About ARDI - Methods

Alcohol-Related Disease Impact (ARDI) software generates estimates of alcohol-related deaths and Years of Potential Life Lost (YPLL) due to alcohol consumption. To do this, ARDI either calculates or uses pre-determined estimates of Alcohol-Attributable Fractions (AAFs)—that is, the proportion of deaths from various causes that are due to alcohol. These AAFs are then multiplied by the number of deaths caused by a specific condition (e.g., liver cancer) to obtain the number of alcohol-attributable deaths.

A Scientific Work Group, comprised of experts on alcohol and health, was convened to guide development of the ARDI software. The Work Group's tasks included:

Alcohol-Attributable Fractions (AAFs)

100% Alcohol-Attributable Conditions

Certain conditions (e.g., alcoholic cirrhosis of the liver) are, by definition, caused by alcohol consumption. These conditions are classified as being 100% alcohol-attributable and are reported in ARDI as having an AAF of 1.00.

Direct Estimates

For some conditions, especially injuries, ARDI includes direct estimates of AAFs. Direct estimates of AAFs are based on direct observations about the relationship between alcohol and a given health outcome. These estimates typically come from studies that have assessed the proportion of persons dying from a particular condition at or above a specified blood alcohol concentration (BAC) (i.e., 100 mg/dL), or from follow-up studies that have specifically assessed prior alcohol use among persons dying of specific conditions (e.g., liver cirrhosis) based on medical record review and interviews with next-of-kin.

For motor-vehicle traffic deaths, we used direct estimates of alcohol involvement in fatal crashes from the Fatality Analysis Reporting System (FARS), which is administered by the National Highway Traffic Safety Administration (NHTSA) (http://www.nhtsa.dot.gov). The FARS methodology is described in detail elsewhere (NHTSA, 2002). Briefly, FARS provides annual estimates of alcohol involvement for all fatal traffic crashes that occur on United States roadways in a given year. In addition to relying on the results of alcohol testing, FARS also uses an imputation formula to estimate alcohol involvement for a driver or non-occupant who was killed in a crash, when BAC measurements were not available. The FARS protocol for determining alcohol involvement in a crash does not require separate measurements of alcohol levels for all persons who died in the crash; as a result, ARDI includes age-specific estimates of the AAFs for all persons who were killed in traffic crashes in a given year regardless of their role in the crash itself (e.g., driver versus occupant). In fact, this is the only condition in ARDI where age-specific AAFs were available and were used to calculate alcohol-attributable deaths. For the purposes of ARDI, a crash death was judged to have been alcohol-attributable if either a driver or non-occupant involved in a fatal crash had a BAC ≥ 100 mg/dL (0.10 g/dL). The FARS data that were used in ARDI was selected to match the death data and life expectancy data for that year. For example, if the most recent death data was for 2001, then 2001 FARS data was used to obtain the AAFs for motor-vehicle traffic deaths.

For non-traffic injury deaths, we relied on direct estimates of AAFs from a meta-analysis performed by Smith et al. (1999). In this study, the researchers judged a fatal non-traffic injury to have been alcohol-attributable if the decedent had a BAC ≥ 100 mg/dL (0.10 g/dL) at the time of death.

For most chronic conditions, we calculated indirect estimates of AAFs using pooled risk estimates. Consistent with a recent analysis of alcohol-related deaths in the United States completed by Mokdad et al. (2004), we used a large meta-analysis by English et al. (1995) as the primary source of risk estimates for most of the chronic conditions included in ARDI. However, this study did not specifically provide risk estimates for deaths due to unspecified liver cirrhosis; thus we relied on a direct estimate of the AAF for this condition, and other closely related conditions (e.g., esophageal varices) from a follow-back study performed by Parrish et al. (1993).

Indirect Estimates

For some conditions, particularly chronic conditions, ARDI calculates indirect estimates of AAFs. These calculations use pooled risk estimates obtained from large, systematic reviews of the scientific literature, known as meta-analyses, on the relationship between alcohol and various alcohol-related health outcomes (e.g., liver cancer) as well as data on the prevalence of alcohol use at specific consumption levels (i.e., low, medium, and high). As mentioned above, most of the pooled risk estimates used in ARDI were drawn from a study done by English et al. (1995). However, the Scientific Work Group also used risk estimates from other meta-analyses if they felt these studies better characterized the relationship between alcohol and a particular health outcome than the English et al. study.

Having selected risk estimates, indirect estimates of AAFs are calculated in ARDI using the following formula:

(Prevalence)(Relative Risk - 1)
1+(Prevalence)(Relative Risk - 1)

Where Prevalence is the percentage of United States men and women (age 20 years and older) consuming alcohol at a specified level of average daily alcohol consumption within a given year, and Relative Risk is the likelihood of death from a particular condition at a specified level of average daily alcohol consumption. In accordance with the methods used by English et al. (1995), when evaluating the relationship between medium and high alcohol use and deaths from various causes, the risk estimates for these consumption levels were divided by the risk estimate for low alcohol consumption, thus making those in the low average daily consumption group the reference population rather than abstainers.

Prevalence Data

Annual estimates of average daily alcohol use at various levels (i.e., low, medium and high) for US adults (age 20 years and older, stratified by gender) were obtained from the Behavioral Risk Factor Surveillance System (BRFSS) (http://www.cdc.gov/brfss/index.htm). The BRFSS data used in ARDI were selected to match both the death data and the life expectancy data that are being used in the system. For example, if the most recent death data was for 2001, then 2001 BRFSS data was used to obtain the prevalence of average daily alcohol consumption.

Indexing Average Daily Alcohol Consumption

The BRFSS currently includes three questions on the prevalence of alcohol use. These questions assess drinking days, the average number of drinks per day on drinking days, and the frequency of binge drinking (defined as 5 or more drinks on one or more occasions), all of which are based on alcohol use during the past 30 days.

Using BRFSS data, average daily alcohol consumption may be calculated by multiplying the number of drinking days by the average number of drinks per day, and then dividing this product by 30. However, because BRFSS respondents appear not to include binge drinking when reporting regular drinking patterns (Mokdad et al, 2004), an indexing procedure developed by Armor and Polich (1982) was used to assure that self-reported information on binge drinking was included in calculations of average daily alcohol consumption. The average number of drinks consumed per binge drinking episode was obtained from the BRFSS Binge Drinking Module, which was used by 13 states in 2003.

Cutpoints for Alcohol Use

The cutpoints used to define different levels of average daily alcohol consumption (i.e., low, medium, and high) were specified in the meta-analyses that were used to obtain risk estimates for a given condition. These cutpoints were typically reported as grams of alcohol per day, and then converted into drinks per day, using a 13.7 gram per drink conversion factor. For those conditions where the Scientific Work Group selected the meta-analysis performed by English et al. or Ridolfo and Stevenson as the source of risk estimates, the following cutpoints were used:

*Excludes those who were abstinent or had < 0.2 drinks/day on average within a 30-day time frame.

For those conditions where the meta-analyses by either Corrao et al. or Bagnardi et al. were used to obtain risk estimates, the following cutpoints were used:

*Excludes those who were abstinent or had < 0.1 drinks/day on average within a 30-day time frame. Cutpoints are the same for males and females.

Data Sources for Mortality and Life Expectancy

Total Deaths Data Set

To calculate alcohol-attributable deaths, the AAFs for a specific condition were multiplied by the number of deaths in a given category. The death data were obtained from the National Vital Statistics System managed by the National Center for Health Statistics (http://www.cdc.gov/nchs/Default.htm). Deaths were coded using the International Classification of Diseases, Tenth Revision (ICD-10). See Alcohol Related ICD Codes.

The death data were stratified by age and gender using standard 5-year age groupings. In general, ARDI assesses deaths due to chronic conditions beginning at age 20 and deaths due to acute conditions starting at age 15. However, death data were also collected on persons who were less than 15 years of age at the time of death if they died from alcohol-related conditions that specifically affect children. These conditions include fetal alcohol syndrome; fetus and newborn affected by maternal use of alcohol; child abuse; and low birth weight, prematurity, and intrauterine growth retardation. Deaths due to motor-vehicle traffic crashes among persons less than age 15 years were also included in the system, because data on alcohol involvement in these deaths are available through FARS.

Life Expectancy Data Set

Data on life expectancy are obtained from the National Vital Statistics System managed by the National Center for Health Statistics (http://www.cdc.gov/nchs). Life expectancy data were also stratified by age and gender using standard 5-year age groupings. These life expectancy data were, in turn, used to estimate the YPLL for alcohol-attributable deaths.

Since YPLL is based on the age at death, the YPLL for a particular alcohol-related condition is directly related to the age distribution of the persons who typically die of that condition. As a result, YPLL generally tends to be higher for conditions that disproportionately affect youth and young adults (e.g., motor-vehicle traffic deaths) and lower for conditions that primarily affect older adults (e.g., ischemic heart disease).

Limitations

ARDI may underestimate the actual number of alcohol-related deaths and YPLL in the United States for several reasons. First, BRFSS data on alcohol use, which are used to calculate indirect estimates of AAFs, are based on self-reports, which tend to underestimate the true prevalence of alcohol use because of sampling noncoverage—that is, the inability to reach certain high-risk populations, such as youth and young adults—and the underreporting of alcohol use by survey respondents (Nelson, 2001). Second, BRFSS prevalence estimates are based on alcohol use during the past 30 days. As a result, former drinkers, who may have discontinued drinking because of health problems, are not included in the calculation of AAFs. Third, ARDI does not include estimates of alcohol-attributable deaths for several conditions (e.g., tuberculosis, pneumonia, and hepatitis C) for which alcohol is widely believed to be an important risk factor but where the Scientific Work Group was unable to find a suitable pooled risk estimate. Fourth, ARDI exclusively uses the underlying cause of death from vital statistics to identify alcohol-related conditions and does not consider contributing causes of death that may also be alcohol-related. Finally, age-specific estimates of AAFs were only available for motor-vehicle traffic deaths even though alcohol-involvement is known to vary widely by age, particularly for acute conditions, and is generally much greater for deaths involving youth and young adults. This limitation is likely to have resulted in a substantial underestimate of YPLL from deaths due to acute conditions.

References

National Highway Traffic Safety Administration, Traffic Safety Facts 2001, Washington DC, National Center for Statistics and Analysis; 2002.

Smith G, Branas C, Miller T. Fatal nontraffic injuries involving alcohol: a meta-analysis. Annals of Emergency Medicine 1999; 33(6): 659-68.

English DR, Holman CDJ, Milne E, Winter MG, Hulse GK, Codde JP, et al. The Quantification of Drug Caused Morbidity and Mortality in Australia, 1995 edition. Canberra, Australia: Commonwealth Department of Human Services and Health 1995.

Parrish K, Dufour M, Stinson F, Harford T. Average daily alcohol consumption during adult life among decedents with and without cirrhosis: the 1986 National Mortality Followback Survey. Journal of Studies on Alcohol 1993; 54(4): 450-6.

Mokdad AH, Marks JS, Stroup DF, Gerberding JL. Actual causes of death in the United States, 2000. JAMA 2004; 291(10): 1238-45.

Bagnardi V, Blangiardo M, La Vecchia C, Corrao, G. A meta-analysis of alcohol drinking and cancer risk. British Journal of Cancer 2001; 85(11): 1700-05.

Corrao G, Bagnardi V, Zambon A, Arico S. Exploring the dose-response relationship between alcohol consumption and the risk of several alcohol-related conditions: a meta-analysis. Addiction 1999; 94(10): 1551-73.

Ridolfo B, Stevenson, C. The Quantification of Drug-Caused Morbidity and Mortality in Australia, 1998. Drug Statistics Series no. 7. AIHW cat. no. PHE 29. Canberra, Australia: AIHW. 2001.

Armor DJ, Polich JM. Measurement of alcohol consumption. In: Mansell Pattison E, Kaufman E, (editors) Encyclopedic Handbook of Alcoholism, New York: Gardner Press; 1982: 72-80.

Nelson DE. Reliability and validity of measures from the Behavioral Risk Factor Surveillance System (BRFSS). Social and Preventive Medicine, 2001; 46 (Supplement 1): S3-S42.