METHOD

Participants

After institutional review board approval, 70 children with TBI were selected retrospectively from a 4-year series of consecutive referrals to a Midwestern regional rehabilitation facility. All of these children had been included previously in an investigation pertaining to the factor structure of the WRAML-S in children with TBI, 17 but that study did not address any gender issues. Specifically for the purposes of the current investigation, and to get an impression of other cognitive abilities in addition to the memory variables that were provided by the WRAML-S, we focused on children who had also completed the Wechsler Intelligence Scale for Children-Third Edition (WISC-III) 18 as part of the same neuropsychologic evaluation that included the WRAML-S. Although additional psychometric measures were administered to the participants, they differed with age of the child; only the WRAML-S and the WISC-III were normed for the entire age range (6-16 years) that was the target of this study. Children with prior psychiatric, neurologic, special education, or personal abuse histories had been excluded from the original sample of 100 children, 17 which had included a larger age range (up to 18 years) and children who had been evaluated with the Wechsler Adult Intelligence Scale-Revised. 19

We were able to locate 35 boys and 35 girls in the original sample who met all the current inclusion criteria. A control group (n = 70) was subsequently obtained from the standardization sample of the WRAML-S. Control group participants were matched in a pairwise manner to the clinical patients on the variables of gender, ethnicity, and age, and none of them had a history of neurologic or psychiatric dysfunction. It was not possible to match the clinical and the control participants on measures of socioeconomic status (e.g., parental education or occupation) because that information was not available for a substantial proportion of the WRAML-S standardization sample. Details of the four groups of participants are presented in Table 1. Some variables (e.g., ethnicity) were dichotomized to have sufficient participants in each cell for the statistical analyses. Length of coma was defined as the number of days until the child responded to verbal commands. Injury severity was classified on the basis of previously established criteria that included the lowest postresuscitation Glasgow Coma Scale within 24 hours after injury and the results of neuroimaging studies of the brain. 20 Information about parental occupations was available only for the children with TBI. Thirty-nine of these children (55.71%) had parents who held professional or clerical jobs, and 31 of them (44.29%) had parents whose jobs involved skilled or unskilled labor.

There were no statistically significant differences between the two groups of clinical participants in terms of age, ethnicity, injury type, injury severity, length of coma, or time since injury (p > .10 on all variables). This was considered to be important because, otherwise, any analyses of gender effects would have been confounded by neurological or demographic variables. There were 22 children in each of the two clinical groups who had neuroimaging evidence for a diffuse (e.g., edema) or focal (e.g., contusion) intracranial lesion, or both. More specific lesion analyses were not performed because of small numbers in each cell (e.g., left, right, anterior, posterior) and because we often had to rely on recorded reports and did not have access to the original scans for more precise location or volumetric measurements.

Measurements

The WRAML-S is an individually administered memory test that consists of four subtests that require the recall of pictorial (social scene, geometric design) or verbal (word list, narrative paragraph) stimuli. Performance on these tasks (Picture Memory, Design Memory, Verbal Learning, Story Memory) is expressed in scaled scores (M=10, SD=3), with higher scores reflecting better performance. Median reliabilities for these subtests are adequate and range from .78 to .86. 15 They can also be combined into a composite Screening index, which is expressed as a standard score (M=100, SD=15), with higher scores reflecting better performance. Because of the fact that previous research 17 has suggested that this index represents a single general memory construct (as opposed to separate verbal and nonverbal factors) in children with TBI, the majority of the analyses in this investigation focused on the composite Screening index and not on the individual subtests.

The WISC-III is an individually administered test of psychometric intelligence. Although a number of variables can be obtained from this instrument, previous research in our laboratory 21,22 has suggested that the four factor index scores provide the most diagnostically useful information when evaluating children with TBI. These are conceptualized as Verbal Comprehension, Perceptual Organization Freedom from Distractibility, and Processing Speed (PS). All these variables are expressed as standard scores (M=100, SD=15), with higher scores reflecting better performance. Average reliabilities for these indexes are excellent, ranging from .85 to .94. 18 We were particularly interested in the PS variable because it has demonstrated great sensitivity to severity of TBI in previous research. 23,24

Procedure

The WRAML-S and WISC-III were administered and scored according to standardized procedures as part of neuropsychologic evaluations that had been requested by attending physicians during the rehabilitation of the children. Assessments were performed only when the children were medically stable and could recall meaningful information from day to day. Informed parental consent was obtained for all neuropsychologic evaluations.

RESULTS

There was a statistically significant group difference on the WRAML-S composite index, F(3, 136)=4.69, p < .01, with the boys with TBI performing worse than the other groups (see Table 1). The associated pattern of scores on the four component subtests is presented in Figure 1. Inspection of this figure suggests that the boys with TBI performed relatively poorly on all WRAML-S subtests (including both visual and verbal stimuli), not just on one or two of them. For this reason, and because of the known unifactorial structure of the WRAML-S in children with TBI, 17 post-hoc statistical analyses were limited to the composite index of the instrument. We decided a priori that a traditional Bonferroni correction would be too conservative in light of sample size. Therefore, we decided to (1) perform only those comparisons that were of greatest theoretic interest and (2) use a measure of effect size ([eta]²) as the criterion for interpretation. Specifically, we specified a priori a minimum level of [eta]² >= .05, which would mean that at least 5% of the variance in the dependent variable was explained by the independent variable.

The boys with TBI performed worse than their respective counterparts from the standardization sample on the WRAML-S index, F(1, 68)=6.15, p < .05, [eta]²=.08, whereas the performance of the girls with TBI appeared to be very close to that of their matched controls on this variable, F(1, 68)=.14, p > .10, [eta]² < .01. In the clinical sample, there was a statistically significant effect of gender on the WRAML-S index, with girls outperforming boys, F(1, 68)=8.17, p < .01, [eta]²=.11. In contrast, the gender effect in the standardization control group did not approach even liberal standards of statistical significance on this same variable, F(1, 68)=.86, p > .10, [eta]²=.01. The gender effect in the clinical sample was also statistically significantly greater than that found in the standardization control sample, z=5.55, p < .01.

We wanted to explore if gender added incremental validity to the prediction of the WRAML-S composite score in the clinical sample, over and beyond the amount of variance accounted for by injury severity and demographic variables. For this purpose, we conducted a series of hierarchical regression analyses in which we first evaluated the predictive power of the selected variables (i.e., age, length of coma, and the presence or absence of an intracranial lesion on neuroimaging) and then determined if the addition of gender to this model would account for a statistically significant amount of additional variance.

The regression model for the WRAML-S index is presented in Table 2. When gender was not included, the model accounted for 36% of the variance, F(3, 66)=12.22, p < .0001. Longer length of coma, presence of a distinct intracranial lesion on computed tomography or magnetic resonance imaging scan, and younger age were associated with lower WRAML-S index scores. The addition of gender to this model yielded a statistically significant increase of an additional 9% of the variance accounted for (p < .01), for a total amount of 45%, F(4, 65)=13.15, p < .0001. Thus gender explained a considerable amount of variance in the memory test performance of the children with TBI, even after accounting for the effects of injury severity and age.

We also wanted to determine if the observed relative superiority of the girls on the WRAML-S index in the clinical sample could be due to gender differences in overall verbal or other intellectual ability. For this purpose, we evaluated the children's scores on the four factor indexes of the WISC-III. These data are presented in Figure 2. There were no statistically significant differences between the gender groups on Verbal Comprehension, F(1, 68)=.88, p > .10, or on Freedom from Distractibility, F(1, 68)=2.36, p > .10, and a trend for a group difference on Perceptual Organization fell short of statistical significance, F(1, 68)=3.17, p < .10. This suggests that overall language, perceptual, or working memory skills cannot account for the differential effect of TBI on memory performance, as assessed by the WRAML-S, in boys as opposed to girls. However, the group difference on the Processing Speed index was statistically significant, F(1, 68)=5.21, p < .05, with girls outperforming boys on this variable and this was associated with a small but potentially relevant effect size, [eta]²=.07.

The Processing Speed index is based on two paper-and-pencil tasks that emphasize accurate and speeded association of visual stimuli. The fact that girls with TBI were able to do this much faster than boys with TBI raised the question of whether differences in speed of information processing might be a factor contributing to their respective memory scores on the WRAML-S. Several of the WRAML-S subtests involve only brief exposure to the stimuli. For example, under standardization guidelines, the geometric patterns on Design Memory are presented to the child for only 5 seconds. Other WRAML-S subtests present information at a fairly rapid pace. For example, the words on Verbal Learning are supposed to be read to the child with only a 1-second pause between items. Thus it appeared plausible that if children could not process the information fast enough, then they might not be able to remember it efficiently later on.

To explore formally the possibility that the relatively reduced memory performance of the boys with TBI might be due to difficulties with speed of information processing, we reexamined the difference between the gender groups on the WRAML-S index while using the PS standard score as a covariate. Under these circumstances, the gender effect was reduced to a statistically not significant trend, F(1, 67)=3.29, p < .10. Thus it appears that reduced speed of information processing is at least partly responsible for the relative difficulty of the boys, as compared with the girls, on the WRAML-S after TBI.

DISCUSSION

The main purpose of this investigation was to determine the degree to which the effect of TBI on children's memory is moderated by gender status. Hypothesis 1 was confirmed: The performance of boys with TBI was more than two-thirds of a standard deviation below that of girls with TBI on the WRAML-S index. Hypothesis 2 was also confirmed: This gender difference was statistically significantly greater than that observed in a sample of healthy control children. These results are consistent with recent findings in our laboratory with a different memory test in a completely independent sample, 11 and improve on that investigation by including a memory test with both pictorial and verbal stimuli, as well as a demographically matched control group.

It is worth noting that gender was certainly not the only variable that affected the memory test performance of the children with TBI in this investigation. In fact, the majority of the variance in the WRAML-S index was explained by injury severity, especially length of coma, with additional contributions from neuroimaging findings and child age. At the same time, gender explained 9% of incremental variance in memory outcome, even after accounting for those other variables. This effect size is somewhat small from a statistical point of view but potentially significant when viewed according to conventional clinical standards. 25,26

The pediatric literature makes a distinction between moderator variables and mediator variables when considering the outcome of children with a particular condition such as TBI. According to Holmbeck, 27 a moderator variable specifies the circumstances under which a condition results in a particular outcome, whereas a mediator variable explains the process by which a condition results in such an outcome. Consistent with this distinction, we consider gender to be a moderator variable when it comes to the effect of pediatric TBI on memory outcome, as assessed by the WRAML-S. The nature of the relationship between severity of TBI and performance on this pediatric memory test varies as a function of gender, with the effect being more detrimental on boys than on girls. Whether this is due to hormonal, neuroanatomic, or other differences will require additional research.

The results from this investigation also raise the possibility that speed of information processing may play a mediating role in this entire process. When performance on the WISC-III Processing Speed index was used as a covariate in the analyses, the gender difference on the WRAML-S index in the clinical sample was reduced to a statistically not significant trend. The fact that Processing Speed had a strong correlation with length of coma (r=-.57, p < .0001) is relevant in this regard. These findings may suggest that severity of TBI is associated with a negative effect on speed of information processing, which is in turn associated with a negative effect on memory performance when information is presented only briefly or at a rapid pace, as is the case on several of the WRAML-S subtests. Again, the strength of this association appears to be moderated by gender.

These interpretations are speculative at this time because of the small sample size. In this context, it could also be argued that there methods other than analysis of covariance have been proposed to evaluate the statistical probability of mediating variables. 28 However, those methods are known to have low statistical power. 29 A much larger sample (n >= 150) would be needed to submit these kinds of data to path analysis or some other kind of structural equation modeling. However, the application of such advanced statistical techniques in other recent studies of pediatric outcome following TBI 30 suggests that this is a feasible goal for future research.

A recent meta-analysis of gender differences in outcome after TBI in adults led to the conclusion that women actually tended to do slightly worse than men. 31 However, the authors noted that very few published reports in the literature described outcome separately for gender, and they were unable to find any gender studies that involved children under the age of 12. More significantly, the largest effect size that the authors reported was for the duration of postconcussion symptoms after mild head injury, with persistent complaints being far more common in women. With adults, this phenomenon may not necessarily be related to cerebral parameters. 32,33 Longitudinal studies would be needed to address the possibility of a developmental change in the relative differences between the genders in memory performance after TBI.

We also want to emphasize that gender differences do not necessarily occur in all areas of outcome after pediatric TBI. In fact, in a prior study in our own laboratory that focused on functional activities of daily living, gender did not add to the prediction of adaptive behavioral outcome after accounting for variables such as cerebral perfusion pressure and premorbid psychosocial history. 34 Part of this difference in findings may be related to the fact that the children in that study were evaluated at time points 1-5 years after injury, whereas the children in the current investigation were all evaluated within 12 months. Again, longitudinal studies with comprehensive assessments are needed to determine how the relationship among preinjury, injury, and postinjury characteristics changes in a different way over time for boys and girls.

Possible limitations of this investigation must also be considered. Our participants were recruited from a referred convenience sample that likely included a greater proportion of children with moderate and severe injuries than would be found with consecutive emergency room visits. At the same time, this guaranteed sufficient variance of injury severity to guard against restriction of range effects. A potentially more significant limitation is that we did not have the opportunity to perform morphometric analyses of the neuroimaging data, which has shown some promise in the analysis of outcome in other studies. 35,36 The absence of data on socioeconomic characteristics for the control participants from the WRAML standardization sample is also regrettable. Future research should address the possible interaction of gender with socioeconomic status in the recovery of pediatric TBI. On the other hand, relative strengths of this investigation are that we had screened the sample for confounding premorbid factors, that the two clinical gender groups did not differ significantly in length of coma, and that we used a demographically matched control group.

CONCLUSION

The findings from this investigation suggest that memory for new information after pediatric TBI is moderated by gender and that boys may have relatively worse outcome in this regard during the first year after injury. This could allow for "flagging" of these children for closer or longer term monitoring, in collaboration with the school system. 37-39 This should not be misconstrued as a reason to neglect the rehabilitation needs of girls with TBI: children with severe TBI and children with poor performance on the WRAML-S (not to mention both) should be offered intensive rehabilitation early, regardless of their gender.

REFERENCES