Alcohol Use at Time of Injury and Survival Following Traumatic Brain Injury: Results From the National Trauma Data Bank

Abstract

Objective:

Premised on biological evidence from animal research, recent clinical studies have, for the most part, concluded that elevated blood alcohol concentration levels are independently associated with higher survival or decreased mortality in patients with moderate to severe traumatic brain injury (TBI). This study aims to provide some counterevidence to this claim and to further future investigations.

Method:

Incident data were drawn from the largest U.S. trauma registry, the National Trauma Data Bank, for emergency department admission years 2002–2006. TBI was identified according to the National Trauma Data Bank’s definition using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), codes. To eliminate confounding, the exact matching method was used to match alcohol-positive with alcohol-negative incidents on sex, age, race/ethnicity, and facility. Logistic regression compared in-hospital mortality between 44,043 alcohol-positive and 59,817 matched alcohol-negative TBI incidents, with and without causes and intents of TBI and Injury Severity Score as covariates. A sensitivity analysis was performed within a subsample of isolated moderate to severe TBI incidents.

Results:

Alcohol use at the time of injury was found to be significantly associated with an increased risk for TBI. Including varied causes and intents of TBI and Injury Severity Score as potential confounders in the regression model explained away the statistical significance of the seemingly protective effect of alcohol against TBI mortality for all TBIs and for isolated moderate to severe TBIs.

Conclusions:

The null finding shows that the purported reduction in TBI mortality attributed to positive blood alcohol likely is attributable to residual confounding. Accordingly, the risk of TBI associated with alcohol use should not be overlooked.

Several animal and laboratory studies have suggested that alcohol might have a neuroprotective effect on the brain following traumatic brain injury (TBI) through such mechanisms as inhibiting N-methyl-D-aspartate (NMDA) receptor-mediated excitotoxicity (Cebere and Liljequist, 2003; Chandler et al., 1993; Is et al., 2005; Türeci et al., 2004), decreasing the degree of uncoupling between glucose metabolism and cerebral blood flow (Kelly et al., 2000), inhibiting the production of pro-inflammatory cytokines (Gottesfeld et al., 2002), attenuating TBI-induced hyperthermia (Taylor et al., 2002), and blunting the sympa-thoadrenal response in TBI (Opreanu et al., 2010). Premised on the evidence from animal research, one recent study confirmed that patients who were intoxicated at the time of injury had better acute neuropsychological outcomes when recovering from TBI than did nonintoxicated patients (Lange et al., 2008). Further, an increasing number of clinical studies have concluded that elevated serum ethanol levels are independently associated with higher survival or decreased mortality in patients with moderate to severe TBI (O’Phelan et al., 2008; Salim et al., 2009b; Talving et al., 2010; Tien et al., 2006).

Chief among these are the studies based on a large, countywide trauma database and a national database of trauma incidents. Analyzing isolated moderate to severe TBI incidents in the Los Angeles County Trauma System database, Berry and colleagues (2010, 2011) found that patients who tested positive for blood alcohol on admission had a significantly lower mortality rate than their alcohol-negative counterparts and that the mortality rate decreased with increasing blood alcohol concentration (BAC) level. Likewise, Salim and colleagues (2009a) reported that, of 38,019 isolated moderate to severe TBI patients identified from the National Trauma Data Bank (NTDB), 14,419 patients who tested positive for alcohol had a significantly lower mortality rate than patients who tested negative for alcohol. This finding remained robust after a multivariable adjustment for injury severity and other confounders. Some related trauma studies (Plurad et al., 2010; Yaghoubian et al., 2009), although not exclusively focusing on TBI, further lent support to the neuroprotective effect of alcohol on TBI mortality.

Despite these emerging findings, a comprehensive review (Opreanu et al., 2010) of the research into the influence of alcohol on TBI mortality has called into question the translation of knowledge from the animal studies to clinical studies because of the limitations in the size and anatomic complexity of animal models. Moreover, because previous clinical studies mostly focused on moderate to severe TBI or isolated TBI, it is unclear whether their findings can be generalized to all types of TBI. An important note is that clinical data of 1,529 TBI patients from the National Study on Costs and Outcome from Trauma, when adjusted for injury severity, did not show alcohol intoxication to be significantly associated with a reduction in mortality following TBI (Shandro et al., 2009). This rare finding suggested that the protective effect of alcohol against TBI mortality may not apply universally to all populations, or else it could be a statistical artifact attributable to inadequate adjustments for confounding factors.

Previous research has indicated that the mechanism of injury is independently predictive of mortality and functional impairment of blunt-trauma patients at hospital discharge (Haider et al., 2009). Therefore, we hypothesized that the mechanism of injury is one of the potential sources of confounding, although the association between acute alcohol consumption and the mechanism of injury has not been well established in the literature (Watt et al., 2006).

In the interest of public heath, the present study sought to investigate this issue further with a twofold objective: (a) to determine the relationship between alcohol use at the time of injury (as indicated by the presence of blood alcohol) and the risk of TBI and (b) to examine injury severity and varied causes and intents of injury as plausible confounders in explaining why in-hospital mortality from TBI ostensibly differed in the presence of blood alcohol.

Method

Sample

This study analyzed the NTDB Research Data Set Version 7.1 (Committee on Trauma, American College of Surgeons, 2008), which comprises incident data of emergency department admission years 2002–2006 submitted to the American College of Surgeons Committee on Trauma by participating hospitals with trauma registries. Data files received from contributing hospitals are checked by the NTDB for completeness, logical consistency, and proper formatting. Any files not passing the checking system are either rejected or flagged, based on the seriousness of the file’s errors. During the period from 2002 to 2006, 1,926,245 records of incident data were submitted to the NTDB by 770 hospitals, which included 175 Level I trauma centers, 188 Level II trauma centers, 54 Level III trauma centers, 43 Level IV trauma centers, and 310 Level V or unspecified trauma centers. Based on the 2007 NTDB Annual Report (Clark and Fantus, 2007), we estimate that 89% of Level I, 76% of Level II, and 19% of Level III trauma centers in the United States contributed data to the version of NTDB used in the present study. Although the NTDB may not be representative of all trauma hospitals nationwide, it is the largest aggregation of trauma registry data ever compiled in the United States. The large number of injury incidents provided a unique opportunity to study TBI in relation to alcohol use at the time of injury.

To reduce selection bias, the NTDB patient inclusion criteria, based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM; National Center for Health Statistics, 2010), included the following discharge diagnosis codes: 800.00–959.9 excluding 905–909 (late effects of injury), 910–924 (blisters, contusions, abrasion, and insect bites), and 930–939 (foreign bodies). Among the cases in the NTDB, 1,116,265 trauma incidents met the NTDB inclusion criteria, and 396,722 trauma incidents met the NTDB’s definition of TBI. Of those cases that met the inclusion criteria, 415,306 patients were tested for blood alcohol at the emergency department and were the pool of our case selection.

At the outset of the analyses, we opted to use a cohort matching method to select the study sample in a deliberate attempt to completely eliminate confounding by the selected matching factors (Cummings et al., 2003). The exact matching method implemented in the statistical package MatchIt (Ho et al., 2011) was used to match each alcohol-positive trauma incident to all possible alcohol-negative trauma incidents on selected matching factors, with unmatchable cases being excluded from subsequent analyses. Additionally, we excluded 14,738 trauma incidents in which patients had chronic alcohol abuse and dependence as a pre-existing comorbid condition because previous research suggested that their trauma outcomes could be different from those of patients with acute alcohol intoxication (Jurkovich et al., 1993).

This left 137,950 alcohol-positive and 262,618 alcohol-negative trauma incidents available for matching. The matching of alcohol-positive to alcohol-negative trauma incidents produced two matched samples for analysis. First, matching on sex, single year of age (i.e., age in years as originally reported without grouping), race/ethnicity, and facility (i.e., each uniquely identified participating hospital) among all trauma incidents yielded a matched trauma sample of 104,654 alcohol-positive and 166,711 alcohol-negative trauma incidents from 382 hospitals. Second, matching on sex, single year of age, race/ethnicity, and facility among the TBI incidents yielded a matched TBI sample of 44,043 alcohol-positive and 59,817 alcohol-negative TBI incidents from 314 hospitals.

Measures

Traumatic brain injury incident.

Following the NTDB’s definition (Fantus and Stone, 2008), we used discharge diagnoses to identify trauma incidents involving TBI, which was head injury, including brain injury (ICD-9-CM: 850–854.19, 803.1–803.49, 800.1–800.49, 800.6–800.99, 801.1–801.49, 801.6–801.99, 803.6–803.99, 804.1–804.49, and 804.6–804.99) and skull injury (ICD-9-CM: 800.0–800.09, 800.5–800.59, 801.0–801.09, 801.5–801.59, 802–803.09, 803.5–803.59, 804–804.09, and 804.5–804.59).

Alcohol use.

As the main factor of interest, alcohol use at the time of injury was determined by the blood alcohol test that took place in the emergency department. The presence or absence of alcohol in the blood reported for the test result was coded as alcohol-positive or alcohol-negative, respectively.

Causes and intents of traumatic brain injury.

These two factors were treated as potential confounders. In the NTDB, each TBI incident was associated with one ICD-9 external cause of injury code (E-code), if it was reported. Based on the E-codes, TBI incidents were classified into mutually exclusive categories of causes and intents of injury in accordance with the framework of E-code groupings for presenting injury mortality and morbidity data recommended by the Centers for Disease Control and Prevention (1997). Causes of injury included the following 18 categories: cut/pierce; drowning/submersion; fall; fire/burn; firearm; machinery; motor vehicle traffic (MVT); pedal cyclist, not MVT; pedestrian, not MVT; transport, not MVT; natural/environment; overexertion; poisoning; struck by/against; suffocation; other specified and classifiable; other specified, not elsewhere classified; and unspecified. Intents of injury included the following four categories: unintentional, self-inflicted, assault, and undetermined/other.

Severity of injury.

Another potential confounder considered to be highly predictive of survival was injury severity, commonly measured by Injury Severity Score (ISS). The ISS was defined as the sum of squares of the three highest 5-point Abbreviated Injury Scale (AIS) scores for six body regions (i.e., head/neck, face, chest, abdomen and pelvic contents, extremities or pelvic girdle, and external), with a range from 0 to 75. A higher ISS indicates greater trauma severity (Baker et al., 1974).

In-hospital death.

This is the main outcome of interest pertaining to whether a patient survived a TBI incident. It was defined by the patient’s vital status (i.e., dead or alive) at discharge from hospital.

Isolated moderate to severe traumatic brain injury.

For a sensitivity analysis, the matched TBI sample was further restricted to incidents of isolated moderate to severe TBI (n = 42,196). To select this subsample, we regenerated the AIS based on ICD codes because a staggering 72% of the incidents in the matched TBI sample had missing data on the original AIS provided in the NTDB. This was done by translating the anatomic ICD-9-CM diagnosis codes (i.e., the nature-of-injury codes) to AIS using the algorithm implemented in ICD Programs for Injury Categorization (ICDPIC) Version 3.0 (Clark et al., 2009). The resulting ICD-based AIS corresponded to six ISS body regions. The criterion for isolated moderate to severe TBI was having a score of 3 or greater for the head/neck region and a score of 3 or less for each of all other body regions. Note that the ICD-based AIS generated by ICDPIC did not separate the head region from the broad head/neck region.

Statistical analysis

Analyses were conducted on the two matched samples using Stata (StataCorp LP, College Station, TX). All estimates incorporated the weights created by MatchIt as sampling weights. Because each alcohol-positive trauma incident could potentially match with more than one alcohol-negative trauma incident, the weights ensured that alcohol-negative trauma incidents resembled alcohol-positive trauma incidents in the sense that these two groups had the same weighted distributions for all of the matching covariates. We estimated the percentage distributions of TBI, causes and intents of TBI, and vital status at discharge as well as means of ISS. Kernel density estimation was used to depict the distribution of TBI across age. Logistic regression with fractional polynomials (i.e., an extension of polynomials of degree n that allows for n integer/fractional powers) procedure of degree 2 implemented in the multivariable fractional polynomial interaction procedure (Royston and Sauerbrei, 2008) was used to generate the predicted values and the associated 95% confidence intervals (CIs). This procedure allows for flexibly modeling the shape of age trend of TBI incidents and TBI deaths on a continuous scale and for examining whether the age trends differed by alcohol use at the time of injury.

Central to addressing our research question was the comparison between two logistic regression models that assessed the effect of alcohol use at the time of injury on the risk of dying at the hospital with and without controlling for causes and intents of TBI and ISS as confounders. A sensitivity analysis restricting the sample to isolated moderate to severe TBI incidents was conducted similarly in an effort to compare the findings with those of other studies that focused on this more severe subset of TBI incidents. A second sensitivity analysis used the multivariable fractional polynomial interaction procedure to model the relationship between injury severity (as measured by ISS and ICD-based head/neck AIS) and the vital status at discharge in the total matched TBI sample. The purpose was to examine whether this relationship differed by alcohol use at the time of injury.

For all analyses, the significance level was set at .05 for chi-square tests; alternatively, the nonoverlapping 95% confidence intervals were used to evaluate the significance of differences observed in the estimates.

Results

The prevalence of TBI in the matched trauma sample was higher among the alcohol-positive trauma incidents (49.5%, 95% CI [49.2%, 49.8%]) than among their alcohol-negative counterparts (40.2%, 95% CI [39.8%, 40.6%]). A further examination by sex and race/ethnicity revealed that this was generally true across all of these demographic subgroups, although the effect size did not appear to be uniform. As shown in Figure 1a, males on average had a higher prevalence of TBI among the alcohol-positive trauma incidents (49.4%, 95% CI [49.1%, 49.8%]) than among their alcohol-negative counterparts (39.6%, 95% CI [39.2%, 40.0%]); the same was true for females (49.9%, 95% CI [49.1%, 50.6%] vs. 43.0%, 95% CI [42.2%, 43.8%]). Similarly, significant differences between alcohol-positive and alcohol-negative trauma incidents existed among Whites, Blacks, Hispanics, and “other” races; conversely, no such differences existed among Native Americans/Alaskan Natives or among Asians/Pacific Islanders.

Figure 1.

(a) Prevalence (%) of traumatic brain injury (TBI), by sex and race/ethnicity; (b) prevalence of TBI, by age; and (c) odds ratio for TBI, by age, among matched alcohol-positive (n = 104,654) and alcohol-negative (n = 166,711) trauma incidents, National Trauma Data Bank, 2002–2006. CI = confidence interval.

As shown in Figure 1b, the evidence of effect modification by age was particularly prominent. Within the younger and older cohorts, a large majority of the trauma incidents involved TBI. For both alcohol-positive and alcohol-negative trauma incidents, the percentage involving TBI decreased precipitously with age before age 40. Beginning with age 40, an upward trend took place, although the subsequent increase was only about half as steep as the preceding decline. Despite a similar convex shape, the two age trends differed by the degree of curvature, with alcohol-positive trauma incidents showing a less curving trend. In terms of odds ratio (OR) for TBI by alcohol use (Figure 1c), the age trend assumed a concave shape with the peak value above 1.5 around age 40, indicating that the effect of alcohol involvement on TBI was a function of age (for interaction, p < .0001). It is worth noting that the odds for TBI were significantly higher among alcohol-positive trauma incidents relative to their alcohol-negative counterparts throughout the age span from 16 to 66 years, when alcohol use was most common.

In-hospital mortality in the matched TBI sample was compared between TBI incidents with and without alcohol involvement. Overall, a slightly lower percentage of alcohol-positive TBI incidents ended in death at discharge than did their alcohol-negative counterparts (6.39%, 95% CI [6.16%, 6.62%] vs. 6.99%, 95% CI [6.70%, 7.28%]). A closer look at the difference across age (Figure 2a and 2b) revealed that the reduction in mortality among alcohol-positive TBI patients largely resulted from the higher mortality of children and the elderly in the alcohol-negative TBI patients (for interaction, p < .0001). However, these segments of the patients contributed relatively smaller proportions of the overall TBIs (Figure 2c). At ages close to 40 years, the difference in in-hospital mortality essentially was nonexistent (i.e., ORs close to 1). Between age 36 and 54, the observed in-hospital mortality appeared slightly higher among alcohol-positive TBI incidents (6.48%, 95% CI [6.08%, 6.89%]) than among their alcohol-negative counterparts (6.32%, 95% CI [5.84%, 6.83%]), although the difference was not significant.

Figure 2.

(a) Proportion of traumatic brain injury (TBI) death, by age; (b) odds ratio for TBI death, by age; and (c) probability density function of age, among matched alcohol-positive (n = 44,043) and alcohol-negative (n = 59,817) TBI incidents, National Trauma Data Bank, 2002–2006. CI = confidence interval.

Further analysis examined the causes and intents of TBI among alcohol-positive versus alcohol-negative TBI incidents. As presented in Table 1, the majority of TBIs resulted from MVT and other transportation-related causes, followed by fall and being struck by/against an object. On average, TBIs resulting from MVT had substantially higher injury severity scores (∼17) than TBIs from being struck by/against an object (∼11). Alcohol-positive TBI incidents, compared with their alcohol-negative counterparts, were less commonly caused by MVT (56.8%, 95% CI [56.3%, 57.2%] vs. 58.3%, 95% CI [57.7%, 58.9%]) but more commonly caused by being struck by/against an object (13.4%, 95% CI [13.1%, 13.7%] vs. 9.31%, 95% CI [8.97%, 9.66%]).

Table 1.

Percentage distribution of external causes and intents of traumatic brain injury (TBI) and the associated Injury Severity Score (ISS), by the presence of alcohol in blood, National Trauma Data Bank, 2002–2006

	Blood alcohol
	Positive (n = 44,043)		Negative (n = 59,817)
Causes and intents of TBI	% [95% CI]	ISS [95% CI]	% [95% CI]	ISS [95% CI]
Causes
Cut/pierce	0.94 [0.86, 1.04]	12.17 [11.18, 13.16]	0.54 [0.47, 0.63]	13.71 [12.17, 15.25]
Drowning/submersion	0.04 [0.02, 0.06]	19.88 [15.15, 24.61]	0.10 [0.06, 0.16]	13.71 [10.93, 16.49]
Fall	14.03 [13.70, 14.36]	13.91 [13.67, 14.15]	16.11 [15.69, 16.54]	15.41 [15.12, 15.70]
Fire/burn	0.09 [0.06, 0.12]	16.77 [11.20, 22.35]	0.09 [0.06, 0.13]	16.76 [11.56, 21.95]
Firearm	2.51 [2.36, 2.66]	22.33 [21.53, 23.13]	3.13 [2.93, 3.34]	22.94 [22.07, 23.81]
Machinery	0.08 [0.06, 0.12]	17.50 [13.93, 21.07]	0.61 [0.52, 0.71]	19.39 [16.82, 21.96]
MVT	56.75 [56.29, 57.22]	17.11 [16.95, 17.27]	58.29 [57.71, 58.87]	17.02 [16.82, 17.21]
Pedal cyclist, not MVT	1.21 [1.11, 1.32]	11.56 [10.84, 12.28]	1.41 [1.28, 1.55]	13.37 [12.55, 14.19]
Pedestrian, not MVT	0.26 [0.22, 0.31]	19.00 [16.77, 21.23]	0.24 [0.19, 0.31]	18.55 [15.57, 21.52]
Transport, not MVT	6.37 [6.14, 6.61]	16.02 [15.59, 16.45]	6.98 [6.67, 7.30]	15.43 [14.97, 15.90]
Natural/environment	0.07 [0.05, 0.10]	14.52 [9.14, 19.89]	0.31 [0.26, 0.37]	11.94 [10.56, 13.32]
Overexertion	0	—	0.01 [0.00, 0.02]	19.51 [13.97, 25.05]
Poisoning	0.04 [0.02, 0.06]	9.93 [5.99, 13.88]	0.08 [0.04, 0.16]	12.86 [9.55, 16.17]
Struck by, against	13.40 [13.08, 13.73]	10.41 [10.22, 10.60]	9.31 [8.97, 9.66]	11.23 [10.91, 11.56]
Suffocation	0.07 [0.05, 0.10]	13.10 [9.55, 16.66]	0.15 [0.11, 0.20]	17.02 [13.69, 20.35]
Other specified and classifiable	0.51 [0.45, 0.59]	13.54 [12.13, 14.95]	0.82 [0.73, 0.93]	16.02 [14.63, 17.40]
Other specified, NEC	0.49 [0.43, 0.56]	10.56 [9.44, 11.68]	0.35 [0.29, 0.43]	11.98 [10.18, 13.77]
Unspecified	3.13 [2.97, 3.30]	12.03 [11.55, 12.51]	1.48 [1.34, 1.63]	13.53 [12.68, 14.38]
Intents
Unintentional	79.65 [79.27, 80.03]	16.31 [16.19, 16.44]	87.24 [86.84, 87.64]	16.36 [16.21, 16.51]
Self-inflicted	1.42 [1.32, 1.54]	21.09 [20.00, 22.18]	1.82 [1.65, 2.00]	23.00 [21.67, 24.33]
Assault	18.24 [17.87, 18.60]	11.56 [11.36, 11.75]	10.36 [10.00, 10.73]	13.16 [12.81, 13.52]
Undetermined/other	0.69 [0.61, 0.77]	14.47 [13.24, 15.69]	0.59 [0.50, 0.68]	16.19 [14.44, 17.95]

Notes: Causes and intents categories are based on the framework of E-code groupings for presenting injury mortality and morbidity data recommended by the Centers for Disease Control and Prevention (1997).

MVT = motor vehicle traffic;

NEC = not elsewhere classified.

With regard to injury intent, the vast majority (>80%) of TBIs among both alcohol-positive and alcohol-negative incidents were unintentional injuries. Self-inflicted injuries constituted only a small fraction (<2%). Although assault injuries made up a sizable proportion of TBIs, they were much less frequent than unintentional TBIs and were associated with lower ISS. Comparison of alcohol-positive and alcohol-negative TBIs found that the percentage of unintentional TBIs was lower (79.7%, 95% CI [79.3%, 80.0%] vs. 87.2%, 95% CI [86.8%, 87.6%]) and the percentage of assault TBI was higher (18.2%, 95% CI [17.9%, 18.6%] vs. 10.4%, 95% CI [10.0%, 10.7%]) among the alcohol-positive TBIs. Taken together, these disparities of TBI composition in terms of causes and intents of injury offer a good explanation as to why the severity of alcohol-positive TBIs on average was lower than that of their alcohol-negative counterparts (ISS = 15.5, 95% CI [15.4, 15.6] vs. ISS = 16.1, 95% CI [16.0, 16.3]).

The logistic regression presented in Table 2 used the matched TBI sample to examine the extent to which the potential confounders affected the observed relationship between in-hospital mortality and alcohol use at the time of injury. Without adjusting for covariates, the alcohol-positive TBI incidents were associated with reduced in-hospital mortality, compared with their alcohol-negative counterparts (OR = 0.91, p = .001). However, when causes and intents of TBI and ISS were included as covariates in the model, which increased R² from near 0 to .31, alcohol-positive TBI incidents became nonsignificantly associated with a very slight increase in in-hospital mortality (OR = 1.05, p = .205). Results from the sensitivity analysis restricting the matched TBI sample to isolated moderate to severe TBI incidents (Model 4) showed the same null finding (OR = 1.03, p = .412).

Table 2.

Logistic regression models predicting the risk of death at discharge as associated with the presence of blood alcohol in the matched traumatic brain injury (TBI) sample and a subsample of isolated moderate to severe TBI, National Trauma Data Bank, 2002–2006

	Any TBI (n = 103,860)				Isolated moderate to severe TBI (n = 42,196)
	Model 1 (McFadden’s R2 = .0003)		Model 2 (McFadden’s R2 = .3098)		Model 3 (McFadden’s R2 = .0005)		Model 4 (McFadden’s R2 = .2117)
Independent variable	OR	(SE)	OR	(SE)	OR	(SE)	OR	(SE)
Blood alcohol (ref.: negative)
Positive	0.91**	(0.03)	1.05	(0.04)	0.89**	(0.03)	1.03	(0.04)
Cause*** (ref.: cut/pierce)
Drowning/submersion			5.17*	(3.64)			1.90	(1.14)
Fall			1.17	(0.38)			1.06	(0.37)
Fire/burn			0.33	(0.24)			0.49	(0.55)
Firearm			9.18***	(2.87)			6.81***	(2.25)
Machinery			1.25	(0.53)			1.10	(0.53)
MVT			0.59	(0.19)			0.62	(0.21)
Pedal cyclist, not MVT			0.55	(0.21)			0.54	(0.21)
Pedestrian, not MVT			0.95	(0.40)			0.95	(0.43)
Transport, not MVT			0.55	(0.19)			0.57	(0.21)
Natural/environment			0.37	(0.27)			0.51	(0.32)
Overexertion			6.65	(7.33)			4.38	(5.00)
Poisoning			0.49	(0.53)			1.25	(1.33)
Struck by, against			1.06	(0.34)			0.86	(0.29)
Suffocation			4.30**	(2.02)			3.58*	(1.94)
Other specified and classifiable			1.52	(0.55)			1.25	(0.47)
Other specified, NEC			1.64	(0.67)			1.16	(0.52)
Unspecified			1.99*	(0.65)			1.61	(0.56)
Intent*** (ref.: unintentional)
Self-inflicted			1.67***	(0.20)			1.69***	(0.23)
Assault			0.48***	(0.05)			0.52***	(0.06)
Undetermined/other			1.07	(0.20)			1.18	(0.23)
Injury Severity Score			1.12***	(0.00)			1.10***	(0.00)

Notes: Bold indicates the odds ratio of positive to negative blood alcohol concentration, which was statistically significant in Models 1 and 3 but became nonsignificant when confounders were included in Models 2 and 4. Sample sizes would be slightly reduced because of listwise deletion of missing data (2.30% on cause and intent of injury and 0.71% on the Injury Severity Score for any TBI cases; 2.41% on cause and intent of injury and 0.56% on Injury Severity Score for isolated moderate to severe TBI cases). OR = odds ratio; ref. = reference; MVT = motor vehicle traffic; NEC = not elsewhere classified.

^*p < .05;

^**p < .01;

^***p < .001.

The sensitivity analysis presented in Figure 3 shows that injury severity (measured by either ISS or ICD-based head/ neck AIS) was highly predictive of the vital status at discharge. However, there was no conclusive evidence that this relationship differed significantly by alcohol involvement (for interaction, p = .754 for ICD-based head/neck AIS and p = .071 for ISS).

Figure 3.

(a) Proportion of traumatic brain injury (TBI) death, by International Classification of Diseases—based head/neck Abbreviated Injury Scale (AIS) score; (b) proportion of TBI death, by Injury Severity Score (ISS); and (c) odds ratio for TBI death, by ISS, among matched alcohol-positive (n = 44,043) and alcohol-negative (n = 59,817) TBI incidents, National Trauma Data Bank, 2002–2006. CI = confidence interval.

Discussion

To our knowledge, this is one of the few studies that have used the largest trauma registry database in the United States to examine the risk of TBI in relation to alcohol use at the time of injury. The results of this study showed that, on balance, alcohol use at the time of injury significantly increased the risk of TBI, particularly throughout the adult life span when more frequently alcohol was consumed in excess of recommended drinking limits.

As in previous studies supporting the hypothesis that alcohol was neuroprotective in TBI, the bivariate relationship showed that, on the surface, alcohol-positive TBI incidents had a slightly lower risk of in-hospital mortality. However, when confounders such as ISS and causes and intents of TBI were taken into account by means of regression modeling, the protective effect of alcohol no longer existed, and the adjusted in-hospital mortality even appeared to increase very slightly in the presence of blood alcohol. Our results confirmed that ISS and causes and intents of TBI were important confounders. A post hoc analysis further matching alcohol-positive and alcohol-negative TBI incidents (7,788 and 8,620, respectively) on ISS and causes and intents of TBI produced a relative risk of 0.99 (p = .896) for death at discharge. This alternative analysis strategy, while reducing the sample size, confirmed the finding from the regression modeling. In either case, our analysis did not show that the presence of blood alcohol significantly reduced in-hospital mortality for patients who sustained TBI. This finding is in line with two recent trauma studies (Hadjizacharia et al., 2011; Zeckey et al., 2011), although these studies were not focused particularly on TBI.

In contrast to the analysis of an earlier version of the NTDB conducted by Salim and colleagues (2009a), our sensitivity analysis restricted to a subsample of moderate to severe TBI incidents also reached a null result. An additional sensitivity analysis looking into the differential risk of death at discharge between alcohol-positive and alcohol-negative TBI incidents across the full spectrum of injury severity similarly failed to find a significant reduction in mortality in association with blood alcohol, except for a slight reduction among those with ICD-based head/neck AIS of 3 or greater.

Other than the analytic methods used and confounding factors considered, several potential factors may have contributed to the different findings. First, Salim and colleagues (2009a) used an earlier version of the NTDB that covered the years 2000–2005. Second, that study reported using the head AIS to define moderate to severe TBI; it was not clear whether the case selection followed the NTDB inclusion criteria for trauma injury because the AIS definition for TBI is not consistent with the NTDB’s definition. In the version of the NTDB analyzed by the present study, 53% of the injury incidents with a valid head AIS (i.e., with values 1–6) did not meet the NTDB patient inclusion criteria for trauma injury, and 63% did not qualify for TBI as defined by the ICD-9-CM. Conversely, 59% of trauma incidents and 65% of TBI incidents defined by the NTDB did not even record AIS. Moreover, the NTDB’s definition of TBI included ICD-9-CM diagnosis code 802 (fracture of face bones), which was considered an injury to the face as opposed to the head/neck of ISS body regions. These differences in the TBI definition and sample selection make it difficult to pinpoint the exact factors that contributed to the discrepancy between the present study and previous reports. Nevertheless, we share the view of a discussant (Salim et al., 2009a) on the earlier study that residual confounding may be a reasonable explanation for the seemingly protective effects of blood alcohol in TBI.

Previous literature has noted that alcohol-positive TBI patients may have a different profile from alcohol-negative TBI patients with respect to major clinical and demographic characteristics (Berry et al., 2010; Salim et al., 2009a). Our use of the exact matching method has an advantage over regular regression in assessing the region of common support between alcohol-positive and alcohol-negative trauma incidents (Morgan and Harding, 2006). The matched sample helps to fully control for the selected matching factors, thereby reducing biases. Although this approach may have caused some loss in efficiency because of sample size reduction by excluding incomparable or unmatchable cases, the matched TBI sample still included more than 100,000 incidents. Because of this large sample size, it is unlikely that this null finding is attributable to insufficient statistical power. Thus, our findings bear credible counterevidence to the purported beneficial effect of alcohol on reducing TBI mortality that has been reported by some observational studies.

There are several limitations in our study. First, a large proportion of the TBI incidents in the NTDB did not include blood alcohol data and, therefore, were excluded from the analysis. In most cases, the NTDB data are not missing completely at random; therefore, our point estimates are subject to bias. The extent to which the missing BAC data may bias the results depends on how missing data are distributed between alcohol-positive and alcohol-negative TBI incidents. Examination of cases missing on BAC data indicated that they had a higher proportion of dead-on-arrival (0.46%, 95% CI [0.43%, 0.49%] vs. 0.18%, 95% CI [0.16%, 0.20%]) or died-in-emergency department (1.61%, 95% CI [1.55%, 1.66%] vs. 0.85%, 95% CI [0.81%, 0.90%]) incidents than the alcohol-tested cases. This raises the possibility that the decision for blood alcohol testing may be influenced by the emergency department disposition or survival prognosis.

In terms of causes and intents of injury, the cases without alcohol test data appeared to be more similar to the alcohol-positive TBI incidents than to the alcohol-negative TBI incidents. Therefore, with their relatively high mortality, exclusion of these cases should not have contributed to our null finding. However, if the missing data were disproportionately from the BAC-negative group, our result may be biased toward underestimating the mortality in the BAC-negative group.

Although we have no way to determine the distribution of missing BAC data, a methodological study of the NTDB data by Roudsari et al. (2008) provides some support to the validity of our conclusion. The study compared results based only on cases with complete information (i.e., ignoring cases missing on BAC) with results based on all cases (i.e., including cases with missing BAC imputed by multiple imputation). Its finding indicates that analysis based on nonmissing BAC cases tends to underestimate the odds of death in the BAC-positive group relative to the BAC-negative group. Still, future studies are needed to corroborate our finding with more complete alcohol testing data.

Second, because no quantitative measurements of BAC are available from the NTDB, we cannot rule out the possibility that low to moderate BACs indeed may have some neuroprotective effect against TBI. It is possible that our null finding reflects the cancellation of the neuroprotective effect of alcohol at low dosage with its neurotoxic effect at high dosage. The dose-response relationship between alcohol use and TBI mortality remains to be examined in future studies.

Third, because the NTDB is a convenient sample obtained through hospitals’ voluntary participation, our finding can only apply to the patient population admitted to hospitals that have a trauma registry and submit data to the NTDB; its generalizability to other trauma centers outside the coverage of the NTDB could be limited, especially to non-Level I and II trauma centers and nontrauma centers where the NTDB has poor representation.

Finally, our analysis focused on the confounding effect of ISS and causes and intents of TBI and therefore did not comprehensively examine all prognostic factors related to in-hospital mortality (e.g., Glasgow Coma Scale score, hy-poxia, other substance uses) (Kim, 2011). Future studies may discover more explanatory factors.

In conclusion, the finding that no statistically significant differences in TBI mortality existed between alcohol-positive and alcohol-negative cases in the present study casts doubt on the claim that blood alcohol is independently associated with decreased mortality in patients with TBI or in those with isolated moderate to severe TBI. According to a Centers for Disease Control and Prevention estimate (Faul et al., 2010), the annual number of TBI-related emergency department visits, hospitalizations, and deaths was close to 1.7 million between 2002 and 2006. Evidently, the NTDB only captures a portion (<5%) of the overall TBI cases annually, but in light of the gravity of TBI morbidity and mortality, it certainly is inadvisable to consume alcohol or administer alcohol to patients either before or after TBI for the purpose of reducing mortality. Even if many of these alcohol-related TBIs do not result in near-term mortality, they still can lead to poorer neurological, medical, neuropsychological, and functional outcomes (Parry-Jones et al., 2006); have long-term consequences for mental, physical, and social functioning (Safaz et al., 2008); and increase the risk of recurrent TBI (Winqvist et al., 2008) and premature death (McMillan and Teasdale, 2007).