Critique 127: Estimation of alcohol-attributable and alcohol-preventable mortality in Denmark — 6 November 2013

Eliasen M, Becker U, Grønbæk M, Juel K, Tolstrup JS.  Alcohol-attributable and alcohol-preventable mortality in Denmark: an analysis of which intake levels contribute most to alcohol’s harmful and beneficial effects. Eur J Epidemiol 2013; DOI 10.1007/s10654-013-9855-2

Authors’ Abstract

The aim of the study was to quantify alcohol attributable and -preventable mortality, totally and stratified on alcohol consumption in Denmark 2010, and to estimate alcohol-related mortality assuming different scenarios of changes in alcohol distribution in the population.

We estimated alcohol-attributable and -preventable fractions based on relative risks of conditions causally associated with alcohol from meta-analyses and information on alcohol consumption in Denmark obtained from 14,458 participants in the Danish National Health Survey 2010 and corrected for adult per capita consumption.  Cause-specific mortality data were obtained from the Danish Register of Causes of Death.

In total, 1,373 deaths among women (5.0 %of all deaths) and 2,522 deaths among men (9.5% of all deaths) were attributable to alcohol, while an estimated number of 765 (2.8 %) and 583 (2.2%) deaths were prevented by alcohol.  Of the alcohol-attributable deaths, 73 and 81% occurred within the high alcohol consumption group (>14/21 drinks/week for women/men).  A reduction of 50% in the alcohol consumption was associated with a decrease of 1,406 partly alcohol attributable deaths (46%) and 37 alcohol-preventable deaths (3%).  Total compliance with sensible drinking guidelines with a low risk limit (<7/14 drinks/week) and a high risk limit (<14/21 drinks/week) was associated with a reduction of 2,380 and 1,977 alcohol-attributable deaths, respectively.  In summary, 5.0% of deaths among women and 9.5%of deaths among men were attributable to alcohol in Denmark 2010.  The minority of Danish women and men had high alcohol consumption (16 and 26%).  However, the majority of all alcohol-attributable deaths among women and men were caused by high consumption (73 and 81%).

Forum Comments

It has long been realized that alcohol consumption is associated with both adverse and beneficial effects on health.  The present study attempts to judge the net effects of alcohol consumption in Denmark by calculating the occurrence of diseases “attributable to alcohol” and the occurrence of conditions that may be “preventable” by alcohol consumption.

Results from previous studies comparing harmful and beneficial effects of alcohol use in the population have varied widely.  For example, an analysis of deaths in Germany by Konnopka et al (Konnopka A, Hans-Helmut König H-H.  The health and economic consequences of moderate alcohol consumption in Germany 2002.  Value in Health 2009;12:253-261) concluded that there were more alcohol-preventable deaths than alcohol-attributable deaths.  Also, a report from the UK estimated much lower rates for alcohol-attributable deaths (White IR, Altmann DR, Nanchahal K.  Mortality in England and Wales attributable to any drinking, drinking above sensible limits and drinking above lowest-risk level. Addiction 2004;99:749–756).  Further, a previous report that included data for Denmark gave very different results for net alcohol effects in the Danish population, estimating that less than 1% of deaths were attributable to alcohol (Britton A, Nolte E, White IR, Gronbaek M, Powles J, Cavallo F, McPherson K.  A comparison of the alcohol-attributable mortality in four European countries. Eur J Epidemiol. 2003;18:643–51).

Considering all levels of alcohol consumption as a single parameter:  Reviewer Ellison comments that “As is often the case, the initial analyses of this paper consider ‘alcohol’ as the cause of the health effects, both bad and good.  Instead, what relates to health is the ‘consumption of alcoholic beverages by individuals.’  Grouping all alcohol consumption as a single exposure is of limited usefulness; it is preferable when the data are presented as the health risks and benefits of varying levels of drinking, and for different age groups.  The authors ‘back into this’ by stating that the large majority of the adverse effects occur among heavier drinkers.  Still, it is important to present data that can be used for setting policy, and for this it would be better to focus on percentages of good and bad health outcomes according to levels of consumption.

“While the authors point out that the minority of Danish women (16%) and men (26%) had high alcohol consumption, they state that 73 – 81% of alcohol-attributable deaths came from such drinkers.  In their abstract, they do not give the percentages of deaths in Denmark that would result if all drinkers followed sensible drinking guidelines (<7/14 drinks per week for women/men), but only state that there would be ‘2,380 fewer alcohol-attributable deaths.’”

The study did not have data on alcohol consumption previous to “the past 12 months,” so lifetime abstainers and former heavy drinkers who had stopped drinking would be in the same category (“No consumption”).  The study used only average alcohol consumption (which they had adjusted upward to reflect a calculated “under-estimation” of intake) to relate to the alcohol-attributable conditions, and could not factor in the pattern of drinking, including binge drinking.

Questions about the analysis:  The authors noted that the reported consumption by participants in this study was considerably lower than estimates from alcohol “disappearance” in Denmark (from sales, export, and import data); they therefore “adjusted” the alcohol intake of their subjects.  This led to fewer subjects in the light drinking categories and more with higher intake, which can be problematic since the relation between the specific amount consumed by an individual cannot be related directly to his/her risk of mortality.

Other questions relate to the lack of data on deaths from chronic pulmonary disease, dementia, and other common causes of death that have been shown to be reduced by moderate alcohol intake.  Were they included under “preventable conditions,” the net effects could be markedly different than those reported.

Further, it appears that under alcohol-preventable conditions, there is a reduction of 17.5% for diabetes (whereas epidemiologic data show the reduction to be 30%) and 13.7% for ischemic heart disease (the most common cause of death), while most studies show a greater protective effect.  The effects of alcohol on the risk of death are not presented by age group, whereas violent deaths are most common among the young (due primarily to binge drinking) while cardiovascular deaths are almost exclusively among older adults.  Presenting all ages together obscures such differences.

It is interesting that the percentage of deaths estimated to be attributable to alcohol vary dramatically from study to study.  A previous study from Denmark estimated less than 1.0% of deaths to be alcohol-attributable, and estimates from other European countries also vary markedly.  When one realizes how key the assumptions are as to the contribution to disease by alcohol, it is easy to see how large differences may occur.  For example, even modest changes in the proportion of cases of ischemic heart disease and diabetes that were estimated to be protected by moderate consumption would lead to very different results from this study.  This calls into question how important such estimations of effect can contribute to setting alcohol policy.

The authors’ estimates of effects if all individuals reduced their intake by 50% are questionable; if such a reduction occurred only among heavy drinkers there would be large beneficial effects, but if primarily among light-to-moderate drinkers (as has been reported to generally be the case) there may well be adverse net effects on the health of the population.

Forum member Zhang stated: “My major concern is the relatively high non-response rate (39%) of the survey sample.  In addition, the average alcohol consumption based on the survey sample is different from the per capita consumption based on sales, export and import data.  The way the authors dealt with this issue is to apply weights based on sex, age, civil status, ethnicity, education, income, working conditions, hospital admissions and medical visits.   This approach assumes that non-response can be predicted by all listed variables (i.e., a kind of missing by random).  I am not fully convinced that this approach will appropriately handle such a high non-response rate.  Other than that, the paper looks fine to me.”

Reviewer Djoussé commented: “While the computation of preventable and attributable fractions is reasonable, a major caveat is the inability to provide a measure of precision for the reported effect size (as the authors briefly acknowledged in the discussion).  It is difficult to interpret a 5% excess in deaths that are attributable to heavy drinking or prevented by moderate drinking without any knowledge about the lower bound of the 95% confidence limit.

“The conclusion appears extremely strong in light of the quality and limitations of the data at hand.  AAF and APF are estimated based on major assumptions including causal relation between alcohol use and evaluated outcomes, ability to completely eliminate alcohol use (AAF) in the population, absence of interactions between alcohol and other factors (i.e., statin use) on outcomes [here AAF is based on RR of alcohol main effects), etc.  For public health messages, a conservative approach would be to consider the generalized impact factor (GIF) that estimates excess outcomes based on realistic and achievable reduction in exposure prevalence in the population (i.e., reduction of heavy drinking by 10% or 15%, rather than eliminating heavy drinking).  Nonetheless, the paper contains valuable information on alcohol and health.”

Other comments by Forum members:  Stated reviewer Finkel: “This paper will provoke a lot of comment.  It comes from a distinguished source.  I am, however, worried about the 40 percent who didn’t participate, and I am uneasy about attributing some of the adversities to alcohol without information on individual cases.  That said, the logical take-away lesson seems in concord with what we know: moderate drinking is healthy; heavy drinking is dangerous.  Therefore, if it were possible to convert the heavy drinker to moderation, there would be large benefits.”

Forum member Skovenborg had a number of comments on the study.  “The high quality of the demographic data is made possible by the use of the excellent Danish Civil Register System, which since 1968 has kept a register of all people living in the country. This study shows the scientific and societal value of such high-quality national registries.  Many years ago sensible drinking limits were introduced to the Danish population and widespread knowledge of the drinking limits has been safeguarded by annual campaigns.  Some years ago the ‘moderate’ drinking limits [21 drinks (12 g alcohol) per week for men and 14 drinks per week for women] were supplemented with ‘low risk’ sensible limits (14 drinks per week for men and 7 drinks per week for women).

“The beneficial effects of a low to moderate consumption of alcohol have always been taken into consideration in the determination of ‘low,’ ‘moderate,’ and ‘high’ risk drinking.  Total abstinence has never been an official recommendation in Denmark, and it has been recognized that the evidence is missing for the argument that the protective effects of low-to-moderate alcohol consumption may be fully replaced by other health behaviours, e.g., physical activity.”

Reviewer Skovenborg continues: “The conclusion of the study is no surprise: the minority of Danish women and men report high risk alcohol consumption (16 and 26%); however, the majority of alcohol-attributable deaths are caused by high risk alcohol consumption (73 and 81%).  In contrast, only minor differences in the number of deaths were observed between low and moderate alcohol consumption.

“Among the weaknesses of the study are (1) the relative low response rate of 60.7% of the national survey used to calculate average consumption, and (2) the difference between the average alcohol consumption of survey participants and the adult per capita consumption based on information on sales, export and import data from Statistics Denmark.”

Reviewer Waterhouse added: “I think the health benefits of low and moderate alcohol consumption are not properly accounted for in this study, despite a suggestion of that in the title.  This seems to actually be a study of the health problems associated with high alcohol consumption.  Without a comparison between never consumers and moderate alcohol consumption, it is not possible to derive the public health effect of moderate consumption.”

Forum Summary

In an attempt to judge the harmful and beneficial health effects related to alcohol consumption, Danish scientists have carried out analyses comparing alcohol-attributable and alcohol-preventable mortality in Denmark.  They have used estimates of the potentially harmful effects of alcohol use on more than 20 diseases, giving 100% values to “alcohol use disorders,” although the specific causes of death are not known for this category.  Most of the other attributions for harm are realistic, but the alcohol preventable attributions for diabetes and ischemic heart disease appear to be low.  The authors conclude that 5.0% of deaths among women and 9.5% of deaths among men were attributable to alcohol in Denmark in 2010, with the majority of all alcohol-attributable deaths caused by high consumption.  They attribute only between 2 and 3% of deaths to be preventable by alcohol.

Previous estimates of alcohol-attributable and alcohol-preventable effects have varied widely, emphasizing the importance of the assumptions made by the investigators in any set of analyses.  For example, an analysis of deaths in Germany by Konnopka et al (Konnopka A, Hans-Helmut König H-H.  The health and economic consequences of moderate alcohol consumption in Germany 2002.  Value in Health 2009;12:253-261) concluded that there were more alcohol-preventable deaths than alcohol-attributable deaths and a report from the UK also estimated much lower rates for alcohol-attributable deaths (White IR, Altmann DR, Nanchahal K. Mortality in England and Wales attributable to any drinking, drinking above sensible limits and drinking above lowest-risk level. Addiction 2004;99:749–756).  Further, a previous report that included data for Denmark gave very different results for net alcohol effects, estimating that less than 1% of deaths were attributable to alcohol (Britton A, Nolte E, White IR, Gronbaek M, Powles J, Cavallo F, McPherson K.  A comparison of the alcohol-attributable mortality in four European countries. Eur J Epidemiol. 2003;18:643–51).  It is probable that different underlying assumptions of alcohol effects on various diseases are the prime reason for such differences.

There is no question that heavy alcohol consumption contributes to a large number of disease conditions, and the findings of this study emphasize the magnitude of the problem.  On the other hand, if the potentially beneficial effects of moderate alcohol consumption on many common diseases are underestimated, a net unfavorable result, as in the present study, is unavoidable.  It will be interesting to see what assumptions are made in other studies such as this, and how such assumptions affect the outcome of analyses.

*                      *                      *

Comments on this critique were provided by the following members of the International Scientific Forum on Alcohol Research:

Luc Djoussé, MD, DSc, Dept. of Medicine, Division of Aging, Brigham & Women’s Hospital and Harvard Medical School, Boston, MA, USA

R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA

Erik Skovenborg, MD, Scandinavian Medical Alcohol Board, Practitioner, Aarhus, Denmark

Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA

Andrew L. Waterhouse, PhD, Marvin Sands Professor, Department of Viticulture and Enology, University of California, Davis; Davis, CA, USA

Yuqing Zhang, MD, DSc, Epidemiology, Boston University School of Medicine, Boston, MA, USA

Gordon Troup, MSc, DSc, School of Physics, Monash University, Victoria, Australia

Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway