenotype performed within the average range and the Impaired phenotype performed within a diminished capacity.Thus, while a pattern of generalized cognitive impairment appears to characterize youth with epilepsy overall, especially in traditional epilepsy versus control/sibling comparisons (Fig. 1) -latent trajectory analysis reveals discrete groups, several with very favorable profiles.The Resilient and Average groups together represented over 80 % of the total sample and performed at or above average suggesting that a significant proportion of children with epilepsy indeed exhibit intact neuropsychological function.On the other hand, ~15 % of the sample performed significantly below average (~1.5 SD below average) which serves to significantly plunge mean cognitive findings such as those depicted in Fig. 1.These findings also suggest that a specific subset of epilepsy patients would benefit from focused early intervention and/or accommodations.Our findings furthermore indicate that these significant differences between the three phenotypes persist across all cognitive domains and over 36 months, indicating the stability of the phenotypic groupings with no evidence of either deterioration or improvement.This implies that it may be critical to identify youth at risk as soon as possible after diagnosis as their performance on general intellectual ability, specific cognitive domains, and school academics will remain essentially unchanged in the context of usual standard epilepsy care.
Risk factors that predicted these phenotype trajectories included neurologic exam and EEG abnormalities at the baseline visit and earlier age at the time of first seizure.These risk factors could potentially be employed to delineate those individuals who might encounter cognitive challenges from those who would be less at risk.In addition to the clinical epilepsy characteristics that predicted cognitive phenotypes, sociodemographic background at baseline also played a significant role in predicting phenotypic trajectories.The sociodemographic background, which incorporates four sociodemographic factors -maternal education, household income, marital status, and self-identified race, was examined as a composite index score.The role of social determinants of health in epilepsy has become more evident in the past decade [16,53,54] and is an area that should be pursued further to fully understand the extent of its impact on epilepsy co-morbidities in addition to epilepsy treatment [55,56] and seizure development [57,58].
Application of unsupervised machine learning approaches to cognitive data in children and adults with epilepsy have served to underscore and harness the existence of cognitive heterogeneity in youth and adults with epilepsy.These phenotypic differences have received more attention amongst adults with epilepsy, however, emerging research in children with epilepsy has similarly informed the presence and nature of cognitive heterogeneity in youth with epilepsy and its associated predictors [32,33, 53,].This approach may become more promising and lead to insights underlying the inconsistencies/variabilities reported in cognitive dysfunction in children with epilepsy.The existing cognitive phenotype papers have been cross-sectional to date, which has limited the option for prognostication from prospective trajectories.Our findings extend the literature as we not only corroborate prior findings -that categorizing children with new-onset seizures into specific cognitive phenotypes to better appreciate their characteristics and associated risk factors can provide a more accurate understanding than examining the aggregate of patients as a whole [57], but we also identify significant risk factors and make more accurate prognostications using long-term evidence.Consequently, the field of epilepsy can utilize this technique to target specific individuals with more potential risk by the long-term trajectory/prognostic analyses.In addition, trajectories of this heterogeneity could potentially lead to interventions and treatments that can be tailored to each child's specific presentation for best long-term  outcomes [12].Within the field of pediatric epilepsy, latent trajectory analysis has been employed to guide treatment and prognostication [31,59,60].These studies have primarily illustrated the use of latent trajectory analysis to evaluate health-related quality of life in children with epilepsy.However, our findings here are among the first to use latent trajectory analysis to evaluate cognitive function in children with new onset seizures longitudinally.Phenotyping has helped to contribute to our understanding of epilepsy as a disorder with a wide scope of cognitive abnormalities ranging from a debilitating/very impaired subset of patients to a subset of substantial proportions of patients with very favorable profiles.This range of abnormality suggests a need for early identification of the subset/phenotype of patients requiring substantial attention, not just in the context of seizure burden.While all children with epilepsy are at a greater risk for cognitive impairment, classifying children by cognitive phenotype allows for more specific care that is focused on each child's most likely trajectory.
It is important to note that our study indicates that cognitive phenotypes are persistent and stable over 3 years, indicating that children can be targeted for specialized treatment based on their baseline assessment.This is specifically imperative for children in the impaired group who may require higher levels of intervention from caregivers, physicians and educators.With the rise of precision medicine, personalized treatment protocols are becoming integral to the practice of medicine [12].Rather than waiting for children in the severely Impaired group to decline in academic performance or fall behind their peers, efforts can be made to blunt the effects of their cognitive phenotype by intervening early with early intervention or accommodations.However, future research would have to specifically determine if these interventions do indeed improve cognition long-term, especially for those within the impaired phenotype.
This study has limitations that should be mentioned.First, evaluation of academic achievement was limited in scope.We also did not assess disorders such as dyslexia and attention deficit disorders, which can adversely affect cognitive and academic performance.Second, the specific epilepsy syndromes evaluated here were limited.We did not evaluate any epileptic encephalopathy syndromes and other disorders such as Lennox-Gastaut syndrome.As a consequence, the inferences of our findings are not generalizable to all pediatric epilepsies.The majority of our patients had focal epilepsy; however, about one-third of our patients had generalized seizures, which may have a genetic component and increases familial risk of epilepsy.Genetic testing was not performed in this study however, siblings with any known or suspected history of seizures or seizure-like activity were excluded.Furthermore, the cause and inciting factors that precipitated the epilepsy were not assessed and may have played a role in the cognitive findings we presented.In addition, the course and treatment details along with early intervention/ accommodation programs can vary between individuals and also over time; and can play a significant role in cognitive and academic performance.We do not have these data and could not include this information in our analyses.Finally, group-based trajectory modeling has a large-sample requirement; consequently, our study may be underpowered to detect other distinct subgroups and non-linear trajectories in this longitudinal cohort.
Future studies can be aimed at developing a method to easily identify cognitive phenotype groups at baseline to ensure that children at high risk are quickly identified.One facet of early identification may involve neuroimaging; as such identifying the relationship between neuroimaging and cognitive phenotypes should be explored.Additional studies should also focus on evaluating the best treatments and interventions for high-risk phenotypes.This should include randomized clinical trials to investigate which interventions have the greatest effect on cognitive function over time.

Table 3
Characteristics of the three cognitive phenotypes using clinical epilepsy characteristics at baseline.The Impaired cognitive phenotype shows significant differences in age of onset, EEG findings, neurologic exam findings and sociodemographic disadvantage compared to the other two phenotype categories.Data presented as mean scores (SD).*p < 0.05, **p < 0.001.

Fig. 1 .
Fig. 1.Cognitive Performance over a 3-year period.Children with new-onset seizures perform significantly poorer compared to sibling controls in all cognitive domains.This pattern remains persistent over the three visits.Data presented as mean factor scores with SE bars.*p < 0.05, **p < 0.001.

Fig. 2 .
Fig. 2. Latency trajectory analysis from the three time points (baseline, M18 (18 months from baseline) and M36 (36 months from baseline) resulted in three distinct phenotype clusters.The Resilient group performed similar to sibling controls over the 3-year period, while the Impaired group performed consistently ~ 1.5SD below normal range.

Fig. 3 .
Fig. 3. Latency trajectory analysis performed in each cognitive domain from the three time points (baseline, M18 (18 months from baseline) and M36 (36 months from baseline) resulted in three distinct phenotype clusters in each domains -Language, Processing Speed, Executive Function, and Verbal Memory.Phenotypes for each cognitive domain were similar with language (18.3 %) and processing speed (19.4 %) showing the highest number of patients in the impaired groups.

Fig. 4 .
Fig. 4. Cognitive Performance of the different cognitive phenotype groupings over 36 months.Data presented as Standardized Scores -Mean = 100, Standard Deviation = 15.Consistency is noted in both performance and period of time over the 3 years.SS -Symbol Search, RT -Reaction Time, %Persev -%Perseveration, PA -Phonological Awareness, PM -Phonological Memory, RN -Rapid Naming.
Other less commonly prescribed medications included levetiracetam, ethosuximide, zonisamide, and gabapentin.The epilepsy syndromes were divided into two groups: Primary Generalized (generalized tonic-clonic, absence, and myoclonic epilepsy syndromes) and Focal/ Localization-Related (focal unaware and focal aware seizures with or without secondary generalization).In this cohort, MRI abnormalities included multiple various abnormalities (e.g., bilateral or unilateral hippocampal atrophy/sclerosis, ventricular enlargement, volume loss,

Table 4
Predictors of class membership using baseline clinical epilepsy characteristics and sociodemographic disadvantage.Data presented as Standardized β coefficients, standard error, significance, along with confidence intervals.*p < 0.05, **p < 0.01.

Table 5
Academic performance and global intellectual ability by cognitive phenotypes.The differences remain significant by the cognitive phenotypes and remain consistent over 36 months.Data presented as mean scores (SD).**p < 0.001.