INJURY PREVENTION/ORIGINAL CONTRIBUTION

Eatal Nontraffic Injuries Involving Alcohol: A Metaanalysis

From the Johns Hopkins School of Hy~ene and Public Health Center for Injury Research and Policy, Baltimore,* and the National Public Services Research Institute, Landover, ~ MD.

Received for publication February 18, 1998. Revision received September 28, 1998, Accepted for publication November 16, 1998.

Presented at the 125th Annual Meeting of The American Public Health Association, Indianapolis, IN, November 1997.

Supported by grant no. AA09812 from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) and in part by NIAAA grant no. AA07700; additional support was provided to Dr Smith by the University of Auckland Injury Prevention Research Centre, which is jointly funded by the Health Research Council of New Zealand and the Accident Rehabilitation and Compensation Insurance CorporaFton.

Address far reprints: Gordon S Smith, MB, ChB, MPH, Center for Injury Research and Policy, Johns Hopkins School of Hy~ene and Public Health, 624 North Broadway, Baltimore, MD 21205; 410-955-7980,fax 410-614-2797; E-mail gsmith@jhsph.edu.

Copyright �9 1999 by the American College of Emergency Physicians.

0196-0644/99/$8.00 + 0 47/1/97292

Gordon S Smith, MB, ChB, MPH* Charles C Brenas, Ph9* Ted R Miller, PhD*

See editorial, p. 699.

Study objective: Estimates of alcohol involvement in fatal injuries vary widely. For injuries ether than those involving motor vehicles, no national data exist and the quality of regional data is limited. This study synthesizes US medical examiner studies of nontraffic fatalities for the purpose of esti- mating alcohol involvement by injury mechanism and intent.

Methods: We reviewed 331 medical examiner studies pub- lished between 1975 and 1995 that reported nontraffic injury fatalities. These studies were identified from computerized bib- liographic databases and by hand searching of reference lists from 26 review publications and the subject indices of 7 promi- nent journals. A final total of 65 articles (19.6%) met our inclu- sion criteria for analysis. National data on motor vehicle fatalities were used for comparison.

Results: A total of 7,459 unintentional injury deaths, 28,696 homicide cases, and 19,347 suicide cases were aggregated. The aggregate percentage tested for blood alcohol concentra- tion (BAC) was highest among homicide cases (88.2%), fol- lowed by unintentional injury deaths (84.0%) and suicide cases (81.7%). The aggregate percentage determined to be intoxi- cated (BAC, _100 mg/dL) was highest among homicide cases (31.5%), followed by unintentional injury deaths (31.0%) and suicide cases (22.7%). Mean and median comparisons pro- duced comparable findings. Fewer than one quarter of the 65 articles reported gender- and age-specific rates. Inconsistent reporting of gender and age-specific rates in fatal nontraffic injuries prevented their analysis by cause of death. Fatally injured motor vehicle drivers, however, demonstrated distinct variation in alcohol involvement by age and gender.

Conclusion: This metaanalysis is the first systematic attempt to estimate alcohol involvement in fatal nontraffic injuries at the national level. It demonstrates that alcohol is an important

JUNE 1999 33:6 ANNALS OF EMERGENCY MEDICINE 659

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Branas & Miller

factor in many fatal injuries and that its importance varies by cause of injuw.

[Smith GS, Branas CC, Miller TR: Fatal nontraffic injuries involving alcohol: A metaanalysis. Ann Emerg MedJune 1999;33:659-668.]

INTRODUCTION

Although detailed national data on alcohol involvement in motor vehicle fatalities are assembled annually, there are no comparable data for nontraffic injuries. 1 Consequently, national estimates of alcohol involvement in nontraffic injuries from the early 1980s continue to be widely cited even though they lack methodologic documentation and data on blood alcohol concentration (BAC) for specific injuries. 2,3

Medical examiners and coroners provide valuable information on the manner and cause of injury deaths and often include test results for alcohol and other drugs. 4,5 In the absence of a national database, regional medical examiner studies provide the only population-based data on BAC and nontraffic injury fatalities. However, no sys- tematic synthesis of medical examiner reports for the pur- pose of investigating nontraffic fatalities and BAC has been completed. This study is a metaanalysis of fatal non- traffic injuries involving alcohol in the United States that seeks to provide statistically robust estimates of alcohol involvement. 6 Such information is essential for formulat- ing policies to prevent alcohol-related injuries.

MATER ALS AND METHODS

We reviewed the English language literature for medical examiner studies of non'traffic injury fatalities. A total of 331 articles published over a 20-year period (1975 to 1995) were obtained from computer-based bibliographic databases and by hand searching of reference lists from 26 review publications and the subject indices of 7 promi- nentj ournals. Eleven of the review publications were obtained from the computer-based search, and the remaining 15 were already in our personal files as refer- eed publications, books, conference proceedings, or gov- ernment documents. The subject indices of the following journals were selected for hand searching as a result of their prominence in the computer-based search: American Journal of Drug and Alcohol Abuse, American Journal of Forensic Medicine and Pathology, American Journal of Public Health, Forensic Science International,

Journal of Forensic Sciences,Journal of Studies on Alcohol, and Morbidity and Mortality Weekly Reports. Other fugitive literature sources, such as dissertation abstracts, were not accessed because they were not refereed by multiinstitu- tional experts and were in limited circulation.6 Computer-based searching contributed 291 articles (77.4%), and hand searching added an additional 40 arti- cles (22.6%).

Computer-based bibliographic searches incorporated the following automated search engines: MEDLINE; Health Planning and Administration/Health Services, Technology, Administration and Research (HEALTH- STAR); Cumulative Index to Nursing and Allied Health (CINAHL); and PsychINFO. For each search engine, the same medical subject headings (MESH) and text words (TW) were used to download articles with the following Boolean search algorithm: ("MESH: wounds or injuries" AND "TW: death or fatality or fatalities or mortality or mortahties" AND "MESH: alcohol, ethyl") OR ("MESH: wounds or injuries" AND "TW: death or fatality or fatali- ties or mortality or mortalities" AND "TW: BAC or blood alcohol or BAL or blood ethanol") OR ("MESH: wounds or injuries" AND "TW: death or fatality or fatalities or mor- tality or mortalities" AND "MESH: substance abuse"). Duplicate articles were then removed. The same algo- rithm was followed as closely as possible during hand searching of review publications and journal subject indices.

Seven additional criteria were used to limit the initial 331 articles. Studies not involving human subjects and studies outside the United States were excluded. Publications involving only military personnel, vehicular deaths, or occupational deaths were excluded, as were articles involving alcohol-related aircraft and railway fatalities, because previously published government reports provide more valid sources of information.7-9

Summary estimates of effect that use different reports from the same study cohort are not independent and therefore violate the statistical assumptions that consti- tute the basis of the procedures for aggregating data. 6 To prevent this problem, the article list was sorted by loca- tion and study period and inspected for repetitive study populations. When these were found, it was our policy to keep the study with the largest decedent population and the longest duration within a group of repetitive studies. This process eliminated 2 full duplicate articles. From the initial article list, a final list of 65 articles (19.6%) remained for analysis, lo-74

Data from 65 articles were abstracted by cause of death. Twenty-two unique cause-of-death categories

660 ANNALS OF EMERGENCY MEDICINE 33:6 JUNE 1999

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL 5mith, 13ranas & Miller

were created based on the International Classification of Diseases, 9th revision, Clinical Modification (ICD-9-CM) external cause-of-injury codes and information gleaned during the abstraction. The 22 categories were broadly separated into unintentional injury death, homicide, and suicide (Table) but were limited by the categories used in published studies. No attempt was made to blind the ini- tial reviewer of the final articles. Retrieval of information from figures was rare and occurred only when this infor- mation was not presented elsewhere in the article and could be reasonably estimated. 6

Data collection forms were developed before the data abstraction began and were subsequently revised based on the results of pilot testing. Similar procedures have been used previously to enhance data reliability. <75

BAC, measured in milligrams o f ethyl alcohol per deciliter of blood, was used as the measure of choice for alcohol content of decedent groups. Postmortem urine alcohol and brain alcohol measurements were excluded as measures of alcohol content because of their limited use as prima facie evidence in most US jurisdictions and their lack of comparability with the more commonly used blood-alcohol measurement. 76,77

For each cause of death, a study identifier, year of study, location of study, duration of study in months, number of decedents, and percentage of testing for alco- hol among the decedent groups were recorded. Five BAC categories (>0, >100, >_200, >_300, and >400 mg/dL) were also recorded, along with their rates within age and gen- der categories, mean BAC, and median BAC for each cause of death. BAC cutoff levels were selected based on previous comparisons of actual ethyl alcohol consump- tion and the resulting symptoms and impairment in non- tolerant users. 77,78 Blood alcohol levels greater than zero were considered positive, and deceased persons with a BAC of 100 mg/dL or higher were considered to have been intoxicated; these levels are concurrent with the cutoff values used in the articles and with values typically used for driving-while-intoxicated determinations.

One primary reviewer abstracted data from all 65 arti- cles. In an effort to assess interabstractor reliability before the final data were analyzed, each article was then reab- stracted by i of 2 independent, secondary reviewers. 6 Detailed abstracting instructions and rules were devel- oped. The authors, institutional affiliations, journal of publication, and funding sources for each study were removed and replaced with a number for identification. The proportion of agreement between the primary and secondary reviewers was determined in relation to both the total number of articles and the total number of causes

of death. Differences between reviewers were discussed, and mistaken data were corrected before the analysis was performed. 79

Information for a specific cause of death was not recorded if it was collectively determined by both abstrac- tors that none was ascertainable. In 2 of the 65 final arti- cles, data for 3 causes of death were available only by mechanism (eg, gunshot) and not by intent (eg, suicide). The authors of both articles were contacted, and supple- mental data on intent were subsequently forwarded (and used) for 1 of the articles. 59

Three summary estimates--aggregated, mean, and median--were selected, because each conveyed different yet meaningful information. The total number of deceased persons, number of deceased persons tested, and the number of deceased persons within each of the 5 BAC levels were aggregated across all studies. Aggregated cases were then used to calculate the percentages of deceased persons tested and within each BAC level for each cause of death. Aggregated percentage estimates accounted for sample size variation among articles and were equivalent to means weighted by sample size. It should be noted that for 5 causes of death the aggregated percent positive was less than the percent intoxicated. Because many medical examiner articles did not consis- tently report information for both positive and intoxi- cated deceased persons, articles were analyzed separately for positive and intoxicated data within each cause of death. Therefore, for some causes of death, the number of articles reporting positive BACs was less than the number reporting intoxicated BACs, resulting in an unavoidable inconsistency.

The percentages of deceased persons tested for alcohol and the percentage of decedents within each of the 5 BAC levels were then used to obtain mean and median esti- mates by cause of death. Mean estimates represented the average percentage affected and were calculated as the arithmetic means and 95% confidence intervals (CIs) of percentages in articles reporting a cause-of-death cate- gory. Mean estimates discerned cause-of-death categories for which there were few articles and broadly dissimilar reported percentages (namely, those with large CIs). Median estimates represented the typical percentage affected and were calculated as the middle value and interquartile range of a series of percentages. Median esti- mates were unaffected by skewed distributions of per- centages and indicated outlier percentages (those well outside the interquartile range). CIs and boxplots of each cause of death were then drawn for positive and intoxi- cated deceased persons. Cause-of-death categories for

JUNE 1999 33:6 ANNALS OF EMERGENCY MEDICINE 66 1

FATAL NONTRAFFIC I N J U R I E S I N V O L V I N G ALCOHOL Smith, Branas & Miller

which only 1 report was abstracted were also displayed, but only as means or medians without CIs or interquartile ranges, so

The reporting of gender and age categories was also assessed. Studies that reported age of deceased persons within categories were compared for general consistency by cause of death. Studies that did not distinguish age cat- egories were not included in this part of the analysis.

Data on motor vehicle driver fatalities were analyzed for comparative purposes using the 1993 Fatality Analysis Reporting System (FARS) of the National Highway Traffic Safety Administration (NHTSA), which contains data on all fatal crashes in the United States.1 This was done to illustrate variations in BAC by age and gender, since it was apparent that such variations were not nationally ascertainable for nontraffic injuries from the medical examiner studies. The FARS database con- tains both actual and estimated BAC data (based on an inference model applied to cases with unknown BACs). We used estimated BAC data, because these are usually reported by NHTSA and others.

All data were managed with the use of Microsoft Excel Version 5.0a (Microsoft Corporation; Redmond, WA). Exploratory and descriptive statistics were analyzed by SPSS for Windows Release 6.1.2 (SPSS, Incorporated, Chicago, IL). The study was approved by our Committee on Human Research.

R E S U L T S

Sixty-five final articles produced 141 total causes of death in 22 categories (Table). Accounting for articles that addressed multiple intent-of-injury categories, 38 arti- cles investigated unintentional injury deaths, 28 investi- gated homicide cases, and 27 investigated suicide cases. Alcohol was the primary subject o f 29 articles (44.6 %). All 65 articles were observational studies of historical cohorts; none evaluated actual injury risk. Seven articles ( 10.8 %) used data covering 1 year or less, and 27 articles (41.4%) used cohorts accumulated over 5 or more years. A majority of the articles were from the south Atlantic states (47.7%), followed by the mid-Atlantic states

Table. Aggregated estimates of alcohol involvement by cause of injury death (intent and mechanism).

Cause of Death No. of Articles No. of Subjects No. Tested % Tested % Positive % Intoxicated*

Unintentional injury 38 7,388 6,203 84.0 38.5 31 .O Burn/Fire 15 1,677 1,425 85.0 37.9 41.9 Cold/Hypothermia 2 185 156 84.1 90.0 40_9 Drowning 15 3,618 2,648 73.2 49.2 34.2 Fall 7 363 285 78.5 63.3 32.2 Gunshot 1 54 39 72.2 48.7 20.5 Poisoning by solid, liquid, or gas 5 150 149 99.3 26.6 29.3

Homicide 28 28,696 25,315 88.2 47.1 31.5 Asphyxiation, hanging, strangulation, or suffocation 3 224 185 82.6 29.7 160 Burn/Fire 1 26 22 84.6 36.4 18.2 Drowning 2 15 12 80.2 50.0 500 Fists and feet, beating, bludgeoning, or blunt object 3 523 455 87.0 40.7 24.9 Gunshot 9 6,145 5,556 90.4 38.9 30.6 Stabbing, cutting, or piercing 6 1,558 1,335 85.7 57.0 43.0

Suicide 27 19,347 15,814 81.7 29.0 22.7 Asphyxiation, hanging, strangulation, or suffocation 5 97 97 100.0 18.5 22.8 Burn/fire 1 24 14 58.3 14.3 0 Drowning 5 133 122 91.9 39.7 14.0 Fall/jumping 3 91 79 86.8 21.5 4.3 Gunshot 8 601 594 98.8 33.1 31.4 Poisoning by solid, liquid, or gas 6 337 331 98.2 23.3 34.9 Stabbing, cutting, or piercing 1 NA 14 NA 7.1 NA

Motor vehicle crash 1 23,134 NA NA 39.7 32.8

NA. not available. *Blood alcohol concentration >100 rng/dL

6 6 2 ANNALS OF EMERDENCY MEDICINE 33:6 .JUNE 1999

F A T A L N O N T R A F F I C I N J U R I E S I N V O L V I N G A L C O H O L Smith, Branas & Mi[~er

(21.5%), Pacific states (10.8%), north central states (9.2%), south centralstates (7.7%), and mountain states (3.1%). No articles were from the New England states.

Information was abstracted from figures in 8 articles (12.3%) and for 18 causes of death (12.8%) The reviewers agreed on abstracted values in 49 of the 65 articles and 113 of the 141 separate causes of death, making the respective proportional agreement values 75.4% and 80.1%.

The aggregated percentage of deaths, mean percentage of deaths, and median percentages of deaths tested for alcohol, positive for alcohol, and intoxicated were gener- ally concordant within each of the 22 cause-of-death cate- gories. 1=if teen articles (23.1%) reported time-dependent

Figure 1. Percentage of decedents positive for alcohol by cause-of-death category (mean and 95% CI).

criteria for postmortem blood-alcohol testing. In 9 of these articles, 6 hours was the limit after which BACs were considered invalid. One article used 2 hours, 2 articles used 4 hours, and the remaining 3 articles used 8 hours as their cutoffpoint. Mean BAC was reported in only 4 arti- cles, and median BAC in 3.

As shown in the Table, the aggregate percentage tested was highest among homicides (88.2%; range, 69.0% to 100%), followed by unintentional injury deaths (84.0%; range, 40.3% to 100%) and suicides (81.7%; range, 58.3% to 100%). Among homicide cases, those who died from gunshots had the highest rates of alcohol testing, fol- lowed by those who died from stabbing. Among suicide cases, asphyxiation deaths had the highest rates of test- ing, followed by gunshot deaths. Among unintentional deaths, those resulting from poisoning had the highest testing rate, followed by those caused by fire. Mean and median comparisons o f percentages of deceased persons tested for alcohol produced comparable findings.

Unintentional -O-

Fire Cold �9

Drown

Fall =

Gunshot �9

Poison ~.

Homicide

Suffocate

Fire �9

Drown ~.

Beat Gunshot

Stab

Suicide

Suffocate

Fire

Drown

Jump

Gunshot Poison

Stab F i i i i i

2O 4O 60 80 100

Figure 2. Boxplots showing percentage of decedents positive for alcohol by cause of death category (mean and interquartile range).

Unintentional Fire t I I I

Cold Drown �9

Fall * 03 ' Gunshot I

Poison ~-~ H

Homicide

Suffocate Fire

Drown Beat

Gunshot .Stab

I I I I

I I I

I I i I !

* [ ]

Suicide H t t Suffocate I---t I

Fire i

Drown t I I

Jump

Gunshot ; ~ i

Poison i I ~ N

Stab I 5 15 25 35 45 55 65 75 85 95

JUNE 1999 33:6 ANNALS OF EMERGENCY MEDICINE 6 6 3

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Branas &MilIer

The aggregate percentages of deceased persons posi- tive for alcohol varied by injury mechanism and intent (Table). The percentage was highest among homicides (47.1%), followed by unintentional injury deaths (38.5 %) and suicides (29.0%). The aggregate percentage of deceased persons who were intoxicated was also highest among homicide cases (31.5 %), followed by uninten- tional injury deaths (31.0%) and suicide cases (22.7%). Mean and median comparisons produced comparable findings (Figures i through 4). Within specific intent cat- egories, there was considerable variation in reported per- centages, although some cause-of-death categories had large CIs or outlier percentages because there were few articles with widely dissimilar percentages.

Thirteen articles (20.0%) reported deaths with a BAC of 200 mg/dL or higher; the mean percentage in this cate-

Figure 3. Percentage of decedents intoxicated for alcohol (BAC, >100 mg/dL) by cause-of-death category (mean and 95% CI).

Unintentional Fire

Cold Drown

Fall Gunshot

Poison

Homicide

Suffocate Fire

Drown Beat

Gunshot Stab

Suicide

Suffocate Fire �9

Drown Jump

Gunshot Poison

I

e

_=

i

2O

i

4O

i

6O

i

8O

gory for all causes of death combined was 17.6%. Eight articles (12.3%) reported deaths with a BAC of 300 mg/dL or higher, and the mean percentage of all deaths in this category was 10.3%. Six articles (9.2%) reported deaths with a BAC of 400 mg/dL or higher, with the mean per- centage of all deaths in this category being 2.3%.

Overall, 12 (18.5 %) articles contained ascertainable gender-specific data, and 15 (23.1%) contained ascer- tainable age-specific data. Because of these small num- bers and the varying age intervals used, formal analyses by cause of death could not be completed, although age and gender variations were qualitatively noted. The data for motor vehicle deaths clearly demonstrated that alco- hol involvement varies widely by both age and gender

Figure 4. Boxplots showing percentage of decedents intoxicated for alcohol ~AC, >100 mg/dL) by cause-of-death category (median and interquartile range).

Unintentional . [ ] * Fire, I "t I I--4

Cold I Drown ~ I e

Fall I Gunshot I

Poison t

Homicide I Suffocate I

Fire I Drown

Beat r - - r - - I Gunshot I - r ~ l - I

Stab I - I - - [ - - ]

Suicide

Suffocate I Fire I

Drown

Jump I

Gunshot

Poison I-i--I

5 15 25 35 45 55 65 75 85 95

6 6 4 ANNALS OF EMERGENCY MEDICINE 33:6 JUNE 1999

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Branas &MiIIer

(Figure 5). Data from the FARS indicated that 37% of male drivers and 19% of female drivers killed in fatal crashes in 1993 had BACs of 100 mg/dL or more. The pro- portion of fatally injured drivers who were intoxicated varied from 48%, for age group 30 to 34 years, to 8% for drivers older than 65 years of age; the figure was 32% for all ages combined. Men aged 25 to 34 years had the high- est proportion intoxicated (51%), and women older than 65 years had the lowest (4%).

DISCUSSION

Medical examiner reports are an important and often overlooked source of data for designing and evaluating injury prevention programs.5 Of particular value is their often reliable recording of the involvement of alcohol in injury fatalities. This is evidenced by the reasonably high number of medical examiner studies of alcohol and injury fatalities accessed in completing this work.

The lack of prospective cohort or case-control studies in the study of nontraffic injuries prohibits the estimation of relative risks for specific injuries by alcohol use. Therefore, it is not possible to calculate population attributable risk (based on the application of cause-specific relative risks to the total US population) based on the medical exam- iner data. In the absence of relative risk information, we could either have taken any evidence of elevated BAC as indicative of significant alcohol involvement or used BAC levels above a certain level. We chose to use 100 mg/dL as a BAC indicative of significant alcohol involvement because this is the level at which the relative risk for motor vehicle crashes increases exponentially. 81 This cutoffvalue is also a commonly accepted level of intoxication that has been widely used in other studies to approximate attributable fractions for alcohol abuse (attributable fractions are proportions of disease occurrence that could poten- tially be eliminated should exposure to a risk factor be

82 3 83 87 removed ). �9 - In these approximations, the increased risks for BACs below 100 mg/dL are assumed to balance out other exogenous factors that may have contributed to injuries in victims with BACs above 100 mg/dL.

Based on our analysis, the best estimates of those non- traffic injury fatalities in which alcohol is a maj or contribut- ing factor are represented by the "Percentage Intoxicated" column in the Table. These percentages were 32 % overall for homicide cases, 31% for unintentional injury deaths, and 23 % for suicide cases. The percentage of intoxicated deceased persons involved in motor vehicle crashes, 33%, is similar to that for homicide cases and unintentional injury deaths.

There was also considerable variation by mechanism of injury within each intent category (Table). Among the unintentional deaths, those caused by burns and by hypothermia had the highest intoxication rate (41% to 42%) and firearm deaths (many of which presumably occurred while hunting) had the lowest (21%). Among homicide cases, the highest intoxication rate was for drownings (50%), although the number of studies and number of deaths were small; the lowest proportion was for those who died by asphyxiation or strangulation (16%). One limitation of this and all other homicide stud- ies is that few data are available on the role of alcohol among the perpetrators of homicide, although it is often thought that because perpetrators come from the same environment as victims, their pattern of alcohol use may be similar.2

Our study both provides updated estimates of alcohol involvement for specific causes of fatal nontraffic injuries and, unlike previous estimates, also provides detailed methodologic documentation and literature referencing. In addition, our estimates are derived with the use of stan- dard statistical techniques, whereas previous studies appear to have used less rigidly calculated estimates accompanied by sparse methodologic documentation.

In examining a number of other studies evaluating the role of alcohol in injury mortality, we find a surprising

Figure 5. Percentage of fatally injured drivers with a BAC of 100 mg/dL or higher, by age group and gender (FARS data, 1993). Broken lines represent mean values across all age categories.

60 Percentage

t --,,-Male -e.-Female 50 --Male & Female

. . . . . . . . . . . . . . . . . .

: 3 0 ~

? 20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . e..an

lO %

"no o 15-19 20-24 25-29 30-34 35-39 40-44 45-49 50-54 55-59 60-64 65+

Age group

JUNE 1999 33:6 ANNALS OF EMERGENCY MEDICINE 665

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Branas & Miller

lack of critical appraisal of available data and an almost total reliance on 2 separate estimates of alcohol- attributable fractions for specific injuries. 3,84 For exam- ple, McGinnis and Foege,83 in an extensively cited study of major causes of death in the United States, rely on esti- mates reported by the Centers for Disease Control and Prevention,85 which in turn used a computer software program developed by the authors of an Alcohol, Drug Abuse, and Mental Health Administration (ADAMHA) study of the costs of alcohol abuse, drug abuse, and men- tal illness. 86 Examination of the latter report shows heavy reliance on the estimates of alcohol-attributable fractions from a study of the costs of alcohol abuse in Minnesota. s4 For motor vehicle injuries, all crashes in which a driver or pedestrian had a BAC exceeding 50 mg/dL were assumed to be attributable to alcohol, whereas BACs exceeding 100 mg/dL were used for other injuries.

For the most part, Parker et a184 appeared to rely on lit- erature reviews available in 1985 to judge the percentage of deaths involving BACs above the thresholds they selected, but they did not generally cite the studies under- lying their choices of these values. The only study they cited was Harwood et al, 87 which reported rates dissimi- lar from their own using their Minnesota data for drown- ing and boating fatalities. The more recent study for ADAMHA 86 apparently did not update the work by Parker et al, nor did it provide support for choosing the estimates from this work over other markedly different estimates, such as those provided by Ravenholt. 3 For example, Parker et al estimated that 41% of fatal uninten- tional falls and 42 % of fatal unintentional fires are attributable to alcohol, well above Ravenholt's estimates of 25 % in each category. It was the marked discrepancies between these studies that provided the primary impetus for our study, which found 32% of unintentional fall deaths and 42 % of unintentional fire and burn deaths to involve intoxication of the deceased person.

A maj or weakness of the alcohol and injury studies available to us was that few data were provided regard- ing alcohol involvement by gender and age. These modi- fying factors are necessary when formulating the relation between alcohol and injuries and to identify specific gender and age groups that are at unusually high risk.

Another weakness of the available studies was that they did not use strict definitions of the interval from injury to time of testing. This is important because alco- hol continues to be metabolized even in severely injured persons. 81 Data obtained from toxicology testing at the time of autopsy are most commonly available, because medical examiners are more likely to test BAC at autopsy

than are hospitals, which less frequently test for BAC.88 Of particular concern are elderly patients who may die days or weeks after a fall injury. 81

The ability to use published mortality studies to build a database across a number of injury causes also was ham- pered by a lack of comparability among medical examin- ers. In some cases, the BAC used for reporting intoxicated persons was inconsistent, although many studies did use a cutoff value between 80 and 100 mg/dL. Furthermore, very few medical examiner studies provided data on the proportion of deceased persons with extremely high BACs. We strongly recommend that, even if studies report different BAC categories, they should at least report data for BACs of 100 mg/dL or higher, because this is the most commonly accepted level indicating intoxication.

Another limitation of the aggregation of studies used in this report was that there were low numbers of reports for certain causes of death. This detracts from the reliabil- ity of some of our estimates and the generali7ability of these estimates on a national level, particularly because the geographic variation in study sites was also limited. For many of the injury causes, hDwever, there was reason- able variation in study sites across the country, reducing the likelihood that local variations in alcohol consump- tion and hazardous drinking behaviors would restrict the generalizability of findings.

A final limitation that deserves consideration in any metaanalysis is publication bias: studies with significant results are more likely to be submitted and published than those without such results. 6,89'9~ In our study, how- ever, publication bias may be less of a concern because the articles considered were almost exclusively descriptive in nature and did not report statistically significant findings. Furthermore, the potential for publication bias may have worked somewhat in our favor in that only areas with high rates of alcohol testing are likely to report their findings in the literature. Areas of the country with poor or no regular BAC testing of injury victims may be underreported.

This s tudy is the first systematic examination of alcohol involvement in fatal nontraffic injuries at the national level. It was completed in accordance with recently published recommendations.9 t Our findings corroborate the view of earlier studies that alcohol plays an important role in injury mortality. The improved estimates of alcohol involvement presented here expand our understanding of the impact of alcohol on society and can be used to generate better cost estimates of alcohol use in nontraffic injuries. However, future studies are needed to more precisely examine the gender- and age- specific role of alcohol in nontraffic injuries, because

666 ANNALS OF EMERGENCY MEDICINE 33:6 JUNE 1999

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Branas & Miller

there is probably variation in alcohol involvement by both age and gender.

We wish to acknowledge the assistance of Farah Shaikh, MD, and Judy Kleppel, MD, in reabstracting portions of the articles used in this study.

REFERENCES 1. National Highway Traffic Safety Administration: Traffic safety facts I993.'A compilation of motor vehicle crash data from the Fatal Accident Reporting System and the General Estimates System. Washington DC: US Department of Transportation, October 1994.

2. Roizen J: Estimating alcohol involvement in serious events, in National Institute on Alcohol Abuse and Alcoholism: Alcohol and Health Monograph I: Alcohol Consumption and Related Problems. Washington DC: US 6overnment Printing Office, 1982:179-219.

3. Ravenholt RT: Addiction mortality in the United States, 1980: Tobacco, alcohol, and other substances. Popul Dev Rev 1984;10:697-724.

4. Combs DL, Parrish RG, Ing RI: Death Investigations in the United States and Canada, 1995. Washington DC: US Department of Health and Human Services, 1995:13-21.

5. Hanzlick R, Parrish R6: The role of medical examiners and coroners in public health surveillance and epidemiologic research. Annu RevPublic Health 1996;17:383-409.

6. Petitti DB: Met&analysis, Oecision Analysis, and Cost-Effectiveness Analysis: Methods for Quantitative Synthesis in Medicine, ed 1. New York: Oxford University Press, 1994:3-130.

7. McLean GA, Wilcox BC, Canfield DV: Selection Criteria forAIcohol Detection Methods. ADA240441. Washington DC: US Department of Transportation, Federal Aviation Administration, Office of Aviation Research, 1991:91-112.

8. Collins WE, Mertens HW: Age, Alcohol and Simulated AItitude: Effects on Performance and Breathalyzer Scores. ADA190642. Washington DC: US Department of Transportation, Federal Aviation Administration, Office of Aviation Research, 1988:88-92.

9. Department of Transportation, Federal Railroad Administration, Office of Safety Research: Enhancing Rail Safety Now and into the 21st Century. The Federal Railroad Administration's Safety Programs andlnitiatives. A report to Congress, October 1996.

10. Abel E, Zeidenberg P, Regan S. Alcohol and violent death--Erie County, New York, 1973-1983. MMWR Morb Mortal Wkly Rep 1984;33:226-227.

11. Abel EL, Zeidenberg P. Alcohol and homicide: A comparison between Erie County, New York, and Los Angeles County, California. Am JDrugAIcoholAbuse 1986;12:121-129.

12. Abel EL, Zeidenberg P. Age, alcohol and violent death: A postmortem study. JStud Alcohol 1985;46:228-231.

13. Abel EL: Guns and blood alcohol levels among homicide victims. Drug Alcohol Depend 1986;18:253-257.

14. Barillo D J, Rush BF Jr, 6oode R, et el: Is ethanol the unknown toxin in smoke inhalation injury? Am Surg 1988;52:841-845.

15. Bastes ML, 6alante L: Toxicological findings in victims of traumatic deaths. J Forensic Sci 1976;21:176-186.

16. Berkelman RL, Herndon JL, Callaway JL, et al: Fatal injuries and alcohol. Am J PrevMed 1985;1:21-28.

17. Birky MM, Clarke F8. Inhalation of toxic products from fires. Bull N YAcad Mad 1981 ;57:997-1013.

18. Brent DA, Perper JA, AIIman C J: Alcohol, firearms, and suicide among youth: Temporal trends in Allegheny County, Pennsylvania, 1960 to 1983. JAMA 1987;257:3369-3372.

19. Budd RD: Cocaine abuse and violent death. Am JDrugAIcoholAbuse 1989;15:375-382.

20. Budd RO: The incidence of alcohol use in Los Angeles County homicide victims. Am J Drug Alcohol Abuse 1982;9:105-111.

21. Cherpitel C J: Alcohol and casualties: A comparison of emergency room and coroner data. Alcohol AIcoho11994 ; 29:211- 218.

22. Copeland AR: Accidental death by gunshot wound: Fact or fiction? Forensic Sci Int 1984;26:25-32.

23. Copeland AR: Accidental fire deaths: The 5-year Metropolitan Dade Country experience from 1979 until 1983. JLeg Med1985;94:71-79.

24. Copeland AR: Deaths during recreational activity. Forensic Sci Int 1984;25:117-122.

25. Copeland AR: Homicide by fire. J Leg Med1985;95:59-65.

26. Copeland AR: Multiple homicides. Am J Forensic Mad Patho11989;10:206-208.

27. Copeland AR: Non-commercial, accidental water transport (boating) fatalities. J Leg Med 1986;96:291-296.

28. Copeland AR: Police shootings. Am J Forensic Med Patho11986;7:39-45.

29. Copeland AR: Suicidal fire deaths revisited. J Leg Med 1985;95:51-57.

30. Copeland AR: Suicide among nonwhites: The Metro Dade County experience, 1982-1986. Am J Forensic Med Patho11989;10:10-13.

31. Copeland AR: Suicide by drowning. Am J Forensic Mad Patho11987;8:18-22.

32. Copeland AR: Suicide by jumping from buildings. Am J Forensic Mad Patho11989;10:295- 298.

33. Copeland AR: Teenage suicide: The five-year metro Dade County experience from 1979 until 1983. Forensic Sci Int 1985;28:27-33.

34. Costatino JP, Kuller LH, Perper JA, et al: An epidemiologic study of homicides in Allegheny County, Pennsylvania. Am J Epidemio11977;106:314-324.

35. Dames R: Hypothermia-related deaths--Cook County, Illinois, November 1992-March 1993. MMWR Morb Mortal Wkly Rep 1993;42:917-919.

35. Davis S, Ledman J, Kilgore J: Drownings of children and youth in a desert state. WestJ Mad 1985;143:196-201.

37. Fishbain DA, Fletcher JR, Aldrich TE, et el: Relationship between Russian roulette deaths and risk-taking behavior: A controlled study. Am J Psychiatry1987;144:563-567.

38. Ford AB, Rushferth N8, Rushforth N, et el: Violent death in a metropolitan county: II. Changing patterns in suicides (1959-1974). Am J Public Health 1979;69:459-484.

39. Friedman IM: Alcohol and unnatural deaths in San Francisco youths. Pediatrics 1985;76:191-193.

40. Gallahar MM, Fleming DW, Berger LR, et el: Pedestrian and hypothermia deaths among Native Americans in New Mexico: Between bar and home. JAMA 1992;267:1345-1348.

41. Garriott JC: Drug use among homicide victims: Changing patterns. Am J Forensic Med Pathol 1993;14:234-237.

42. Goodman RA, Istre GR, Jordan FB, et el: Alcohol and fatal injuries in Oklahoma. JStud Alcohol 1991;52:158-161.

43. Goodman RA, Mercy JA, Loya F, et el: Alcohol use and interpersonal violence: Alcohol detected in homicide victims. Am J Public Health 1986;76:144-149.

44. Haberman PW, Baden MM: Alcohol Other Drugs and Violent Death. New York: Oxford University Press, 1978:75-93.

45. Haberman PW, Natarajan 6: Trends in alcoholism and narcotics abuse from medical examiner data. J Stud AIcoho11985;47:316-321.

46. Haberman PW: Alcohol and alcoholism in traffic and other accidental deaths. Am J Orug Alcohol Abuse 1987;13:475-484.

47. Hain JR, Ryan D, Spitz W: Fatal accidents and blood ethanol levels in adolescents and adu Its: The Wayne County experience. Am J Forensic Med Pathol 1989;16:187-192.

48. Harruff RC, Francisco JT, Elkins S, et al. Cocaine and homicide in Memphis and Shelby County: An epidemic of violence. J Forensic Sci 1988;33:1231-1237.

49. Harruff RC, Llewellyn AL, Clark MA, et al: Firearm suicides during confrontations with police. J Forensic Sci 1994;39:402-411.

50. Kirkpatrick JB, Pearson J: Fatal cerebral injury in the elderly. J Am Geriatr Soc 1978;26:489-497.

51. Levine MS, Radford EP: Fire victims: Medical outcomes and demographic characteristics. Am J Public Health 1977;67:1077-1080.

52. Lowry PW, Hassig SE, Gunn RA, et al: Homicide victims in New Orleans: Recent trends. Am J Epidemiol 1988;128:1130-1136.

53. Marzuk M, Tardiff K, Leon AC, et el: Prevalence of cocaine use among residents of New York City who committed suicide during a one-year period. Am J Psychiatry 1992;149:371- 375.

54. McBride DC, Burgman-Havermehl C, Alpert J, et al: Drugs and homicide. Bull N YAcad Med 1985;62:497-508.

JUNE 1999 33:6 ANNALS OF EMERGENCY MEDICINE 667

FATAL NONTRAFFIC INJURIES INVOLVING ALCOHOL Smith, Bran~ & Miller

55. McKelway R, Vieweg V, Westerman P: Sudden death from acute cocaine intoxication in Virginia 1988. Am J Psychiatry 1990;147:1667-1669.

56. Mierley MC, Baker SP. Fatal house fires in an urban population. JAMA 1983;249:1466-1468.

57. Muscat JE, Huncharek MS: Firearms and adult, domestic homicides: The role of alcohol and the victim. Am J Forensic Med Patho11991;12:105-110.

58. Nashold RD, Naor EM: Alcohol-related deaths in Wisconsin: The impact of alcohol on mor- tality. Am J Public Health 1981 ;71:1237-1241.

59. Nichter MA, Everett PB: Profile of drowning victims in a coastal community. J Fla Med Assoc 1989;76:253-256.

60. Patetta M, Biddinger P, Freeman J, etal: Current trends--North Carolina drownings, 1980-1984. MMWR Morb Mortal Wkly Rep 1986;35:635-638.

61. Patetta M J, Biddinger PW: Characteristics of drowning deaths in North Carolina. Public Health Reports 1988;103:406-414.

62. Patetta M J, Cole TB: A population-based descriptive study of housefire deaths in North Carolina. Am J Public Health 1990;80:1116-1117.

63. Riddick L, Luke JL: Alcohol-associated deaths in the District of Columbia: A post mortem study. J Forensic Sci 1978;23:493-502.

64. Rutledge R, Messick W J: The association of trauma death and alcohol use in a rural state. J Trauma 1992;33:737-742.

65. Schuman SH, Ollison CW, Wannamaker CC, et al: Which drownings are preventable? An indepth study of 62 deaths in Charleston County, South Carolina. J S C MedAssoc 1978;74:361- 364.

66. Sikes RK. Alcohol and fatal injuries--Fulton County, Georgia, 1982. MMWR Morb Mortal Wkly Rep 1983;32:573-578

67. Smith GS, Re K, Tarantino S, et al: Drink, drugs, and drowning: A dilemma for prevention. Paper presented at the 18th Annual Alcohol Epidemiology Symposium, Toronto, Ontario, Canada, June 1-5, 1992.

68. Smith SM, Goodman RA, Thacker SB, et al: Alcohol and fatal injuries: Temporal patterns. Am J Prev Med 1989;5:296-302.

69. Tardiff K, Marzuk PM, Leon AC, et al: Cocaine, opiates, and ethanol in homicides in New York City: 1990 and 1991. J Forensic Sci 1995;40:387-390.

70. Welte JW, Abel EL: Homicide: Drinking by the victim. J Stud AIcoho11989;50:197-201.

71. Wetli CV, Wright RK: Death caused by recreational cocaine use. JAMA 1979;241:2519- 2522.

72. Wetli CV: Changing patterns of methaqualone abuse: A survey of 246 fatalities. JAMA 1983;249:621-626.

73. Wintemute G, Kraus J, Teret S, et al: The epidemiology of drowning in adulthood: Implications for prevention. Am J Prev Med 1988;4:343-348.

74. Wintemute G J, Teret SP, Kraus JF, etal: Alcohol and drowning: An analysis of contributing factors and a discussion of criteria for case selection. AccidAnal Prev 1990;22:291-296.

75. Stock WA, Okun MA, Haring M J, et al: Rigor in data synthesis: A case study of reliability in meta-analysis. Educ Res 1982;11:10-20.

76. Fatteh A: Consideration of individual poisons, in Fatteh A (ed): Handbook of Forensic Pathology. Philadelphia: JB Lippincott, 1973:283-287.

77. Flanagan R J, Braithwaite RA, Brown SS, et al: Basic Analytical Toxicology. Geneva: World Health Organization, 1995:131-134.

78. David D J, Spyker DA: The acute toxicity of ethanol: Dosage and kinetic nomograms. Vet Hum Toxicol 1979;21:272-276.

79. Milne R, Thorogoed M: Hand searching the Journal of Epidernielogy and Community Health as part of the Cocbrane Collaboration. J Epidemiol Community Health 1996;50:178-181.

80. Rice JA: Mathematical Statistics and Data Analysis, ed 3. Belmont, CA; Wadsworth, 1988:336-338.

81. Smith GS, Kraus JF: Alcohol and residential, recreational and occupational injuries: A review of the epidemiologic evidence. Annu Rev Public Health 1988;9:99-121.

82. Kelsey JL, Thompson WD, Evans AS: Methods in Observational Epidemiology. New York: Oxford University Press, 1986:39-40.

83. McGinnis JM, Foege WH: Actual causes of death in the United States. JAMA 1993;270:2207-2212.

84. Parker DL, Shultz JM, Gertz L, etal: The social and economic costs of alcohol abuse in Minnesota, 1983. J Public Health 1987;77:982-986.

85. Centers for Disease Control and Prevention: Alcohol-related mortality and years of potential life lost--United States, 1987. MMWR Morb Mortal Wkly Rep 1990;39:173-178

86. Rice DP, Kelman S, Miller LS, et al: The Economic Cost of AIcohol and Drug Abuse and Mental Illness, 1985. Rockville, MD: US Department of Health and Human Services, Alcohol, Drug Abuse, and Mental Health Administration, 1990.

87. Harvvood H J, Napolitando DM, Kristinsen RL, etal: Economic Costs to Society of Atcohol and Drug Abuse and Mental Illness, 1980. Research Triangle Park, NC; Research Triangle Institute, 1984.

88. Soderstrom CA, Dailey JT. Kems T J: Alcohol and other drugs: An assessment of testing and clinical practices in US trauma centers. Arch Surg 1994;36:68-73.

89. Lewin DI: Meta-analysis: A new standard or clinical fool's gold? J NIH Research 1996;8:30- 31.

90. Mann C: Meta-analysis in the breech. Science 1990;249:476-480.

91. Breslew RA, Ross SA, Weed DL: Quality of reviews in epidemiology. Am J Public Health 1998;88:475-477.

668 ANNALS OF EMERGENCY MEDICINE 33:6 JUNE 1999