Future Research On Adolescent Brain Injury
Investigations to identify effective approaches to the rehabilitation of children with TBI may benefit from the literature in other related fields. Future research could be guided by themes that have emerged across many disability groups. Although TBI has its unique features, it shares many characteristics with other disabilities. The task is to identify the shared characteristics, and include what has been learned in other fields when designing interventions. One example is social skills training. Certain models for social skills training and cognitive rehabilitation have been shown to be ineffective with people who have other, similar disabilities, yet these models are being used in TBI rehabilitation. At the very least, the failure of these interventions in other fields should call into question their effectiveness with TBI. Similarly, we should pay attention to, and systematically test, successful approaches in other fields.
Three gaps in the literature pertaining to child and adolescent TBI define priorities for future research.
1. Insufficient evidence exists about the natural history of TBI in this population. Longitudinal, observational studies with large samples are needed to provide this information. Such studies could help us understand and define the subsets of severity categories, assessments, and interventions. Without distinct subsets, we will continue to pool diverse groups into the same research sample and produce results of questionable value.
2. Interventions must be tested with experimental study designs.
3. Because of the strong influence of development on all aspects of life for this population, both longitudinal and experimental studies must incorporate concepts of human development presented in this paper, as well as sophisticated methods of analysis capable of accounting for individual variation.
The field of developmental psychology provides a technology for designing studies and analyzing data that accounts for individual variation. The following are recommendations for research about children and adolescents with TBI that incorporates this technology.
Life-Span Developmental Psychology and the Study of Child and Adolescent TBI
Application of Life-Span Developmental Psychology to child and adolescent TBI introduces the study of predictors of individual growth curves (both stability and change), including recovery as a function of an intervention, spontaneous recovery, and short- and long-term outcomes. Using a life-span approach to study the effectiveness of interventions in pediatric TBI requires that we:
1. Distinguish between stability and change that would have occurred without intervention from that which is a direct result of the intervention (Montado and Schmitt, 1982).
2. Determine whether there are life periods in which certain interventions are more or less efficacious.
3. Attempt to identify which individual characteristics (e.g., age, education level, etc.) are associated with improved outcomes from treatment.
Research regarding the effectiveness of interventions in child and adolescent TBI should be designed to collect information that could determine if individual factors (e.g., child temperament or intelligence), historical/environmental events (e.g., family or school environment), and/or nonnormative events (e.g., death in the family or divorce) account for longitudinal profiles of recovery, problems, and outcomes. In such studies, interventions following child and adolescent TBI would be considered as systems in past and/or present environments, and TBI would be considered one very significant nonnormative event.
Consideration of these potential interacting influences on individuals throughout the life course would result in a developmental effectiveness study that would ideally follow children with TBI and a comparison group throughout the remainder of life and would repeatedly measure outcomes of interest. To go beyond linear growth curves to understand nonlinear change, including acceleration or deceleration of growth (e.g., Thompson, Francis, Stuebing, et al., 1994), outcomes of interest would need to be measured a minimum of three times in addition to a baseline measure. Given the difficulty of completing such studies, the perspective of life-span developmental psychology suggests some standards for research that can strengthen studies of child and adolescent TBI.
Elements of Research Designs
Prospective Longitudinal Research Designs
More information is needed about changes among the pediatric TBI population in domains such as cognitive, motor, emotional, and social performance. In all cases, performance must be conceptualized as time-dependent. However, we cannot rely on retrospective studies using registries or medical chart reviews to test hypotheses about age changes following interventions (Montado and Schmitt, 1982). To adequately test hypothesized age changes as a result of interventions, prospective longitudinal research designs are required. Hypotheses that could be tested with these designs include predictions about the effects of interventions on social or cognitive growth curves, predictions about the differential effectiveness of a variety of interventions on individual profiles of change or patterns of profiles, and predictions about interrelationships between changing environments (including interventions) and changing individuals.
Internal Validity and Comparison Groups
Threats to internal validity are defined as those that make it difficult to determine "whether or not the relationship observed is accurately or validly identified or interpreted" (Baltes, Reese, and Nesselroade, 1977, p. 37). Baltes and colleagues note that "maturation" (age change that naturally occurs) is usually considered a threat to internal validity. When studying the effectiveness of interventions, the goal is to differentiate the maturation effects from the links between the intervention and time-dependent outcomes. Therefore, maturation effects complicate the study of the effectiveness of pediatric TBI interventions, and complex research designs, such as quasiexperimental research and advanced statistical techniques, are needed to tease apart change in outcomes related to recovery from change due to maturation.
Since the use of quasiexperimental designs with closely matched controls does not account for all potential bias and error (Campbell and Stanley, 1966) and because randomized controlled trials have not been done, some researchers have suggested the use of "natural experiments" (Michaud, 1994). Natural experiments compare the efficacy of treatment approaches by grouping children with TBI by their natural access to acute care or rehabilitative services (e.g., as a result of regional differences in standards of care). Given the apparent variety in managing children following TBI, natural experiments could yield valuable information, especially if they are designed to assess intraindividual change. Adequate control groups would still be necessary to differentiate maturation that would have occurred regardless of intervention from growth patterns that are a direct result of intervention efforts.
The approaches we suggest will result in complex research designs and data sets in which children with and without TBI are followed for extended periods of time with multiple times of assessment. In addition, the children, and possibly their families, teachers, or friends, would be asked to complete multiple assessments at each time of measurement. The focus becomes predicting individual growth curves from antecedent variables such as age at injury, initial child status, environmental factors (e.g., family functioning), and intervention techniques. Hence, the focus is on the prediction of growth curves and trajectories in performance among children following TBI compared with control children and/or children with TBI participating in alternative interventions. To maximize these data, several statistical techniques are available including growth curve modeling, cluster analysis, and time-series analysis.
Growth Curve Models
One technique used for repeated measurements of intervally scaled data is growth curve modeling (sometimes called general linear mixed models, multilevel models, or hierarchical linear models [HLMs]) (Bailey, Burchinal, and McWilliam, 1993; Bryk and Raudenbush, 1992; Burchinal and Appelbaum, 1991; Francis, Fletcher, Steubing, et al., 1991; Thompson, Francis, Stuebing, et al., 1994; Willett, Ayoub, and Robinson, 1991). Several software applications are available to complete growth curve modeling, including a mixed modeling procedure (PROC MIXED) available in SAS/STAT software1 (SAS Institute, Inc., Cary, NC; Littell, Milliken, Stroup, et al., 1996), HLM (Bryk and Raudenbush, 1987), and structural equation modeling software (see McArdle and Epstein, 1987; Willett and Sayer, 1994).
Typically, these methods have been used to describe growth curves among a population or to compare the average growth curve of one group to another (e.g., an intervention and a control group or males and females) (Bailey, Burchinal, and McWilliam, 1993; Francis, Fletcher, Stuebing, et al., 1991). Other, less common applications of this method can answer questions about interindividual differences in intraindividual change. For example, the target of study can be variation among individuals in their patterns of change over time. Sample questions include whether the initial level of an antecedent variable (e.g., initial child status, injury severity, or acute rehabilitative efforts) launches the trajectory of an outcome of interest (e.g., performance IQ or visual-motor tasks). This has been called a launch relation. One also can ask whether a measured variable must be maintained at some level to promote a more positive change pattern (e.g., some component of rehabilitation; an ambient-level relation). Finally, one can investigate associations between multiple change processes occurring simultaneously (e.g., time-dependent cognitive performance and time-dependent social performance; a change-to-change relation) (Connell and Skinner, 1990; Skinner, Zimmer-Gembeck, and Connell, 1998; Zimmer-Gembeck, 1998).
A recent example of the use of HLM to study launch relations was completed by Thompson and colleagues (1994). In this study, researchers expected that pupillary status at admission, Glasgow Coma Scale 24 hours postinjury, the duration of impaired consciousness, and age at injury (6 to 15 years) would predict intraindividual change in motor, visual-spatial, and somatosensory skills. Throughout the following 5 years, motor, visual-spatial, and somatosensory skills of children with TBI were assessed up to seven times. Instead of comparing mean outcomes between groups of children formed based upon predictor variables, individual growth curves in performance were the outcomes of interest. The focus of the study was shifted from "the differences between performance at fixed time points to the individual growth trajectory" (Thompson, Francis, Stuebing, et al., 1994, p. 333).
Thompson and colleagues (1994) claimed that the measurement of outcome as a single endpoint might result in artificially weak relationships between injury characteristics, other hypothesized predictors, and outcome because of the insensitivity to change processes. Therefore, they designed their study to analyze intraindividual patterns of change. They demonstrated that growth curves in motor, visual-spatial, and somatosensory tasks were launched by age at injury, initial status, and characteristics of injury. They also demonstrated that rate of recovery differed by age at injury, and age at injury interacted with injury severity to predict rate of recovery for some outcomes. Controlling for initial status improved estimates of the influences of injury severity and age at injury on growth curves. However, this study did not attempt to differentiate recovery from maturation, and no data on treatment or interventions were included. Future research on the effectiveness of interventions in pediatric TBI should build upon the approach of Thompson and colleagues (1994) by including control groups and the assessment of interventions (see also Yeates, Taylor, Drotar, et al., 1997 for another example).
A second promising approach to take advantage of repeated measures of intervally scaled data is cluster analysis (Bergman, Eklund, and Magnusson, 1991). Cluster analysis can be used to empirically capture distinct profiles based upon patterns of change. Clustering algorithms are used to classify children based upon repeated measures of outcomes of interest; these clusters could be compared to determine whether they differ on characteristics of interest such as intervention techniques, initial status measures, family characteristics, or other outcomes of interest. Although we did not locate a study that used this technique to evaluate pediatric TBI, we found a study of self-esteem during adolescence that is an example of the utility of the method. Hirsch and DuBois (1991) used cluster analysis to identify patterns of change in self-esteem with four times of measurement over a 2-year period. They identified four distinctly different patterns of change in self-esteem and demonstrated that boys were overrepresented in the group with sharply increasing self-esteem while girls were overrepresented in the group with sharply decreasing self-esteem. This design might be useful in discovering what variables (e.g., sex or age) associate with patterns of recovery in child and adolescent TBI.
Time-series analysis could be used to investigate causal links between intervention components and performance among individual children over time (Baltes, Reese, and Nesselroade, 1977). The use of time-series analysis also could improve our understanding of the particular process that occurs during interventions and illustrate whether and by what means interventions directly affect the performance of children with TBI. Time series analysis relies on an almost continuous collection of data and results in a high number of measurements of outcomes of interest (50 or more is ideal). Therefore, this technique is quite detailed, but studies could be designed to repeatedly measure a small number of children participating in interventions that are specifically designed to improve particular outcomes. This technique could provide needed detail on the influences of interventions on performance and recovery of children with TBI. Again, we found no studies in the pediatric TBI literature that used this technique. In a study by Schmitz and Skinner (1993), a small subset of children (ages 9 to 12) completed 50 or more measurements of their objective and subjective effort in school, objective and subjective performance in school, attributions they made regarding the reasons for their performance, and perceived control in the classroom. Measures were completed both before and after every graded assignment in school. Time-series analysis was used to determine causal links between perceived control, effort, and performance. An application of this technique in the study of the effectiveness of pediatric TBI interventions might be to study the relation between actual and subjective level of children's and families' efforts in intervention activities and outcomes predicted to occur as a consequence of the intervention activities.
In order to study growth curves and trajectories, it is ideal to have outcome measures that are intervally scaled and developmentally appropriate. One of the problems in studying growth curves is the difficulty in finding intervally scaled measures that are equally applicable, reliable, and valid in children of various ages and that are known to measure the same construct at different ages (Bergman, Eklund, and Magnusson, 1991; Taylor and Alden, 1997). In developmental psychology, this is known as measurement equivalence (Baltes, Reese, and Nesselroade, 1977). For example, Bergman and colleagues (1991) used a simple mathematical problem to measure inductive-deductive ability at age 10 and numerical ability at age 15. The use of measures like this results in growth curves that indicate the measurement of changing constructs as opposed to recovery or development of a single construct. In other words, "is it the people themselves or what the test is measuring that has changed?" (Baltes, Reese, and Nesselroade, 1977, p. 157). A related problem is the use of measures that are easier for older children or less severely injured children to complete than younger or more severely injured children. As a result, measurement could easily become confounded with age at injury, age at testing, or injury severity.
Clearly, when studying intraindividual change, much thought must precede the selection of measurement instruments, and the choice of measures should be based on sound theoretical constructs. Because measurement equivalence is difficult to validate, researchers should include techniques that provide some evidence of equivalence but acknowledge the possibility of measurement problems. Given the lack of standard measures in assessing child and adolescent TBI, measurement development is necessary in order to expand the study of interventions to emphasize intraindividual change and interindividual differences in intraindividual change (Michaud, 1995; Taylor and Alden, 1997; Ward, 1994).
Large longitudinal studies and controlled experimental research that incorporate important concepts from developmental science are needed to investigate specific questions about the effectiveness of treatments for TBI in children and adolescents. Methods for designing studies and evaluating data provided by life-span developmental psychology may be well suited for the study of recovery from TBI in this population.