Published online: 05 Nov 2013.

Transcription

1 This article was downloaded by: [University of California, Los Angeles (UCLA)] On: 07 November 2013, At: 06:39 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: Registered office: Mortimer House, Mortimer Street, London W1T 3JH, UK Applied Neuropsychology: Child Publication details, including instructions for authors and subscription information: A Comparison of IQ and Memory Cluster Solutions in Moderate and Severe Pediatric Traumatic Brain Injury Nicholas S. Thaler a, Jennifer Terranova a, Alisa Turner a, Joan Mayfield b & Daniel N. Allen a a Psychology, University of Nevada-Las Vegas, Las Vegas, Nevada b Neuropsychology, Our Children's House at Baylor, Dallas, Texas Published online: 05 Nov To cite this article: Nicholas S. Thaler, Jennifer Terranova, Alisa Turner, Joan Mayfield & Daniel N. Allen, Applied Neuropsychology: Child (2013): A Comparison of IQ and Memory Cluster Solutions in Moderate and Severe Pediatric Traumatic Brain Injury, Applied Neuropsychology: Child, DOI: / To link to this article: PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content ) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at

2 APPLIED NEUROPSYCHOLOGY: CHILD, 0: 1 11, 2014 Copyright Taylor & Francis Group, LLC ISSN: print/ online DOI: / A Comparison of IQ and Memory Cluster Solutions in Moderate and Severe Pediatric Traumatic Brain Injury Downloaded by [University of California, Los Angeles (UCLA)] at 06:39 07 November 2013 Nicholas S. Thaler, Jennifer Terranova, and Alisa Turner Psychology, University of Nevada-Las Vegas, Las Vegas, Nevada Joan Mayfield Neuropsychology, Our Children s House at Baylor, Dallas, Texas Daniel N. Allen Psychology, University of Nevada-Las Vegas, Las Vegas, Nevada Recent studies have examined heterogeneous neuropsychological outcomes in childhood traumatic brain injury (TBI) using cluster analysis. These studies have identified homogeneous subgroups based on tests of IQ, memory, and other cognitive abilities that show some degree of association with specific cognitive, emotional, and behavioral outcomes, and have demonstrated that the clusters derived for children with TBI are different from those observed in normal populations. However, the extent to which these subgroups are stable across abilities has not been examined, and this has significant implications for the generalizability and clinical utility of TBI clusters. The current study addressed this by comparing IQ and memory profiles of 137 children who sustained moderate-to-severe TBI. Cluster analysis of IQ and memory scores indicated that a fourcluster solution was optimal for the IQ scores and a five-cluster solution was optimal for the memory scores. Three clusters on each battery differed primarily by level of performance, while the others had pattern variations. Cross-plotting the clusters across respective IQ and memory test scores indicated that clusters defined by level were generally stable, while clusters defined by pattern differed. Notably, children with slower processing speed exhibited low-average to below-average performance on memory indexes. These results provide some support for the stability of previously identified memory and IQ clusters and provide information about the relationship between IQ and memory in children with TBI. Key words: cluster analysis, IQ, memory, pediatric, TBI Pediatric traumatic brain injury (TBI) often results in significant impairment in neurocognitive, behavioral, and interpersonal domains. The extent of these deficits is heterogeneous due to variability in the mechanism, severity, and location of injury, secondary factors, and individual differences among patients (Allen, Thaler, Cross, & Address correspondence to Daniel N. Allen, Psychology, University of Nevada-Las Vegas, 4505 S. Maryland Pkwy., Las Vegas, NV Daniel.allen@unlv.edu Mayfield, 2013; Babikian & Asarnow, 2009; Fay et al., 2009; Roman et al., 1998; Schwartz, et al., 2003; Yeates et al., 2005). This variability has proven challenging in classifying cases of TBI into subgroups that clinically differ. General classification methods based on severity of injury, length of coma, and postinjury amnesia may fail to fully capture differential outcomes on important variables, such as behavior, academic achievement, and adaptive functioning. For example, some cases that are initially classified as severe using the Glasgow Coma

3 2 THALER ET AL. Scale (GCS; Teasdale & Jennett, 1974) may ultimately exhibit little long-term disability (Fay et al., 2009), while others may develop significant behavioral disturbances subsequent to their injury (Allen, Leany et al., 2010). Certain characteristics of the injured child, such as premorbid functioning, age and developmental level at time of injury, socioeconomic status, and development of secondary complications, are associated with outcome after TBI (Babikian & Asarnow, 2009; Max et al., 1999). The impact of these variables contributes further to the disparate outcomes after TBI, and general classification approaches may therefore be assisted by additional methods that describe more homogeneous subsets of patients and predict specific outcomes. Multivariate grouping techniques, such as cluster analysis, can provide an empirical and mathematical method to classify groups in a manner that extends beyond observation and description of symptoms and behavior. To that end, previous studies have used cluster analysis to identify discrete subgroups within TBI that demonstrate differential levels and patterns of performance on neuropsychological tests (Crawford, Garthwaite, & Johnson, 1997; Curtiss, Vanderploeg, Spencer, & Salazar, 2001; Donders & Warschausky, 1997; Mottram & Donders, 2006), which have since been found to be associated with different profiles of impairment across a number of lifefunctioning domains (Allen, Leany et al., 2010; Thaler, et al., 2010). Neuropsychological variables may provide a unique approach to classification because they reflect the impact of injury on involved brain regions and are also useful predictors of some important outcomes, such as educational placement, social functioning, and general functional impairment (Miller & Donders, 2003; Rassovsky et al., 2006; Yeates et al., 2004). The extant neuropsychological cluster-analytic literature on pediatric TBI indicates that: (a) neurocognitive clusters identified in children with TBI differ markedly from those identified in normal populations (Allen, Leany et al., 2010; Donders, 1996, 1999; Donders & Warschausky, 1997; Mottram & Donders, 2006); (b) cluster solutions for IQ tests exhibit adequate generalizability across different samples of children with TBI (e.g., an IQ cluster that emerges in pediatric TBI populations that has marked impairments in perceptual organization and processing speed; Donders & Warschausky, 1997; Thaler et al., 2010); and (c) clusters derived from IQ and memory tests show meaningful associations with behavioral disturbances and disruption of other neurocognitive abilities, with children falling in severely impaired clusters consistently exhibiting the worst neurocognitive, academic, and behavioral outcomes (Allen, Leany et al., 2010; Thaler et al., 2010). However, one area that has not yet received attention involves the impact of test selection on the clusters identified in pediatric TBI samples. Test selection is, of course, an important consideration in such studies. For example, tests that are insensitive to the sequelae of childhood TBI cannot be expected to distinguish level or pattern of performance differences from the general population. Goldstein, Allen, and Seaton (1998) made this point in relation to cluster-analytic studies of schizophrenia, where selection of tests also had the potential for significantly influencing cluster-analytic results. Even in cases where tests are selected that assess different but relevant neurocognitive domains, different cluster solutions may result if cluster membership was more a function of the psychometric properties of the tests and the abilities they measure, rather than the detection of unique and meaningful subgroups of patients. It follows, then, that analyses that generate comparable cluster solutions using a variety of pertinent measures speak to the stability of the clusters and their subsequent utility in identifying discrete profiles of neurocognitive impairment (Goldstein et al., 1998). Moreover, by demonstrating that these empirically derived subgroups of TBI correspond to current conceptualizations of brain injury presentations and recovery patterns, cluster-analytic studies may provide clinical benefits or additional theoretical understanding of disorders. Therefore, the purpose of the present study was to determine whether neurocognitive clusters derived using different tests generalize across these tests. Another purpose was to characterize a sample of children with TBI referred for a pediatric neuropsychological evaluation on their IQ and memory profiles so that clinicians who work in such settings may be informed about the cognitive impairments expected in this population. Tests were selected from the IQ and memory studies by Allen and colleagues (Allen, Leany et al., 2010; Thaler et al., 2010). Examining clusters through other tests that were not used in deriving the clusters themselves provides a better understanding of the relationships among the tests. If the identified clusters show similar levels and patterns of performance when plotted across the other tests, it would confirm that these clusters are stable and generalizable and were not unduly influenced by psychometric issues. Based on the clinically meaningful solutions obtained in previous cluster-analytic studies (Allen, Leany et al., 2010; Thaler et al., 2010), we predicted that four or five cluster solutions will be the best fit for the IQ and memory tests and that clusters will demonstrate stability when cross-plotted on the other tests. Specifically, IQ clusters similar to those previously identified in children with TBI (Donders & Warschausky, 1997; Thaler et al., 2010) will include a cluster with near-normal IQ functioning, a cluster with impaired functioning, and a cluster with selective weaknesses in processing speed. Regarding memory clusters, in line with Allen, Leany, and colleagues (2010), we anticipated there would be clusters that have differential patterns of performance in verbal and

4 nonverbal memory subtests, as well as clusters that are universally impaired or performing in the average range. It was also hypothesized that IQ and memory clusters would correspond in a clinically meaningful way (e.g., a high verbal memory cluster will have high verbal IQ), therefore further cross-validating the cluster solutions. METHOD Participants The sample consisted of 137 children who sustained a TBI and were selected from a consecutive series of cases seen for neuropsychological evaluation during a 5-year period. All children were referred to a pediatric specialty hospital specifically because they were either inpatients at the hospital or because they had a medical disorder, including TBI. Participants were included in the current study if appropriate neuroimaging, laboratory, and examinational findings determined a primary diagnosis of TBI and if they had been administered the Test of Memory and Learning (TOMAL) and Wechsler Intelligence Scale for Children-Third Edition (WISC-III) as part of their neuropsychological evaluation. Individuals with preinjury neurological or neurodevelopmental disorders were excluded, as were individuals who underwent a previous evaluation with the WISC-III. Participants in the current study were selected from those used in previous memory and IQ studies (Allen, Leany et al., 2010; Thaler et al., 2010). Of the 137 participants who qualified, 61% were boys with an average age of 11.7 years (SD = 3.0, range = years). The causes of injuries included motor vehicle accident (48.9%), being struck by a motor vehicle (25.5%), gunshot (5.1%), fall (2.2%), four-wheeler accident (5.1%), bike accident (1.5%), skiing accident (5.8%), and other causes (5.1%). The majority of participants had closedhead injuries (93.4%). The children were assessed at an average of 12.1 months (SD = 16.5, range = months) following injury and were clinically stable and capable of cooperating with testing procedures. At the time of assessment, participants had a mean WISC-III Full-Scale IQ of 83.4 (SD = 14.4) and a mean TOMAL Composite Memory score of 81.3 (SD = 14.8). GCS scores were available for 89 of the participants. The median GCS score was 7 (range = 3 15), which indicates the injuries acquired were generally severe in nature. Measures Wechsler Intelligence Scale for Children-Third Edition (Wechsler, 1991). The WISC-III was used to assess intellectual ability. The four index factors, composed of 10 subtests, include Verbal Comprehension Index (VCI), NEUROCOGNITIVE HETEROGENEITY IN TBI 3 Perceptual Organization Index (POI), Freedom from Distractibility Index (FDI), and Processing Speed Index (PSI) scores. The index scores have a mean of 100 and a standard deviation of 15. Test of Memory and Learning (Reynolds & Bigler, 1994). The TOMAL, composed of 14 subtests (10 that are core and 4 that are supplementary), is used to assess memory ability in children ages 5 to 19 years old. The 10 core subtests include Memory for Stories, Facial Memory, Word Selective Reminding, Visual Selective Reminding, Object Recall, Abstract Visual Memory, Digits Forward, Visual Sequential Memory, Paired Recall, and Memory for Location. The 4 supplementary subtests include Letters Forward, Digits Backward, Letters Backward, and Manual Imitation. Core index scores, calculated by subtest scores, include the Verbal Memory Index (VMI), Nonverbal Memory Index (NMI), Composite Memory Index, Delayed Recall Index (DRI), and Attention/ Concentration Index (ACI). The index scores have a mean of 100 and standard deviation of 15. Glasgow Coma Scale score (Teasdale & Jennett, 1974). The GCS was used to evaluate the severity of injury and was administered shortly after injury either by first responders at the scene or in the emergency room of the hospital. The GCS uses three methods to determine severity: Best Eye Response (score range = 1 4), Best Verbal Response (score range = 1 5), and Best Motor Response (score range = 1 6). The sum of the three scores is then used to classify TBI severity as either mild (range = 13 15), moderate (range = 9 12), or severe (range = 3 8). Procedures Tests were administered following standardized procedures by a pediatric neuropsychologist or doctoral-level technicians who were supervised by the pediatric neuropsychologist. The TOMAL and WISC-III were administered to all children, unless there was a reason not to administer the test (age of child, incapable of cooperating for the tests, etc.). Data Analysis Two separate hierarchical cluster analyses were conducted, one for IQ index scores selected from the WISC- III and a second for memory factor scores identified in the TOMAL from a previous study (Allen, Leany et al., 2010). Although TOMAL factor scores were analyzed, results were plotted across TOMAL index scores to allow a more direct clinical interpretation. Four-, five-, and sixcluster solutions were specified in each case. Both analyses used Ward s method as a clustering method, as it is

5 4 THALER ET AL. commonly used in neuropsychological studies and is useful with quantitative variables in accounting for outliers and for generating reliable clustering solutions (Donders, 2008; Goldstein, Allen, & Caponigro, 2010; Milligan & Hirtle, 2003). Squared Euclidian distance served as the similarity measure, as is also consistently done in other neuropsychology studies of this sort to provide a direct measure of Euclidean space (Everitt, Landau, & Leese, 2001). The number of clusters per battery was determined using the selected approach detailed by Aldenderfer and Blashfield (1984). Specifically, visual inspection of the dendograms and subsequent plotting of clusters in discriminant function space served as a preliminary method to determine clusters that have minimal or no overlap. Attributes for cluster solutions also served as predictors of cluster membership to examine classification rates using discriminant function analysis. A second partitional clustering method (K-means) was also examined for each cluster solution with starting centroids specified as the mean index and factor scores of the clusters derived by the hierarchical method. Agreement between K-means and Ward s clusters were then calculated using Cohen s kappa. Cluster solutions with excellent agreement rates (e.g., >.75; Fleiss, 1981) were deemed stable. Final cluster solutions were determined based on such stability, separation among cluster centroids, and theoretical interest. External validity of the solutions was evaluated across demographic and clinical variables not included in the cluster analysis, including age, gender, months since injury, age at injury, and GCS scores. IQ clusters were then plotted across TOMAL index scores rather than factor scores to aid in clinical interpretability. Memory clusters were plotted across WISC- III index scores. Solutions were additionally cross-tabulated and analyzed with chi-square and Cohen s kappa to examine consistency of cluster membership between the two procedures. Cluster Analysis RESULTS IQ clusters. Ward s method was compared with the K-means method for the four-, five-, and six-cluster solutions. In all cases, agreement between the methods was excellent (kappas =.80,.80,.85, respectively). When plotted in discriminant function space, the four- and fivecluster solutions had clear separation among clusters, with classification rates at 86.9% and 91.2%, respectively. However, the separation for the six-cluster solution was not as clear and there was a drop in classification rate from the five-cluster solution at 89.1%. Cluster solutions were next plotted across IQ index scores. The four-cluster solution yielded: a cluster with low-average (i.e., ) standard scores across the WISC-III index scores (C1); a second cluster with average scores slightly above (C2); a third cluster with average scores that were slightly below (C3); and a fourth cluster with low-average VCI and FDI scores and impaired ( ) POI and PSI standard scores (C4). The five-cluster solution split one of the average clusters (C3) into a cluster with average VCI, POI, and FDI scores and low-average PSI scores (C3) and a cluster with lowaverage VCI and POI scores and average FDI and PSI scores (C5). The six-cluster solution again split the same cluster (C3) into a similar cluster with average VCI, POI, and FDI scores and low-average PSI scores, and a new average cluster. The four-cluster was selected for further analysis, as it had wide separation in discriminant function space and comparable kappa agreement with the five-cluster solution and was consistent with previous findings (Donders & Warschausky, 1997; Thaler et al., 2010). The five-cluster solution s new clusters appeared to represent children who performed in the intermediate range between average and low-average clusters and was not particularly of interest. Although the six-cluster solution had a higher kappa agreement across the two clustering methods, it was of less interest because of the overlap in multidimensional space, as well as the fact that the new sixth cluster was essentially an average-functioning cluster that was similar to an existing cluster. See Figure 1 for a plot of the four-cluster solution. Clusters were named for their level and pattern of performances across index scores. As such, the C1 was identified as below average, as all four index scores were between a standard score of 74 and 80. C2 was identified as near normal because all index scores were at or above the mean. C3 had an average PSI index score that was below 90 while other index scores were between 90 and 100, and so C3 was identified as low normal. The fourth cluster (C4) had below-average VCI and FDI scores and impaired POI and PSI scores, and so was coined impaired POI/PSI. As seen in the figure, one of the clusters differed mainly by level of performance, while three had differences in pattern of index score performance. See Table 1 for a complete list of demographic and IQ data for the clusters. Analyses of variance (ANOVAs) revealed no differences among the clusters for age, months since injury, or GCS scores (p >.39 in all cases). Chi-square analysis indicated no significant differences for gender, ethnicity, open versus closed injury, or mechanism of injury (p >.21 in all cases). The clusters therefore appear to be independent of other clinical variables that might influence outcome. Memory clusters. Again, four-, five-, and six-cluster solutions were examined via Ward s method and were compared to the K-means iterative partitional method.

6 NEUROCOGNITIVE HETEROGENEITY IN TBI 5 Downloaded by [University of California, Los Angeles (UCLA)] at 06:39 07 November 2013 FIGURE 1 IQ and memory performance on respective solutions. VCI = Verbal Comprehension Index; POI = Perceptual Organization Index; FDI = Freedom from Distractibility Index; PSI = Processing Speed Index; VMI = Verbal Memory Index; NMI = Nonverbal Memory Index; ACI = Attention/ Concentration Index; DRI = Delayed Recall Index. Variables TABLE 1 Descriptive and Clinical Variables of the IQ Four-Cluster Solution Near Normal (n = 21) Below Average (n = 45) Clusters Low Normal (n = 55) Impaired POI/ PSI (n = 16) Total (n = 137) Gender (% Male) TBI Type (Closed) M SD M SD M SD M SD M SD Age (years) Time Elapsed (months) GCS WISC-III VCI POI FDI PSI FSIQ TOMAL VMI NMI ACI DRI CMI GCS = Glasgow Coma Scale; WISC-III = Wechsler Intelligence Scale for Children-Third Edition; TOMAL = Test of Memory and Learning; VCI = Verbal Comprehension Index; POI = Perceptual Organization Index; FDI = Freedom from Distractibility Index; PSI = Processing Speed Index; FSIQ = Full-Scale IQ; VMI = Verbal Memory Index; NMI = Nonverbal Memory Index; ACI = Attention/Concentration Index; DRI = Delayed Recall Index; CMI = Composite Memory Index.

7 6 THALER ET AL. Downloaded by [University of California, Los Angeles (UCLA)] at 06:39 07 November 2013 Agreement between the two methods for the clusters was excellent (.79,.88,.86, respectively) in all cases. When plotted in multidimensional space, again the four- and five-cluster solutions demonstrated a separation of cluster centroids while the six-cluster solution had significant overlap with two clusters. Discriminant function classification accuracy was comparable across solutions with 87.6% for the four-cluster solution, 89.8% for the five-cluster solution, and 91.2% for the six-cluster solution. Although the memory clusters were derived from TOMAL factors identified in the study by Allen, Leany, and colleagues (2010), they were plotted across TOMAL VMI, NMI, ACI, and DRI scores to ease interpretability for theoretical and clinical application. The four-cluster solution produced a cluster with below-average performance on all index scores (C1); a cluster with near-normal performance on all index scores except the ACI, which had a mean low-average standard score of 89.5 (C2); a cluster with impaired functioning across all indexes (C3); and a cluster with average VMI and DRI scores and below-average NMI and ACI scores (C4). The five-cluster solution split the below-average cluster (C1) into a below-average cluster and a cluster with average NMI scores and below-average scores on the other indexes. The six-cluster solution again split the below-average cluster into a below-average cluster and a second average Variables Near Normal (n = 45) TABLE 2 Descriptive and Clinical Variables of the Memory Five-Cluster Solution Below Average (n = 49) cluster. The memory five-cluster solution was deemed the most useful for interpretation, as it added a new cluster of clinical and theoretical interest (nonverbal cluster) while the six-cluster solution was redundant, with its second average cluster. Refer to Figure 1 for a plot of the five-cluster memory solution. TOMAL clusters were named based on their levels and patterns of performance on index scores. C1 had belowaverage scores on all memory indexes and was so named below average. C2 was identified as near normal, as average performance on all index scores approached the population mean. C3 was identified as impaired, given that all index scores were below a standard score of 70. C4 was notable for having average performance on the VMI and DRI but below-average performance on the NMI and ACI and was therefore identified as verbal. Finally, C5 showed reverse findings an average NMI score but below-average scores on other indexes and was identified as nonverbal. See Table 2 for demographic and memory data of the clusters. ANOVAs among the memory clusters revealed no significant differences for age, months since injury, and GCS scores (p >.09 in all cases). Chi-square analyses showed no significant differences for gender, open versus closed head injury, and mechanism of injury (p >.33 in all cases). However, there were significant differences for ethnicity (p <.01). Clusters Impaired (n = 20) Verbal (n = 13) Nonverbal (n = 10) Total (n = 137) Gender (% Male) TBI Type (Closed) M SD M SD M SD M SD M SD M SD Age (years) Time Elapsed (months) GCS TOMAL VMI NMI ACI DRI CMI WISC-III VCI POI FDI PSI FSIQ GCS = Glasgow Coma Scale; WISC-III = Wechsler Intelligence Scale for Children-Third Edition; TOMAL = Test of Memory and Learning; VCI = Verbal Comprehension Index; POI = Perceptual Organization Index; FDI = Freedom from Distractibility Index; PSI = Processing Speed Index; FSIQ = Full-Scale IQ; VMI = Verbal Memory Index; NMI = Nonverbal Memory Index; ACI = Attention/Concentration Index; DRI = Delayed Recall Index; CMI = Composite Memory Index.

8 NEUROCOGNITIVE HETEROGENEITY IN TBI 7 Downloaded by [University of California, Los Angeles (UCLA)] at 06:39 07 November 2013 FIGURE 2 IQ and memory clusters plotted on counterpart batteries. VCI = Verbal Comprehension Index; POI = Perceptual Organization Index; FDI = Freedom from Distractibility Index; PSI = Processing Speed Index; VMI = Verbal Memory Index; NMI = Nonverbal Memory Index; ACI = Attention/Concentration Index; DRI = Delayed Recall Index. Comparisons between the IQ and memory clusters. Cluster memberships of the IQ and memory clusters were next plotted against their counterpart batteries, as seen in Figure 2. Three clusters from each solution appeared to match, as they represent three discrete levels of general intellectual and memory functioning. Specifically, the near-normal clusters, the below-average clusters, and the impaired POI/PSI IQ cluster and impaired memory cluster correspond visually with similar level and pattern of performance across batteries. Along with these distinctions, additional unique relationships were identified. The low-normal IQ cluster had generally low-average memory performance with TABLE 3 Relations Between IQ and Memory-Based Cluster Analyses scores across TOMAL indexes that ranged from 84.9 to 87.3 (see Table 1). The verbal memory cluster had nearaverage scores on the VCI, FDI, and PSI but low-average scores on the NMI, while the nonverbal cluster had below-average VCI and PSI scores and average POI and FDI scores, consistent with their performance on the memory indexes. To determine the extent to which the IQ and memory clusters classified the cases into comparable groups, cluster memberships were cross-tabulated. The clusters that were chiefly defined by level of performance were matched together with percentages of agreement rates provided. Results are presented in Table 3. IQ-Based Clusters Near Normal Low Normal Below Average Impaired POI/PSI Total Memory-Based Clusters Below Average Near Normal Nonverbal Verbal Impaired Total Note. Percentages reflect number of cases correctly classified into corresponding level of performance on the other battery. Percentages were not calculated for clusters with unique pattern of performance because no corresponding cluster was available on the other battery.

9 8 THALER ET AL. Degree of agreement was fair (kappa =.29, p <.01), which might be expected given the imbalance in number of clusters between measures. Inspection of the table reveals that there was some correspondence in which cases fell into the below-average clusters, with 55.6% of cases in the IQ cluster and 51.0% of cases in the memory cluster agreeing on membership. Of interest, 34.7% of the cases classified as below average in memory were classified as low normal in IQ. Although 71.4% of the cases in the near-normal IQ cluster were classified as near normal in memory, only 33.3% of near-normal memory cases were classified as near normal in IQ, while 55.6% of these cases were instead classified as low normal. Regarding the impaired POI/PSI cluster, 68.8% were classified as impaired in memory, while the remaining 31.2% were classified as below average. For the impaired memory cluster, 55.0% were identified as impaired POI/PSI, while the remaining 45.0% were classified as low PSI. For the clusters that had no apparent corresponding cluster, classifications were more diffuse. The low-normal cluster had 30.9% of its cases classified as below average in memory, 45.5% of its cases classified as near normal, 10.9% as verbal, and 12.7% as nonverbal. For the nonverbal cluster, 60% were classified as low normal and the remaining 40% were classified as either below average or near normal. For the verbal cluster, 54.8% were classified as low normal with the others being classified as below average or near normal. DISCUSSION Findings from this study demonstrate that cluster analysis of IQ and memory batteries identifies somewhat distinct subgroups of children with TBI, though the patterns of performance on the IQ and memory batteries made theoretical sense and were consistent enough to confirm that clusters represent actual levels of ability as measured by the tests. Our hypothesis that the cluster solutions identified in previous studies would be replicated was met. Demographic and clinical variables, including age, gender, and mechanism of injury, did not influence cluster membership, confirming that these clusters were based primarily on neurocognitive outcome. There were TOMAL cluster differences due to ethnicity, which may be related to family socioeconomic status, access to rehabilitation services, or other environmental factors that we were not able to directly address. Differences on the GCS were nonsignificant, though there was a trend for the TOMAL clusters toward differentiating (p =.09) with a medium effect size (partial η² =.08), suggesting that a larger sample might identify differences in which the most impaired cluster initially had the severest injury, as might be expected. The identified profiles appear to reflect specific cognitive subtypes of functioning across IQ and memory. However, cluster analysis uses mathematical models to classify cases and will do so even with random data (Goldstein et al., 1998). It is therefore necessary to determine the external validity of the clusters by comparing them on variables that were not included in the cluster analysis. In this case, we cross-examined clusters on corresponding batteries in which IQ clusters were plotted on memory indexes and memory clusters on IQ indexes. IQ and memory share approximately 40% to 50% of the same variance (Reynolds & Voress, 2007), so it was expected that interpretable clusters would share similar profiles across respective batteries, albeit with some unique insights. Both the IQ and memory tests classified the sample into three corresponding levels of general performance specifically, a group that performed near the average range, a group that performed below average, and a group that exhibited moderate-to-severe impairment. The near-normal cluster made up nearly a third of the total memory sample (32.8%) but considerably less of the IQ sample (15.3%). A relatively large proportion of the participants were in the below-average clusters for both the IQ and memory batteries (32.8% and 34.3%, respectively), while fewer participants were in the severe clusters (11.7% and 14.5%, respectively). The near-normal, below-average, and impaired clusters appeared stable when cross-plotted. For example, the near-normal IQ cluster also performed in the average range on the memory indexes, while the near-normal memory cluster was in the average range on the IQ indexes. The memory subgroup with impaired performance exhibited severe impairments in the POI and PSI indexes when plotted across the WISC-III indexes, resembling the corresponding IQ subgroup with severe impairment on the POI and PSI and suggesting that severe impairments in processing speed and perceptual organization may be associated with a similarly severe impairment in memory functioning. Not all clusters had corresponding solutions. One IQ cluster (low normal) and two memory clusters (verbal, nonverbal) had unique profile patterns that were not replicated by the other battery. As these clusters represented selective cognitive deficits, they provide additional insights on the relationship between IQ and memory when cross-plotted. The low-normal cluster had overall low-average performance across all memory indexes, approaching one standard deviation below the mean, though it had bet ter scores on the VCI and FDI. This suggests that mild weaknesses in processing speed and perceptual organization may contribute to a drop in general memory functioning. The verbal cluster exhibited a relative weakness on the POI index and average to low-average performance on the other IQ indexes, while the nonverbal cluster had a reverse relationship,

10 indicating that these clusters patterns of performance were consistent across batteries and providing evidence of modality-specific deficits in verbal/nonverbal processing, respectively. When the two solutions were compared for similarities between group membership, the agreement rate was fair. This is inevitable as the number of clusters between the two measures differed, but it also reflects that IQ and memory capture separate cognitive processes that differentially vary. Most of the children who were classified as below average on one cluster were also classified as below average on the corresponding cluster, though more than a third (34.6%) of the cases in the below-average memory cluster were classified instead as low normal in IQ. Most of the children who performed near the average range with IQ were classified in the average range with memory. However, a substantial proportion of children who were classified in the near-normal cluster in memory were classified as being in the low-normal cluster in IQ (55.6%), as might be expected given that both clusters reflect similar levels of functioning. The two impaired clusters were either classified into the corresponding cluster or otherwise put into the below-average cluster, confirming that these children are underperforming on both IQ and memory measures. Finally, the verbal and nonverbal memory clusters were spread out on IQ clusters, though a majority of each was classified as low normal. All in all, clusters defined by level of performance generally concurred, while those defined by pattern of performance were diffused across the other battery s cluster groups. These findings are particularly relevant for clinicians working with children referred for TBI, as they provide expected profile variations with corresponding profiles for each cluster, as well as an indication of differences that might be expected on other measures among the clusters. As these data reflect a representative sample of children referred for evaluation during a 5-year period, they capture a range of neurocognitive functioning that might be observed in a pediatric hospital setting. Profiles observed here in turn can inform clinicians about expected neurocognitive functioning in their patients, and by extension, expected outcomes, as have been identified in other studies (Allen, Leany et al., 2010; Thaler et al., 2010). In general, children with near-normal cognition following TBI perform equally well on intellectual and memory tests. However, children with IQ scores that are roughly two thirds of a standard deviation below average performed close to a full standard deviation below the mean on memory measures. Below-average and impaired IQ scores also reflect below-average and impaired memory scores, with the POI and PSI indexes particularly corresponding with overall impaired memory performance. Memory clusters representing overall level of performance reflect similar levels on IQ clusters. The NEUROCOGNITIVE HETEROGENEITY IN TBI 9 verbal memory cluster had an 8-point difference between the VCI and POI favoring the VCI, while the nonverbal cluster had a 7-point difference favoring the POI. The current results are limited in a number of ways. This study relied on archival data using outdated measures such as the WISC-III and TOMAL. Future studies should examine the generalizability of these findings to the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV; Wechsler, 2003) and the Test of Memory and Learning-2 (Reynolds & Voress, 2007), especially in light of findings identifying differences in WISC- III and WISC-IV performance in children with TBI (Allen, Thaler, Donohue, & Mayfield, 2010). However, the purpose of this article was not to examine the use of the measures themselves, but rather the constructs they measure. Symptom validity data were unavailable, although patients were not assessed for compensatory or disability claims and trained pediatric neuropsychologists took care to obtain accurate data whenever possible during assessments. It was also impossible to ascertain how premorbid functioning might have contributed to cluster membership, as premorbid data were unavailable. Although all children underwent standard neuroimaging protocols, data were not collected for research purposes and therefore were also not available for analysis. The sample primarily consisted of children with severe injuries and consequently lower IQ scores than those observed in other cluster-analytic studies of TBI (Donders & Warschausky, 1997; Goldstein et al., 2010). As such, the results should primarily generalize to other children who were classified with severe injuries in a pediatric care setting. However, some (4%) children included in this study had sustained milder injuries with GCS scores ranged from 3 to 15, and data were missing from others. The inclusion of a smaller portion of children with less severe injuries may have had some impact on the obtained cluster solution, though the purpose of this study was to characterize a population of pediatric patients referred for TBI and results should be generalizable to clinicians working in similar settings. Similarly, length of time since injury ranged from 3 to 127 months, indicating that some children included in the study may still have been experiencing spontaneous recovery while others recovery had likely plateaued; however, given that clusters did not differ among this variable, this does not appear to have significantly influenced membership. Some of the clusters consisted of only 7% to 12% of the entire sample, limiting their power to detect cluster differences. However, as clusters replicate those from previous studies (Allen, Leany et al., 2010; Thaler et al., 2010), they appear reliable. Finally, no control sample was available to compare clusters, though we have evidence that IQ and memory clusters presented in children with severe TBI substantially differ from those in healthy controls (Allen, Leany et al., 2010; Donders & Warschausky, 1997).

11 10 THALER ET AL. In summary, the cognitive heterogeneity evidenced in children with TBI manifests in consistent levels and patterns across memory and IQ batteries. The comparison between the WISC-III and TOMAL batteries identified four or five clusters in the solutions, with clusters replicating profiles represented in other studies (Allen, Leany et al., 2010; Donders & Warschausky, 1997; Thaler et al., 2010). Both solutions identified a subgroup of patients with average performance on IQ and memory measures, consistent with previous clinical studies of adult patients with TBI and schizophrenia (Goldstein et al., 1998, 2010). Both solutions also identified children with overall below-average and impaired cognitive performance across measures. Allen, Leany, and colleagues (2010) and Thaler and colleagues (2010) both found significant behavioral and emotional dysfunction in their impaired clusters, suggesting that subtyping cognitive performance may assist in providing insight into the functional outcome of patients. Demographic variable analyses indicate that most of the differences among clusters are independent of age, onset and mechanism of injury, and GCS scores, and so these clusters appear to capture neurocognitive differences that are not related to these characteristics. Results confirm that subtyping cognitive heterogeneity in TBI produces consistent profiles across a continuum of impairment severity. Even with this consistency, the inherent differences between intelligence and memory translate to subtle variations that may further define the nature of cognitive impairment resulting from injury. ACKNOWLEDGEMENTS Some of the data presented in this article were used in related cluster-analytic studies on pediatric TBI including those by Allen, Leany et al. (2010) and Thaler et al. (2010). REFERENCES Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster analysis. Beverly Hills, CA: Sage. Allen, D. N., Leany, B. D., Thaler, N. S., Cross, C., Sutton, G. P., & Mayfield, J. (2010). Memory and attention profiles in pediatric traumatic brain injury. Archives of Clinical Neuropsychology, 25, Allen, D. N., Thaler, N. S., Cross, C., & Mayfield, J. (2013). Classification of traumatic brain injury severity: A neuropsychological approach. In D. N. Allen & G. G. Goldstein (Eds.), Cluster analysis in neuropsychological research: Recent applications (pp ). New York, NY: Springer. Allen, D. N., Thaler, N. S., Donohue, B., & Mayfield, J. (2010). WISC- IV profiles in children with traumatic brain injury: Similarities to and differences from the WISC-III. Psychological Assessment, 22, Babikian, T., & Asarnow, R. (2009). Neurocognitive outcomes and recovery after pediatric TBI: Meta-analytic review of the literature. Neuropsychology, 23, Crawford, J. R., Garthwaite, P. H., & Johnson, D. A. (1997). WAIS-R subtest pattern clusters in closed-head injured and healthy samples. The Clinical Neuropsychologist, 11, Curtiss, G., Vanderploeg, R. D., Spencer, J., & Salazar, A. M. (2001). Patterns of verbal learning and memory in traumatic brain injury. Journal of the International Neuropsychological Society, 7, Donders, J. (1996). Cluster subtypes in the WISC-III standardization sample: Analysis of factor index scores. Psychological Assessment, 8, Donders, J. (1999). Cluster subtypes in the standardization sample of the California Verbal Learning Test-Children s Version. Developmental Neuropsychology, 16, Donders, J. (2008). Subtypes of learning and memory on the California Verbal Learning Test-Second Edition (CVLT-II) in the standardization sample. Journal of Clinical and Experimental Neuropsychology, 30, Donders, J., & Warschausky, S. (1997). WISC-III factor index score patterns after traumatic head injury in children. Child Neuropsychology, 3, Everitt, B. S., Landau, S., & Leese, M. (2001). Cluster analysis (4th ed.). London, England: Arnold. Fay, T. B., Yeates, K. O., Wade, S. L., Drotar, D., Stancin, T., & Taylor, H. G. (2009). Predicting longitudinal patterns of functional deficits in children with traumatic brain injury. Neuropsychology, 23, Fleiss, J. L. (1981). Statistical methods for rates and proportions (2nd ed.). New York, NY: John Wiley & Sons. Goldstein, G., Allen, D. N., & Caponigro, J. M. (2010). A retrospective study of heterogeneity in neurocognitive profiles associated with traumatic brain injury. Brain Injury, 24, doi: / Goldstein, G., Allen, D. N., & Seaton, B. E. (1998). A comparison of clustering solutions for cognitive heterogeneity in schizophrenia. Journal of the International Neuropsychological Society, 4, Max, J. E., Roberts, M., Koele, S. L., Lindgren, S. D., Robin, D. A., Arndt, S., & Sato, Y. (1999). Cognitive outcome in children and adolescents following severe traumatic brain injury: Influence of psychosocial, psychiatric, and injury-related variables. Journal of the International Neuropsychological Society, 5, Miller, L. J., & Donders, J. (2003). Prediction of educational outcome after pediatric traumatic brain injury. Rehabilitation Psychology, 48, doi: / Milligan, G. W., & Hirtle, S. C. (2003). Clustering and classification methods. In J. Schinka & W. Velicer (Eds.), Handbook of psychology: Research methods in psychology (Vol. 2, pp ). New York, NY: Wiley. Mottram, L., & Donders, J. (2006). Cluster subtypes on the California Verbal Learning Test Children s Version after pediatric traumatic brain injury. Developmental Neuropsychology, 30, Rassovsky, Y., Satz, P., Alfano, M. S., Light, R. K., Zaucha, K., McArthur, D. L., & Hovda, D. (2006). Functional outcome in TBI: I. Neuropsychological, emotional, and behavioral mediators. Journal of Clinical and Experimental Neuropsychology, 28, doi: / Reynolds, C. R., & Bigler, E. D. (1994). Test of Memory and Learning: Examiner s manual. Austin, TX: Pro-Ed. Reynolds, C. R., & Voress, J. K. (2007). Test of Memory and Learning: Second edition. Austin, TX: Pro-Ed. Roman, M. J., Delis, D. C., Willerman, L., Magulac, M., Demadura, T. L., de la Peña, J. L., Kracun, M. (1998). Impact of pediatric traumatic brain injury on components of verbal memory. Journal of Clinical and Experimental Neuropsychology, 20(2), Schwartz, L., Taylor, H. G., Drotar, D., Yeates, K. O., Wade, S. L., & Stantin, T. (2003). Long-term behavior problems following pediatric traumatic brain injury: Prevalence, predictors, and correlates. Journal of Pediatric Psychology, 28,

12 NEUROCOGNITIVE HETEROGENEITY IN TBI 11 Teasdale, G., & Jennett, B. (1974). Assessment of coma and impaired consciousness: A practical scale. Lancet, 2, Thaler, N. S., Bello, D. T., Randall, C., Goldstein, G., Mayfield, J., & Allen, D. N. (2010). IQ profiles are associated with differences in behavioral functioning following pediatric traumatic brain injury. Archives of Clinical Neuropsychology, 25, Wechsler, D. (1991). Manual: Wechsler Intelligence Scale for Children- III. New York, NY: The Psychological Corporation. Wechsler, D. (2003). Wechsler Intelligence Scale for Children-Fourth Edition. San Antonio, TX: The Psychological Corporation. Yeates, K., Armstrong, K., Janusz, J., Taylor, H., Wade, S., Stancin, T., & Drotar, D. (2005). Long-term attention problems in children with traumatic brain injury. Journal of the American Academy of Child & Adolescent Psychiatry, 44, doi: /01.chi Yeates, K., Swift, E., Taylor, H., Wade, S. L., Drotar, D., Stancin, T., & Minich, N. (2004). Short- and long-term social outcomes following pediatric traumatic brain injury. Journal of the International Neuropsychological Society, 10, doi: /s