This is the first report which outlines the burden of mortality and PYLL attributable to alcohol consumption in the US differentiated by race, age and sex, and it reveals that alcohol consumption is a large contributor to the burden of mortality. As previously hypothesized, this variation in premature mortality provides evidence that a health disparity in alcohol-attributable harms exists across races in the US. In particular, Native Americans, and to a lesser extent Black Americans, when compared to other races, have a higher standardized rate of alcohol-attributable mortality and PYLL. Thus, alcohol consumption can be seen as a main contributor underlying the known health disparities in the US 
. In addition to racial differences in alcohol-attributable mortality, men in every racial group experienced more than three times the amount of mortality when compared to women.
The observed differences in alcohol-attributable mortality across races may not be attributable to alcohol consumption alone as health-care utilization and the underlying population risk for alcohol-related diseases, injuries and conditions also impact on differences in the burden of alcohol-attributable mortality across races 
. Thus, effective interventions and policies aimed at addressing this disparity should address race differences in health plan coverage and health-care utilization for alcohol-related conditions 
, and should include race-specific interventions aimed at reducing the volume of alcohol consumed and deterring harmful alcohol consumption patterns 
This analysis has certain limitations, such as the quality of health outcomes data 
. Information concerning cause of death has long been seen as containing inaccuracies 
, and more recent studies still confirm considerable degrees of error in such information 
. Additionally, the exposure estimates for drinking status and binge drinking patterns used in our analysis were measured in 2001, whereas the outcomes were measured in 2005; however, the length of time between 2001 and 2005 should not greatly affect the alcohol-attributable mortality and PYLL estimates as alcohol consumption in the US remained relatively stable from 2001 to 2005 
. The estimates of alcohol consumption used in our analysis were also cross-sectional, i.e., measured more or less concurrently with deaths and PYLL, whereas long-term patterns of alcohol consumption impact the risk of some chronic diseases such as cancer 
Furthermore, our analysis did not include all aspects of harms to others (such as motor vehicle accidents, and assaults), which recently have been shown to constitute a large proportion of the burden of injury attributable to alcohol 
; this exclusion was due to an absence of a methodology to calculate these harms by race, age and sex 
We also did not estimate the number of deaths for people in the US over the age of 64 due to the unreliability of data relating to cause of death in the elderly 
. If the age group above 64 years had been included in our analysis, we estimated that the number of deaths and PYLL attributable to alcohol would have increased from 55,974 deaths and 1,288,700 PYLL to 82,213 deaths and 1,557,030 PYLL. Our analysis focused on premature mortality; however, in a case where all alcohol-attributable mortality is examined, injury may play a smaller part, as its role in causing death in people 65 years of age and older is relatively smaller than its same role in younger age groups 
This analysis was based only on race and does not provide alcohol-attributable estimates by ethnicity or socio-economic status. Alcohol consumption has been shown to vary by both ethnicity and socio-economic status 
and, thus, the alcohol-attributable harms are expected to vary by these variables as well 
. The exclusion from this paper of analyses of these variables was due to the unavailability of data differentiated by ethnicity and socio-economic status.
This analysis is also based on RR functions that were usually differentiated by sex and adjusted for age and smoking status, and in some cases for a variety of other risk factors. While the use of adjusted RR functions may introduce bias (see 
) most of the published literature on risk factors only report adjusted RRs and, thus, they are the basis of almost all comparative risk analyses (
; for alcohol see the publications outlined in Appendix S2
). For the risk estimates for alcohol in particular, most analyses show no marked differences after adjustment for the usual confounders and effect measure modifiers tested (see 
, and the meta-analyses cited there). However, there may be a need for adjustment to the RRs for alcohol if future research indicates that other dimensions of alcohol consumption, such as irregular heavy drinking occasions, impact the risk estimates.
For our analysis, we corrected the survey estimates of consumption so that the coverage of the alcohol consumption data used was equal to 80% of the US per capita consumption for 2005 (the per capita consumption of alcohol in 2005 was 9.5 liters of pure alcohol per person). If we had not triangulated the survey data based on total adult per capita consumption, the survey coverage rate would have been 49.7% for 2005. If unadjusted alcohol consumption survey data were used to calculate the burden of alcohol consumption, we estimated that alcohol would be responsible for 49,788 deaths and 1,120,740 PYLL for people aged 15 to 64 years. These results are similar to those which we calculated using a coverage rate of 80%, where it was estimated that 55,974 deaths and 1,228,700 PYLL were attributable to alcohol consumption.
The incomplete coverage of per capita
consumption in the NESARC 2001–2002, typical of survey-based consumption estimates, may have been due to disproportionately high levels of consumption among non-responders and to not capturing in the sampling frame people who were homeless and not living in shelters (about half of all people who are homeless in the US) 
. This may be a concern as a relatively small proportion of the population is responsible for the majority of the alcohol consumed. For instance, in the NESARC 2001–2002 sample, 6.7% of White male drinkers consumed 33% of the overall consumption, so excluding or undersampling of small groups with high consumption may result in a large degree of undercoverage 
. However, given that the unsheltered, homeless population represents a small fraction of the total population (0.1% of those people 15 years of age and older), their inclusion in the NESARC 2001–2002 would have increased the survey coverage rates by less than 1% (on the basis of the assumptions in Shield and Rehm 
). Thus, almost all of the undercoverage results from incomplete reporting of consumption among survey respondents and disproportionately high levels of consumption among the non-responders who were part of the sampling frame.
It should be noted that our analysis did not take into account morbidity attributable to alcohol consumption. As alcohol consumption has a greater impact on morbidity (as measured by Years Lived with Disability (YLD), a metric which combines the duration lived with a disease or injury and the severity of the disease or injury) than on mortality or premature mortality (as measured in PYLL), metrics such as Disability Adjusted Life Years (DALYs) (a measure that combines PYLL and YLD) are required to accurately characterize the burden of alcohol consumption.
Danaei and colleagues estimated that alcohol consumption was responsible for 64,000 deaths (45,000 for men and 20,000 for women) for all ages in the US in 2005 
. These estimates are substantially lower than our study’s estimates of 82,213 deaths (61,539 for men and 20,674 for women) for all ages. Our updated estimates of the burden of alcohol consumption show that alcohol is a greater risk factor for mortality in the US than was previously thought. Differences between our estimates and those of Danaei and colleagues may be explained by our use of better modeling methods for alcohol consumption and its associated risks, and our use of alcohol consumption data corrected for undercoverage. In addition, we included alcohol-attributable causes of death not included in the study by Danaei and colleagues, such as infectious diseases 
Comparison to Other Risk Factors
Our updated estimate of the burden of alcohol consumption in terms of mortality for the US is still lower than the burden estimated for tobacco, and poor diet and physical inactivity 
; tobacco use was responsible for an estimated 435,000 deaths, and poor diet and physical inactivity were responsible for an estimated 400,000 deaths in the US in 2000 (there were no age restrictions used when calculating these estimates.