Comparison of the Traumatic Brain Injury (TBI) Model Systems National Dataset to a Population-Based Cohort of TBI Hospitalizations

Methods 

The primary objectives of the TBIMS are to: (1) develop and demonstrate a model system of care for persons with TBI, stressing continuity and comprehensiveness of care; (2) track the long-term outcomes of persons who go through these systems; and (3) establish and maintain a centralized, standardized database designed to achieve the first 2 objectives. Prior to 1998, no more than 5 centers were funded for participation in the TBIMS national dataset. In 1998, 12 new centers were funded, increasing enrollment in the dataset to approximately 500 new cases per year. Since 2002, 16 centers have been funded throughout the United States. Although each local TBIMS center has unique attributes, the common denominator throughout the program is that all research participants experience both acute care and acute inpatient rehabilitation within designated hospitals in the local system of care. In addition, all centers collect a common core dataset consisting of variables that characterize the participant throughout inpatient care and at follow-up intervals. The dataset includes demographic and injury-related characteristics; diagnoses; selected medical, neuropsychologic, and psychologic conditions; and functional outcomes, including quality of life and community integration. The core dataset has varied over the years and currently consists of 112 variables collected during initial hospitalization and 88 variables collected at follow-up years 1, 2, 5, and every 5 years thereafter.

The definition of TBI used for the TBIMS national dataset has been based on the CDC’s original case definition: “damage to brain tissue caused by an external mechanical force, as evidenced by loss of consciousness due to brain trauma, posttraumatic amnesia, skull fracture, or objective neurologic findings that can be reasonably attributed to TBI on physical examination or mental status examination.”9(p2) Eligible research participants provide informed consent (or have consent provided by an authorized proxy). After consent, a subject’s acute hospital medical record is abstracted for information about injury-related characteristics, timing of milestones in recovery of consciousness, and medical complications encountered. Data are collected contemporaneously during the comprehensive rehabilitation stay, and include demographic characteristics, premorbid functioning, functional status at admission and discharge, and cost of services received. The follow-up interviews include sociodemographic status, subsequent injury and hospitalization, functional independence, employment, substance use, and subjective well-being. At the time of the present study, the TBIMS database included over 5000 participants.

CDC’s TBI surveillance activities have the overarching purpose of collecting data from acute care hospitals to ascertain the incidence, etiology, risk factors, severity, and outcomes of TBI in a statewide population.3 In accordance with the CDC case definition, TBI is defined for surveillance purposes as any discharge with a primary or secondary diagnosis of skull fracture and/or intracranial injury, including concussion.11, 12 Acknowledging a need to acquire additional information about TBI outcomes, in 1995 CDC funded the Colorado Department of Public Health and Environment to conduct population-based follow-up of residents hospitalized with a diagnosis of TBI. In 1998, CDC funded the South Carolina Department of Health and Environmental Control as the second state to implement a statewide TBI follow-up system. The SCTBIFR became an extension to South Carolina’s surveillance of persons who are hospitalized in acute care facilities or treated and released from the emergency department after acquiring TBI. SCTBIFR’s specific objectives were: (1) to determine the proportion of persons with TBI who report cognitive, behavioral and emotional, and neurosomatic deficits during the follow-up period, (2) to identify specific functional limitations (disability) by person and injury characteristics, (3) to determine level of life satisfaction, (4) to assess the mortality experience of the cohort during the follow-up period, and (5) to identify unmet needs and barriers to receiving services.

SCTBIFR defined a case of TBI in accordance with the CDC case definition for TBI.11, 12 For case identification purposes, nature of injury codes (N-codes) 800.0–801.9, 803.0–804.9, 850.0–854.1, and 959.01 from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), were included. The SCTBIFR cohort comprised a random sample of all persons, age 15 and older, in the South Carolina TBI Surveillance System who were discharged alive from acute care hospitals in the previous year. Potential participants were stratified into high (Abbreviated Injury Scale [AIS] score for the head ≥3) or low (AIS head score of 2) TBI severity. The initial sample had a sampling ratio of 2:1 high to low severity. Health information specialists from the South Carolina Department of Health and Environmental Control acquired extended variables from the eligible participants’ hospital medical records to validate the existing data and to gather predischarge history and additional medical information. SCTBIFR conducted the first annual interview approximately 12 months after the participant’s hospital discharge. The domains assessed included general health status, pre-existing health conditions, difficulties performing activities of daily living and instrumental activities of daily living, depressive symptoms, services and unmet needs, alcohol and drug use, pre- and postinjury employment, living situation, marital status, education level, health insurance coverage, personal income, transportation, life satisfaction, cognition, social integration, and social support. The SCTBIFR eligible sample consisted of 3715 of 9688 randomly selected patients discharged from acute care facilities from January 1, 1999, through June 30, 2002. First year follow-up interviews were conducted on 2118 participants from May 2000 through July 2003. Proxy respondents reported for 285 participants.

We used the 1-year cohorts for both the SCTBIFR and the TBIMS for this study. The limited number of variables available from hospital discharge data used in the South Carolina TBI Surveillance System are augmented through medical record review. At follow-up, cases included in the SCTBIFR had substantially more data collected including variables necessary for the comparisons needed in the current study. For the SCTBIFR, 43% of the participants eligible for follow-up could not be located, refused to participate, died, became incarcerated, or otherwise were unable to complete the follow-up interview. Characteristics of nonrespondents were accounted for in the weighting of the year 1 cohort.12 Effects of potential bias due to loss to follow-up were further mitigated by also using the year 1 follow-up cohort for the TBIMS, which had a 74.6% follow-up rate. Based on Corrigan et al,13 comparison of dropout rates among 3 longitudinal studies, including the TBIMS, variables most likely to be associated with loss to follow-up were socioeconomic disadvantage, a history of substance abuse, and a violent injury etiology. They also observed that severe motor deficits made it more likely that a person will be found for follow-up after 1 year.

To integrate the SCTBIFR and TBIMS datasets, we developed a process to determine which variables in both datasets were comparable. Working from codebooks of variables for each program, we further examined those with apparently similar domains for their actual content and methods used to elicit responses. Corresponding levels for comparable variables in each dataset, including instances in which levels were combined to establish equivalence, were identified for an integrated dataset. A variable was created from available information to indicate whether a participant was expected to be working prior to their injury (whether or not they actually were). A participant was expected to be working if they were between the ages of 18 and 65, were not a student, homemaker, resident of specialized housing, or retired and not working. Prior competitive employment was defined as the number of hours worked per week before their injury for participants expected to be working (if not expected, competitive employment was shown as missing rather than zero). Postinjury competitive employment was the number of hours worked per week after their injury, again, if they were expected to be working.

For each case, we determined injury severity by translating the ICD-9-CM diagnosis codes into an AIS score for the head region using ICDMAP-90 software.a ICDMAP-90 is severity scoring software that computes the severity of the trauma from ICD-9-CM nature of injury codes using an internal translation scheme that takes the patient’s age and clinical descriptors (the fifth digits of ICD-9-CM codes) into account.14 When the fifth digit was missing, we assigned the least severe rating. The AIS classifies over 2000 injuries according to the body region of injury (eg, head, chest, extremity), type of structure injured (eg, nerve, vessel, bone), location of injury within the body region (eg, femur, tibia, talus), and nature of injury (eg, abrasion, burn, crush).14 Consistent with the manually assigned AIS, ICDMAP-90 derived AIS scores grade each injury according to its associated threat to life on an ordinal scale from 1 (minor) to 6 (unsurvivable).14, 15, 16

Results 

Univariate Comparisons 

Comparisons between crude frequencies for the 2 cohorts were computed for 21 variables. Results of these univariate comparisons are shown in table 2. Analyses revealed multiple differences between the SCTBIFR and TBIMS cohorts. SCTBIFR had greater numbers of Medicare and uninsured patients, whereas the TBIMS had greater numbers of Medicaid patients. A significant difference in education was observed for the unknown category, with the TBIMS having a higher rate. The 2 cohorts differed significantly for persons expected to be working, with 75% of the TBIMS cohort expected to be working, whereas 60% were expected in the SCTBIFR. The 2 cohorts were similar in marital status, though the SCTBIFR cohort had more persons who had been widowed, whereas the TBIMS had more persons who had never been married. Differences in age for the 2 cohorts appeared to be a primary contributor to these differences.

Table 2. Characteristics of the SCTBIFR and TBIMS Cohorts

	Variables	SCTBIFR (N=2118)		TBIMS (N=3125)
		%	95% CI	%	95% CI
	Age (y)
	15−19	15.5	12.9–18.1	16.3	13.7–18.9
	20−24	10.6	8.2–13.0	16.0	13.6–18.4
	25−34	13.2	10.6–15.8	19.0	16.4–21.6
	35−44	14.7	12.0–17.4	19.6	16.9–22.3
	45−54	13.7	11.2–16.2	13.1	10.6–15.6
	55−64	8.8	6.8–10.8	7.8	5.8–9.8
	65−74	8.6	6.8–10.4	4.5	2.7–6.3
	75+	15.0	12.8–17.2	3.6	1.4–5.8
	Mean (95% CI)	44.9	43.9–45.7	36.7	35.6–37.8
	Sex
	Male	60.6	57.2–64.0	72.6	69.2–76.0
	Female	39.4	36.0–42.9	27.4	24.0–30.8
	Race
	White	76.4	73.4–79.6	67.4	64.2–70.6
	Black	21.2	18.2–24.2	22.8	19.8–25.8
	Asian	0.4	−0.4to1.2	2.4	1.6–3.2
	Hispanic	1.8	0.5–3.1	6.0	4.7–7.3
	Other	0.3	−0.2to0.9	1.5	0.9–2.1
	Payer
	Medicare	24.4	21.7–27.1	8.5	5.8–11.2
	Medicaid	14.3	11.5–17.1	26.5	23.6–29.2
	Worker’s compensation	3.7	2.2–5.2	5.8	4.3–7.3
	Other government	1.5	0.7–2.3	0.7	−0.1to1.5
	Private/commercial	43.5	39.9–47.1	50.2	46.6–53.8
	Uninsured	12.0	10.0–14.0	4.2	2.2–6.2
	Other	0.6	−0.4to1.6	4.1	3.1–5.1
	Education
	≤8y	9.3	7.3–11.3	6.7	4.7–8.7
	9−11y	24.2	21.1–27.3	25.0	21.9–28.1
	12y or GED	32.8	29.4–36.2	30.9	27.5–34.3
	Some college	21.3	18.3–24.3	23.1	20.1–26.1
	College graduate	11.7	9.4–14.0	10.3	8.0–12.6
	Unknown	0.7	−0.4to1.8	4.0	3.0–5.0
	Expected to be working preinjury
	Yes	59.9	56.5–63.3	75.9	72.5–79.3
	No	40.1	36.7–43.5	24.1	20.7–24.5
	Marital status
	Married	37.5	32.3–39.1	31.2	27.8–34.6
	Divorced	11.2	8.9–13.5	12.0	9.7–13.3
	Widowed	11.9	9.9–13.9	3.2	1.2–5.2
	Separated	4.4	2.9–5.9	4.1	2.6–5.6
	Never been married	36.7	33.2–40.2	49.6	46.0–53.1
	Cause of injury
	Vehicular	51.7	48.1–55.3	58.1	54.5–61.7
	Violence	7.9	5.7–10.1	14.9	12.6–17.1
	Sports	2.1	1.1–3.1	1.3	0.3–2.3
	Pedestrian	2.2	0.8–3.6	7.2	5.8–8.6
	Fall/flying object	31.1	28.0–34.2	17.7	14.6–20.8
	Other	5.0	3.7–6.3	0.8	−0.4to2.1
	AIS head region
	2	31.2	28.2–34.2	12.5	9.5–15.5
	3	22.5	19.7–25.3	11.3	8.5–14.1
	4	42.8	39.2–46.4	58.6	55.0–62.2
	5−6	3.5	1.4–5.6	17.6	15.5–19.7
	GCS score
	3−8	9.8	7.3–12.3	19.5	17.0–22.0
	9−12	6.4	4.5–8.3	10.0	8.1–11.9
	13−15	40.6	37.2–44.0	25.4	22.0–28.8
	Other/unknown	43.2	39.6–46.8	45.1	41.5–48.7
	CT
	Normal	28.8	25.9–31.7	10.4	7.5–13.3
	Abnormal	55.7	52.5–58.9	79.6	76.3–82.9
	Unknown	15.5	13.3–17.7	10.0	7.6–12.6
	Work hours preinjury⁎	43.4	42.8–44.0	31.7	30.4–33.0
	Acute LOS⁎	8.1	7.3–9.0	21.2	20.3–22.1

Abbreviation: GED, General Educational Development diploma.

⁎ Mean (95% CI).

The 2 cohorts were similar in the proportions injured in vehicular crashes or sports-related incidents. More persons in the TBIMS cohort had been injured by violent means or as pedestrians. In contrast, more persons in the SCTBIFR cohort were injured in falls. The AIS scores differed at all levels, with the SCTBIFR having a higher proportion of AIS 2 and 3 injuries, whereas the TBIMS had a higher proportion of AIS 4, 5, and 6 injuries. For the Glasgow Coma Scale (GCS), the TBIMS had a higher proportion of participants with scores of 3 to 8 and a lower proportion with scores of 13 to 15. The proportion of participants in the TBIMS who did not have a GCS score due to chemical paralysis or intubation was comparable to the SCTBIFR proportion of participants for whom GCS scores were lacking for various reasons. As found with other measures associated with severity, the TBIMS cohort had a larger proportion of participants with an abnormal CT finding, whereas the SCTBIFR had more participants with normal or unknown findings. Comparisons between continuous variables indicated that the mean age was higher for the SCTBIFR, but acute hospital LOS was significantly lower. Among participants expected to be working prior to their injury, the SCTBIFR cohort worked an average of 10 hours more a week prior to their injury.

Multivariable Analyses 

Multivariable analyses allowed comparison between cohorts when all other variables were held constant. As shown in table 3, there were significant differences in cohort membership; however, many of the differences observed in the crude comparisons were no longer significant. The most important variables for distinguishing cohort membership were the severity indices. As one would expect from the composition of the reference populations from which the cohorts were drawn, the TBIMS cohort was more likely to have greater initial injury severity, abnormal CT scan, and extended LOS in acute care than the SCTBIFR cohort. These 3 severity indices as a group explained well over 75% of the cohort membership. The mean acute hospital LOS, after controlling for severity and insurance status, was 20.3 days (95% CI, 19.7−21.0) for TBIMS versus 9.9 days (95% CI, 9.2−10.6) for SCTBIFR. In the multivariable logistic model, the parameter estimate for acute hospital LOS indicated that the TBIMS cohort was twice as likely as the SCTBIFR cohort to have a mean difference of 10 days in acute care hospital stay after adjusting for other covariates. Likewise, the odds of having an abnormal CT scan were 2.4 times higher in the TBIMS cohort than in the SCTBIFR. Initial GCS score was not included in the final multivariable model due to its multicolinearity with AIS scores. Alternative models including either GCS or AIS indicated that the AIS score contributed more to distinguishing cohort membership.

Table 3. Odds Ratios From Logistic Regression for Membership in the TBIMS Cohort Compared With the SCTBIFR Cohort

	Variable	Adjusted Odds Ratio	95% CI	P
	Age⁎	0.99	0.99−1.00	.015
	Race
	White	Reference
	Black	0.64	0.43−0.95	.028
	Hispanic	4.04	1.48−11.03	.007
	Other	1.76	0.46−6.69	.405
	Education
	≤11y	Reference
	12y or GED	0.84	0.71−1.01	.057
	Some college or more	1.04	0.87−1.25	.664
	Unknown	7.12	3.83−13.24	<.001
	Marital status
	Married	Reference
	Divorced	0.79	0.66−0.97	.015
	Never been married	1.17	0.96−1.43	.121
	Expected to be working preinjury
	No	Reference
	Yes	1.59	1.33−1.90	<.001
	Payer
	Worker’s compensation	Reference
	Medicare	0.44	0.29−0.65	<.001
	Medicaid	0.72	0.51−1.03	.069
	Private/commercial	0.69	0.50−0.96	.028
	Uninsured	0.20	0.14−0.30	<.001
	Unknown	3.31	1.64−6.68	<.001
	Cause of injury
	Vehicular	Reference
	Violence	1.58	1.25−2.01	<.001
	Fall/flying object	0.76	0.63−0.93	.006
	Other	0.85	0.66−1.08	.181
	CT scan
	Normal	Reference
	Abnormal	2.36	1.92−2.91	<.001
	Unknown	1.03	0.79−1.34	.839
	AIS score
	2	Reference
	3	0.64	0.51−0.81	<.001
	4	1.47	1.20−1.80	<.001
	5−6	2.77	2.01−3.82	<.001
	LOS	2.09	1.94−2.23	<.001

⁎ Significant before rounding.

Beyond injury severity, the only additional characteristics that had strong explanatory power to differentiate cohort membership were racial and ethnic background and payer status. There was a 1.5 times greater likelihood that a black participant would be a cohort member of the SCTBIFR, whereas Hispanic participants were 4 times more likely to be in the TBIMS. These differences were consistent with ways in which South Carolina differs in racial and ethnic composition from the United States overall, with the TBIMS cohort appearing more similar to national estimates.1 For payer source there was nearly a 5-fold increased likelihood for an SCTBIFR cohort member to be without health insurance when compared with TBIMS participants, and a 2.3 times increased likelihood to be covered under Medicare. There was more than a 3-fold greater likelihood that TBIMS payer source was unknown. There were no significant differences in the likelihood of members of the 2 cohorts having Medicaid or private insurance coverage.

The 2 cohorts varied only minimally in other aspects. A difference was observed for cause of injury, with the TBIMS cohort showing a modestly greater likelihood of violence-related etiologies, but the SCTBIFR was slightly more likely to include persons who had fallen or been struck by a flying object. Persons who were expected to be working before sustaining TBI were modestly more likely to be cohort members of TBIMS. Though significant, there was only a slight difference between the cohorts in marital status. A large but uninterpretable difference was evident for participants whose education was unknown.

Discussion 

A long-expressed concern about the TBIMS has been its sampling strategy and the impact of that strategy on statistical findings from the TBIMS data, especially the extent to which results from the TBIMS reflect the more general population of persons receiving comprehensive rehabilitation for TBI. Our results showed that the TBIMS database differed from the population-based cohort of the SCTBIFR in expected ways, in most respects. The most apparent difference was in injury severity, as assessed by the AIS, GCS, CT scan, and LOS in the acute hospital. Injury severity variables accounted for 75% of the difference in cohort membership. The proportion of TBIMS subjects in the most severe AIS strata was 4 times that for SCTBIFR (17% vs 4%). In contrast, the SCTBIFR had twice the proportion of subjects in the least severe AIS strata (31% vs 14%). Among subjects with GCS scores, the TBIMS had almost twice the proportion of moderate and severe injuries as the SCTBIFR (29% vs 16%). Furthermore, though approximately 45% of both cohorts did not have GCS scores, for the TBIMS subjects the missing values were due to chemical paralysis and intubation, suggesting a much higher proportion of more serious TBI. Although the current study could not empirically test the representativeness of the TBIMS cohort relative to the population of patients receiving acute rehabilitation, it does confirm that severity is the primary source of difference from the population of patients receiving acute hospitalization.

Assessing the role of injury severity in these 2 cohorts was complicated by several factors, including the TBIMS and SCTBIFR having emphasized different measures of injury severity and the inherent limitations of all injury severity scales. Although the TBIMS uses the GCS as one of its primary measures of injury severity, the substantial proportion of patients who are acutely intubated or chemically paralyzed complicates use of the standard GCS rating for research purposes. GCS data are essentially missing for this group of TBIMS participants. In the SCTBIFR, a similarly high rate of missing GCS data occurs because the data are not present in the medical record. Although the GCS has been quite useful in predicting mortality and identifying patients more likely to benefit from trauma care, it has been somewhat more limited in predicting functional outcome.19, 20 In addition, GCS is dynamic, changing dramatically during the early, acute phase. There can be substantial variability due simply to how and when GCS is collected, adding “noise” to GCS data. In contrast, with the ability to convert ICD-9-CM diagnoses into AIS scores for the head, this indicator of severity was available for the vast majority of subjects in both cohorts. We observed variability across TBIMS centers in the thoroughness of ICD coding—some centers submitted more ICD codes per case than others. Additionally, not all centers routinely provided the fifth digit subclassification in the ICD coding. It is not possible to convert ICD codes that are missing fifth digit without making assumptions (in our case by using the lowest severity) about the underlying injury severity that cannot be independently verified. These factors may have decreased the precision of the AIS in the current analyses. Although the greater completeness of the AIS variable makes it far more suitable for statistical modeling, questions remain regarding whether it has predictive utility to assess long-term disability.

The results clearly suggest that the TBIMS should further evaluate both the methodology and the utility of the AIS as calculated with ICDMAP-90 software. Although the AIS has been shown to be associated with mortality, LOS, and return to employment, it has not been widely used in medical rehabilitation. The TBIMS database would benefit from another index of injury severity that is both valid and available for most participants. A compelling reason is the inability to assess GCS when patients are intubated, chemically paralyzed, heavily intoxicated, and when there are massive craniofacial injuries. Furthermore, the findings suggested that multiple indicators of injury severity may be needed. For example, even after controlling for insurance status, the mean acute hospital LOS remained substantially higher for the TBIMS cohort across all levels of AIS. Co-occurring injuries and comorbid conditions certainly contribute to longer LOS, if not to long-term outcomes.21 There are several valid indices of both comorbid conditions and co-occurring injury that can be compiled from diagnostic codes22, 23; however, the TBIMS database has historically captured only ICD-9-CM head and neck injury diagnostic codes. A tremendous amount of additional injury information could be obtained if all ICD-9-CM diagnoses were collected from medical records. Capturing these additional injury and comorbid data could assist in interpreting the GCS and AIS, and provide a more comprehensive picture of injury severity and other medical conditions that may impact recovery after TBI. These additions to the TBIMS would involve minimal cost and allow coding of 9 AIS body regions affected by the injury.

Our findings suggested that there were differences between the TBIMS and the SCTBIFR that were not attributable to injury severity alone. Univariate comparisons of the crude frequencies revealed that, in comparison to the TBIMS cohort, persons in the SCTBIFR cohort were more likely to be older, white, injured in a fall, retired or otherwise not expected to be working, and either to have Medicare or no insurance. On the other hand, persons in the TBIMS cohort, when compared with the SCTBIFR cohort, were more likely to be younger, expected to be working, from minority groups other than black, to have had a violence-related injury, and to have some type of health insurance. Multivariable analyses suggested that most of these differences were correlates of injury severity. For example, older adults who were injured due to falls received acute hospitalization but not comprehensive rehabilitation due to the lesser severity of their injury. Likewise, the preponderance of whites in the SCTBIFR cohort as was noted in the univariate analysis showed reversal in the multivariate analysis due to the confounding effect of the other covariates in the model. The degree of confounding indicated a 2.7-fold underestimation of the odds that blacks would be more likely to be represented in SCTBIFR than the TBIMS cohort. Likewise, multivariable analyses indicated Hispanics were under-represented in the SCTBIFR. The lower composition of blacks and higher composition of Hispanics in the TBIMS are more consistent with national population estimates and do not appear to be a basis for concern regarding the representativeness of TBIMS data.

Differences between the 2 cohorts in insurance payer also could not be accounted for by injury severity. The SCTBIFR cohort was 5 times more likely to be uninsured than the TBIMS cohort, and more than 2 times as likely to have Medicare. The higher level of admission of Medicare patients for acute hospital care may be attributed to the vulnerability of the elderly for serious complications and the extra healing time needed for good recovery.24 Given that all other variables were held constant including age and severity, however, this finding could also represent an overpropensity to provide acute hospital care solely due to Medicare being a favorable payer source.25 For rehabilitation care, Medicare coverage is not as unequivocally advantageous, especially for patients with more severe illness and for whom there may be alternate placement available in skilled nursing facilities.26 The finding for those who have no insurance is both more pronounced and more difficult to interpret. Although many rehabilitation centers, including TBIMS sites, admit patients who have no insurance at time of injury and subsequently assist them to become eligible for Medicaid, a corresponding greater likelihood of TBIMS participants being covered by Medicaid was not observed. Indeed, though not statistically significant, SCTBIFR members were also more likely to be on Medicaid. Other studies have found that after controlling for injury severity, uninsured persons have significantly lower rates of acute hospital admission when compared with patients with insurance.27, 28, 29, 30 One might extrapolate from these findings that a proportion of persons with TBI who do not have health insurance coverage do not receive the comprehensive rehabilitation required by their injuries. If true, this finding represents a likely health disparity related to socioeconomic status—some uninsured people in need of comprehensive rehabilitation are not receiving it. Although it is not possible to determine if the under-representation of the uninsured in the TBIMS is due solely to inability to pay, this finding warrants further investigation on a prospective basis where additional clinical and sociodemographic parameters can be measured and examined.

A major challenge in comparing the TBIMS and SCTBIFR was deriving a common set of variables on which to base comparisons. Although both projects had different aims and objectives, standard demographic, injury-related, and outcome variables were collected. Yet, these standard variables were often scaled differently and it was not always possible to recode the variables to make them comparable across databases. Data collection is obviously time-consuming and expensive. Consideration might be given to developing a core set of evidence-tested variables that could be recommended for use in all TBI studies. The proliferation of idiosyncratic, “boutique” databases precludes meaningful comparisons across projects. A major limitation of this study was our inability to compare functional outcome between the cohorts. Both databases assessed functional outcome but the variables were not comparable. Measuring functional outcome is crucial for understanding the relationship between injury severity and morbidity, assessing the effectiveness of interventions, and assisting in resource allocation.

The ability to “anchor” the TBIMS national dataset, or other samples based on use of postacute services, would be greatly facilitated by accurate and complete information on services received after acute hospitalization. The analyses conducted in this project indicate that receiving rehabilitation may not be solely a function of injury-related consequences. Having population-based estimates of patients with TBI receiving postacute rehabilitation services would provide additional insight into the epidemiology of service use. Future TBI outcome surveillance projects should evaluate the feasibility of collecting information on inpatient, subacute, and outpatient rehabilitation services, including points in time when services were received. The feasibility of ascertaining the actual facilities or services used should also be considered, so that programmatic characteristics such as admission policies, medical and therapy services provided, and accreditation by the Commission on Accreditation of Rehabilitation Facilities or Medicare certification could be evaluated.

The utility of the integrated dataset created for this project extends beyond attempts to characterize the samples involved. Combining datasets focusing on acute hospitalization and acute rehabilitation provides an opportunity for examining outcomes that are highly affected by injury severity. The combined dataset includes large numbers of participants representing both the most and least severe cases. Neither dataset alone has the sample size for examining the full continuum of injury severity. For instance, an examination of hours employed 1 year after injury reveals very different results for the 2 samples, with relatively small numbers of participants in the worst and best outcome categories for the SCTBIFR and TBIMS, respectively. The combined dataset also includes a more complete representation of subjects at the more severe ends of the spectrum because the SCTBIFR includes some patients not captured via rehabilitation samples. Together, there is adequate sample size to robustly characterize risk factors across the range of severity. Further analyses of comparable outcome variables may shed new light on the consequences of TBI and provide a more comprehensive perspective for public policy recommendations.

Conclusions 

Given the confines of the comparisons that were possible, results of this study provide support for the representativeness of the TBIMS dataset when studying factors that impact recovery after TBI rehabilitation. The study further underscores the relative benefits of both population-based follow-up registries that have been funded by the National Center for Injury Prevention and Control at CDC and the longitudinal dataset for persons with more severe TBI that has been funded through the NIDRR TBIMS program. A benefit of the TBIMS dataset is its focus on persons with more severe injuries who are likely to have longer-term consequences. Whereas the current study compared only year-1 cohorts, the TBIMS is designed to provide longitudinal data on outcomes spanning many years. Sampling designed to represent the population of persons with TBI (in this case, hospitalized patients in South Carolina) is essential for identifying issues related to demographic distributions and, as suggested by the current results, social disparities. The study showed the potential for complementary use of longitudinal and population-based data collection protocols. Finally, the current study also suggested areas for methodologic improvements in both the TBIMS dataset and future population-based outcome surveillance.18