Salim A, Ley EJ, Cryer HG, Margulies DR, Ramicone E, Tillou A. Positive Serum Ethanol Level and Mortality in Moderate to Severe Traumatic Brain Injury. Arch Surg. 2009;144(9):865-871. doi:10.1001/archsurg.2009.158
Ethanol exposure is associated with decreased mortality in patients with moderate to severe traumatic brain injury.
Retrospective database review.
Trauma centers contributing to the National Trauma Data Bank (NTDB).
Version 6.2 of the NTDB (2000-2005) was queried for all patients with moderate to severe traumatic brain injury (head Abbreviated Injury Score ≥3) and ethanol levels measured on admission. Demographics and outcomes were compared between patients with traumatic brain injuries with and without ethanol in their blood. Logistic regression analysis was used to investigate the relationship between mortality and ethanol.
Main Outcome Measures
Mortality and complications.
A total of 38 019 patients with severe traumatic brain injuries were evaluated. Thirty-eight percent tested positive for ethanol. Ethanol-positive patients were younger (mean [SD], 37.7 [15.1] vs 44.1 [22.0] years, P < .001), had a lower Injury Severity Score (22.3 [10.0] vs 23.0 [10.3], P < .001), and a lower Glasgow Coma Scale score (10.0 [5.1] vs 11.0 [4.9], P < .001) compared with their ethanol-negative counterparts. After logistic regression analysis, ethanol was associated with reduced mortality (adjusted odds ratio, 0.88; 95% confidence interval, 0.80-0.96; P = .005) and higher complications (adjusted odds ratio, 1.24; 95% confidence interval, 1.15-1.33; P < .001).
Serum ethanol is independently associated with decreased mortality in patients with moderate to severe head injuries. Additional research is warranted to investigate the potential therapeutic implications of this association.
Two million individuals each year sustain traumatic brain injury (TBI) in the United States, resulting in 56 000 deaths and 80 000 impaired individuals at an estimated cost of $56 billion.1 Therapeutic interventions that inhibit the brain's deterioration after TBI have made little impact in clinical practice. More than 21 multi-center clinical trials aimed to assess the value of various treatments for TBI have been conducted since 1985 with none demonstrating a convincing benefit for TBI-directed treatment.2
Alcohol use has been implicated as a risk factor for all types of trauma. Thirty-nine percent of all traffic-related deaths during 2004 were alcohol related3 and up to 50% of patients hospitalized for trauma are intoxicated at the time of injury.4 Despite these associations between alcohol and trauma, very little is known regarding the pathophysiologic implications of ethanol on TBI. Experimental animal studies suggest that ethanol may have a neuroprotective effect, though results are conflicting.5,6 Recently, several small clinical studies have demonstrated a survival advantage in patients with TBI who tested positive for serum ethanol compared with ethanol-negative patients.4,7 The objective of this study was to examine the effect of a positive serum ethanol level in patients with moderate to severe TBI using data from a large national trauma database. Our hypothesis is that serum ethanol is protective in patients with moderate to severe TBI.
This is a retrospective review of trauma patients included in version 6.2 of the National Trauma Data Bank who were injured between 2000 and 2005. Patients with isolated moderate to severe TBIs (defined as having a head Abbreviated Injury Score [AIS] of 3 or higher with an AIS of 3 or less for all other body regions [chest, abdomen, and extremity]) who had a serum ethanol level measured on admission were identified as the study population. Routine demographic data (age, sex, Injury Severity Score, AIS, and Glasgow Coma Scale score) and outcome data (complications, hospital and intensive care unit length of stay, days spent on a ventilator, discharge disposition, functional independence score, and mortality) were recorded for all patients. Patients were excluded for being aged 13 years or younger, death on arrival or within 24 hours, transfer to another facility, admission with a burn diagnosis, AIS greater than 5 for any body region, and missing data on age, sex, or head AIS.
Frequency and percentages for the categorical variables; simple means; and standard deviations for the continuous variables are reported. Owing to the large number of patients not tested for ethanol, patient factors between ethanol tested and not tested were analyzed. A comparison between the presence of and absence of ethanol with respect to patient characteristics and outcomes was also performed. The equality of distributions between the 2 groups was tested by the 2-sample Wilcoxon test. For categorical variables and continuous variables, which have been regrouped into ordinal or binary scales, the distributions or proportions of patients between the 2 groups were compared by using the χ2 test. If any of the patient characteristics were significantly different between the presence and absence of ethanol at a tolerance (α) level of .05, they were used as a covariate (adjusting factor) in logistic regression in comparing the death rates or complication rates between the 2 groups. The covariates used for adjustments were age group (14-20, 21-45, 46-55, 56-65, 66-75, 76-89, and >89 years); race (white, black, Hispanic, Asian, American Indian, and other); sex; Glasgow Coma Scale score (≤ or >8); blunt or penetrating injury; Injury Severity Score (≥16 or <16); AIS (>3 or ≤3) in the head, thorax, or abdomen; any comorbidity (yes or no); systolic blood pressure (not dichotomized); face AIS; upper and lower extremity AIS; and unspecified AIS in original scale. The same set of covariates were also used in multiple regressions for comparing the adjusted means of hospital and intensive care unit length of stay.
Logistic regression analysis was used to investigate the relationship between mortality (and morbidity) and a set of risk factors, each adjusting for other factors. Significance level for entry into the model was P = .2; to remain in the model, P = .05. Variables at P < .2 with less than 10% missing data were preselected as variables available at the initial stage of stepwise logistic regression. They were presence or absence of alcohol; age in 7 groups (14-20, 21-45, 46-55, 56-65, 66-75, 76-89, and >89 years); race (white, black, Hispanic, Asian, American Indian, and other); blunt or penetrating injury; Glasgow Coma Scale score (≤8 or >8); systolic blood pressure (<90 or ≥90 mm Hg); Injury Severity Score (≥16 or <16); AIS (>3 or ≤3) in the head, face, lower extremity, neck, thorax, abdomen, and spine; any comorbidity; and any complication (yes or no). The significance level of removal from the model was set at P = .05. Adjusted odds ratios (AORs) and the 95% confidence intervals (CIs) were calculated. All statistical analysis was performed using SAS, version 9.1.3 (SAS Institute Inc, Cary, North Carolina). Data are presented as mean (standard deviation) unless otherwise specified. This study was determined to be exempt from institutional review board approval by the institutional review board of the Cedars-Sinai Medical Center.
A total of 72 294 patients with isolated moderate to severe TBI were evaluated for the presence of ethanol. Nearly 53% of patients (38 019 patients) had measured serum ethanol levels and comprised the study population. Table 1 compares demographic data and outcomes between patients with TBI tested and not tested for ethanol. The 2 groups were statistically quite different in every aspect with the exception of head AIS. Table 2 presents the demographics of the total study population. Nearly 38% (14 419 patients) tested positive for serum ethanol, and overall mortality was 8.9%. Table 3 compares the demographic data between patients with and without positive serum ethanol levels. Ethanol-positive patients were significantly younger (37.7 [15.1] vs 44.1 [22.0] years, P < .001), had a significantly lower Injury Severity Score (22.3 [10.0] vs 23.0 [10.3], P < .001), and had a lower Glasgow Coma Scale score (10.0 [5.1] vs 11.0 [4.9], P < .001) compared with their ethanol-negative counterparts. As expected, ethanol-positive patients were more likely to have a positive toxicology for an illicit drug (55.2% vs 46.6%, P < .001).
Table 3 also compares clinical outcomes between the 2 groups. Ethanol-positive patients had a lower mortality (7.7% vs 9.7%, P < .001; AOR, 0.87; 95% CI, 0.79-0.96; P = .005), more complications (12.9% vs 9.8%, P < .001; AOR, 1.28; 95% CI, 1.18-1.38; P < .001), shorter time on a ventilator (7.9 [10.0] vs 8.7 [10.3] days, P < .001), and shorter adjusted intensive care unit lengths of stay (7.3 [0.1] vs 7.7 [0.1] days, P < .001) with no difference in adjusted hospital length of stay (12.4 [0.1] vs 12.5 [0.1] days, P = .33). After adjusting for risk factors by logistic regression, we found that ethanol was associated with reduced mortality (AOR, 0.88; 95% CI, 0.80-0.96; P = .005) (Table 4) and more complications (AOR, 1.24; 95% CI, 1.15-1.33; P < .001) (Table 5).
In the largest database review on this topic to date, patients with isolated moderate to severe TBI who present to a trauma center with a positive serum ethanol level died less frequently than their ethanol-negative counterparts. For every 100 patients without ethanol positivity who died, only 88 with ethanol died; this mortality benefit reached significance. The decrease in mortality was tempered by an apparent increase in complications for patients with positive serum ethanol levels, a finding consistently reported in the literature.8- 10
How ethanol reduces mortality in patients with isolated moderate to severe TBI is unknown. Animal studies have demonstrated both positive and negative effects of ethanol on TBI depending on the dose administered.5,6,11- 13 Several TBI animal studies demonstrated improved mortality, reduced cognitive impairments, decreased contusion volume, and improved motor performance with the preinjury administration of low doses of ethanol compared with placebo.6,11- 13 The mechanisms are multifactorial but include a marked attenuation of immediate postinjury hyperglycolysis with a decreased degree of uncoupling between glucose metabolism and cerebral blood flow,12 a decrease in pro-inflammatory cytokine production,14 and the inhibition of N-methyl-D-aspartate receptor–mediated excitotoxicity.5,6 These beneficial effects are apparently lost at higher doses.
Animal studies that demonstrate an ethanol-mediated decrease in survival after TBI are also plentiful.15- 19 Ethanol-treated animals demonstrated early respiratory depression, decreased cerebral perfusion pressure, and decreased cerebral blood flow.16,17 Furthermore, the deleterious effect of ethanol was found to be greater when TBI was accompanied by cerebral hemorrhage.18,19 Fabian and Proctor20 uniquely concluded that mortality was not affected by ethanol after TBI, provided that cardiopulmonary support was given. They did observe an elevation in systemic and cerebral venous lactate in the ethanol-TBI group, which did not appear to affect outcome. It is important to note that most of these animal models did not follow the entire life of the animal with TBI after ethanol administration. Their conclusions about how ethanol affects survival after TBI were based on secondary parameters, such as early respiratory depression, cerebral perfusion pressure, and cerebral blood flow; thus, variations in these parameters might not describe how TBI affects mortality.
Clinical reports on the subject also vary. Fabbri et al21 found that a positive blood ethanol level was associated with an increased risk of multiple and more severe injuries in general. Pories et al22 found an association between positive admission serum ethanol levels and worse severity of injury scores. Alexander et al23 related decreased cerebral blood flow to positive ethanol level on admission, though no increase in mortality or long-term outcome as measured by the Glasgow Outcome Score was demonstrated. Cunningham et al24 found that ethanol potentiates the severity of TBI as determined from head computed tomography of individuals in motor vehicle collisions. In one of the only studies that examined the role of catecholamines after trauma, Woolf et al25 demonstrated a reduction in the catecholamine response to TBI with ethanol use. They found that this ethanol-induced blunting of the catecholamine response was most marked in those who were severely injured. More recently, O’Phelan et al7 demonstrated improved survival in ethanol-intoxicated patients with TBI (OR, 0.23; 95% CI, 0.10-0.56; P = .001) compared with their unintoxicated counterparts. Finally, Tien et al4 demonstrated a dose-response association between ethanol and severe TBI. Patients with brain injuries with low to moderate blood alcohol levels had improved survival (OR, 0.76; 95% CI, 0.52-0.98), whereas a high blood alcohol level was associated with worse survival (OR, 1.73; 95% CI, 1.05-2.84).
It is difficult to interpret the wide range of results regarding the effects of ethanol on patient outcomes after TBI. Although current explanations are inadequate, the blunting of the catecholamine response after TBI with the use of β-blockers is also described in a number of recent studies.26- 30 No randomized prospective trial exists that can answer this question fully, and it is doubtful that ethanol will be used to treat TBI. Our study does suggest a survival benefit with ethanol (AOR, 0.88; 95% CI, 0.80-0.96; P = .005). Not surprisingly, ethanol was associated with a significant increase in complications (AOR, 1.24; 95% CI, 1.15-1.33; P < .001).
The sociologic implications are important and have been raised previously.4 It is important to note that we examined in-hospital mortality as our outcome measure. It is well established that alcohol contributes to nearly 40% of traffic fatalities3,31 and the risk of dying is obviously increased while driving under the influence. However, the finding of reduced mortality in TBI patients with preinjury ethanol raises the intriguing possibility that administering ethanol to patients with brain injuries may improve outcome.
There are a number of limitations to this study. The retrospective design of the trauma registry used for data analysis limits the conclusions that can be established. Our study analyzed only patients with TBI who had an ethanol level measured. Trauma centers vary in terms of the percentage of patients with measured ethanol levels. Only 53% of the patients with moderate to severe TBI in the National Trauma Data Bank had a measured serum ethanol level. This obviously introduces a level of bias that is unavoidable. The comparison between patients tested and not tested for ethanol demonstrates that the 2 patient populations were quite different. This makes it difficult to apply our results to all patients with moderate to severe brain injuries. Several studies suggest a dose-response curve with different results according to the level of ethanol. Unfortunately, our analysis only includes its presence or absence and not its specific level. Our outcomes included only mortality and complications. Obtaining detailed neurological and functional outcomes would have strengthened the study. We also did not have adequate data on chronic vs acute alcohol consumption, a potentially important difference, as it may represent 2 distinct patient populations. Lastly, we did not analyze the effects of alcohol consumption on specific types of brain injury. It is possible that the beneficial effect we observed may not apply to all types of injury. Despite these limitations, the study represents the largest analysis of the effects of ethanol on TBI in the literature and provides some useful information.
We found that a positive serum ethanol level is independently associated with increased survival in patients with isolated moderate to severe TBI. Additional research is warranted to investigate reasons for this association and the potential therapeutic implications.
Correspondence: Ali Salim, MD, Department of Surgery, Cedars-Sinai Medical Center, 8700 Beverly Blvd, Ste 8215N, Los Angeles, CA 90048 (email@example.com).
Accepted for Publication: May 15, 2009.
Author Contributions:Study concept and design: Salim and Ley. Acquisition of data: Cryer and Tillou. Analysis and interpretation of data: Salim, Margulies, and Ramicone. Drafting of the manuscript: Salim and Ley. Critical revision of the manuscript for important intellectual content: Cryer, Margulies, Ramicone, and Tillou. Statistical analysis: Ramicone. Administrative, technical, and material support: Salim. Study supervision: Cryer and Tillou.
Financial Disclosure: None reported.
Previous Presentation: This paper was presented at the 80th Annual Meeting of the Pacific Coast Surgical Association; February 16, 2009; San Francisco, California; and is published after peer review and revision. The discussions that follow this article are based on the originally submitted manuscript and not the revised manuscript.
Edward E. Cornwell III, MD, Washington, DC: Dr Salim and colleagues have queried a more than 1.4 million–patient database and performed a multivariate analysis in order to determine if ethanol use is associated with an altered risk of death or complications. This manuscript is important because it is emblematic of the challenges associated with outcomes research, which has become such a large part of both academic medicine and health policy.
When seeking to address a topic such as ethanol use and its impact on trauma outcomes, where prospective randomized studies are either impossible or impractical, the authors correctly resort to multivariate logistic regression analysis to account for differences between ethanol-positive and -negative groups seen on univariate analysis.
In this study, ethanol-positive patients were younger and less severely injured than their ethanol-negative counterparts and not surprisingly had a lower mortality on univariate analysis. When the authors then apply logistic regression analysis to adjust for all observed differences and report that ethanol positivity was associated with a 12% lower risk of death, it becomes the reader's responsibility to offer constructive criticism and questions to uncover any other unstudied factors that may account for their findings. I have a few questions:
Your study sample of 38 000 patients with isolated moderate to severe TBI and a measured ethanol level is only about 2.5% of the 1.4 million–patient NTDB [National Trauma Data Bank]. Even when accounting for your other exclusion criteria, this raises the question of a selection bias, with only a small proportion of patients having serum ethanol checked. The impact of such a bias would be pure speculation, but one wonders if better-equipped trauma centers (with better patient outcomes) were more likely to comply with screening guidelines and therefore identify all their patients who were ethanol-positive.
In an analysis of NTDB by our group presented at PCSA [Pacific Coast Surgical Association] and published in Archives of Surgery, insurance status was a powerful predictor of outcome in trauma patients—even more powerful than race. Subsequent analysis appears to suggest that it is also more important than gender or mechanism of injury. Reasons for this are unclear, but we believe insurance status is a surrogate for access to a primary physician—knowledge of and treatment of comorbidities and other positive health behaviors before patients are injured. Do you have information on health insurance status and its distribution/impact on the outcomes in the groups?
We have found controlling for comorbidities as you attempted in your logistic regression analysis to be difficult because these data are so often missing in NTDB. How much did your sample size drop when you included any comorbidity? In other words, how many of your 38 000 patients were included in your final multivariate analysis?
Dr Salim: Of 72 000 patients with moderate to severe traumatic brain injury, only about 55% were screened for ethanol. I absolutely agree that an inherent selection bias is a major limitation of our study.
It also is very possible ethanol testing is done by the better-equipped trauma centers. Also, not every trauma center submits its patients to the data bank, so maybe it's the better-equipped trauma centers that actually do so. This is a limitation of all studies using the NTDB.
We did not look at insurance status. In a paper presented at last year's AAST [American Association for the Surgery of Trauma] meeting, insurance status had no effect on outcome. As the uninsured are more likely to have positive toxicology for alcohol and other substances, insurance status may have an influence on our results.
Regarding comorbidity, again we are limited by our data. In our patient population, approximately one-third of the patients had evidence of comorbidity. I don't know how many patients had to be excluded in the final regression analysis.