Hundreds of studies document the frequent involvement of alcohol in fatal and nonfatal injury (National Institute on Alcohol Abuse and Alcoholism, 1997; Lipsey et al., 1996; Roizen, 1988; Smith et al., 1999a; Smith et al., 1999b). These studies show that people often are injured while drinking and that heavy drinkers have above-average injury risks. Several studies document injury victims’ self-reports that alcohol consumption caused injury (Cherpitel et al., 2006; Sommers et al., 2000; Stephens, 1987). Consistent with those reports, laboratory studies show that alcohol impairs coordination and the ability to perceive and respond to hazards (Normand et al., 1994) and that hangover impairs neurocognotive performance and pyschomotor vigilance (Howland et al., 2010; Prat et al., 2008; Rohsenow et al., 2010). It is difficult, however, to assess the impact on injury risk in the uncontrolled setting of real life (Hingson et al., 2001; Morrow et al., 1991; Tornros and Laurell, 1991; Wolkenberg et al., 1975; Yesavage and Leirer, 1986).

Early epidemiologic studies estimated how likely a group of people classified by drinking behavior were to suffer injuries relative to a base group. These studies demonstrated linkage only weakly (Heien and Pittman, 1989; Lipsey et al., 1996). They cannot tell us whether the excess injury risk associated with alcohol results from alcohol impairment or from unmeasured factors that differentiate the groups being compared. One factor is a lifestyle that includes risk-taking and heavy drinking. These studies established correlation, not causation. They included injuries that involved alcohol but would have occurred even if the victim were alcohol negative. For example, a drunk driver rear-ended by an alcohol-negative driver while stopped at a red traffic light suffers an alcohol-involved injury unrelated to alcohol. So does a sleeping drunk whose home collapses in an earthquake.

Epidemiologists estimate attributable risk ratios to reduce the effects of such extraneous events (Rothman and Greenland, 2005). Except for impaired driving (Borkenstein et al., 1974; Lloyd, 1992), only recently have studies shown injury risk rises while drinking. Some studies use case-crossover designs. These studies ask injured people in emergency departments (EDs) about their drinking a week prior to and at the time of their injury (Borges et al., 2004a, 2006, 2008; Pledger et al., 2007; Vinson et al., 1995, 2003). An overlapping group of studies use case-control designs with people visiting the ED for reasons other than injury serving as a comparison group (Cherpitel et al., 2005; Kuendig et al., 2008). A third group are case-control studies with a community comparison group (Borges et al., 1998; McLeod et al., 1999; Stockwell et al., 2002; Vinson et al., 2003; Watt et al., 2004). Some studies use alcohol tests in the ED to replace or supplement self-report. Here we introduce a less accurate but also far less expensive approach; we construct the comparison group from existing consumption surveys. The method is especially useful in analyzing attribution for severe injuries that result in death or hospital admission because interviewing the severely injured can be impractical (e.g., those with brain injury, quadriplegia, or a memory gap around the injury incident). It also yields insight into attribution of criminal acts where self-report by apprehended criminals and impressions of victims describes alcohol involvement but comparison with other time periods or with contemporaneous alcohol use by similar people can be difficult.

Some studies assess if injury risk is higher for those who drink or drink heavily. They confound injuries caused by alcohol with injuries caused by lifestyle and risk-taking differences between groups whose drinking patterns differ. Studying whether alcohol causes injury requires differentiating injuries during alcohol-positive and alcohol-negative hours.

This study probes the roles of alcohol impairment and lifestyle in the risk of injury. We test three hypotheses: (1) the hospitalized injury risk of both regular heavy drinkers and other drinkers rises when they are alcohol positive (i.e., alcohol appears to cause injury); (2) regular heavy drinkers have higher hospitalized injury risks than other people when alcohol negative (i.e., heavy drinker lifestyle is riskier and/or hangover is risky); and (3) being alcohol positive raises the hospitalized injury risk of people who are not heavy drinkers more than it raises the risk of heavy drinkers (i.e., regular heavy drinkers tolerate alcohol better than most drinkers).

The innovative seed for our analysis is simple; dividing existing data on alcohol consumption by data on alcohol metabolism rates yields less costly exposure estimates than breath testing. Once exposure is known, estimating the risk ratios simply requires writing some standard epidemiologic formulas related to relative risk (Szklo and Nieto, 2000). A literature search and some secondary data analyses yield the data for the ratios. Annual hours alcohol positive versus negative are used as the exposure measures. Risk attributable to alcohol is measured by estimated relative risk when alcohol positive versus negative.

The equations used to meet the objectives appear in Table 1. They are developed below. Table 2 lists the best estimates and sensitivity analysis values for the equations. We adjusted the data to achieve better consistency in age range and definition of regular heavy drinkers. Sensitivity analysis assessed how value choices influenced the results. We used wide ranges when available data were weak or inconsistent or values varied over time.

An estimated 27% of hospitalized injury victims are alcohol positive (Table 3); this includes nearly half of hospitalized pedestrian and near-drowning injury victims. The relative risk of hospitalized injury when alcohol positive versus negative is especially high for assault, pedestrian, and near-drowning incidents. An estimated 21% of hospitalized injuries are attributable to alcohol (from equation 2, 27% * 3.5/4.5; sensitivity analysis range 14% to 29%). Multiplying the 21% rate times the 1998 injury count from the National Hospital Discharge Survey suggests that almost 540,000 people were hospitalized for injuries caused by alcohol.

Table 4 presents relative risk estimates for hospitalized injury by drinker type. All estimates use other drinkers’ hospitalized injury rate when alcohol negative as the comparison rate. Heavy drinkers are an estimated 3.2 times (4.33/1.36) more likely to have a hospitalized injury when alcohol positive than negative. Other drinkers are at even greater risk. When alcohol positive, they are 6.8 times more likely to have a hospitalized injury than when negative. Notably, during alcohol-negative hours, the risk of hospitalized injury among heavy drinkers is an estimated 1.4 times greater than the risk among other alcohol-negative people.

The pattern is similar for hospitalized burn and spinal cord injuries. The estimated risk among heavy drinkers is 3.2 times greater when alcohol-positive than negative. During alcohol-negative hours, heavy drinkers are an estimated 1.6 times more likely to suffer a hospitalized spinal cord injury compared to other alcohol-negative people. The elevated risk of hospitalized spinal cord injury among other drinkers is striking: the estimated risk is 13.6 times higher when alcohol positive than negative.

In sensitivity analysis (Table 5) we computed 578 sets of relative risk estimates using all combinations of alternate values for the variables in Equations 1 and 3 and the injury rate calculation for heavy drinkers versus other people and considered two boundary conditions. The conclusions of the sensitivity analysis are quite strong. Relative to alcohol-negative people who do not regularly drink heavily, heavy drinkers have an elevated risk of hospitalized injury when alcohol negative (true in all but 2.2% of the 578 sets). The injury risk of heavy drinkers rises when alcohol positive. Relative to their risk when alcohol negative, non-heavy drinkers face greater injury risks per alcohol-positive hour than heavy drinkers (true in all but 3.5% of the 578 sets).

Table 6 estimates all-hours relative risk of hospitalized injury by drinker class, regardless of whether alcohol positive or negative. These relative risk estimates were computed with data from Miller, Lestina, and Smith (2001). They control for differences in age, sex, white-versus-blue collar employment, and region of residence between regular heavy drinkers and other adults. The estimated all-hours injury risk of moderate drinkers relative to abstainers is modestly elevated.

The three hypothesized patterns emerge. First, both heavy and moderate drinkers appear to have elevated risks of hospitalized injury when alcohol positive. Prior studies agree that even moderate drinking significantly increased injury risk (Cherpitel, 1999). Second, regular heavy drinkers appear to have higher hospitalized injury risks even when alcohol negative. Third, alcohol appears to raise injury risks of regular heavy drinkers less than that of other drinkers.

This paper’s estimates suggest that alcohol causes serious injury. By comparing injury risks of relatively homogeneous groups of drinkers during alcohol-negative hours relative to alcohol-positive hours, we control for lifestyle and demographic differences between heavy drinkers and light to moderate drinkers. Alcohol-impaired psychomotor skills pose a risk. Laboratory studies find that alcohol impairs reaction time, ability to divide attention, motor skills, coordination, memory, and decision-making (Howland et al., 2010; Normand et al., 1994). Whether the impairment results in injury in real life depends on factors such as the dose and timing of alcohol intake, the drinker’s ability to perform under the influence (functional tolerance), and the task being performed.

The elevated injury risk estimate for heavy drinkers when alcohol negative may reflect a risk-taking lifestyle. Cherpitel (1999) examined the relationship between substance use, risk-taking, and injury among a nationally representative sample of Americans. She developed a scale for risk taking disposition. Both substance users and risk-takers were more likely to be injured. However, only risk-taking was a significant predictor of injury after controlling for substance use.

The association of drinking habits with injury also may be due to low residual blood alcohol levels or to hangover (Howland et al., 2001; Morrow et al., 1991; Wolkenberg et al., 1975; Yesavage and Leirer, 1986). Declining health, or other lingering effects of regular heavy drinking, also may elevate the risk of injury when alcohol negative.

As well as estimating overall injury risk, this article estimates risks of hospital-admitted or fatal spinal cord and burn injuries. We chose these risks because the incidents are serious, they are readily identified from diagnosis codes in medical records, and their rate of alcohol involvement has been well documented in an entire state. Our burn estimates, however, have not untangled possible interactions between alcohol-related and smoking-related risks.

Our estimates that heavy drinkers experience a smaller injury risk than other drinkers when alcohol positive but have higher background risks also are consistent with the literature (Cherpitel et al., 2004; Borges et al., 2006; Spurling and Vinson, 2005). Alcohol’s smaller estimated impact on hospitalized injury risk for heavy drinkers compared to other drinkers also supports previous findings that regular drinkers develop tolerance to the impairing effects of alcohol (Chesher and Greeley, 1992). Their brain functions adapt to the physical disruption caused by the drug in both behavior and physical function. Consequently, the causative link between alcohol and hospital-admitted injury seems to be stronger for people who are not heavy drinkers.

We estimate 24.2% of hospitalizations for motor vehicle injury are alcohol-attributable. Alcohol involvement in crashes rises with crash severity. In 1994, 38.4% of drivers in fatal crashes were alcohol-positive. Applying the formula used in Table 3, the estimated attributable fraction is 33.3%. By comparison, applying the odds ratios from the classic Borkenstein (1974) data to the 1994 driver blood-alcohol distribution, the attributable fraction is 31.2% (13390 drivers at .08 and above * 0.895 + 2236 drivers at .04-.079 * 0.326). So our formula closely approximates the more exact measurement. Our hospitalization estimate also appropriately lies between narrower estimates from Miller et al. (1998) that 26.7% of fatal injuries and 9.5% of nonfatal injuries in U.S. crashes are alcohol attributable injuries to occupants of drinking drivers’ vehicles.

We estimate that the relative risk of hospital-admitted injury after drinking in the United States is 4.5. A pooled case-crossover analysis that incorporates data from 28 ED clusters in 16 countries (including 6 U.S. clusters) finds a relative risk of ED-treated injury of 5.2 before statistical adjustment for covariates and 5.7 after adjustment (Borges et al., 2006). Additional large U.S. case-crossover studies in a university town in Missouri find odds ratios of 4.0 to 5.6 for any alcohol in the six hours prior to injury (computed using the discordant pairs method) (Vinson et al., 1995, 2003). A pooled relative risk estimate for all 8 U.S. studies is 3.8 and for all 30 studies is 4.8. Across the six U.S. clusters in Borges et al. (2006), 21.3% of ED-treated injuries were alcohol-involved and the pooled relative risk estimate is 3.8. The formula used in Table 3 yields a relative risk ratio of 3.3 for ED-treated injury. Thus our estimates are reasonably consistent with but possibly a bit below the case-crossover estimates..

Estimates from case-control methods are less consistent with our estimates for admitted patients. The studies with a community control group arrive at odds ratios of 6.7 for Mexico and 2.1-3.4 for Australia and the U.S. with a pooled odds ratio of 2.8 (Borges et al., 1998; McLeod et al., 1999; Stockwell et al., 2002; Vinson et al., 2003; Watt et al., 2004). Our rough pooling of published data on ED-comparison groups from 19 ED clusters in 8 countries yields odds ratios of 1.6-1.9 for all injury (1.2-1.6 for 6 U.S. studies) depending on whether one relies on blood alcohol level (BAL) test results or on self-report (Cherpitel et al., 2005; Kuendig et al., 2008). In these data, 10.5%-11% of US injury patients were alcohol-positive. Our formula yields a relative risk ratio of 1.4-1.5, which matches the case-control estimates. This method assumes that all patients going to the ED due to acute or chronic illness were well enough to drink normally in the six hours prior to arriving at the ED, which is questionable and may bias the risk ratio. That may explain why the ratios with non-injury ED controls are lower than those from other methods.

Our estimated relative risk ratio for interpersonal violence is 8.4, meaning 88% of assaults were caused by alcohol if the victims were alcohol positive. By comparison, a simple case-weighted average of relative risk ratios from 13 studies with ED controls yielded a relative risk ratio of 7.4 (Cherpitel et al., 2005). The ratio was 5.6 with a range from 4.1 to 10.4 for the subset of five U.S. studies (four of them in California). A Swiss study with ED controls yielded an odds ratio of 9.2 (Kuendig et al., 2008). A case-crossover study in three Latin American countries yielded a matched pair odds ratio of 15.0 (Borges et al., 2008). A U.S. study yielded a community-case control odds ratio of 10 and a case-crossover odds ratio of 34 (Vinson et al., 2003). Thus our estimate is consistent with most others. We attribute an estimated 36.1% of hospitalized assaults to victim alcohol use, with 41% of victims alcohol positive. In the U.S., perpetrators report being alcohol positive in 41% of assaults (Miller et al., 2006). Conventional wisdom has been to conservatively assume half of those were attributable to alcohol. We recommend raising the attributable fraction to 36%, at least for medically attended assaults. That estimate is conservative since surely some alcohol-attributable incidents involve an alcohol positive perpetrator and an alcohol negative victim. Indeed, with 24.4% to 44.3% of apprehended offenders qualifying as abusing or dependent on alcohol (Compton et al., 2010), our formulas suggest 30.3%-34.6% of violent crime and 9.0%-14.7% of property crime is partially attributable to perpetrator alcohol use (e.g., H=(.244 * 31.156% of heavy drinker’s hours alcohol positive + (1-.244) * 2.845% of other people’s hours alcohol positive) = .0975. I = 0.23 for property crime from Miller et al. (2006). RR= (.23/.0975)/(.77/.9025) = 2.764. AF = .23 * 1.764/2.764 = 14.7%). Collins and Schlenger (1988) estimated that the relative risk that a crime was violent given alcohol use when it was committed was 1.73, somewhat below our estimate of 2.33 (ratio of relative risks = 6.430/2.764).

For suicide acts, our relative risk estimate is 5.1 for hospital-admitted nonfatal cases ages 18 and over (4.5 for those ages 14 and over). It would have been 6.6 had we used the 35.3% alcohol involvement rate reported for the pooled 102 suicide attempters in 7 North American and Australian ED studies (Borges et al., 2004b). Based on a case-crossover design, the relative risk reported for those studies was higher, 9.6, with a 95% confidence interval of 5.7-16.3. Powell et al. (2001) found a 26.5% alcohol involvement rate among 153 near-lethal suicide acts (76.5% of ED-treated acts) in Houston TX. The crude odds ratio was 7.0 relative to community controls. That ratio rose to 8.0 after statistical adjustment for sample differences.

This paper shares the limitations of the studies that supplied data for the computations. The alcohol involvement rate for hospital-admitted heavy drinkers is for injury victims treated at a single trauma center in Maryland that only sees seriously injured patients (Soderstrom et al., 1997). The injury risks for heavy drinkers relative to controls may be underestimates because heavy drinkers who cannot be identified from a 3-year medical record are misclassified into the lower-risk control group (Miller et al., 2001). Also, this paper assumes the injury profile of heavy drinkers diagnosed with alcohol problems applies to all regular heavy drinkers.

The attribution estimates by injury type require additional assumptions, making them more speculative. Furthermore, the Oklahoma data are restricted to cases exceeding a severity threshold (Levy et al., 2002). Less severe injuries, notably burns not triaged to burn centers may have different causal chains. Still, these estimates are consistent with the all-hours injury estimates. They suggest that, controlling for lifestyle effects, alcohol raises hospitalized injury risk for these subsets.

The relative risks estimated with Equation 1 are spread across all hours in a drinking status, including the minimal risk during hours when someone is sleeping off the alcohol they consumed or at very low positive blood alcohol concentrations (BACs). The alcohol-negative injury risk estimate also includes sleep hours and most working hours. Alcohol impairment is not uniform across those hours; it rises more than linearly with alcohol level (e.g., Kuendig et al., 2008). Moreover, drinking pace affects peak BAC. Drinking sessions often last 4-5 hours (Beirness et al., 2004; Lange and Voas, 2001; Thombs et al., 2003). Consequently, BACs of half of people consuming 5 drinks or more were at .06 or lower when they stopped drinking. Similarly, Cranford et al. (2006) finds that 10% of those who drink 5 or more drinks in a drinking session drink slowly enough that their BACs remain moderate. Data were not available to adjust Equation 1 to reflect impairment variations due to changing alcohol levels. Similarly, exposure to injury hazards is not uniform across those hours. It is lowest when sleeping and relatively low when working.

Except for heavy drinkers, this paper assumes that hospital-admitted injury risks while alcohol negative do not differ systematically with a person’s drinking habits. That assumption is weak. For example, some people who are not classified as regular heavy drinkers may drink heavily occasionally. Those episodic drinkers may be risk-takers generally. Hence the relative injury risks for heavy drinkers when alcohol positive versus alcohol negative are the more perfect fingerprints. Controlling for lifestyle by focusing on a homogeneous group of drinkers, the findings demonstrate that average injury risks are elevated during alcohol-positive waking and sleeping hours.

A further limitation is the cobbled nature of our estimates. They are sewn from data from different locations and time periods. To the extent that alcohol consumption in the states represented in the data about alcohol involvement in injury differs from national average consumption, the estimates will be skewed.

These limitations are offset by the robustness of the results in extensive sensitivity analysis on the imperfect parameter estimates. Estimates from the sensitivity analyses are consistent with the main estimates. The estimated size of the relative risks shifts, but not their direction.

Our attribution rate estimates consider only victim alcohol involvement. Some alcohol-negative victims are injured, however, because someone else was drinking (Levy et al., 2002). For example, crashes attributable to alcohol in the U.S. injure an estimated 257,000 alcohol-negative pedestrians, drivers, and passengers per year (Miller et al., 1998). These victims account for 7.6% of deaths, 3.9% of nonfatal injuries, and 25,000 hospitalized injuries in crashes. Adding them would raise the alcohol-attributable percentage of hospital-admitted injuries from 21% to 22%.

To refine and validate our estimates, future research needs to measure blood alcohol levels of a more representative sample of hospitalized injury victims and assess their drinking behavior. Segmenting the injury rates by alcohol involvement for those who occasionally drink heavily versus those who always drink moderately seems critical. Broadly representative data are needed on alcohol involvement in the injuries of heavy and dependent drinkers. Breaking out risk by injury cause and drinker demographics, especially sex and race, is important. Most Asians lack secondary alcohol metabolism channels. Research on their relative risks when alcohol positive might be especially accurate and informative because the calculations would be free from metabolism-related uncertainty. Another priority is an analysis for youth accounting for their lower total body water and accelerated rate of ethanol elimination (Donovan, 2009).

Our estimates have strong policy implications. They set the stage for injury warning labels on alcohol bottles. They also suggest what percentage of public injury cost justifiably could be recovered through alcohol taxes.

Moderate drinking has not traditionally been considered hazardous. From an injury viewpoint, it appears to be more hazardous per drink than regular heavy drinking. Heavy drinkers are alcohol positive for 3 to 6 times as many hours as other drinkers. Thus, they suffer more alcohol-positive injuries. Other drinkers, however, suffer more injuries per hour alcohol positive. Intervention rarely has been targeted to this group, because its high risk was hidden. Revealing that risk sets a difficult public debate into motion. Estimates that moderate drinking with an occasional binge greatly raises hospitalized injury risks raise doubts about U.S. nutritional guidelines that tout moderate wine drinking (U.S. Department of Agriculture, 1990). Given the historical failure of prohibition, harm reduction strategies seem the logical course of action.