Health Care Costs Associated with Traumatic Brain Injury and Psychiatric Illness in Adults

Abstract

A cohort design was used to determine the contribution of traumatic brain injury (TBI) and psychiatric illness to health care costs for adolescents and adults in the 3 years following mild or moderate-to-severe TBI compared to a matched cohort without TBI, controlling for confounders. In all, 3756 subjects 15 years or older from a large health maintenance organization database were examined. We identified subjects who sustained a TBI in 1993 (n=939) and selected three control subjects per TBI-exposed subject (n=2817), matched for age, sex, and enrollment at the time of injury. Unadjusted mean costs in 2009-adjusted dollars were compared using Kruskal-Wallis tests and Mann-Whitney U tests, and adjusted mean costs were compared using gamma regression analyses. Average costs were 76% higher in the 3 years after injury for the mild TBI group, and 5.75 times greater for the moderate-to-severe TBI group compared to controls. The presence of psychiatric illness was associated with more than doubling of total costs for both inpatient and outpatient non-mental health care. Gamma regression analyses confirmed significantly higher costs in patients with TBI or psychiatric illness. A significant interaction between moderate-to-severe TBI and psychiatric illness indicated a 3.39 times greater cost among patients with both exposures compared with those exposed to moderate-to-severe TBI without psychiatric illness. TBI and psychiatric illness were each associated with significant increases in health care costs; those with the combination of moderate-to-severe TBI and psychiatric illness had much higher costs than any other group.

Introduction

T raumatic brain injury (TBI) is the leading cause of injury-related morbidity and mortality in adults throughout the world. The costs of TBI are difficult to estimate, in part because mild TBI often goes unreported, especially when co-occurring injuries are considered primary (Hyder et al., 2007). In the United States 1.7 million people per year experience head trauma, at a yearly economic cost of more than $56–60 billion per year (Faul et al., 2010). The estimated average lifetime cost of a severely injured TBI patient is between $600,000 and $1.875 million (Centers for Disease Control and Prevention, 2010). In addition, long-term neuropsychiatric sequelae of TBI, such as cognitive dysfunction, depression, and anxiety, are common among both civilians and veterans, and may require psychiatric treatment (Warriner and Velikonja, 2006). Even mild TBI can be associated with significant and long-lasting neuropsychological impairment, but there is a clear association between severity of TBI and neuropsychological cost (Hessen et al., 2007; Massagli et al., 2004).

Past research has shown that both mild TBI and moderate-to-severe TBI are associated with an increased risk of subsequent psychiatric illness (Bombardier et al., 2010; Fann et al., 2004), and that TBI patients compared with matched control subjects self-report more mental health problems (Hawthorne et al., 2009). However, our literature review did not locate prior studies that examined the association between the severity of TBI and the associated cost of psychiatric illness in adolescents and adults. Related research on rehabilitation costs, in which TBI was one of the possible reasons for admission, demonstrated significantly higher inpatient rehabilitation costs in adult rehabilitation patients with major depressive disorder (Dobrez et al., 2010).

The present research examines the associations among TBI of varying severity including a comparison group not exposed to TBI, psychiatric illness, and health care costs over a 3-year period. We assess whether the association between costs and psychiatric illness were modified by the presence of TBI.

Methods

This prospective cohort study used computerized records of enrollees in a large staff model health maintenance organization (HMO), Group Health Cooperative of Puget Sound (GHC). Details of the study design have been previously described (Fann et al., 2004). GHC serves approximately 500,000 members in 6 counties in the Puget Sound region of Washington State. The health plan members are generally representative of the region's population in terms of sex, age, race, and marital status (Saunders et al., 1994). The data included information on all inpatient and outpatient visits and diagnoses, all prescriptions provided from GHC pharmacies, and demographic information. Diagnoses in 1992 were recorded on 95% of all visit records (Saunders et al., 1994). Only about 7% of plan members have dual insurance coverage, so determination of medical service use in the study is nearly complete. GHC also has a low rate of disenrollment (13.1% for 1992–1993). The institutional review boards of GHC and the University of Washington approved this study.

Participants

Subjects were persons 15 years or older who were diagnosed with mild or moderate-to-severe TBI in the emergency department (ED), hospital, or outpatient clinic in 1993, and who were enrolled in GHC but had not had an International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) diagnosis of TBI in the year prior to the study. TBI exposure was defined by the diagnostic categories used by the United States Centers for Disease Control and Prevention (CDC) in their surveillance studies (Thurman et al., 1995). Severity of TBI was dichotomized into mild or moderate-to-severe TBI using the categorization criteria of the CDC (Centers for Disease Control and Prevention, 2010). The TBI was considered mild if ICD-9-CM codes (National Center for Health Statistics, 1991) indicated brief (<1 h) or no loss of consciousness and no traumatic intracranial lesions; and moderate-to-severe if ICD-9-CM codes indicated prolonged loss of consciousness or documented traumatic intracranial or brain lesions. In all, 939 GHC members who met our inclusion criteria were diagnosed with TBI, 803 with mild TBI and 136 with moderate-to-severe TBI.

Three TBI-unexposed comparison subjects (controls; n=2817) per TBI-exposed subject were selected at random from the GHC enrollment files and were frequency matched to participants with TBI by sex, age group, and enrollment at the time of the TBI reference date. Participants with TBI and controls were required to have been continuously enrolled in GHC for at least 1 year prior to the reference date, and to not have had an ICD-9-CM diagnosis of TBI in the year before the reference date. In the first 6 months after enrollment, 52 subjects (9 controls and 43 cases) disenrolled from GHC, resulting in a lack of cost data for those subjects, and creating a final sample of 3704.

Psychiatric illness determination

Computerized records of all study subjects were examined to establish psychiatric illness over the 1-year period before the TBI reference date, and the 3-year period after that date. Based on automated HMO data, psychiatric illness was considered present if the subject met at least one of the following criteria: an ICD-9 code of a psychiatric diagnosis, using mental health services, or filling a prescription for a psychiatric medication (Fann et al., 2004).

Psychiatric illness and use of mental health services

ICD-9-CM codes made by any GHC provider were used to identify psychiatric diagnoses (Thurman et al., 1995). The specific ICD-9-CM codes used are provided in Fann and associates (2004), and were categorized as: acute reaction to stress or adjustment reaction; alcohol or drug intoxication, withdrawal, or dependence; anxiety; depression; hyperkinetic syndrome of childhood; malaise or fatigue; organic psychotic mental disorders; organic nonpsychotic mental disorders; schizophrenia, hallucinations, or paranoia; somatoform disorders; or other psychiatric disorders (Fann et al., 2004). Subjects were considered to have had a psychiatric prescription filled if GHC pharmacy data indicated a prescription for medications in any of the following classes: antidepressants, antipsychotics, anxiolytics, lithium, or psychostimulants. Because some of these medications are prescribed for indications other than psychiatric diagnoses, antidepressants and anxiolytics were considered to be for a psychiatric indication if they were filled within 60 days of a diagnosis of depression or anxiety, respectively. Surveys indicate that more than 90% of medications prescribed at GHC are filled at GHC pharmacies because of prescription insurance coverage in the GHC plan (Saunders et al., 1994).

Use of mental health services was determined based on computerized records of inpatient psychiatric hospitalization, outpatient mental health clinic visits, and alcohol or drug treatment in inpatient or outpatient settings. Out-of-plan use of mental health services is rare because of the comprehensive mental health services and small co-payments available to patients in GHC.

Based on these psychiatric illness indicators, we present data on the 3 major clinical categories: affective disorders (depressive or anxiety disorder diagnosis; antidepressant, lithium, or anxiolytic prescription), psychotic disorders (schizophrenia, hallucinations, or paranoia or organic psychotic disorder diagnosis; or antipsychotic prescription), and substance abuse disorders (alcohol or drug intoxication, withdrawal, or dependence diagnosis) for subgroup analyses (these diagnostic categories were also examined in Fann et al., 2004). Patients with psychiatric illness outside of these three major categories constituted the “other disorders” group, which we used as a reference group in the subgroup analyses.

Cost data

Costs were determined using the HMO's automated cost accounting system. This system tracks all health care services provided for or paid for by GHC during the study period. These included outpatient services for general medical or mental health care, inpatient medical and mental health services, emergency visits, pharmacy costs, laboratory and radiology costs, and other health care services paid for by GHC. All GHC outpatient and inpatient services were assigned costs on the basis of health care plan accounting records (including the actual personnel, supply, and overhead costs), following procedures used by other investigators (Ciechanowski et al., 2003; Walker et al., 2003). Services purchased by GHC from external providers were assigned costs equal to the amount reimbursed by GHC. All costs were adjusted to 2009 dollars using a conversion factor obtained from the Consumer Price Index (U.S. Department of Labor, 2006), based on 2009 costs of health services in the Seattle/Puget Sound region, and were discounted 3% per year for years 2 and 3 (Brouwer et al., 2005; Ortendahl and Fries, 2002), as recommended by a consensus of experts in medical cost analysis (Gold et al., 1996). We did not include direct out-of-pocket costs and the indirect costs of lost time or income.

Statistical analysis

We examined demographic and clinical variables in the TBI-exposed and TBI-unexposed groups using chi-square tests. We compared the unadjusted health care costs in the TBI-exposed and TBI-unexposed groups (mild and moderate to severe) within specific cost categories (i.e., primary care, specialty care, ED, pharmacy, laboratory and radiology, outpatient mental health, and inpatient care) using Kruskal-Wallis rank tests. We also compared the unadjusted health care costs of subjects with psychiatric illness in the 3 years following the TBI reference date versus those without, using a Mann-Whitney U test.

Total costs and costs of specific components of medical care were examined using a two-part model (Diehr et al., 1999). The two-part model was developed to adjust for the proportion of subjects with no health care costs and the non-normal distribution of cost data. In part 1 of the two-part model, we used multivariable logistic regression models to examine differences between groups in the odds of having non-zero health care costs in each category by TBI and psychiatric illness status. In part 2 of the model, only subjects with nonzero costs were included, and we focused on total costs to avoid having cells that had zero counts (e.g., mental health costs in patients without psychiatric illness). Differences in mean cost for TBI-exposed subjects versus TBI-unexposed subjects, and for subjects with psychiatric illness after the TBI reference date versus those without, were estimated based on regression analyses, with the model [(total positive cost)=TBI+covariates] or [(total positive cost)=psychiatric illness+covariates]. In addition, a separate analysis was performed which tested the interaction between the presence of a TBI and the presence of psychiatric illness [(total positive cost)=TBI+psychiatric illness+interaction+covariates]. Generalized linear modeling with gamma regression and log link was chosen based on heteroscedasticity considerations using the modified Park test. More specifically, we first conducted ordinary least squares (OLS) regression on log-transformed costs, then performed a Cook-Weisberg test (Cook and Weisberg, 1983) for heteroscedasticity on the OLS residuals. When heteroscedasticity was detected in some explanatory variables, we employed generalized linear models (GLM). We then ran a gamma regression with log link on the cost data as an initial model. Next, we applied the modified Park test to the results of this initial model to determine which specific GLM to use. This was done by regressing the log of the raw-scale squared residuals on the log of the raw-scale predictions. When the slope coefficient was estimated to be close to 2, a gamma model with log link was indicated (Manning and Mullahy, 2001). Additional advantages for this model choice included that gamma regression with log link is the most commonly used GLM specification for health care cost analysis (Blough et al., 1999; Manning et al., 2005), and its use allows comparison with other studies of costs in the medical literature (Asche et al., 2010; Becker et al., 2011; Rockhill et al., 2010); and it avoids needing to re-transform the data for interpretation.

Covariates for the regression models in parts I and II of the cost analyses included sex; age; physical injury requiring medical attention within 6 months before or after the TBI reference date; chronic medical illness (ambulatory care group; Smith and Weiner, 1994; Weiner et al., 1992); insurance type (government funded versus other); total health care costs and the presence of psychiatric illness in the year prior to the TBI reference date; and the number of 6-month intervals for which each subject had data available in the study.

Results

Table 1 demonstrates that significant differences were seen between patients with mild TBI, moderate-to-severe TBI, and comparison participants in all demographic and clinical characteristics with the exception of gender. Although TBI-exposed and TBI-unexposed subjects were matched by age, separation of mild and moderate-to-severe TBI patients resulted in an age difference between groups, with older age seen in the moderate-to-severe TBI group. Subjects in the moderate-to-severe group were also more likely to have had a physical injury within 6 months of the TBI reference date. TBI-unexposed subjects had less psychiatric illness in the year prior to the study. In the 3 years after the study, TBI-unexposed persons had the least psychiatric illness, followed by mild TBI-exposed subjects, and the moderate-to-severe TBI group had the most.

Table 1.

Demographic Characteristics of the Sample

	TBI-unexposed controls	Mild TBI-exposed subjects	Moderate-to-severe TBI- exposed subjects
	n=2817	n=803	n=136
Demographic characteristic	n (%)	n (%)	n (%)	X² value
Male	1371 (48.7)	394 (49.1)	63 (46.3)	0.35
Female	1446 (51.3)	409 (50.9)	73 (53.7)
Age (y)				103.8^***
15–24	810 (28.8)	250 (31.1)	20 (14.7)
25–44	783 (27.8)	244 (30.4)	17 (12.5)
45–64	576 (20.4)	170 (21.2)	22 (16.2)
65–99	648 (23.0)	139 (17.3)	77 (56.6)
Physical injury within 6 months of TBI reference date	250 (8.9)	315 (39.2)	101 (74.3)	703.8^***
Psychiatric illness in the year prior to the TBI reference date	479 (17.0)	229 (28.5)	51 (37.5)	77.55^***
Psychiatric illness in the 3 years after the TBI reference date^a	1255 (44.7)	457 (58.0)	77 (71.3)	67.16^***

***

p<0.001.

For this variable, n's were reduced to 2808, 788, and 108 due to missing data.

TBI, traumatic brain injury.

Of the 1789 subjects with psychiatric illness in the 3 years after the TBI reference date, 569 had identified psychiatric illness in the year prior, and 1220 subjects did not; 41.1% of the subjects without evidence of psychiatric illness in the prior year were identified as having a psychiatric condition in the subsequent 3 years (data not shown, Pearson χ²=304.5, p<0.001), and 77.8% of the patients identified with a psychiatric condition in the prior year had a psychiatric condition in the subsequent 3 years. Among subjects with cost data for at least the first 6 months after the TBI reference date, affective disorders were identified in 337 (12%) of the TBI-unexposed group, and 195/896 of the TBI-exposed group (21.8%; 174 with mild TBI, 21 with moderate-to-severe TBI); psychosis was identified in 104 (3.7%) of the TBI-unexposed group and 68 (7.6%; 46 with mild TBI, 22 with moderate-to-severe TBI) of the TBI-exposed group; and substance abuse disorder was identified in 134 (4.8%) of the TBI-unexposed group and 103 (11.5%; 91 with mild TBI, 12 with moderate-to-severe TBI) of the TBI-exposed group.

Table 2 shows the unadjusted mean and median costs in the 3-year period after the TBI reference date, by TBI exposure and psychiatric illness status, for all subjects who remained enrolled at GHC for at least 6 months after the TBI reference date. The mean total cost of mild TBI-exposed patients was 76% greater than the mean for the TBI-unexposed control subjects ($12,990 versus $7377), whereas the cost for moderate-to-severe TBI-exposed patients was 5.75 times that of the TBI-unexposed control subjects ($42,441 versus $7377), with most of the costs incurred for the moderate-to-severe group in inpatient care. The mean total cost for the patients with psychiatric illness in the 3 years of the study was more than double the mean for the subjects without psychiatric illness ($13,383 versus $6053).

Table 2.

Mean Costs for Each Service Category by TBI Status and Psychiatric Illness Status

	TBI-unexposed controls	Mild TBI-exposed subjects	Moderate to severe TBI-exposed subjects		Subjects without psychiatric illness in the 3-years after reference date	Subjects with psychiatric illness in the 3-years after reference date
Service category	n=2808	n=788	n=108	Kruskal-Wallis rank test	n=1915	n=1789	Mann-Whitney U test
	Mean costs [95% CI]			χ² (d.f.=2)	Mean costs [95% CI]		Z
Primary care	1415	2383	1998	157.87^{*** b}	1328	1969	7.41^***
	[1356,1474]	[2201,2566]	[1569,2426]		[1267,1389]	[1861,2077]
Specialty care	1217	1639	2860	68.74^{*** a}	1039	1693	5.78^***
	[1133,1301]	[1467,1811]	[2186,3533]		[960,1117]	[1560,1825]
Emergency department	267	1052	1864	700.83^{*** a}	305	668	9.12^***
	[226,308]	[925,1178]	[1403,2325]		[260,351]	[590,746]
Pharmacy	925	1320	2491	85.20^{*** a}	743	1387	10.44^***
	[858,991]	[1163,1476]	[1642,3339]		[670,817]	[1279,1496]
Laboratory/radiology	712	1138	1403	142.24^{*** b}	609	1052	8.05^***
	[659,766]	[1004,1273]	[1019,1787]		[556,662]	[963,1141]
Outpatient mental health	168	398	244	64.66^{*** a}	0	453	26.82^***
	[137,198]	[268,527]	[95,392]		[0,0]	[380,527]
Total ambulatory care	4704	7930	10859	270.88^{*** b}	4025	7223	11.99^***
	[4472,4935]	[7329,8530]	[8824,12893]		[3808,4241]	[6819,7627]
Inpatient care	2674	5060	31583	185.88^{*** a}	2029	6160	10.68^***
	[2267,3081]	[3850,6270]	[17856,45309]		[1600,2458]	[5049,7272]
Total health service costs	7377	12990	42441	298.90^{*** a}	6053	13383	12.27^***
	[6818,7937]	[11482,14498]	[28040,56842]		[5493,6614]	[12081,14685]

***

p<0.001.

Post-hoc least significant difference (LSD) tests indicate that control costs are lower than mild TBI costs, which are lower than moderate-to-severe TBI costs.

Post-hoc LSD tests indicate that control costs are lower than mild TBI costs and moderate to severe TBI costs, which do not significantly differ from each other.

Costs adjusted to 2009 dollars and discounted 3% per year for years 2 and 3.

Logistic regression analyses controlled for age, sex, insurance type, any prior psychiatric illness, physical injury sustained within 6 months of TBI reference date, chronic medical conditions, total costs in the year prior to the TBI reference date, and amount of time in the study.

TBI, traumatic brain injury; CI, confidence interval.

The mean incremental cost associated with psychiatric illness in control subjects was $4851 ($10,060−$5209); in mild TBI-exposed patients was $7504 ($16,100 versus $8696); and for moderate-to-severe TBI-exposed patients was $30,285 ($51,421 versus $20,136). These large differences suggested possible strong interactions between TBI exposure and psychiatric illness, which we tested later using regression analysis.

Among all TBI-exposed subjects, more than 42.1% of the total health care costs in the 3-year period were incurred in the first 6 months after the TBI index date, whereas in the control group 15.8% of the total costs were incurred in the first 6 months.

Table 3 displays part 1 of the two-part model examining costs associated with health care utilization. TBI-exposed subjects had a higher proportion of respondents with nonzero cost in each category of care than TBI-unexposed subjects. Subjects with psychiatric illnesses had a higher proportion of nonzero cost in ED care, outpatient mental health care, and inpatient care, compared to subjects without psychiatric illness. Logistic regression models suggested that moderate-to-severe TBI patients had much higher odds of incurring costs in emergency care, pharmacy, and inpatient care. Patients with psychiatric illnesses appear to incur costs in primary care, pharmacy, and ambulatory care at higher odds than patients without psychiatric illness.

Table 3.

Logistic Regression Examining Odds of Incurring Any Health Care Costs in the Three Years following the TBI Reference Date by TBI Exposure Status and Psychiatric Illness Status

	TBI-unexposed controls	Mild TBI-exposed subjects	Moderate-to-severe TBI-exposed subjects			Subjects without psychiatric illness in the 3-years after the reference date	Subjects with psychiatric illness in the 3-years after the reference date
	n=2808	n=788	n=108			n=1915	n=1789
	n (%)	n (%)	n (%)	OR mild TBI versus controls	OR moderate-to-severe TBI versus controls	n (%)	n (%)	OR with psychiatric illness versus without
Service category	n (%) with nonzero cost
Primary care	2463 (87.7)	759 (96.3)	100 (92.6)	3.45^***	1.28	1733 (90.5)	1589 (88.8)	2.88^***
Specialty care	1975 (70.3)	611 (77.7)	90 (83.3)	1.41^*	1.31	1377 (71.9)	1299 (72.6)	1.49^***
Emergency department	563 (20.0)	463 (58.8)	100 (92.6)	4.97^***	30.08^***	468 (24.4)	658 (36.8)	1.65^***
Pharmacy	2462 (87.7)	740 (93.9)	106 (98.1)	1.79^*	5.76^*	1707 (89.1)	1601 (89.5)	2.82^***
Laboratory/radiology	2271 (80.9)	724 (91.9)	98 (90.7)	2.72^***	1.80	1599 (83.5)	1494 (83.5)	2.60^***
Outpatient mental health	359 (12.8)	190 (24.1)	25 (23.1)	1.71^***	1.83^*	0 (0.0)	574 (32.1)	NA
Total ambulatory care	2604 (92.7)	788 (100.0)	108 (100.0)	NA	NA	1823 (95.2)	1677 (93.7)	3.00^***
Inpatient care	520 (18.5)	256 (32.5)	68 (63.0)	2.01^***	4.37^***	305 (15.9)	539 (30.1)	2.32^***
Total health service costs	2605 (92.8)	788 (100.0)	108 (100.0)	NA	NA	1824 (95.2)	1677 (93.7)	3.01^***

p<0.05; ^*** p<0.001.

Logistic regression analyses controlled for age, sex, insurance type, any prior psychiatric illness, physical injury sustained within 6 months of the TBI reference date, chronic medical conditions, total costs in the year prior to the TBI reference date, and amount of time in the study.

TBI, traumatic brain injury.

We applied OLS regression to log-transformed costs, while adjusting for the same covariates used in previous logistic regression models. Cook-Weisberg heteroscedasticity test showed that the OLS residuals on the log-scale were heteroscedastic in some of the explanatory variables. Based on recommendations from Manning and Mullahy (2001), GLM was employed. After running a gamma model with log link and TBI status as the main predictor, we conducted the modified Park test. The slope coefficient estimate from the Park test was 1.92 (p<0.001). Since this coefficient was close to 2, gamma regression with log link was indicated. We further conducted a sensitivity analysis using Poisson model and log link; the Park test coefficient was 1.77 (p<.001), which also supported our choice of the gamma model. The Park test coefficient was 1.88 when psychiatric illness was the main predictor, again supporting gamma model specification.

Table 4 presents a summary of the results of gamma regression analyses examining differences in total cost by TBI status and psychiatric illness status during the 3 years following the TBI reference date, controlling for covariates. Compared to the TBI-unexposed group, exposure to mild TBI was associated with an increased cost ratio of 1.81, whereas exposure to moderate-to-severe TBI was associated with 7.27 times higher average cost. In comparison with the costs of subjects who did not have psychiatric illness in the 3-year period, the presence of psychiatric illness was associated with 2.13 times higher average cost. Subgroup analyses of costs within the psychiatric illness group revealed that significantly increased costs were associated with affective disorders and psychotic disorders, but not substance abuse disorders, compared to disorders outside these three major categories. Having psychotic disorder was associated with the highest increase in mean cost. We also tested for an interaction between TBI exposure and psychiatric illness. A significant interaction was found for the combination of moderate-to-severe TBI and psychiatric illness, with an increased cost ratio of 3.39 for the interaction between moderate-to-severe TBI and psychiatric illness.

Table 4.

Summary of Gamma Regression Analyses Examining the Mean Costs Associated with TBI and Psychiatric Illness in the Three Years after the TBI Reference Date among Participants with Nonzero Costs

		Unstandardized beta (robust SE)	z	Mean cost ratio [95% CI]
Total health service costs
Mild TBI versus TBI-unexposed		0.59 (0.08)	7.46^***	1.81 [1.55,2.11]
Moderate-to-severe TBI versus TBI-unexposed		1.98 (0.25)	7.83^***	7.27 [4.42,11.94]
Psychiatric Illness versus no Psychiatric Illness		0.75 (0.07)	10.73^***	2.13 [1.85,2.44]
Psychiatric illness	Affective disorder versus other disorders	0.26 (0.12)	2.20^*	1.30 [1.03,1.63]
	Psychotic disorder versus other disorders	0.95 (0.19)	5.03^***	2.59 [1.79,3.75]
	Substance abuse disorder versus other disorders	0.18 (0.12)	1.47	1.19 [0.94,1.51]
Interaction between TBI and psychiatric illness
Interaction of mild TBI and psychiatric illness		−0.02 (0.14)	−0.12	0.98 [0.75,1.29]
Interaction of moderate-to-severe TBI and psychiatric illness		1.22 (0.35)	3.52^***	3.39 [1.72,6.69]

p<0.05; ^*** p<0.001.

Costs adjusted to 2009 dollars and discounted 3% per year for years 2 and 3. Regression analyses controlled for age, sex, insurance type, prior psychiatric illness (any prior psychiatric illness for the comparison of the whole sample; prior affective disorder, psychotic disorder, or substance use disorder for the analyses specific to those diagnoses), physical injury sustained within 6 months of the TBI reference date, chronic medical conditions, total costs in the year prior to the TBI reference date, and the amount of time in the study.

TBI, traumatic brain injury; CI, confidence interval; SE, standard error.

Discussion

In this study we found that TBI exposure was associated with higher total health care costs in the 3-year period following a TBI in adolescents and adults, after controlling for relevant covariates. Both the mild and moderate-to-severe categories of TBI exposure were associated with substantial increases in the proportion of subjects who had nonzero cost in a variety of medical areas, and both were associated with increased mean total health care costs after controlling for relevant covariates. As expected, moderate-to-severe TBI was associated with greater increases in costs than mild TBI. There was a striking increase in psychiatric care costs among patients with moderate-to-severe TBI, supporting evidence that mental health sequelae of more severe TBIs has significant health care consequences (Hyder et al., 2007; Centers for Disease Control and Prevention, 2010).

The results also indicated that psychiatric illness in the 3 years following the index date was associated with a significant increase in the proportion of subjects who had nonzero cost in ED care, outpatient mental health care, inpatient care, and total health care costs.

Gamma regression showed no interaction between mild TBI and psychiatric illness, indicating that, although mild TBI-exposed subjects were more likely to have psychiatric illness, the additional cost of care for subjects with mild TBI did not significantly differ between those with and without psychiatric illness. In contrast, a significant interaction between moderate-to-severe TBI and psychiatric illness indicated that the additional cost of care for subjects with moderate-to-severe TBI was 3.4 times greater among subjects with psychiatric illness compared to those with moderate-to-severe TBI but no psychiatric illness.

It is important to note that in this cohort study, there were pre-existing differences between the TBI-exposed and TBI-unexposed subjects. Namely, TBI-exposed subjects were more likely to have a physical injury in the 6 months before or after the TBI reference date, and to have evidence of psychiatric illness in the year prior to the TBI reference date than the age- and sex-matched controls. Although the overall age composition of the TBI-exposed versus TBI-unexposed groups were matched, the age distribution between the TBI severity groups differed, with older mean age in the moderate-to-severe group. Therefore, although we controlled for age, this study is not a randomized controlled trial, and therefore we cannot fully separate the impact of moderate-to-severe TBI and older age on medical costs. To statistically control for pre-existing group differences, we used psychiatric illness in the year before the TBI reference date and physical injury in the 6 months before or after the TBI reference date as covariates in subsequent regression models. We did not exclude cases with multiple injuries, nor did we have detailed data on injury severity. Other important covariates were also included in the model.

Although we examined the difference in cost associated with psychiatric illness after TBI after controlling for prior psychiatric illness, these data do not examine the issue of whether new psychiatric diagnoses were a direct result of the TBI exposure. It is possible that in some cases the increased medical attention that results from TBI exposure increased diagnosis of pre-existing but previously undiagnosed psychiatric problems.

Information on the increased costs associated with treatment of TBI will be important to care providers in planning for appropriate resource allocation for patients who sustain mild and moderate-to-severe TBIs. Possible reasons for the increase in cost associated with psychiatric illness include lowered tolerance for aversive post-TBI symptoms such as headache and fatigue (Fann et al., 1995), and cognitive impairment impacting symptom tolerance and problem-solving (Slomine et al., 2002). This information adds to concern that the costs of providing treatment (e.g., inpatient rehabilitation services) are often greater than reimbursement from Medicare's prospective payment system (Dobrez et al., 2004), and are increased when psychiatric illness such as depression is comorbid (Dobrez et al., 2010). Few studies have examined the effectiveness or costs of depression treatment in the setting of TBI (Fann et al., 2009). Future research is needed to examine factors associated with increased cost in patients exposed to TBI.

The extremely high cost of care for patients with moderate-to-severe TBI is striking. Hospitalizations for these patients comprise the majority of the costs. Ongoing developments in outpatient treatments for this population may decrease total health care costs. We are currently in an exciting time for researchers interested in learning more about treatment responses in patients with TBI. A consortium of researchers has created the IMPACT database of approximately 10,000 TBI cases, which was collected from eight randomized controlled trials and three observational surveys, and are endeavoring to examine issues related to treatment effectiveness that had heretofore been impossible to examine due to a lack of adequate sample size (Maas et al., 2007). That database has the potential to further illuminate our finding that moderate-to-severe TBI modifies the incremental costs associated with psychiatric illness, by allowing examination of patient and treatment characteristics when patients have a combination of moderate-to-severe TBI and psychiatric illness. In this study, we found that affective disorders and psychotic disorders contributed to particularly high cost seen in the group of those having experienced psychiatric illness.

The present study had several limitations related to the use of administrative data. For example, it is possible that coding or entry errors may be present. In addition, mental health diagnoses were likely underreported, as is common in administrative datasets (Lofgren et al., 2006), and some subjects with TBI coded in 1993 could have sustained their TBI at an earlier date, thus making differences between groups more conservative. Our use of mental health service utilization and psychiatric medication prescription indicators may have partially counteracted this limitation by facilitating ascertainment of psychiatric illness. Also, the use of enrollment files resulted in a higher proportion of participants in the control group compared to the TBI-exposed group who lack costs during the examined time period. We addressed this issue by using a two-part model, in which the second part of the model includes only adults who have nonzero costs during the time period examined. The present data included patients from a large HMO, and may not be generalizable to other health care settings. In addition, some TBI severity ratings may have been misclassified, given that TBI is rated on a continuum and it is often difficult to make clear distinctions between the mild and more severe categories (Appleton and Baldwin, 2006). In addition, 43 of 52 subjects who were dropped from the study due to disenrollment in GHC in the first 6 months were in the TBI group, suggesting a possible underestimate of costs if those subjects enrolled in another system of care in which their costs accumulated. Finally, the data used in this study were collected from 1992–1996, and changes in medical care over time may reduce the generalizability of the findings. However, recent reviews indicate that the treatment of TBI and its neuropsychiatric sequelae in clinical settings has not changed substantially in the time frame examined (Ashman et al., 2006; Fann et al, 2009; Neurobehavioral Guidelines Working Group, 2006; Nolan, 2005). Furthermore, changes in Medicare's inpatient rehabilitation payment system (e.g., the Prospective Payment System, enforcement of the 60 Percent Rule) primarily affect moderate-to-severe TBIs: only 14% of our study population had moderate-to-severe TBI, and 4% were Medicare subscribers. Moreover, after adjusting to 2009 dollars, Medicare costs per inpatient discharge did not change substantially from 2001 ($16,969) to 2009 ($16,579) (Medicare Payment Advisory Commission, 2010; American Medical Rehabilitation Providers Association, 2010). We were not able to examine the differential impact of various cost factors (e.g., equipment costs and labor costs), and how changes in these factors might have affected overall costs. Future research is needed to examine these issues with newer data, but given that this article is one of the first to examine this issue, subsequent data will be able to build on our findings.

In sum, the present study showed that patients with TBI have greater medical costs during the 3-year period following their injury than control subjects over the same time period, after controlling for demographic, clinical, and prior utilization variables. Subjects with psychiatric illness, regardless of whether or not they had sustained a TBI, had about a doubling of costs during the 3-year period, in comparison with subjects without psychiatric illness after controlling for the same variables. For participants who both sustained a TBI and had psychiatric illness in the 3-year period, each illness contributed to cost. Furthermore, the combination of moderate-to-severe TBI and psychiatric illness resulted in a disproportionately high amount of cost for this subgroup of patients. Implications for health organizations include that both TBI and psychiatric illness can be expected to require a significant investment in care, and that consideration of ways to provide more efficient care for those with psychiatric illness and moderate-to-severe TBI has the potential to save significant health care dollars. Future research in this area should focus on whether screening for psychiatric illness and provision of psychiatric treatment after TBI is cost-effective and has cost-offset potential.

Footnotes

Acknowledgments

This research was supported by a grant to the last author from the National Center for Injury Prevention and Control, Centers for Disease Control and Prevention, Atlanta, Georgia. The funding organization's role was limited to review of the design of the study. We thank David Rubanowice, B.S., for his computer programming assistance, William Hollingsworth, Ph.D., for guidance regarding cost analyses, and Leighton Chan, M.D., M.P.H., for his expertise with Medicare and rehabilitation service payment policies.

Author Disclosure Statement

No competing financial interests exist.

References

American Medical Rehabilitation Providers Association (AMRPA). 2010. State of the Field. Washington, DC.

Appleton

, Baldwin

2006. Management of Brain-Injured Children. Oxford University Press: New York.

Asche

C.V.

, Joish

V.N.

, Camacho

, Drake

C.L.

2010. The direct costs of untreated comorbid insomnia in a managed care population with major depressive disorder. Curr. Med. Res. Opin., 26:1843–1853.

Ashman

T.A.

, Gordon

W.A.

, Cantor

J.B.

, Hibbard

M.R.

2006. Neurobehavioral consequences of traumatic brain injury. Mt. Sinai J. Med., 73:999–1005.

Becker

H.R.

, Iqbal

S.U.

, Kazis

L.E.

, Macarios

, Badamgarav

2011. The association of adherence to osteoporosis therapies with fracture, all-cause medical costs, and all-cause hospitalizations: a retrospective claims analysis of female health plan enrollees with osteoporosis. J. Manag. Care Pharm., 17:25–39.

Blough

D.K.

, Madden

C.W

, Hornbrook

M.C.

1999. Modeling risk using generalized linear models. J. Health Economics, 18:153–171.

Bombardier

, Fann

, Temkin

, Esselman

, Barber

, Dikmen

2010. Rates of major depressive disorder and clinical outcomes following traumatic brain injury. JAMA, 301:1938–1945.

Brouwer

W.B.

, Niessen

L.W.

, Postma

M.J.

, Rutten

F.F.

2005. Need for differential discounting of costs and health effects in cost effectiveness analyses. British Med. J., 331:446–448.

Centers for Disease Control and Prevention. 2010. National Center for Injury Prevention and Control. Injury Prevention and Control: Traumatic Brain Injury. http://www.cdc.gov/TraumaticBrainInjury/index.html. 2011 February 15.

10.

Ciechanowski

P.S.

, Katon

W.J.

, Russo

J.E.

2000. Depression and diabetes: impact of depressive symptoms on adherence, function, and costs. Arch. Intern. Med., 160:3278–3285.

11.

Cook

R.D.

, Weisberg

1983. Diagnostics for heteroscedasticity in regression. Biometrika, 70:1–10.

12.

Diehr

, Yanez

, Ash

, Hornbrook

M.C.

, Lin

D.Y.

1999. Methods for analyzing health care utilization and cost. Ann. Review Public Health, 20:125–144.

13.

Dobrez

, Heinemann

A.W.

, Deutsch

, Durkin

E.M.

, Almagor

2010. Impact of mental disorders on cost and reimbursement for patients in inpatient rehabilitation facilities. Arch. Phys. Med. Rehabil., 91:184–188.

14.

Dobrez

, LoSasso

, Heinemann

2004. The effect of prospective payment on rehabilitative care. Arch. Phys. Med. Rehabil., 85:1909–1914.

15.

Fann

, Burington

, Leonetti

, Jaffe

, Katon

, Thompson

2004. Psychiatric illness following traumatic brain injury in an adult health maintenance organization population. Arch. Gen. Psychiatry, 61:53–61.

16.

Fann

, Hart

, Schomer

2009. Treatment for depression after traumatic brain injury: A systematic review. J. Neurotrauma, 26:2383–2402.

17.

Fann

J.R.

, Katon

W.J.

, Uomoto

J.M.

, Esselman

P.C.

1995. Psychiatric disorders and functional disability in outpatients with traumatic brain injuries. Am. J. Psychiatry, 152:1493–1499.

18.

Faul

, Xu

, Wald

M.M.

, Coronado

V.G.

2010. Traumatic Brain Injury in the United States: Emergency Department Visits, Hospitalizations and Deaths. 2002–2006 Atlanta, GA: Centers for Disease Control and Prevention, National Center for Injury Prevention and Control.

19.

Gold

M.R.

, Siegel

J.E.

, Russel

L.B.

, Weinstein

M.C.

1996. Cost-Effectiveness in Health and Medicine. Oxford University Press: New York.

20.

Hawthorne

, Gruen

, Kaye

2009. Traumatic brain injury and long-term quality of life: Findings from an Australian study. J. Neurotrauma, 26:1623–1633.

21.

Hessen

, Nestvold

, Anderson

2007. Neuropsychological function 23 years after mild traumatic brain injury: a comparison of outcome after paediatric and adult head injuries. Brain Inj., 21:963–979.

22.

Hyder

A.A.

, Wunderlich

C.A.

, Puvanachandra

, Gururaj

, Kobusingye

O.C.

2007. The impact of traumatic brain injuries: A global perspective. NeuroRehabilitation, 22:341–353.

23.

Katon

, Lin

, Russo

, Unutzer

2003. Increased medical costs of a population-based sample of depressed elderly patients. Arch. Gen. Psychiatry, 60:897–903.

24.

Lofgren

, Karpf

, Perman

, Higdon

C.M.

2006. The U.S. health care system is in crisis: Implications for academic medical centers and their missions. Academic Med., 81:713–720.

25.

Maas

, Marmarou

, Murray

, Teasdale

, Steyerberg

2007. Prognosis and clinical trial design in traumatic brain injury: The IMPACT study. J. Neurotrauma, 24:232–238.

26.

Manning

W.G.

, Mullahy

2001. Estimating log models: to transform or not to transform? J. Health Economics, 20:461–494.

27.

Manning

W.G.

, Basu

, Mullahy

2005. Generalized modeling approaches to risk adjustment of skewed outcomes data. J. Health Economics, 24:465–488.

28.

Massagli

T.L.

, Fann

J.R.

, Burington

B.E.

, Jaffe

K.M.

, Katon

W.J.

, Thompson

R.S.

2004. Psychiatric illness after mild traumatic brain injury in children. Arch. Phys. Med. Rehab., 85:1428–1434.

29.

Medicare Payment Advisory Commission (MedPAC). 2010. Inpatient Rehabilitation Facility Services: Assessing payment adequacy and updating payments. Report to the Congress, Medicare Payment Policy, March, 2011.

30.

National Center for Health Statistics. 1991. International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) Publication no. (PHS), 91–1260.

31.

Neurobehavioral Guidelines Working

Group

, Warden

D.L.

, Gordon

, McAllister

T.W.

, Silver

J.M.

, Barth

J.T.

et al. 2006. Guidelines for the pharmacologic treatment of neurobehavioral sequelae of traumatic brain injury. J. Neurotrauma, 23:1468–1501.

32.

Nolan

2005. Traumatic brain injury: A review. Crit. Care Nursing Q., 28:188–194.

33.

Ortendahl

, Fries

J.F.

2002. Time-related issues with application to health gains and losses. J. Clin. Epidemiol., 55:843–848.

34.

Rockhill

C.M.

, Fann

J.R.

, Fan

, Hollingworth

, Katon

W.J.

2010. Healthcare costs associated with mild traumatic brain injury and psychological distress in children and adolescents. Brain Inj., 24:1051–1060.

35.

Saunders

K.W.

, Stergachis

, Van Korff

1994. Group Health Cooperative of Puget Sound. In: Pharmacoepidemiology, 2nd. B.L. Strom (ed). John Wiley and Sons: New York, pp. 171–186.

36.

Slomine

B.S.

, Gerring

J.P.

, Grados

M.A.

, Vasa

, Brady

K.D.

, Christensen

J.R.

et al. 2002. Performance on measures of executive function following pediatric traumatic brain injury. Brain Inj., 16:759–772.

37.

Smith

M.S.

, Weiner

J.P.

1994. Applying population-based case mix adjustment in managed care: the Johns Hopkins Ambulatory Care Group system. Manag. Care Q., 2:21–34.

38.

Thurman

D.J.

, Sniezek

J.E.

, Johnson

, Greenspan

, Smith

S.M.

1995. Guidelines for surveillance of central nervous system injury. Centers for Disease Control and Prevention, National Center for Injury Prevention and Control: Atlanta.

39.

U.S. Department of Labor, Bureau of Labor Statistics. 2006. Consumer Price Index. http://stats.bls.gov/cpi/RetrievedDecember 15, 2006.

40.

Walker

E.A.

, Katon

, Russo

, Ciechanowski

, Newman

, Wagner

A.W.

2003. Health care costs associated with posttraumatic stress disorder symptoms in women. Arch. Gen. Psychiatry, 60:369–374.

41.

Warriner

E.M.

, Velikonja

2006. Psychiatric disturbances after traumatic brain injury: neurobehavioral and personality changes. Curr. Psychiatry Reports, 8:73–80.

42.

Weiner

J.P.

, Starfield

G.H.

, Lieberman

R.N.

1992. Johns Hopkins Ambulatory Care Groups (ACGs) A case-mix system for UR, QA and capitation adjustment. HMO Pract., 6:13–19.