The Role of Alcohol and Other Drugs on Emergency Department Traumatic Injury Mortality in the United States

Abstract

Background

Alcohol and other drugs (AOD) increase the risk of traumatic injury occurring, but data suggest a protective benefit in preventing trauma-related mortality. The objective of this study is to describe the epidemiology of AOD-related traumatic injury in the US over a recent 7 year period and assess the interaction of traumatic injury and AOD on pre-admission fatality on both an additive scale using incidence contrasts and on a statistical multiplicative scale using survey-adjusted logistic regression.

Methods

Using the National Emergency Department Sample (NEDS), we describe the epidemiology of alcohol and substance-related emergency department traumatic injury over a recent period. AOD-related injury was assessed using survey-adjusted counts and means. Ratio estimates and differences were calculated using simulations based on survey-adjusted counts and standard errors. Differences in trends over time were evaluated by comparing the slopes of linear regression equations with year as the predictor variable.

Results

Alcohol and substance-related emergency department injury discharges increased 9.8% during the study period. There was a statistically significant interaction between traumatic injury death and AOD on both an additive scale and multiplicative scale. (Odds Ratio for interaction term = 1.76, 95% CI = 1.53, 2.03).

Conclusions

AOD use does not provide a protective benefit in the setting of trauma, but rather is an important contributor to traumatic injury mortality.

Introduction

Alcohol and other drug (AOD) use is increasing in the United States (US). In 2014, approximately 27 million Americans older than age 12 reported using an illicit drug in the previous month—the highest proportion since 2002 – and approximately 61 million Americans reported binge drinking.¹ Evidence indicates a large overlap between alcohol and other drugs. In 2008, nearly a third of Americans who were heavy alcohol users reported illicit drug use as well, compared to 3% of non-drinkers.²

AOD use is recognized as a risk factor for the occurance of trauma and injury. Injury risk (including motor vehicle and bicycle crashes and falls) increases in a dose-response fashion with blood alcohol content, and drugs such as benzodiazepines and opioids are over-represented among those seeking medical care for injuries.^3,4 The prevalence of AOD use in severely injured patients presenting to Level 1 trauma centers has been estimated to be as high as 71%.⁵ An alcohol level greater than 0.08g/dl increases the risk of severe or fatal bicycling injuries by over twenty-fold (OR = 20.2),⁶ and a blood alcohol level greater than 200 mg/dl in a driver translates to a near certainty that a motor vehicle crash (MVC) resulted from the effects of alcohol on the vehicle operator.⁷

Despite these clear correlations, the secondary role of AOD use in mortality among those who are injured has been the subject of debate. A numer of investigations report that individuals with traumatic injury and AOD are at decreased risk of death when compared to persons with similar traumatic injuries but no AOD use.^8,9,10,11 This has perhaps most frequently been observed in the setting of traumatic brain injury (TBI).^12,13,14 A recent state-based trauma registry study reported a protective effect for AOD on a dose-response basis.¹⁵

There have been other studies showing contradictory data reporting higher injury severity and mortality for those with AOD at the time of injury compared to those without AOD.^{16,17, 18,19} A recent trauma-registry based study found an “inverse U-shaped” response curve with increasing inpatient mortality risk for low or moderately impaired individuals followed by lower rates at higher alcohol levels.²⁰

These discrepancies may be due at least in part to selection bias. Inpatient databases may impart a a survival benefit to those who live to be admitted to a hospital, while those with AOD are more likely to die at the scene or in the ED.² Additionally, the interaction between traumatic injury and AOD has generally been treated from a statisical perspective.

In this study, we attempt to address this issue by analyzing ED discharges, which would be expected to address at least some selection bias by including a larger population-based group or patients. We consider the interaction between traumatic injury and AOD from the perspective of additive interaction, which to our knowledge, has not been applied to this issue and may allow for a more biologically plausible model as well as more clinically relavent measurement and quantification, ³² than the perhaps more commonly seen assessment of statistical interaction.

We begin by presenting the descriptive epidemiology of AOD-related traumatic injury in the US over a recent 7 year period using a comprehensive and nationally-representative ED database to evaluate trends. Our main outcomes and measures are survey-adjusted counts, proportions, means, and rates with standard errors, and 95% confidence intervals. We assess clinically relevant interaction of traumatic injury and AOD on fatality on an additive scale using incidence contrasts. As a point of comparison, we use logistic regression to model the statistical interaction of AOD with traumatic injury in ED fatality controlling for age and gender. The results represent one of the most comprehensive assessments of AOD and traumatic injury in the US in recent years.

Methods

Data

Data were obtained from the US Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) Nationwide Emergency Department Sample (NEDS) for the years 2006 to 2012. HCUP is a group of inpatient and outpatient files created by AHRQ. Based on a 20% stratified single-cluster sample of hospital-based EDs, NEDS is the largest and most representative single publicly available ED database in the US. Core files consist of 100% of visits from the sampled hospitals. Taken together the sampled ED’s account for approximately 66% of all ED discharges for that year. Sampled hospitals are non-federal general and specialty hospitals including public hospitals and academic medical centers. Stratification variables include geographic area, urban/rural, ownership, trauma center and teaching status, and bed size.²¹

Comma-separated text core files for each year were read into an R data-frame, which were then converted to MonetDB column-oriented database files and appended to each other using the R packages “DBI”²² and “MonetDBLite”²³, resulting in a full dataset of 198,102,435 unweighted observations. Traumatic injury discharges and external cause of injury codes were identified using principle or first-listed ICD 9th edition diagnosis codes for acute injury 800-904.9, 909.4, 909.9, 910-994.9, 995.5-995.59, and 995.80-995.85.²⁴ As noted in the HCUP documentation, the ICD-9-CM coding guidelines define principal diagnosis as “that condition established after study to be chiefly responsible for occasioning the admission of the patient to the hospital for care.”²⁵ Fourteen secondary diagnostic codes were searched to create indicator variables for alcohol and substance use as outlined in Table 1. An alcohol or drug related injury was defined as one with discharge with a primary or first diagnostic code for traumatic injury and the presence of an alcohol or drug related diagnostic coded in any of the following 14 codes.

Table 1:

ICD9 Codes for Secondary Alcohol and Drug Related Diagnoses

Category	ICD9 Codes
Alcohol	3050, 30500, 30501, 30502, 3030, 30300, 30301, 30302, 3039, 30390, 30391, 30392, 291, 2910, 2911, 2912, 2913, 2914, 2915, 2918, 29181, 29182, 29189, 2919
Opioid	304, 3040, 30400, 30401, 30402, 3055, 30550, 30551, 30552
Sedative Hypnotic	3041, 30410, 30411, 30412, 3054, 30540, 30541, 30542
Cocaine	3042, 30420, 30421, 30422, 3056, 30560, 30561, 30562
Cannabis	3043, 30430, 30431, 30432, 3052, 30520, 30521, 30522
Amphetamine	3044, 30440, 30441, 30442, 3057, 30570,
Stimulant	30571, 30572
Hallucinogen	3045, 30450, 30451, 30452, 3053, 30530, 30531, 30532
Drug Combination	3047, 30470, 30471, 30472, 3048, 30480, 30481, 30482
Antidepressant	3058, 30580, 30581, 30582
Other Unspecified	3049, 30490, 30491, 30492, 3059, 30590,
Drug	30591, 30592, 3046, 30460, 30461, 30462
Drug-Related Mental	292, 2920, 2921, 29211, 29212, 2922, 2928,
Disorder	29281, 29282, 29283, 29284, 29285, 29289, 2929

Injury severity was quantified using the ICD-derived Injury Severity Score (ICISS) as proposed by Osler et al as a means of estimating injury severity using ICD codes in administratively collected hospital discharge data.²⁶ ICISS is calculated in two steps. First, survival risk ratios (SRRs) for each injury diagnosis in a data set are “…calculated as the ratio of the number of times a given ICD-9 code occurs in (surviving patients) to the total number of occurrences of that code”. Second, the ICISS for an individual patient is calculated as “the product of all the survival risk ratios for each of an individual patient’s injuries (for as many as ten different injuries).”²⁷ The ICISS is then defined as the probability of patient surviving their injuries and ranges from 0 to 1. An ICISS cut-off of less than 0.94 was used to categorize patients into those with the most severe injuries as proposed by Gedeborg.²⁸ This indicator variable identifies patients with a 6% or greater probability of dying, and has performed well in previous analyses, returning an odds ratio of 6.75 (95% CI 6.48, 7.03) in multivariate logistic regression analysis of trauma mortality.²⁹ Fatalities were restricted to those that occurred after arrival in the ED but prior to hospital admission. Trauma center designations were based on the AHRQ “HOSP_TRAUMA” indicator variable found in the NEDS hospital file for that year. Costs were based on charges for each discharge. Unlike the HCUP Nationwide Inpatient Sample (NIS) files, cost-to-charge files are not available for NEDS. As a conservative estimate, we calculated costs as 42% of charges. Costs were then adjusted for inflation and standardized to 2012 US dollars based upon the all-item average yearly consumer price index obtained from the Bureau of Labor Statistics. US census data were obtained from AHRQ as part of the HCUP family of data products.

Analyses

The descriptive epidemiology, including visit counts and census population-based rates and trends, ages, genders, types of injuries and external causes of injury and costs for alcohol and substance-related emergency department traumatic injury dischages was assessed using survey-adjusted counts and means were estimated on the full data set using the R package “sqlsurvey”³⁰. Ratio estimates and differences were calculated using simulations based on survey-adjusted counts and standard errors with each simulation consisting of 1,000 random normal draws. Differences in trends over time were evaluated by comparing the slopes of linear regression equations with year as the predictor variable.

We assessed the interaction between injury and any alcohol or substance use two ways. First, we assessed what we considered to be a more clinically relevant interaction of injury and substance use on the risk of pre-admission emergency department fatality by testing on an additive scale using incidence contrasts as proposed by Darroch³¹ and Rothman and Greenland ³² within the framework of component causes, where

• R_injury. _substance. _unknown - the risk of fatality when traumatic injury and substance use are both present

• R_injury. _unknown - the risk of fatality when injury but not substance use is present

• R_substance. _unknown - the risk of fatality when substance use but not injury is present

• R_unknown - the ”background” risk of fatality in the absence of either injury or substance use

To determine if the observed R_injury. _substance. _unknown exceeds what we might expect if the two risks did not interact, we subtract out R_injury. _unknown and R_substance. _unknown and then add back R_unknown which we subtracted twice, setting up an equality to test the independence of causes:

$$R_{injury\cdot substance\cdot unknown}=R_{injury\cdot unknown}+R_{substance\cdot unknown}-R_{unknown}$$

Any excess risk beyond these inequalities is considered due to interaction. We assessed interaction on an unadjusted additive scale, in terms of absolute risk differences:

$$\left(R_{injury\cdot substance}-R_{unknown}\right)=\left(R_{injury}-R_{unknown}\right)+\left(R_{substance}-R_{unknown}\right)$$

We assessed the significance of the result by running 1000 simulations taking into account survey variation using point estimates and standard errors returned by survey procedures.

Next, as a point of comparison, we tested the direction and statistical significance of an interaction term in a logistic regression equation with pre-admission death as the outcome and variables to control for confounding by age and gender:

$$death={\beta }_{intercept}+{\beta }_{age}+{\beta }_{gender}+{\beta }_{injury}+{\beta }_{substance}+{\beta }_{injury\cdot substance}$$

Where, “age” is measured continuously in years, “gender” is an indicator variable for male (0) vs. female (1), “injury” is an indicator variable for the presence (1) or absence (0) of a primary diagnostic code indicating traumatic injury, “substance” is an indicator variable for the presence (1) or absence (0) of a secondary diagnostic code indicating either alcohol or drug as defined above. The statistical interaction term, injury. substance indicates the difference in the slope for the association of injury and death in the presence or absence of alcohol or drugs.

The study protocol was approved as exempt by the New York University School of Medicine Institutional Review Board. A full set of notes with code to reproduce the analyses is available at http://www.injuryepi.org/resources/Misc/etohSubstanceInjuryNotesOnline.pdf.

Results

Descriptive Epidemiology

The study data consisted of 890,613,941 (s.e. = 9,332) survey-adjusted (ED) discharges for the 7-year period of 2006 to 2012. There were a total of 181,194,431 (s.e. = 9344) discharges with a primary ICD-9 diagnostic code related to traumatic injury, accounting for 20.3% (s.e. = 0.001) of all discharges. A total of 3,499,134 (9,328) or 1.9% (s.e. = 0.005) of these traumatic injury discharges had a secondary diagnostic code for alcohol or substance use as defined for the study. The majority of these, 2,791,471 (s.e. = 8,362), were alcohol-related. While overall ED population-based injury discharges rates declined during the study period, discharge rates for alcohol and drug-related injuries increased (Figure 1). The mean population-based rate of all ED injury discharges decreased 6.5% from the first 3 years of the study period to the final 3 years, while alcohol and substance-related injury discharges increased 9.8% during the same comparison period.

Figure 1:

Total, Alcohol or Drug, Alcohol Only and Drug Only US Population-Based Emergency Department Discharges for Traumatic Injury with Regression Line for Effect of Year on Rate. United States Hospitals, 2006-2012.

The mean age for a person with a primary injury diagnosis and any secondary alcohol or drug diagnosis was 41.0 (s.e. = 0.05) years. The average age for injured individuals with secondary alcohol diagnoses was 41.8 (s.e. = 0.05) years, compared to 37.4 (s.e. = 0.08) for injured individuals with secondary drug diagnoses. Slightly less than half (46.3% ,s.e. = 0.02) of all persons with primary injury diagnoses were female, compared to only 26.6% (s.e. = 0.13) of persons discharged with a traumatic injury and a secondary alcohol or drug diagnosis.

The most common external cause of injury codes for discharges with a secondary alcohol or drug diagnosis involved falls (747,483 discharges, s.e. = 4335) or an unarmed fight or brawl (256,853 discharges, se = 2548), which respectively accounted for 21.4% (s.e. = 0.14) and 7.3% (s.e. = 0.08) of alcohol or drug related injury discharges. This pattern held for those injury diagnoses with a secondary diagnosis limited to alcohol where there were 648,225 discharges (s.e. = 4041) due to falls or 23.2% (s.e. = 0.16) and 219,296 discharges (s.e. = 2357) or 7.9% (s.e. = 0.08) due to unarmed fights. A similar pattern held for injuries with a secondary diagnosis limited to drugs with the addition of motor vehicle crashes as an additional important cause, accounting for 36,047 discharges (s.e. = 967) or 3.8% (s.e. = 0.10) of all such injuries.

A total of 3,880,647 (s.e. = 9846) or 2.1% (s.e. = 0.01) of discharges with a primary injury diagnosis were classified as “severe” or likely to result in death. Of these, 476,582 (s.e. = 3539) or 12.3% (s.e. = 0.10) had a secondary alcohol or drug diagnosis. There were a total of 1,344,950 (s.e.= 2581) pre-admission fatalities during the study period of which 73,655 (s.e. = 1396) or 5.5% (s.e. = 0.10) had a primary diagnosis of traumatic injury. Of pre-admission traumatic injury deaths, 999 (s.e. = 77) or 1.4% (s.e. = 0.11) had a secondary alcohol or drug diagnosis.

The total cost over the study period of ED care for discharges with a primary injury diagnoses was $99.70 (s.e. = 0.06) billion. Traumatic injuries with a secondary diagnosis of alcohol or drugs accounted for $4.14 (s.e. = 0.01) billion or 4.2% (s.e. = 0.01) of the total cost of emergency department injury care in the US. The total cost of alcohol-related injuries increased 67.2% from the first three years of the study period compared to the final three years. In a similar comparison, the total cost of drug-related injuries increased 63.5% (Figure 2).

Figure 2:

Total Yearly Cost of Emergency Department Discharges for Primary Injury Diagnoses with Secondary Alcohol or Drug Diagnoses in Millions of Dollars. US Hospitals 2006 - 2011.

Interaction Between Injury and Alcohol or Drugs

Pre-admission fatality rates per 100,000 population as outlined in Table 2 were entered into an equation to test for inequality indicative of additive interaction for the presence of injury and alcohol or substance. The resulting calculation, which is positive on the left side of the equation and negative on the right side, indicated that a positive interaction between traumatic injury and AOD was also present on an additive scale: (28.6 - 184.5) > (40.9 - 184.5) + (60.0 - 184.5). A simulation comparing the left and right sides of the equations accounting for survey error returned near certainty of the right side of the equation being greater than the left. (p<0.0001)

Table 2:

Contingency Table of Pre-Admission Fatality Rates per 100,000 population for Test of Additive Interaction Between Primary Injury Diagnosis and Secondary Alcohol or Drug Diagnosis. US Emergency Department Discharges 2006-2012.

	Alcohol/Drug	No Alcohol/Drug
Injury	28.6	40.9
No Injury	60.0	184.5

In a logistic regression on the risk of pre-admission ED fatality, controlling for age and gender, both a primary diagnostic code for traumatic injury and a secondary diagnostic code for alcohol or other drug were associated with a decreased risk of death. An interaction term for traumatic injury and AOD was associated with an increased risk of death (OR = 1.76, 95% CI = 1.53, 2.03; Table 3).

Table 3:

Logistic Regression Effect of Primary Injury Diagnosis and Secondary Alcohol or Drug Diagnosis on the Risk of Pre-admission Emergency Department Fatality, Controlling for Age and Gender. US Emergency Department Discharges 2006-2012.

Variable	Odds Ratio (95% CI)
Age	1.04 (1.04, 1.04)
Female	0.64 (0.64, 0.65)
Drug or Alcohol	0.35 (0.34, 0.36)
Injury	0.25 (0.25, 0.25)
Drug or Alcohol*Injury	1.76 (1.53, 2.03)

Discussion

Cognitive and perceptual impairments, poor judgement, and riskier behavior increase injury risk. Contrary to reports of a protective clinical benefit from AOD in the setting of trauma, our results support the conclusion that AOD increases risk of morbidity and mortality across the injury spectrum. Impairment and intoxication complicate the assessment and evaluation of a trauma patient during the immediate post-injury phase, while effects of chronic substance use (e.g. alcohol withdrawal) may complicate medical care, delay recovery, and worsen outcomes.

Our results also demonstrate the ongoing, and in some cases increasing, contribution of alcohol and drugs to trauma-related morbidity and mortality in the US. emergency department discharges of trauma that include secondary diagnoses of alcohol or drugs increased nearly 10% from the beginning of the study period to the end, despite a decline of 6.5% in overall ED discharges for traumatic injury. In addition to the clinical and public health implications, alcohol and drugs contribute to the economic burden of trauma care in the US. While traumatic injuries with a secondary alcohol or drug diagnosis represented 1.9% of all traumatic injury discharges, they accounted for 4.2% of the costs of injury care—a total of $4.1 billion.

While our study supports recent findings by others including Sethi et al., Majid et al., and Jurkovick et al. in confirming the association between alcohol and worse clinical outcomes in trauma, recently publications such as those by Opreanu et al. and Liou et al. have helped perpetuate the notion of alcohol attributing to a protective effect in trauma.^{11,12,16,17,18} The precise reason for this protective effect is not clearly understood and may be analogous to other paradoxical associations in the epidemiological literature, such as lower infant mortality among low birthweight infants born to mothers who smoke cigarettes; smoking itself does not protect low birthweight children from death, but a non-smoking mother is more likely to have a condition with a stronger association with fatality when compared to smoking.³³ Likewise, alcohol does not protect injured individuals from mortality. But, among the population of hospitalized patients, people are more likley to die from other conditions with a stronger association with fatality, e.g. stroke, heart failure, myocardial infarction, sepsis. In our data, the mortality rate for individuals without diagnostic codes for injury or AOD were 3 to 6 times higher than those with any combination of injury and AOD. This highly uneven distribution of mortality rates indicates severe selection effects and confounding by indication. Statistical analyses for interaction assume a baseline probability of the outcome of interest that is similar across groups, and this is not the case in these data. Additional selection bias comes into play if AOD-impaired individuals are more likley to die at the scene of the injury and never even arrive to the ED or hospital.

Limitations

Our study is subject to a number of important limitations. In our data, only 2% of traumatic injury discharges had secondary AOD diagnoses, while other published reports suggest this number may be as high as 47%.^33,34,35 We note that our estimates mirror those reported in studies based on measurement of very high alcohol levels (> 300 mg/dl). We also note that an ICD-9 AOD diagnosis is more likely to be based on toxicological reports and that toxicology is more likely to be ordered in the setting of the most severe trauma and in fatalities. Studies based on toxicology results have their own biases. Because drugs like cannabis and cocaine have longer half-lives, screening results are more likely to be positive, increasing their prevalence. Alcohol, by contrast, declines at a fairly rapid rate of approximately 15 mg/dl per hour (25mg/dl/h for some individuals), and missing data levels can be as high as 71%.¹⁵ In this respect, our results are very much the tip of the iceberg for the effect of AOD on trauma mortality.

Even sizeable datasets can be biased.³⁶ While our study is based on the largest, most representative source of ED visits in the US, it remains observational in nature, and our results may be affected by ascertainment or selection bias. It may be that intoxicated or impaired individuals are more likely to die at the scene of the trauma or, for some other reason, not to present to emergency departments – the database only captures ED discharges and thus misses those other cases. Including medical examiner data could help address this, but it was beyond the scope of our data sources. To help address this issue, and to differentiate our efforts from previous studies based on registries of admitted patients, we chose to focus on pre-admission fatalities.

We used the order of the diagnostic codes as entered into the administrative data set to define a primary discharge diagnosis of traumatic injury in which AOD was a secondary or contributing factor. But traumatic injury may more properly be seen as a step in a causal chain that begins with AOD. The very fact of adjusting for injury can itself introduce bias.¹⁷ While an intoxicated person who then trips and falls would likely receive a primary trauma diagnosis, it is alcohol that precipitated the injury.

Lastly, we did not stratify by region or urban versus rural as this is the topic of ongoing investigation but would nonetheless help inform interventions. We note that neither race nor ethnicity variables are available using NEDS data and were thus not included in the analysis.

Future Directions

Looking at external causes of injury can also help explain discrepancy across studies. Additional analyses should look at blunt vs. penetrating injuries for AOD vs non-AOD use. It may be that severely AOD-impaired individuals are less likely to have the kinds of penetrating injuries, like gunshot wounds, that are more likely to result in immediate mortality. One study of this issue indicated that there were fewer penetrating injuries (9% vs 22%), and lower proportion of Injury Severity Score (ISS) > 16 (14% vs 19%) in individuals with high (>300 mg/dl) vs low (<100 mg/dl) alcohol levels.³⁷ In a trauma registry-based study, moderate alcohol levels were associated with an increased risk of death from penetrating injury, while high alcohol levels were associated with an increased risk of blunt-trauma related mortality.¹⁷ Parsing out these differences would further inform our current understanding.

Conclusions

During our study period, alcohol and substance-related emergency department injury discharges increased 9.8%. There was a statistically significant interaction between traumatic injury death and alcohol and other drugs. Further research should clarify the effect of alcohol versus other drugs in trauma settings, as well as investigate specific injury subtypes and their causes. Additionally, the role of AOD in injury mortality would be further elucidated by including pre-hospital deaths and comparing them to pre-admission and post-admission deaths, basing results on the uniform or random assessment of AOD use, and attempting propensity score matched or case-control studies.

Highlights.

• Alcohol and substance-related emergency department visits increased during a recent 7-year study period

• Alcohol and substance abuse does not offer a protective benefit in the setting of trauma, as has previously been suggested

• Alcohol and substance abuse may increase the risk of death related to traumatic injury