Predictors of Outcome following Traumatic Brain Injury in Young Children

Original Paper Pediatr Neurosurg 2002;36:64–74 Received: July 6, 2001 Accepted: November 9, 2001 Predictors of Outcome following Traumatic Brain Injury in Young Children Mary R. Prasad a Linda Ewing-Cobbs a Paul R. Swank a Larry Kramer b Departments of a Pediatrics and b Radiology, University of Texas Health Science Center, Houston, Tex., USA Key Words Outcome W Glasgow Coma Scale W Glasgow Outcome Scale W Child abuse W Brain injury W Children W Infants W Neuroimaging Abstract The relationship between clinical and neuroimaging variables and multiple outcome measures was examined in a longitudinal, prospective study of 60 children less than 6 years of age who sustained either inflicted or noninflicted traumatic brain injury. Hierarchical multiple regression indicated that the modified Glasgow Coma Scale score, the duration of impaired consciousness and the number of intracranial lesions visualized on CT/MRI accounted for a significant amount of the variance in the Glasgow Outcome Scale (GOS), cognitive and motor scores at baseline, 3- and 12-month evaluations. Inflicted brain injury adversely affected both GOS and cognitive outcomes. Pupillary abnormalities were associated with poorer motor outcome. Neither age at injury nor the Injury Severity Score accounted for significant variability in outcomes. Copyright © 2002 S. Karger AG, Basel ABC Fax + 41 61 306 12 34 E-Mail karger@karger.ch www.karger.com © 2002 S. Karger AG, Basel Accessible online at: www.karger.com/journals/pne Introduction A variety of factors influence outcome from traumatic brain injury (TBI) in young children. Age at injury, type of intracranial damage and acute care variables have been related to both global ratings of outcome and more specific neuropsychological indices of functioning. Although several studies have examined outcome in children aged 0–15 years at injury, there have not been any studies specifically examining predictors of outcome in infants and young children. Within the pediatric population, the etiology and outcome of TBI differ according to the age at injury. Although some studies examining functional/neurobehavioral outcome in age-based cohorts of children and adolescents following nonpenetrating brain injury did not identify relationships with age [1–5], the majority of studies report less favorable rates of mortality and morbidity in infants and preschoolers than in school-age children and adolescents [6–14]. Assessment of injury severity and outcome is complicated in young children by the lack of validated, developmentally appropriate scales. The most widely accepted classification of acute TBI severity is based on the Glasgow Coma Scale (GCS) [15]. The GCS requires modification of several verbal and motor items to assess acute TBI severity in infants and toddlers. Several modifications of the GCS score for infants are used widely [14, 16, 17]. Mary R. Prasad, PhD Department of Pediatrics, University of Texas Health Science Center-Houston 7000 Fannin, Suite 2431 Houston, TX 77030 (USA) Tel. +1 713 500 3876, Fax +1 713 500 3878, E-Mail Mary.R.Prasad@uth.tmc.edu Difficulties with modified coma scales include the use of verbal and motor bahaviors that are not appropriate for children less than 2 years of age or that assume specialized knowledge of complex brain stem and/or ocular responses [18]. The Glasgow Outcome Scale (GOS) [19], which was developed for adults, is the most common global outcome measure used in pediatric studies. However, few studies explicate the criteria used to assign scores to young children [20]. The quality of outcome in infants and young children using the GOS is difficult to equate with adult outcomes due to the less obvious impact of increased impairment and dependency in young children versus adults. For example, TBI resulting in intellectual deficiency is obviously disabling for adults, who are often unable to resume family and vocational responsibilities. Intellectual deficiency in young children may be less obvious as there are fewer expectations for independence in activities of daily living, provision of daily care is perceived as less burdensome because caregivers expect to provide substantial support to young children and numerous programs are available to provide educational and therapeutic services. The GOS should be modified to reflect the unique issues of assessing functional outcome in young children. Modifications of the scale for pediatric populations have been shown to be useful in assessing outcome [20–22]. Studies typically restrict poor outcome on the GOS to children who are severely disabled or are in a persistent vegetative state. However, mild functional impairment in children can impact significantly on development and lead to chronic deficits [20, 23]. Increasing injury severity as reflected in lower GCS scores and increasing duration of impaired consciousness or post-traumatic amnesia is generally associated with decreasing functional competence at the time of discharge [24] as well as 6 months to several years after injury in pediatric populations [6, 13, 25]. Both the total GCS score and the component scores have been significantly related to outcome during early [5] and late stages of recovery [25, 26]. Several investigators have determined that the duration of abnormal scores on the GCS motor scale or total scale is significantly related to outcome [25, 26]. In addition to GCS scores, the predictive validity of neuroimaging findings and variables reflecting the child’s acute neurologic and physiologic status has been examined following pediatric TBI. Acute CT findings of extraaxial hemorrhage, diffuse swelling, focal mass lesion and diffuse axonal injury [26] have been shown to be predictive of poorer outcome; intracerebral hematoma, subdural hematoma and skull fracture have not [5, 26]. Pupillary reactivity has been predictive of outcome in some Predictors of Outcome following Brain Injury samples [26] but not in others [5]. Focal neurologic deficits were not predictive of outcome [26]. Extracranial injuries as indicated by the AIS or Injury Severity Score (ISS) also adversely impact on the quality of outcome [26]. Hypoxia and shock were also associated with a poorer outcome [26]. Other variables assessing acute status in the emergency room such as vital signs and blood gasses [5] have not shown significant relationships with outcome. To identify the variables most predictive of outcome, several studies have developed statistical models. Using multiple stepwise logistical regression, Ong et al. [26] identified the following variables as most predictive of a poor outcome 6 months after pediatric TBI: hypoxia on admission and CT scan findings of subarachnoid hemorrhage, diffuse axonal injury and swelling. The GCS score alone had limited predictive value. Massagli et al. [5] examined GOS scores 4–8 years after pediatric TBI and concluded that coma duration was most strongly related to late GOS outcome. Similarly, McDonald et al. [25] identified that the number of days until a child attained age-appropriate orientation and GCS scores of 15 was strongly predictive of neuropsychological test performance. The mechanism of TBI may contribute additional variability to outcome in young children. Assault is a common external cause of TBI in young children; the rate of inflicted TBI has been reported to range from 4 to 24% of consecutive admissions [16, 27, 28]. In children with similar GCS scores, inflicted TBI is associated with greater initial cognitive and motor deficit than noninflicted TBI [21], as well as poorer late outcomes [29]. Although infants have been included in some studies of predictors of outcome following pediatric injuries, no studies have specifically examined predictors of global and neuropsychological outcomes in very young children. The purpose of the present study was to identify acute care and neuroimaging variables significantly related to (1) global outcomes as indicated by GOS scores and (2) cognitive and motor outcomes at baseline and 3 and 12 months after TBI. Patients and Methods Patients Children aged 1 month to 6 years who had been hospitalized for inflicted (n = 31) or noninflicted TBI (n = 29) were recruited from the Memorial Hermann Children’s Hospital or the Texas Children’s Hospital in Houston, Tex., USA. Inclusion criteria were moderate to severe TBI, gestational age of 632 weeks and age at injury ! 6 years. Pediatr Neurosurg 2002;36:64–74 65 Table 1. Demographic and neurological variables by abuse group Mean age, months** Gender* Females Males Family psychiatric history Baseline Cognitive score* Motor score GOS** good recovery GOS** poor recovery 1-Year outcome Cognitive score* Motor score GOS good recovery GOS poor recovery Unreactive pupils CT/MRI findings, number of lesions/cases Extraaxial hemorrhages** Intraparenchymal hemorrhages Infarct/edema Atrophy, cases** Mean lowest GCS Mean number of days of impaired consciousness Mean ISS Inflicted TBI Noninflicted TBI 10.8 35.13 22 9 11 10 19 17 77.27 78.32 6 25 85.73 83.51 16 13 79.23 82.46 7 19 4 83.51 89.36 11 14 5 64/30 5/4 14/7 11 10.23 3.82 21.29 37/20 15/11 12/7 1 8.82 3.22 19.77 * p ! 0.05; ** p ! 0.01. Moderate TBI was defined as a modified GCS score of 9–12 or 13–15 with positive neuroimaging findings on admission scans. Severe TBI included children with GCS scores of 3–8. At 1 year, 8 children were lost to follow-up because their guardians declined to participate or children were placed with family members who lived out of state. One child in the inflicted TBI group could not be tested because he was in a persistent vegetative state. Children with known preinjury neurologic, metabolic or psychiatric disorders were excluded from the study. Table 1 contains demographic and injury variables for the inflicted and noninflicted TBI groups. The sample in this study was composed of all the children with TBI who participated in this project. Determination of inflicted versus noninflicted TBI was performed by the state protective agency and the children’s protection committee at each hospital. The criteria used by the agencies are comparable to the algorithm developed by Duhaime et al. [27] to detect inflicted injuries, i.e. injuries incompatible with the reported mechanism of injury (e.g. subdural hematoma and retinal hemorrhages from falling off the couch), unexplained injuries (no known cause of injury), delay in seeking treatment and changing histories. All children were enrolled in a prospective, longitudinal study of outcome following early acquired brain injury in accordance with the informed consent procedures at each hospital. Because inflicted brain injury is more common in infancy, there was a significant difference in age at injury between the two TBI groups; children who had an inflicted brain injury were significantly younger than those with noninflicted TBI. 66 Pediatr Neurosurg 2002;36:64–74 The GCS was modified to reflect the abilities of children from birth to 35 months of age. The motor scale item ‘following commands’ was modified to include spontaneous movements in infants 0–6 months of age and goal-directed movements in children 7–35 months of age. The verbal scale items ‘confused’ and ‘oriented’ were regarded as equivalent to ‘cries’ and ‘cries to indicate needs’. A pseudoscore of 1 was assigned if periorbital swelling or intubation precluded the assessment of eye opening or verbal output; the use of pseudoscores does not significantly affect score distributions [30]. To examine serial scores in children treated with sedatives and/or antiparalytics, the child’s hourly GCS scores were recorded and the best level of response was coded for each 24-hour period. Neuroimaging studies included MRI and CT scans, which were performed without contrast. CT was obtained upon admission in all but 1 child. MRI scans were performed within 7 days of admission in 24 children with inflicted and in 14 children with noninflicted TBI. MRI scans were completed on average 2.6 days following admission. All scans were reviewed by a board-certified radiologist specializing in MRI who was blind to the group designation (inflicted or noninflicted). Birth records were obtained for 80% of the sample. The inflicted and noninflicted TBI groups were comparable for Apgar scores, gestational age, number of days hospitalized, length and head circumference. There was a significant difference in birth weight [F(1, 46) = 12.85; p ! 0.0008], as children with inflicted TBI weighed less than those with noninflicted TBI. Due to the reluctance of some families to allow the release of birth records, particularly those in the inflicted group, there were 8 children with missing birth data. Prasad/Ewing-Cobbs/Swank/Kramer Table 2. Correlations between the three Lowest GCS score injury severity predictors Lowest GCS score Duration of impaired consciousness CT/MRI findings Procedure As part of the longitudinal study, each child was evaluated as soon after the injury as possible (M = 1.3 months). Follow-ups were obtained 3 and 12 months after injury in 92 and 84% of the sample, respectively. Statistical Approach Logistic regression analysis and hierarchical multiple regression were used to create models of predictors of categorical and continuous outcome measures, respectively. Predictor variables were retained in each model if p was less than 0.10. Variables assessing the severity of brain injury, injury to other body regions and age at injury were entered in the first step and the group (inflicted vs. noninflicted TBI) was entered in the second step. Predictor Variables The following predictor variables were examined: (1) age at injury; (2) presence/absence of abusive TBI; (3) severity of extracranial injuries based on the ISS, which reflected the severity of anatomic injury as indicated by the squared value of the 3 highest ISSs [31]; (4) pupil reactivity on admission (normal vs. unilaterally/bilaterally unreactive); (5) lowest postresuscitation modified GCS score; (6) duration of impaired consciousness, defined as the number of days the modified GCS motor score was less than 6, and (7) the number of lesions coded from acute CT/MRI scans. Each scan was evaluated for supratentorial and infratentorial extraaxial hemorrhages, intraparenchymal hematoma or edema/infarction, shear injury and chronic changes suggestive of prior brain abnormality (atrophy, ventriculomegaly, subdural hygromas). Because of the greater sensitivity of MRI to the presence of subdural hematomas, infarct/edema, shear injuries and myelination [32], only the findings from the MRI were included if both the CT and MRI findings were available. The total number of lesions was calculated by assigning a value of 1 for the presence of each injury. For example, if a child had a subdural hematoma extending across several lobes, then multiple lesions were coded. Skull fractures were not included in the total lesion score. Age at Injury Because the two brain injury groups (inflicted, noninflicted) differed significantly with respect to age at injury [F(1, 59) = 21.68; p ! 0.0001], including age at injury and the presence of abuse in the same model would eliminate most of the group effect. In order to control for age differences within each abuse group without eliminating the group effect, a residualized age variable was created by regressing age on abuse and subtracting the predicted age from the actual age at injury. This residualized variable was used as the predictor. Predictors of Outcome following Brain Injury 1.00000 –0.55 0.39 CT/MRI findings –0.55 1.00000 0.45 Duration of impaired consciousness –0.39 0.45 1.00000 Severity Factor Three of the injury variables – duration of impaired consciousness, number of lesions on CT/MRI imaging and lowest GCS score – were highly intercorrelated (table 2). If one predictor has 90% of its variance in common with other predictors, this can increase the variance of the parameter estimate (regression coefficient) by ten-fold. Combining the three highly correlated variables into one variable potentially provides a more powerful model. A principal axes factor analysis was performed to create a composite score from the three variables (lowest GCS, number of lesions on CT/MRI imaging and duration of impaired consciousness). This score accounts for the maximum common or shared variance in the set of variables. In order to do this, one must initially estimate the common variance in each variable (called communality). We specified the maximum correlation of each variable with the remaining variables as the initial communality estimate. Two factors were then extracted which accounted for all of the common variance among the three predictor variables. However, the first factor accounted for 93% while the second accounted for 7%. A Varimax rotation was performed followed by an oblique (Promax) rotation to determine if the factors were correlated. The first factor comprised the duration of impaired consciousness and lowest GCS scores. The second factor was mainly composed of the number of lesions on CT/MRI imaging. However, as the two factors were highly correlated (r = –0.71), it was decided to extract a single factor using a principal components procedure. This single factor accounted for 64% of the variance in the three measures and had high correlations with each variable, i.e. –0.81 with lowest GCS score, 0.85 with duration of impaired consciousness and 0.75 with lesions on CT/MRI imaging. Severity factor scores were estimated using the regression method for each subject and these factor scores were used in all succeeding analyses. Severity factor scores represent the estimate of the level of each subject on the underlying factor. Severity factor scores have a mean of 0 and a standard deviation of 1. A higher score indicates longer duration of impaired consciousness, more lesions on CT/MRI imaging and a lower score on the GCS. In figure 1, the profiles or the mean scores of the three predictor variables for the three levels of the severity factor score are displayed. For example, children who received a high severity factor score of 2 had an average GCS score of 3, 14 days of impaired consciousness and 9 lesions on CT/MRI imaging. Outcome Measures The GOS [19] was modified for infants and children. Good outcome referred to the return to age-appropriate or preinjury levels of functioning and the return to full-time classes with no special education services. Moderate disability was assigned based on: (1) a significant reduction in cognitive functioning from estimated premorbid levels; (2) motor deficits including hemiparesis interfering with activ- Pediatr Neurosurg 2002;36:64–74 67 steps. The first step predicted only the baseline outcome measure while the second included outcome measures at baseline, 3 months after injury and 1 year after injury. This procedure was used because of the smaller sample sizes for the later time points. The GENMOD procedure in SAS statistical software was used to analyze the GOS baseline data. Logistic regression for repeated measures was conducted on the follow-up time points. Results Fig. 1. The mean scores of the three predictor variables for the three levels of the severity factor score are displayed. For example, children who received a high severity factor score had an average GCS score of 3, 14 days of impaired consciousness and 9 lesions on CT/MRI imaging. DIC = Duration of impaired consciousness. ities of daily living; (3) referral to outpatient rehabilitation therapies, and (4) attending special education or resource classes. Severe disability was assigned if (1) cognitive functioning was in the deficient range, (2) severe motor deficits were present, such as lack of ageappropriate postural control or ambulation, and/or (3) there was referral for inpatient rehabilitation. The criteria for persistent vegetative state were unchanged and reflected total dependence for daily care. For the purposes of this study, moderate disability, severe disability and persistent vegetative state were classified as ‘poor outcome’. Previous studies have indicated that even mild sequelae in children with TBI can have a great impact on long-term development [20, 23]. By including children with moderate disability in the poor outcome category, this study more accurately reflects the impact of mild and moderate impairments on development. The Bayley Scales of Infant Development Mental and Motor Scales, 2nd edition [33], were used to assess cognitive and motor functions in children under the age of 42 months. Bayley scores were corrected for prematurity for children with gestational ages of 32–37 weeks. For children aged 43–71 months, the Stanford-Binet Intelligence Scale, 4th edition [34], and the McCarthy Scales of Children’s Abilities [30] motor scales were administered. All cognitive and motor scores were converted to standard scores with a mean of 100 and a standard deviation of 15. Multivariate Predictor Models Each outcome measure was analyzed using residualized age at injury, the severity factor score and pupillary reactivity. The analysis was done hierarchically in that all the variables except the presence of abuse were entered into the model initially. The variables which did not contribute significantly to the model were eliminated. The presence of abuse was entered last to reduce the bias in the parameter estimates which may occur when noncontributing factors are included in the model. Each outcome measure was analyzed in two 68 Pediatr Neurosurg 2002;36:64–74 Relationships of Individual Predictors to Outcome Measures To examine the relationships between the injury severity and outcome variables, correlation coefficients were calculated for each time interval. Pearson product moment coefficients are provided for continuous variables (table 3). As displayed in table 2, the three variables reflecting the severity of the injury (lowest GCS, duration of impaired consciousness and number of lesions on CT/ MRI imaging) correlated strongly with all three outcome measures at baseline. Duration of impaired consciousness and lowest GCS correlated with all three outcome measures 1 year after injury. The presence of inflicted TBI was significantly related to cognitive outcome at baseline and 1 year after injury. Glasgow Outcome Scale The GOS score was treated as a dichotomous variable, with good outcome reflecting the category of good recovery and poor outcome reflecting the categories of moderate recovery, severe disability and persistent vegetative state. Correlational analysis indicated that baseline GOS was significantly associated with lowest GCS, duration of impaired consciousness, number of lesions on CT/MRI imaging and the presence of inflicted brain injury. Because of the significant intercorrelations among the three severity variables, a single severity factor score was used (calculation of the severity factor score is detailed above). The severity factor, pupil reactivity, ISS and the residualized age score were used in the baseline model to predict GOS scores. The initial model for the baseline GOS indicated an adequate fit to the data [¯2(52) = 49.21; p = 0.58]. The severity factor was significant [¯2(1) = 17.72; p ! 0.0001]. High factor scores were associated with a decreased likelihood of being classified as having good recovery on the GOS. For example, a factor score 1 standard deviation above average (indicating high severity) would predict a 0.045 chance of good recovery, whereas 1 standard deviation below average (indicating low severity) would predict a 0.726 likelihood of good recovery. ISS, Prasad/Ewing-Cobbs/Swank/Kramer Table 3. Pearson and point-biserial correlations between injury variables and outcome Variables GOS Age at injury Abuse Pupil reactivity ISS GCS DIC Lesions 1-Year IQ Baseline 0.17 0.37** 0.12 –0.21 0.37** –0.30* –0.37** cognitive motor 0.16 0.27* 0.08 –0.12 0.43** –0.43** –0.45** 0.09 0.14 –0.06 –0.12 0.31* –0.46** –0.40** GOS 0.03 0.17 0.31* –0.12 0.54** –0.41** –0.21 cognitive motor 0.26 0.33* 0.22 –0.29* 0.43** –0.33* –0.45** 0.11 0.17 0.23 –0.34* 0.31* –0.50** –0.38** * p ! 0.05; ** p ! 0.01. DIC = Duration of impaired consciousness. pupil reactivity and the age residual were not found to be significantly related to outcome and were removed from the model. The presence of abuse was added to the model and it significantly predicted GOS outcome [¯2(1) = 19.8; p = 0.0001]. The severity factor score also continued to predict uniquely [¯2(1) = 31.34; p ! 0.0001]. The severity factor by group interaction was not significant. Children who had inflicted brain injury had a lower probability of being classified as having good recovery on the GOS when the severity of the injury was controlled for. The result indicates that for individuals who had low severity factor scores at least 1 standard deviation below the mean (higher GCS, little or no unconsciousness and few lesions on CT/MRI imaging), the model predicted a likelihood of 0.986 for good recovery for those in the noninflicted TBI group but only 0.552 for the inflicted TBI group. For a factor score of 0 (average for the sample), we would predict a likelihood of 0.574 for good recovery for the noninflicted TBI group but only 0.023 for the inflicted TBI group. For greater trauma (factor scores 1 standard deviation above the mean), we would predict a likelihood of good recovery on the GOS for noninflicted TBI children of 0.025, compared to 0.00045 for inflicted TBI children. Cognitive Outcome Baseline cognitive scores were found to be highly correlated with the presence of inflicted brain injury, lowest GCS, duration of impaired consciousness and number of lesions on CT/MRI neuroimaging. Because of the intercorrelations between the three severity variables (lowest GCS, duration of impaired consciousness and the number of lesions on CT/MRI), the severity factor score was used in the following analyses. A general linear model of severity factor, residualized age, pupil reactivity and ISS score was used to predict the cognitive outcome at baseline. The Predictors of Outcome following Brain Injury full model was significant [R2 = 0.299; F(4,52) = 5.54; p = 0.0009]. ISS, pupil reactivity and residualized age were not found to contribute significantly to the prediction of baseline cognitive scores above the severity factor and were thus eliminated from subsequent analyses. The severity factor alone accounted for 28.9% of the variance in baseline cognitive outcome [F(1,58) = 23.55; p ! 0.0001]. Addition of the presence of inflicted brain injury to the severity factor resulted in a model R2 of 0.376 [F(2, 57) = 17.17; p ! 0.0001]. The presence of inflicted brain injury predicted an additional 8.7% of the variance [F(1,57) = 7.97; p = 0.0005] above the severity factor [F(1,57) = 27.37; p ! 0.0001]. After controlling for the contribution of the severity factor, children who had been abused scored 9.05 points lower on the cognitive score than children who had not been abused. A repeated-measures hierarchical regression analysis was performed to determine the relation of the severity factor and the presence of abuse to cognitive outcome over time (baseline, 3 months after injury and 1 year after injury). The first model included severity factor, pupil reactivity, ISS and residualized age. There were no significant changes in cognitive scores over time and the interactions between the predictor variables and change in cognitive scores over time were also not significant. Only the severity factor had a significant relationship to cognitive scores over time [F(1,49) = 16.33; p = 0.0002]. Therefore, all of the predictors except the severity factor were removed from the model and the presence of abuse was added to the model. Again, there was no difference in scores over time nor did changes over time relate to either abuse group or the severity factor. Both the presence of abuse [F(1,48) = 12.97; p = 0.0007] and the severity factor [F(1,48) = 23.83; p ! 0.0001] were consistently related to the cognitive scores over time. Figure 2 depicts the recov- Pediatr Neurosurg 2002;36:64–74 69 Fig. 2. There was no significant improvement in cognitive or motor scores from baseline to 1 year after injury for either the inflicted TBI group or the noninflicted TBI group. A repeated-measures hierarchical regression analysis was performed to determine the relation of the severity factor, ISS, residualized age and pupillary reactivity to motor outcome over time. Because ISS scores, residualized age and the presence of abuse were not significantly related to the change in motor scores over time, these variables were dropped. There was no significant change in motor scores over time [F(2,96) = 0.22; p = 0.80]. There was a trend for children who had pupil abnormalities to do worse early on but not at the follow-up time points [F (2,96) = 2.57; p = 0.08]. The severity factor was related to motor scores [F(1,48) = 37.62; p ! 0.0001] and there was a trend for pupillary reactivity [F(1,48) = 3.46; p = 0.069] to be related to motor scores as well. While general severity may be inversely related to motor scores consistently over time, pupil reactivity may enhance the prediction of early motor scores but not late outcome. Discussion ery curve for the longitudinal cognitive and motor scores by injury group. Motor Outcome Motor scores at baseline were found to be significantly correlated with lowest GCS, duration of impaired consciousness and number of lesions on CT/MRI imaging. Because of the significant intercorrelations between the three severity variables, the severity factor score was used (calculation of the severity factor score is detailed above). Severity factor, residualized age at injury, pupil reactivity and the ISS score were used to predict motor outcome at baseline. The full model was significant [R2 = 0.315; F(4,52) = 5.97; p = 0.0005]. However, only the severity factor and pupil reactivity were found to be significantly related to baseline motor outcome. Therefore, ISS and residualized age were removed from the model and the model was reexamined. Children who had a high severity factor score (higher duration of impaired consciousness, lower GCS and more abnormalities on CT/MRI) had lower baseline motor scores [F(1,57) = 24.81; p ! 0.0001]. Children with pupil abnormalities had worse baseline motor outcome than those with bilaterally reactive pupils [F(1,57) = 4.92; p = 0.0305]. The overall model accounted for 30.6% of the variance in motor scores [F(2,57) = 12.58; p ! 0.0001]. Addition of the presence of abuse [F(1,56) = 1.78; p = 0.19] to this model did not increase the predictive power of the model, increasing R2 by only 2.2%. 70 Pediatr Neurosurg 2002;36:64–74 This study examined the relationship between several injury variables, age at injury and the presence of inflicted brain injury and longitudinal outcome following moderate to severe TBI in children under the age of 6 years. Of the seven predictor variables examined, three predictors (lowest GCS, duration of impaired consciousness and number of lesions on CT/MRI imaging) showed robust relationships with all three outcome measures. That is, these three variables were highly predictive of early outcome and outcome 1 year after injury. A single brain injury severity factor consisting of the lowest postresuscitation GCS score, duration of impaired consciousness and the number of intracranial lesions visualized on CT/MRI accounted for a significant amount of the variance in the baseline GOS, cognitive and motor scores and continued to show strong relationships with all outcome measures at 1 year. The presence of abuse was a moderating factor in both GOS and cognitive outcome. Motor scores were related to injury severity and pupillary responsiveness. Age at injury and ISS did not account for significant variability in outcome. Methodological Issues The severity factor was a composite score drawn from lowest GCS, duration of impaired consciousness and number of lesions on CT/MRI imaging. This score reflected the individual subject’s level on three of the predictor variables such that a higher severity factor score indicates lower GCS score, longer duration of impaired Prasad/Ewing-Cobbs/Swank/Kramer consciousness and more lesions on CT/MRI imaging. A factor score was used because these three variables were highly correlated with each other and multicollinearity can mask the relationship of the predictors to the outcome measure. Their high correlations with each other and the outcome measures indicated that no one variable superseded the other two in predictive ability. The severity factor accounted for a significant amount of the variance in the baseline GOS, cognitive and motor scores and continued to show strong relationships with all outcome measures 1 year following the TBI. Using a factor score as opposed to individual predictors allowed for a more powerful statistical model. The use of the GCS [35] and GOS [36] in infants and toddlers is controversial. The GCS uses verbal and motor behaviors that are not appropriate for children less than 2 years of age. The modifications of the GCS used in this study, which are similar to those of Hahn et al. [16], yielded a robust relationship to outcome. Relatedly, the duration of impaired consciousness, which was measured by the GCS motor scale score, was also a robust predictor of outcome. The GOS, like the GCS, was developed for adults and emphasizes return to a preinjury level of functioning. In children, TBI disrupts the maturation or development of skills; thus, using a return to premorbid skill level as a measure of outcome in children fails to address the effect of TBI on development. Consequently, the GOS can significantly underestimate the impact of the brain injury in young children. Similar to our previous study [21], we modified the GOS to reflect child-specific parameters such as the acquisition of age-appropriate skills and the need for rehabilitative therapies or special education services. Although these modifications appear to have increased the sensitivity of the GOS to outcome in young children with TBI, more work is needed to fully validate this modification of the GOS. Prediction of Outcome Despite the strong predictive ability of the severity variables, the severity factor did not account for the majority of variance in any of the outcome measures. The two traumatic brain injury groups (inflicted and noninflicted) did not significantly differ in injury severity as measured by the GCS, duration of impaired consciousness and number of lesions on CT/MRI imaging. The presence of inflicted brain injury was strongly related to outcome. Regarding the GOS score, children with inflicted TBI had significantly less favorable GOS scores at baseline. The presence of inflicted brain injury also mod- Predictors of Outcome following Brain Injury erated cognitive outcome at baseline and 1 year after injury. After accounting for variance attributable to the severity of the brain injury, abuse significantly improved the prediction of cognitive scores. Overall, children with inflicted TBI scored on average 9 points lower at the baseline evaluation than children with noninflicted TBI. This differential in intellectual functioning was stable over the 1-year follow-up and indicated persistent deficits in cognitive development. Our findings are consistent with those obtained in previous studies that indicated poor outcome following inflicted TBI [21, 29]. There were no significant interactions between cognitive outcome and the predictor variables. The two brain injury groups did not differ in regard to socioeconomic status, family history of psychiatric illnesses or the receipt of rehabilitation services following the injury. However, there were significant differences in the types of lesions found on CT/MRI imaging. Children with inflicted brain injury had significantly more extraaxial hemorrhages and atrophy was present in a significant number of children despite no known history of previous neurotrauma. The high rate of atrophy on acute neuroimaging in the inflicted TBI group raises the possibility of previous cerebral trauma for which no medical attention was sought. The unfavorable outcome following inflicted brain injury may be related to an interaction between several factors including previous neurological injury, as reflected in the high rate of atrophy on acute imaging, and environmental differences. It is known that children who have been physically abused or neglected and who have no known brain injuries score significantly lower than their matched controls on measures of cognitive development [37, 38]. Family environment has been shown to play a significant role in outcome following accidental brain injury in school-age children and adolescents [39] and is likely to play a pronounced role in the cognitive development of children with inflicted brain injury. Motor scores were strongly related to the severity of brain injury at baseline; only pupillary responsiveness improved the prediction of motor scores. Motor scores remained stable over time. For our sample, pupillary responsiveness was not strongly related to GOS or cognitive outcomes. Previous studies have reported inconsistent findings regarding the relationship between pupillary responses and outcome. While Massagli et al. [5] reported that impaired pupillary response was not associated with GOS scores obtained at discharge or follow-up, other investigators noted less favorable GOS scores 6 months after TBI in children with pupillary abnormalities [10, 26]. In sharp contrast to the GOS and cognitive scores, the Pediatr Neurosurg 2002;36:64–74 71 presence of abuse was not significantly related to motor outcome. This difference in the impact of abuse on cognitive and motor outcome suggests that cognitive outcome may be more influenced by psychosocial factors whereas motor outcome may be more directly related to neurological variables such as prolonged unconsciousness and the depth of the unconsciousness. However, it is important to note that the model of severity factor score and pupillary reactivity did not account for the majority of variance in motor outcome. Thus, although the presence of abuse did not have a significant role in motor outcome, other environmental factors such as rehabilitative services may be strongly related to motor outcome. Age at injury was not found to significantly predict early or 1-year outcome. This finding is contrary to expectations, as many studies examining outcome across the entire pediatric age range reported higher rates of mortality, morbidity and/or cognitive impairment in infants and preschoolers as compared to school-age children and adolescents [6–14, 40]. It is possible that outcomes vary more between infants and school-age children than between infants and preschoolers. Because of the limited age range used in this study, it is not possible to ascertain whether young children differ from older children in regard to outcome. Additionally, most of the children with inflicted brain injury were between the ages of 2 and 24 months. Thus, age at injury is confounded by the presence of abuse in the youngest age range. Direct comparison of injury characteristics and outcomes in infants, older children and adolescents is essential to identify differences attributable to age at injury. Recovery curves in older children and adolescents indicate significant improvement in IQ scores during the first 3–6 months after TBI followed by a minimal change 1–2 years after injury [41, 42]. Recovery curves depicting the change in IQ scores over time appear flatter in young children than in school-age children and adolescents [21, 43]. In the present study, the minimal improvement in IQ scores over time in young children with inflicted and noninflicted TBI is alarming. Several mechanisms have been hypothesized to account for the greater disruption in cognitive development in children with early brain injury: greater vulnerability of the immature brain to injury, greater effect on subsequent neuronal development, neural degeneration or damage to neural systems responsible for skill acquisition [44]. Young children may be particularly vulnerable to the diffuse cerebral damage associated with TBI. Outcome in young children with inflicted TBI is poor; many children are rendered intellectually deficient and/or have major neurological deficits [21, 29, 45]. 72 Pediatr Neurosurg 2002;36:64–74 It is possible that the modest recovery in younger children is restricted by their smaller and less well-established skill repertoire, which is more prone to disruption by diffuse brain injury. Young children with TBI are at risk of increasing developmental deficits over time due to lower cognitive functioning as well as deficits in the areas of regulation of attention, initiating social interactions and responding to the social behavior of others [22]. Additionally, there is a higher incidence of inflicted TBI in very young children and the presence of inflicted brain injury adversely influences outcome. Inflicted TBI involves a unique mechanism of injury to the brain as well as significant psychosocial factors that can influence outcome. Thus, studies that address the long-term outcome of infants and preschoolers must address the complicated issue of inflicted brain injury. More research is needed to address the factors that may moderate outcome from early brain injury such as family environment and therapeutic interventions. No study to date has specifically addressed the relationship of commonly used measures of severity to outcome in infants and preschoolers. This study showed that the most commonly used measures of injury severity (GCS, duration of impaired consciousness and CT/MRI findings) were strongly related to outcome in young children. A larger sample size would permit the introduction of additional injury variables and would increase the power to detect interactions between predictors and outcome. The differential relationship between the presence of abuse and the three outcome measures suggests that psychosocial factors may play a greater role in cognitive outcome and GOS. Although we know that the presence of abuse is detrimental to cognitive outcome, we do not know the role of rehabilitative services or the postinjury home environment in recovery from TBI in young children. Historically, children with TBI have not been identified for rehabilitative services early in their recovery and as such, often do not receive appropriate services. It is important for clinicians to recognize that the standard adult measures such as the GCS and GOS are not appropriate for young children and that a modified version must be used to fully assess the severity of the brain injury and long-term outcome. The GCS used in this study was modified for young children and was found to be a robust predictor of outcome. Likewise, the GOS was modified to reflect not only return to premorbid skill level but also to account for slowed acquisition of age-appropriate skills. The modified GOS used in this study accounted for developmental expectations and correlated strongly with formal measures of cognitive development. Failure to con- Prasad/Ewing-Cobbs/Swank/Kramer sider developmental issues when assessing severity or outcome can greatly underestimate the impact of brain injury on long-term outcome and may result in a lack of essential rehabilitative and special education services. There is a need for functional outcome measures that address the unique issues of early childhood. Deficits in children under the age of 2 years are often underestimated [20] and impairments can become apparent over time as a child fails to meet developmental expectations. Thus, it is important that functional outcome scales address both current and long-term outcome for young children. Functional outcome measures in young children should address not only cognitive deficits but behavioral changes as well. Behavioral dysregulation such as disinhibition and inattentiveness are common following brain injury in children and these issues can greatly impact on educational and social development. Acknowledgements The participation of the families of the children and the assistance provided by the Harris County Children’s Protective Services in this research is gratefully acknowledged. This research was supported in part by NIH NINDS grant No. R01-NS29462, Accidental and Nonaccidental Pediatric Brain Injury, to L.E.-C. References 1 Berger MS, Pitts LH, Lovely M, Edwards MSB, Bartowski HM: Outcome from severe head injury in children and adolescents. J Neurosurg 1985;62:194–199. 2 Bruce DA, Schut L, Bruno LA, Wood JH, Sutton LN: Outcome following severe head injuries in children. J Neurosurg 1978;48:679– 688. 3 Braakman R, Gelpke GJ, Habbema JDF, Maas AIR, Minderhoud JM: Systematic selection of prognostic features in patients with severe head injury. Neurosurgery 1980;6:362–370. 4 Costeff H, Grosswasser Z, Goldstein R: Longterm follow-up review of 31 children with severe closed head trauma. J Neurosurg 1990;73: 684–687. 5 Massagli TL, Michaud LJ, Rivara FP: Association between injury indices and outcome after severe traumatic brain injury in children. Arch Phys Med Rehabil 1996;77:125–132. 6 Filley CM, Cranberg LD, Alexander MP, Hart EJ: Neurobehavioral outcome after closed head injury in childhood and adolescence. Arch Neurol 1987;44:194–198. 7 Koskiniemi M, Kyykka T, Nybo T, Jarho L: Long-term outcome after severe brain injury in preschoolers is worse than expected. Arch Pediatr Adolesc Med 1995;149:249–254. 8 Kriel RL, Krach LE, Panser LA: Closed head injury: Comparison of children younger and older than 6 years of age. Pediatr Neurol 1989; 5:296–300. 9 Jennett B, Teasdale G, Braakman R, Minderhoud JM, Heien J, Kurze T: Prognosis of patients with severe head injury. Neurosurgery 1979;4:283–289. 10 Levin HS, Aldrich EF, Saydjari C, et al: Severe head injury in children: Experience of the Traumatic Coma Data Bank. Neurosurgery 1992;31:435–444. 11 Luerssen TG, Klauber MR, Marshall LF: Outcome from head injury related to patient’s age. A longitudinal prospective study of adult and pediatric head injury. J Neurosurg 1988;68: 409–416. Predictors of Outcome following Brain Injury 12 Michaud LJ, Rivara FP, Grady MS, Reay DT: Predictors of survival and severity of disability after severe brain injury in children. Neurosurgery 1992;31:254–264. 13 Thakker JC, Splaingard M, Zhu J, Babel K, Bresnahan J, Havens PL: Survival and functional outcome of children requiring endotracheal intubation during therapy for severe traumatic brain injury. Crit Care Med 1997;25: 1396–1401. 14 Raimondi AJ, Hirschauer J: Head injury in the infant and toddler. Coma scoring and outcome scale. Childs Brain 1984;11:12–35. 15 Teasdale G, Jennett B: Assessment of coma and impaired consciousness: A practical scale. Lancet 1974;ii:81–84. 16 Hahn YS, Chyung C, Barthel M, Bailes J, Flannery A, McLone DG: Head injuries in children under 36 months of age. Demography and outcome. Childs Nerv Syst 1988;4:34–40. 17 Reilly PL, Simpson DA, Sprod R, Thomas L: Assessing the conscious level in infants and young children: A pediatric version of the Glasgow Coma Scale. Childs Nerv Syst 1988;4:30– 33. 18 Durham SR, Clancy RR, Leuthardt E, Sun P, Kamerling S, Dominguez T, Duhaime AC: CHOP Infant Coma Scale (‘Infant Face Scale’): A novel coma scale for children less than two years of age. J Neurotrauma 2000;17:729–737. 19 Jennett B, Bond M: Assessment of outcome after severe brain damage. Lancet 1975;i:480– 487. 20 Crouchman M, Rossiter L, Colaco T, Forsyth R: A practical outcome scale for paediatric head injury. Arch Dis Child 2001;84:120–124. 21 Ewing-Cobbs L, Kramer L, Prasad M, Canales DN, Louis PT, Fletcher JM, Vollero H, Landry SH, Cheung K: Neuroimaging, physical, and developmental findings after inflicted and noninflicted traumatic brain injury in young children. Pediatrics 1998;102:300–307. 22 Ewing-Cobbs L, Prasad M, Kramer L, Landry SH: Inflicted traumatic brain injury: Relationship of developmental outcome to severity of injury. Pediatr Neurosurg 1999;31:251–258. 23 Fiser DH, Long N, Roberson PK, Hefley G, Zolten K, Brodie-Fowler M: Relationship of pediatric overall performance category and pediatric cerebral performance category scores at pediatric intensive care unit discharge with outcome measures collected at hospital discharge and 1- and 6-month follow-up assessments. Crit Care Med 2000;28:2616–2620. 24 DiScala C, Grant CC, Brooke MM, Gans BM: Functional outcome in children with traumatic brain injury. Am J Phys Med Rehabil 1992;71: 145–148. 25 McDonald CM, Jaffe KM, Faye GC, Polissar NL, Martin KM, Liao S, Rivara J: Comparison of indices of traumatic brain injury severity as predictors of neurobehavioral outcome in children. Arch Phys Med Rehabil 1994;75:328– 337. 26 Ong L, Selladurai BM, Dhillon MK, Atan M, Lye MS: The prognostic value of the Glasgow Coma Scale, hypoxia and computerised tomography in outcome prediction of pediatric head injury. Pediatr Neurosurg 1996;24:285–291. 27 Duhaime AC, Alario AJ, Lewander WJ, et al: Head injury in very young children: Mechanisms, injury types, and ophthalmologic findings in 100 hospitalized patients younger than 2 years of age. Pediatrics 1992;90:179–185. 28 Hahn YS, Raimondi AJ, McLone DG, Yamanouchi Y: Traumatic mechanisms of head injury in child abuse. Childs Brain 1983;10:229– 241. 29 Duhaime AC, Christian CW, Moss E, Seidl T: Long-term outcome in infants with the shaking-impact syndrome. Pediatr Neurosurg 1996; 24:292–298. 30 McCarthy D: McCarthy Scales of Children’s Abilities. New York, Psychological Corporation, 1972. Pediatr Neurosurg 2002;36:64–74 73 31 Baker SP, O’Neill B, Haddon W Jr, Long WB: The injury severity score: A method for describing patients with multiple injuries and evaluating emergency care. J Trauma 1974;14: 187–196. 32 Sato Y, Yuh WTC, Smith WL, Alexander RC, Kao SCS, Ellerbrook CJ: Head injury in child abuse: Evaluation with MR imaging. Radiology 1989;173:653–657. 33 Bayley N: Bayley Scales of Infant Development, ed 2. San Antonio, Psychological Corporation, 1993. 34 Thorndike RL, Hagen EP, Sattler JM: Stanford-Binet Intelligence Scale, fourth edition: Guide for Administering and Scoring. Chicago, Riverside, 1986. 35 Luerssen TG: Acute traumatic cerebral injury; in McLone DG, Marlin AE, Reigel DH, Scott RM, Walker ML, Steinbok P, Cheek WR (eds): Pediatric Neurosurgery. Philadelphia, Saunders Company, 2001, pp 601–617. 74 36 Koelfen W, Freund M, Dinter D, Schmidt B, Koenig S, Schultze C: Long-term follow up of children with head injuries classified as ‘good recovery’ using the Glasgow Outcome Scale: Neurological, neuropsychological and magnetic resonance imaging results. Eur J Pediatr 1997;156:230–235. 37 Appelbaum AS: Developmental retardation in infants as a concomitant of physical child abuse. J Abnorm Child Psychol 1977;5:417– 423. 38 Hoffman-Plotkin D, Twentyman CT: A multimodal assessment of behavioral and cognitive deficits in abused and neglected preschoolers. Child Dev 1984;55:794–802. 39 Yeates KO, Taylor HG, Drotar D, Wade S, Klein S, Stancin T, Schatschneider C: Preinjury family environment as a determinant of recovery from traumatic brain injuries in school-age children. J Int Neuropsychol Soc 1997;3:617–613. 40 Anderson VA, Catroppa C, Morse S, Haritou F, Rosenfeld J: Recovery of intellectual ability following traumatic brain injury in childhood: Impact of injury severity and age at injury. Pediatr Neurosurg 2000;32:282–290. Pediatr Neurosurg 2002;36:64–74 41 Chadwick O, Rutter M, Brown G, Shaffer D, Traum M: A prospective study of children with head injuries. II. Cognitive sequelae. Psychol Med 1981;11:49–61. 42 Jaffe KM, Fay GC, Polissar NL, Martin KM, Shurtleff HA, Rivara JM, Winn HR: Severity of pediatric traumatic brain injury and neurobehavioral recovery at one year – a cohort study. Arch Phys Med Rehabil 1993;74:587– 595. 43 Anderson V, Morse SA, Klug GL, Catroppa C, Haritou F, Rosenfield J, Pentland L: Predicting recovery from head injury in young children: A prospective analysis. J Int Neuropsychol Soc 1997;3:568–580. 44 Taylor HG, Alden J: Age-related differences in outcomes following childhood brain insults: An introduction and overview. J Int Neuropsychol Soc 1997;3:555–567. 45 Bonnier C, Nassogne MC, Evrard P: Outcome and prognosis of whiplash shaken infant syndrome; late consequences after a symptom-free interval. Dev Med Child Neurol 1995;37:943– 956. Prasad/Ewing-Cobbs/Swank/Kramer Copyright: S. Karger AG, Basel 2002. Reproduced with the permission of S. Karger AG, Basel. Further reproduction or distribution (electronic or otherwise) is prohibited without permission from the copyright holder.