PREDICTION OF OUTCOME following traumatic brain

Transcription

1 300 Outcome After Traumatic Brain Injury: Pathway Analysis of Contributions From Premorbid, Injury Severity, and Recovery Variables Thomas A. Novack, PhD, Beverly A. Bush, PhD, Jay M. Meythaler, JD, MD, Kay Canupp, CRNP ABSTRACT. Novack TA, Bush BA, Meythaler JM, Canupp K. Outcome after traumatic brain injury: pathway analysis of contributions from premorbid, injury severity, and recovery variables. Arch Phys Med Rehabil 2001;82: Objective: To examine the relationship of premorbid variables, injury severity, and cognitive and functional status to outcome 1 year after traumatic brain injury (TBI) and to assess the feasibility of multivariate path analysis as a way to discover those relationships. Design: Prospective, longitudinal. Settings: Level I trauma center, acute inpatient rehabilitation hospital. Patients: One hundred seven subjects (87 men, 20 women; mean age, yr) who had experienced severe TBI, typically from motor vehicle crashes. Interventions: Acute medical and rehabilitation care. Main Outcome Measures: Disability Rating Scale, Community Integration Questionnaire, and return to employment. Evaluated in acute rehabilitation, and at 6 and 12 months postinjury. Results: Path analyses revealed that premorbid factors had significant relationships with injury severity, functional skills, cognitive status, and outcome; injury severity affected cognitive and functional skills; and cognitive status influenced outcome. No significant relationships were found between injury severity and emotional status, injury severity and outcome, emotional status and outcome, and functional skills and outcome. Conclusions: Multivariate analysis is important to understanding outcome after TBI. Injury severity, as measured in this study, is less important to 12-month outcome than the premorbid status of the person and the difficulties (particularly cognitive deficits) exhibited at follow-up 6 months after the trauma. Key Words: Brain injuries, traumatic; Treatment outcome; Rehabilitation by the American Congress of Rehabilitation Medicine and the American Academy of Physical Medicine and Rehabilitation From the Department of Physical Medicine and Rehabilitation, University of Alabama, Birmingham, AL. Accepted in revised form May 22, Supported in part by the Centers for Disease Control (grant no. R49CCR40364) and by the National Institutes of Health (grant no. HD07420). No commercial party having a direct financial interest in the results of the research supporting this article has or will confer a benefit upon the authors or upon any organization with which the author(s) is/are associated. Reprint requests to Thomas A. Novack, PhD, University of Alabama at Birmingham, Spain Rehabilitation Center, Dept of Physical Medicine and Rehabilitation, th Ave S, Birmingham, AL 35233, novack@uab.edu /01/ $35.00/0 doi: /apmr PREDICTION OF OUTCOME following traumatic brain injury (TBI) has proven to be an elusive target, largely because of the complexity of the issues involved. It is generally acknowledged that measuring outcome after TBI is a difficult undertaking in itself, independent of the many variables that might influence such outcome. In some studies, outcome has been measured in broad strokes, such as with the Glasgow outcome score 1 or by viewing return to work as the single outcome variable. When such outcome measures are used, poorer outcome has been associated with increasing injury severity (as measured by Glasgow Coma Scale 2 [GCS] score, duration of coma, length of posttraumatic amnesia [PTA]), longer duration of hospitalization, and presence of cognitive deficits at various intervals after injury. 3-5 The trend in TBI research is toward measuring outcome in multiple ways and employing multivariate techniques to predict outcome. The multivariate approach is a concession to the number of variables that can influence outcome and the fact that those variables often are not independent of one another. Corrigan et al, 6 for instance, employed 8 measures of outcome, including assessment of employment, home activities, satisfaction with life, and postconcussion symptoms expressed by the injured person. The multivariate techniques employed have included variations of multiple regression, in some cases focusing on a few variables, 7,8 and in others assessing a broader range of predictors. 9,10 Return to employment can be predicted with impressive success when multivariate analysis of many contributing factors is performed. 9,10 While a multivariate approach is necessary given the complexity of outcome after TBI, many studies, as identified by Corrigan, 6 lack a theoretical framework for the nature of outcome and the causality and colinearity of factors impacting it. In a complex situation, such as recovery from TBI, the most comprehensive understanding is going to come from measuring outcome in multiple ways while evaluating the impact of (and relationships among) a wide range of variables on that outcome within a theoretical framework that can either be proved or disproved. The occurrence of TBI and recovery from it suggests an inherent theoretical framework. First, are the personal characteristics of the injured individual, which presumably affect everything that follows, including the severity of the injury, cognitive and functional deficits, and the eventual outcome. The next major factor to consider is the injury itself, which relates to the deficits exhibited and the outcome experienced. As a result of premorbid status and injury severity, the person will potentially exhibit a range of physical, cognitive, and emotional difficulties. Finally, there is the outcome, as reflected by resumption of community and home activities, including employment. The sequence of events is a unidirectional path in which factors encountered earlier in the path have the potential to influence, in a causal fashion, later events. For instance, while injury severity influences the physical, cognitive, and emotional deficits exhibited by the person, as well as the ultimate outcome, no one would argue that the deficits and outcome influence the severity of the injury. The logical path

2 TRAUMATIC BRAIN INJURY OUTCOME, Novack 301 Variable Table 1: Characteristics of Subjects Participating in Follow-Up Versus Nonparticipants Follow-Up Group Mean (SD) % (n 107) No Follow-Up Group Mean (SD) % (n 62) Age (yr) (14.15) (19.73).007 African American 40.20% 22.60%.020 Caucasian 59.80% 77.4% Men 81.30% 74.20%.276 Married 45.80% 43.50%.777 Education 12th grade 20.60% 24.20%.582 Employed Preinjury 63.60% 51.60%.128 History of Alcohol Abuse 29.90% 27.40%.731 History of Drug Abuse 28.00% 24.20%.586 MVC Etiology 69.20% 61.30%.297 GCS Score 7.22 (3.53) 8.26 (3.78).076 Acute LOS (15.19) (20.78).957 Rehab LOS (14.71) (37.76).246 Total Discharge FIM (22.57) (27.50).300 Abbreviation: SD, standard deviation; MVC, motor vehicle collision; LOS, length of stay. p described takes on theoretical significance when assumptions are made about the strength of relationships between particular factors relative to other factors. Fortunately, statistical methods are available to delineate paths within a theoretical framework, enabling evaluation of a complex situation, such as recovery from TBI. In the present study we used path analysis, a multivariate, structural equation modeling procedure, to evaluate the causal, predictive relationships contributing to outcome after TBI. Basing our premise on published studies, we hypothesized that premorbid characteristics would influence all aspects of TBI, including the deficits exhibited and the outcome. We postulated that, while injury severity influenced the TBI patients functional, emotional, and cognitive difficulties, its effects on outcome would diminish 1 year postinjury. The strongest predictors of outcome, we believed, would be the functional, emotional, and cognitive difficulties evident at 6 months posttrauma. This is not a trivial series of hypotheses; the emphasis on rehabilitation after TBI is based on an assumption that a strong relationship exists among the functional, emotional, and cognitive difficulties being addressed and the outcome eventually experienced. METHOD Subjects Of an eligible 169 persons admitted to an intensive care unit with a diagnosis of TBI and participating in acute rehabilitation, 107 participated in the present study. Subjects were identified as having TBI if they had experienced loss of consciousness of any duration, skull fracture, PTA of any duration, or had objective neurologic findings such as alteration in mentation, cranial nerve deficits, or hemiparesis. Subjects who completed the study participated in a neuropsychologic evaluation and interview 6 months after TBI onset and were interviewed again at 12 months postonset. Attempts were made to follow all 169 eligible subjects. The 62 who did not participate in the study either did not return for evaluation at 6 months or did not complete an interview at 12 months. Table 1 provides demographic and injury data for the 169 subjects who participated in rehabilitation, and compares the subjects who completed follow-up evaluations with those who did not. No significant difference (based on t test or 2 analysis, as appropriate) existed between patients who completed the follow-up and those who did not in terms of gender, education, preinjury employment, marital status, injury severity on the GCS, length of acute rehabilitation stay, etiology of injury, or history of alcohol or substance abuse. However, the sample that completed follow-up was significantly younger (average, 33.9 vs 41 yrs; t , p.007). Age is a potential bias in the sample that must be considered when interpreting the results. On the other hand, degree of functional impairment during rehabilitation, as measured by FIM discharge scores (70.9 vs 75.2, t , p.300), was comparable, suggesting that the debilitating effect of the trauma was similar for the 2 groups. The follow-up sample had a higher percentage of African Americans compared with patients who did not return for follow-up evaluation (40.2% vs 22.6%). Further analysis did not reveal significant differences in these 2 groups in their age, education, employment history, severity of injury, or outcome at 1 year. However, African-American subjects had significantly (t , p.039) lower discharge FIM scores. The data do not clarify why this was the case and it is not clear if this difference represents a study bias that must be considered in interpreting the results. Measures The present study s database included information collected during acute care and rehabilitation and from follow-up contacts at 6 and 12 months. Specific measures included the GCS score, FIM, 11 Disability Rating Scale 12 (DRS), and Community Integration Questionnaire 13 (CIQ). PTA was determined retrospectively based on interview 14 and was partitioned by its duration into 5 categories: less than 1 hour, 1 to 24 hours, 1 to 7 days, 7 days to 4 weeks, and more than 4 weeks. 2 Pupillary response was defined as absent or with a nonresponsive pupil on either side as present. Computed tomography (CT) scan results were classified according to the presence of abnormal densities, signs of diffuse injury, effacement of cisterns or ventricles, and combinations of these findings (eg, abnormal densities and signs of diffuse injury) to generate a score from 0 (no abnormalities) to 7 (all 3 abnormalities present). Premorbid factors examined in the present study included the person s age, education, and employment status at the time of injury; information about alcohol, drug, and social history was obtained from family members interviewed during acute care. The CAGE questionnaire 15 was used to establish history of alcohol abuse, with a positive response on 2 or more ques-

3 302 TRAUMATIC BRAIN INJURY OUTCOME, Novack tions being considered sufficient to establish abuse. Drug abuse history was established on the basis of a positive response to at least 2 of 4 questions relating to the frequency, severity, and impact of drug use. Premorbid social problems were defined as present if 2 or more positive responses were obtained to questions regarding being expelled from school, a history of suicide attempts, a history of arrests, or a history of incarceration. The neuropsychologic evaluation completed at 6 months post-tbi onset was intended to be a brief (2-hr maximum) battery emphasizing orientation, speed of processing, concentration skills, memory for new information, constructional abilities, and verbal skills. The present study s sample size did not permit us to analyze all the tests. Basing our decisions on correlations between the tests and intending to cover a breadth of cognitive abilities (memory skills, speed of processing, constructional skills, verbal abilities, constructional skills), we selected scores from these instruments for our analysis: Trail Making Part B 16 (TMTB), block design (BD) subtest of the Wechsler Adult Intelligence Scale Revised, 17 Controlled Oral Word Association Test 18 (COWAT), California Verbal Learning Test 19 (CVLT) long-delay free recall (LDFR), and selected items from the Neurobehavioral Rating Scale 20 (NBRS). Raw scores were converted to standard scores (M 100, SD 15) for analysis purposes. When patients were unable to perform or complete a task because of the severity of their cognitive disorder, the lowest possible score was assigned to that test, a procedure that has been used successfully in other studies. 21 The items selected from the NBRS 20 for the analyses included depression, anxiety, emotional lability, disinhibition, and selfappraisal. For statistical reasons it was not possible to include all items of the NBRS. Our intention was to select items that would best capture the potential emotional and behavioral problems exhibited by the TBI sample. 22,23 The NBRS was completed by the neuropsychologist, who interviewed the patient and reviewed the patient s test results. Outcome at 12 months was measured by the CIQ, DRS, and level of productive activity. We defined productive activity as part- or full-time gainful employment or participation in an educational program (eg, high school or college) at the time of follow-up. There was no stipulation regarding the duration of productive activity at follow-up or about it being the same job the person had held before the injury. Procedures On admission to the neurointensive care unit of a university hospital patients were identified as having experienced a TBI. Family members were approached within 7 days to obtain consent to begin data collection. The subjects were followed through acute rehabilitation, during which FIM scores were obtained. Follow-up evaluations that included both physical and neuropsychologic examinations were completed at 6 months postonset, with a window of plus/minus 1 month. Subjects were provided a stipend to defray transportation costs and meal expenses. The 12-month follow-up interview focused on completion of the DRS, CIQ, and determining employment status. Seventy-two of these interviews took place in the context of an appointment at the rehabilitation center and the remainder were completed by telephone by the same researcher (KC) who conducted face-to-face interviews. Data collection ended at 1 year postinjury because of time constraints associated with the study; no assumption was made that recovery would be complete at 1 year postinjury. Data Analysis Path analysis is a multivariate method that allows investigators to examine the direct and indirect relationships between constructs, as measured by multiple variables, based on hypotheses regarding theoretical relationships The constructs under study (Premorbid Factors, Injury Severity, Emotional Status, Cognitive Status, Functional Status, Outcome) are called endogenous variables, and path analysis is concerned with examining the causes of variance in these constructs. The path coefficients generated reflect the longitudinal causal influences among study constructs. The theoretical model is supported if the path coefficients between constructs are significant or not significant, as hypothesized by the model. An understanding of the causal influences among the constructs under study results from inspection of the path coefficients, in a similar manner as inspection of R 2 results in an understanding of the amount of variance accounted for in a variable by 1 or more other variables. Path analysis, however, enables researchers to examine several multiple regression analyses simultaneously, resulting in a more comprehensive clinical understanding of complex interrelationships, and a stronger, more parsimonious statistical result an understanding of the causal relationships among the variables. In the present study, the AMOS statistical package, a version 3.6, generated path analyses that evaluated the causal contributions among the constructs, including their causal contributions to TBI outcome 1 year postinjury. The model under study evaluated the causal relationships among constructs labeled as Premorbid Factors, Injury Severity, Cognitive Status at 6 months, Emotional Status at 6 months, Functional Status at 6 months, and Outcome at 12 months. Observed variables comprising the latent construct Premorbid Factors included age, education, employment status at injury, and alcohol, drug, and social histories. Injury Severity variables included total GCS score, duration of PTA, CT scan results, and pupillary response. Observed scores on neuropsychologic tests (CVLT LDFR, TMTB, COWAT, BD scores) measured the construct Cognitive Status. Emotional Status resulted from ratings on the NBRS for depression, anxiety, disinhibition, self-appraisal, and lability. The following FIM scores measured Functional Status: rehabilitation discharge total FIM, and 6-month FIM locomotion, expression, and feeding scores. Finally, scores on the CIQ, DRS, and employment variables measured Outcome at 12 months postinjury. Because no previous empirical studies have evaluated the proposed models, the path analysis resulted in causal models that are exploratory in nature and need replication and, therefore, we report no fit indexes. Instead, we establish the models ability to support the a priori hypothesized relationships. RESULTS The subjects completing the study (table 1) were predominantly young men with a high school education or less who were injured in motor vehicle collisions. Based on GCS 2 score, a majority (70.1%) of the sample had experienced a severe TBI (GCS score, 3 8); 17.7% experienced a moderate TBI (GCS score, 9 12); and 12.1% experienced a mild TBI (GCS score, 13 15). In the case of mild injuries, subjects exhibited other problems that necessitated admission to the intensive care unit, such as abnormalities on CT scan. The duration of PTA confirmed the severity of trauma, with 63% of the sample having a PTA extending beyond 4 weeks. At 6 months postinjury, the neuropsychologic evaluation revealed significant deficits in cognition, as reflected by standard scores standard deviation of on the COWAT, on the CVLT LDFR, on the TMTB, and on the BD subtest. Moderate (or worse) depression and anxiety were exhibited by 9% and 6%, respectively, of the sample. At 12 months postinjury, 17.8% of

4 TRAUMATIC BRAIN INJURY OUTCOME, Novack 303 Fig 1. Path analysis for the TBI outcome model. the participants were gainfully employed. Scores on the DRS (M 2.85; range, 0 24) and CIQ (M 14.0; range 0 29) reflected continuing problems in community integration. With regard to the path analysis (fig 1), all values are designated as significant at p.05 or as nonsignificant. Reporting the results longitudinally, the statistical model revealed that Premorbid Factors had significant causal relationships with Injury Severity (.23), Functional Skills (.25), Cognitive Status (.27), and Outcome (.28). Better premorbid functioning resulted in better functional skills and outcome. Significant causal relationships were not found for Premorbid Factors and Emotional Status (.12), Injury Severity and Emotional Status (.26), or Emotional Status and Outcome (.03). A significant causal relationship between Injury Severity and Functional Skills (.54) was detected, as well as between Injury Severity and Cognitive Status (.46). However, a nonsignificant relationship was noted between Injury Severity and Outcome (.10) and between Functional Skills and Outcome (.18). On the other hand, a significant relationship existed between Cognitive Status at 6 months and Outcome at 12 months (.47). DISCUSSION The subjects in this study were young, mostly men, and had limited education (usually because they were too young at age of injury to have completed their education). Most injuries were from motor vehicle crashes, and there was a distribution of severity of injury, although severe TBI was predominant. The cognitive difficulties encountered were comparable to what has been described elsewhere, 3,20,27 with deficits in memory functioning and speed of mental processing being particularly evident. In other respects, the present sample differed from others. Judging from their FIM scores during rehabilitation, we found this sample to be more impaired than others described in the literature. 4,28 Outcomes at 6 and 12 months, as measured by the DRS and CIQ, were not as good as noted in other samples. 6,28 Furthermore, return to employment was possible for only 17.8% of the subjects in this study (22.1% for those who were employed before injury), compared with other studies reporting a resumption of employment activity in 20% to 35% of cases of severe TBI in the same time frame. 4,6,9,10,28,29 In summary, the present sample is demographically similar to other samples in the literature, but it appears that degree of impairment was greater, as reflected by lower FIM scores during rehabilitation and less successful outcome. Evaluation of the present data supports the hypotheses regarding the relations among Premorbid Factors, Injury Severity, difficulties in recovery, and Outcome. The findings include significant relationships among premorbid factors and functional limitations, cognitive deficits, and outcome: better Premorbid Factors (particularly being employed) has a positive influence on Functional Skills, Cognitive Status, and Outcome. Injury Severity exhibited a causal relationship with Functional Skills and Cognitive Status, but not with Outcome at 1-year postinjury. The relation between Cognitive Status and Outcome was particularly strong. This underscores the impact of Premorbid Factors and Cognitive Status, relative to Injury Severity, on Outcome after TBI. The model examined reflects the complexity of recovery from TBI by defining many factors that contribute to outcome. The results indicate that consideration of these factors adds to an understanding of the overall pattern of recovery, because they significantly contribute to outcome at 1-year after TBI. It is possible that the impact of injury severity was minimized because the people in the sample were severely injured for the most part; if the full range of TBI severity had been present we might have found a stronger direct relationship between Injury Severity and Outcome. The path analysis results are important because uncovering the causal relationships among Premorbid Factors, Injury Severity, difficulties in recovery, and Outcome

5 304 TRAUMATIC BRAIN INJURY OUTCOME, Novack shows that Outcome cannot be understood or predicted unless factors other than Injury Severity are considered. This suggests that successful rehabilitation interventions addressing difficulties in recovery, particularly cognitive deficits, could be very important in terms of overall outcome after TBI. The statistical model presented 2 somewhat surprising findings. First, no significant relationship existed between Functional Skills and Outcome. Possibly, our measures of Functional Skills, derived from the FIM, were not sufficiently sensitive, because most study participants could ambulate and communicate several months after injury. These abilities may be important in the recovery process, but are not causally related to outcome variables such as employment. The other surprising finding was the lack of a causal relationship between Injury Severity and Emotional Status, and between Emotional Status and Outcome. In part this finding may result from the lack of sensitivity of the measurement of depression and anxiety that we used. Relatively few ( 10%) subjects scored in the moderate to severe range on the NBRS with regard to anxiety and depression. Therefore, the sample was skewed between those with no emotional or psychiatric problems and those with significant problems, resulting in a higher standard of error for that particular path. Another possible reason for the limited effect of emotional difficulties is that patients with TBI at this rehabilitation program are typically administered antidepressant medication early in recovery to address possible disruption of neurotransmitter levels and depression. This treatment may have minimized the occurrence of emotional distress at 6 months postinjury. CONCLUSION The results of the present study underscore the need for researchers to be very circumspect when examining the relationship of individual variables to outcome. Using only a few variables, or examining 1 aspect of TBI, one may identify a relationship that would not fare well in a multivariate study. For instance, the relation between Injury Severity and Outcome is likely to be stronger if no other variables that could contribute to outcome are considered. This misunderstanding might lead to inappropriate emphasis on some aspects of TBI recovery, while overlooking more informative factors in recovery. Because professionals treating persons with TBI must examine multiple variables defining conceptual constructs, structural equation modeling, including path analysis, is a tool well fitted to their needs. The difficulty, if not impossibility, of performing traditional experimental studies in this situation, particularly studies that can account for a broad range of variables, underscores the need to explore the use of newer multivariate techniques. The present study shows the value of path analysis in this regard. Other factors, such as the influence of family support, the availability of postacute or vocational rehabilitation, and patient awareness of difficulty could be studied multidimensionally. Studies examining these variables singularly have indicated an influence on ultimate outcome, but to derive a full understanding of the role these factors play in outcome, the variables need to be included in a multivariate study, such as path analysis. The paths established in the present sample must be replicated in larger samples, possibly with expansion of the areas examined. Using the path coefficients detected in this study, replication studies must examine goodness-of-fit indexes in the evaluation of this model. The present study also emphasizes the need for reliable and valid measurements of patient status before injury and at specific times after injury. The relationships detected in a multivariate study can be strongly influenced by the measures employed, although with multiple variables measuring a single construct the potential influence of a single variable is lessened. Nonetheless, subtleties in recovery and outcome can be lost. For instance, in the present study, return to employment was supposed to be an important outcome measure. But some participants who did not achieve employment 1 year after TBI were able to resume, and sometimes increase, household activities as reported in interviews. Even without employment, this outcome might still be considered a success, particularly by the injured persons and their families. Contradictory outcomes such as this may have been responsible, in part, for the nonsignificant path coefficient between Functional Skills and Outcome in the current study. Capturing these subtleties in a complex multivariate environment while emphasizing reliable and valid measurement is a daunting challenge. References 1. Jennett B, Snoek J, Bond MR, Brooks N. Disability after severe head injury: observations on the use of the Glasgow Outcome Scale. J Neurol Neurosurg Psychiatry 1981;44: Jennett B, Teasdale G. Management of head injuries. Philadelphia: FA Davis; Levin HS, Gary HE, Eisenberg HM, Ruff RM, Barth JT, Kreutzer J, et al. Neurobehavioral outcome 1 year after severe head injury experience of the Traumatic Coma Data Bank. J Neurosurg 1990; 73: Cifu DX, Keyser-Marcus L, Lopez E, Wehman P, Kreutzer JS, Englander J, et al. Acute predictors of successful return to work 1 year after traumatic brain injury: a multicenter analysis. Arch Phys Med Rehabil 1997;78: Hellawell DJ, Taylor R, Pentland B. Cognitive and psychosocial outcome following moderate or severe traumatic brain injury. Brain Inj 1999;13: Corrigan JD, Smith-Knapp K, Granger CV. Outcomes in the first 5 years after traumatic brain injury. Arch Phys Med Rehabil 1998;79: Ponsford JL, Oliver JH, Curran C, Ng K. Prediction of employment status 2 years after traumatic brain injury. Brain Inj 1995; 9: Millis SR, Rosenthal M, Lourie IF. Predicting community integration after traumatic brain injury with neuropsychological measures. Int J Neurosci 1994;79: Dikmen SS, Temkin NR, Machamer JE, Holubkov AL, Fraser RT, Winn HR. Employment following traumatic head injuries. Arch Neurol 1994;51: Gollaher K, High W, Sherer M, Bergloff P, Boake C, Young ME, et al. Prediction of employment outcome one to three years following traumatic brain injury (TBI). Brain Inj 1998;12: Hamilton BB, Granger CV, Sherwin FS, Zielenzny M, Tashman JS. A uniform national data system for medical rehabilitation. In: Fuhrer MJ, editor. Rehabilitation outcomes analysis and measurement. Baltimore: Brookes; p Rappaort M, Hall KM, Hopkins K, Belleza T, Cope DN. Disability Rating Scale for severe head trauma: coma to community. Arch Phys Med Rehabil 1982;63: Willer B, Rosenthal M, Kreutzer JS, Gordon WA, Rempel R. Assessment of community integration following rehabilitation for traumatic brain injury. J Head Trauma Rehabil 1993;8(2): McMillan TM, Jongen E, Greewood R. Assessment of posttraumatic amnesia after severe closed head injury: retrospective or prospective? J Neurol Neurosurg Psychiatry 1996;60: Ewing JA. Detecting alcoholism: the CAGE Questionnaire. JAMA 1984;252: Reitan RM, Wolfson D. The Halstead-Reitan Neuropsychological Test Battery; theory and clinical interpretation. Tucson (AZ): Neuropsychological Pr; Wechsler D. Wechsler Adult Intelligence Scale Revised manual. San Antonio (TX): Psychology Corp; Benton AL, Hamsher KS. Multilingual Aphasia Examination manual of instructions. Iowa City (IA): Univ Iowa Pr; 1983.

6 TRAUMATIC BRAIN INJURY OUTCOME, Novack Delis DC, Kramer JH, Kaplan E, Ober BA. California Verbal Learning Test research edition manual. San Antonio (TX): Psychology Corp; Levin HS, High WM, Goethe KE, Sisson RA, Overall JE, Rhoades HM, et al. The neurobehavioural rating scale: assessment of the behavioural sequelae of head injury by the clinician. J Neurol Neurosurg Psychiatry 1987;50: Dikmen SS, Machamer JE, Winn R, Temkin NR. Neuropsychological outcome at 1-year post head injury. Neuropsychological 1995;9: McAllister, TW. Neuropsychiatric sequelae of head injuries. Psychiat Clin North Am 1992;15: Hibbard MR, Uysal S, Kepler K, Bogdany J, Silver J. Axis I psychopathology in individuals with traumatic brain injury. J Head Trauma Rehabil 1998;13(4): Pedhazur, EJ. Multiple regression in behavioral research: explanation and prediction. 2nd ed. New York: Holt, Rinehart & Winston; Schumaker RE, Lomax RG. A beginner s guide to structural equation modeling. Mahwah (NJ): Lawrence Erlbaum; Wright S. The method of path coefficients. Ann Math Stat 1943; 5: Kreutzer JS, Gordon WA, Rosenthal M, Marwitz J. Neuropsychological characteristics of patients with brain injury: preliminary findings from a multicenter investigation. J Head Trauma Rehabil 1993;8(2): Harrison-Felix C, Newton CN, Hall KM, Kreutzer JS. Descriptive findings from the Traumatic Brain Injury Model Systems National Data Base. J Head Trauma Rehabil 1996;11(5): Ruff RM, Marshall LF, Crouch J, Klauber MR, Levin HS, Barth J, et al. Predictors of outcome following severe head trauma: follow-up data from the Traumatic Coma Data Bank. Brain Inj 1993;7: Leach LR, Frank RG, Bouman DE, Farmer J. Family functioning, social support and depression after traumatic brain injury. Brain Inj 1994;8: Sherer M, Bergloff P, Levin E, High WM, Oden KE, Nick TG. Impaired awareness and employment outcome after traumatic brain injury. J Head Trauma Rehabil 1998;13(5): Willer B, Button J, Rempel R. Residential and home-based postacute rehabilitation of individuals with traumatic brain injury: a case control study. Arch Phys Med Rehabil 1999;80: Supplier a. SmallWaters Corp, 1507 E 53rd St, Ste 452, Chicago, IL