Critique 082: A statistical model attempting to estimate the level of alcohol consumption that is “optimal” for health — 6 June 2012
Nichols M, Scarborough P, Allender S, Rayner M. What is the optimal level of population alcohol consumption for chronic disease prevention in England? Modelling the impact of changes in average consumption levels. BMJ Open 2012;2:e000957. doi:10.1136/bmjopen-2012-000957.
Objective: To estimate the impact of achieving alternative average population alcohol consumption levels on chronic disease mortality in England.
Design: A macro-simulation model was built to simultaneously estimate the number of deaths from coronary heart disease, stroke, hypertensive disease, diabetes, liver cirrhosis, epilepsy and five cancers that would be averted or delayed annually as a result of changes in alcohol consumption among English adults. Counterfactual scenarios assessed the impact on alcohol-related mortalities of changing (1) the median alcohol consumption of drinkers and (2) the percentage of non-drinkers.
Data sources: Risk relationships were drawn from published meta-analyses. Age- and sex-specific distributions of alcohol consumption (grams per day) for the English population in 2006 were drawn from the General Household Survey 2006, and age-, sex-, and cause-specific mortality data for 2006 were provided by the Office for National Statistics.
Results: The optimum median consumption level for drinkers in the model was 5 g/day (about half a unit), which would avert or delay 4579 (2544 to 6590) deaths per year. Approximately equal numbers of deaths from cancers and liver disease would be delayed or averted (~2800 for each), while there was a small increase in cardiovascular mortality. The model showed no benefit in terms of reduced mortality when the proportion of non-drinkers in the population was increased.
Conclusions: Current government recommendations for alcohol consumption are well above the level likely to minimise chronic disease. Public health targets should aim for a reduction in population alcohol consumption in order to reduce chronic disease mortality.
Background: In 1995, Duffy1 applied the all-cause mortality risk estimates from the British Regional Heart Study to the 1987 British population alcohol consumption data for men aged 45–64 years and estimated that there would be fewer deaths among drinkers than would have occurred in an abstinent population. In a meta-analysis in 1999, based on data from 20 cohort studies available at the time, White2 found that the lowest risk of total mortality among men in the UK was among those reporting an average of 12.9 units per week (95% CI 10.8–15.1). Britton and McPherson3 reported in 2001 that in England and Wales, alcohol consumption marginally reduces mortality at a population level. In their study, the favorable mortality balance from alcohol consumption was only found among men aged over 55 years and women aged over 65 years. In another report from 2003, Britton et al4 found that more deaths among both males and females were prevented or delayed than lost in England and Wales from the current drinking levels.
Choosing data upon which to base statistical modeling: Statisticians frequently use “modeling” to attempt to summarize data on the relation of alcohol intake to health outcomes. The investigators in this study used the macro-simulation model to estimate deaths from common chronic diseases that relate, either favorably or adversely, to alcohol consumption. Most Forum reviewers thought that this analytic method is unnecessarily complex. Said one, “This is an extremely complex analysis. In general, the simpler the statistical method, the more trustworthy the results. The complicated statistical method used in this paper depends on a large number of assumptions, yet it is not possible to test how valid most of these assumptions are.” Another Forum reviewer pointed out that while the original data that form the basis for such analyses are collected prospectively, the model is a retrospective analysis of the published retrospective meta-analyses of the original studies.
The results of analyses such as the one used in this paper are very much dependent upon the parameters chosen to reflect the risks and benefits of alcohol. In the present study, alcohol-related health outcomes were selected from the World Health Organization Global Burden of Disease ‘Global Health Risks’ report from 2009.5 Site-specific cancer data were derived from the World Cancer Research Fund Report from 2007.6 The estimates of age- and sex-specific risk ratios assigned in this study were based on meta-analyses or prospective cohort or case-control studies identified in the literature. In general, these sources were appropriate, but there were concerns by Forum reviewers that some of the estimates appear to have been taken from these publications at face value without close examination, as commented on below.
Effects of alcohol on liver cirrhosis: Estimates of effects on alcoholic liver cirrhosis, the most important single fatal chronic disease condition caused by alcohol consumption globally, come from a report by Rehm et al.7 That source is based on studies in the USA, Italy, Denmark, China, and Japan; there were none from the UK, although drinking patterns vary markedly across different countries. In the Rehm et al paper, women reporting 1-2 drinks daily had virtually the identical risk of liver cirrhosis as lifetime abstainers, and no explanation is given why a consumption of 0 – 12 grams of alcohol per day would translate into a RR of 1.9 (95% CI 1.1 – 3.1) for liver cirrhosis mortality in women. For men, relative to lifetime abstainers, men who consumed 1-2 drinks per day actually had a lower risk of liver cirrhosis [RR 0.3 (95% CI 0.2 – 0.4)] while their risk of liver cirrhosis mortality was increased by 60%: RR 1.6 (95% CI 1.4 – 2.0).7
Effects of alcohol on cancer: The data source for alcohol-related cancer mortality was the WCRF/AICR report.6 The data presented in the paper from that source gives the RR of mortality for colorectal cancer as 1.09 (1.03 – 1.14) per 10 grams of alcohol per day. However, the WCRF/AICR report states that increased risk of colorectal cancer is only apparent above a threshold of 30 g/day of ethanol for both sexes. The authors use the same reference for mortality for upper digestive cancer (mouth, larynx, pharynx): RR=1.24 (1.18 – 1.30) per drink per week. However, smoking is a well-known confounder of the alcohol-cancer association, and the “The Million Women Study” found no association between cancers of the upper aerodigestive tract and intake of alcohol in non-smoking women.8
Lack of adjustment for pattern of drinking: A major problem with this study is that the results are based exclusively on the reported average intake of individuals, and do not include data on the pattern in which such alcohol was consumed. For example, subjects consuming 7 drinks on one day of the week, and those consuming one drink/day, were included in the same alcohol category. It has clearly been shown that most health benefits are associated with regular, moderate intake, and such benefits are lost (and adverse effects predominate) with episodic (binge) drinking of large amounts on a single occasion. In many studies, the total amount of alcohol consumed is not as important as the pattern of drinking. As one Forum reviewer commented: “Considering that frequency of drinking is of critical importance because of alcohol’s generally short duration of health effects, it seems to me impractical to suggest that the population consume one-half a unit daily. Alcoholic beverages are not to be used as medicine, dispensed from a measuring spoon. Alcohol consumption is a cultural phenomenon.”
Presentation of the “J-shaped curve” for alcohol and health outcomes: The Forum contends that the implications of the overall association between drinking (or not drinking) alcohol and health are not presented clearly in the current paper. When examining the “J-shaped curve” between alcohol intake and mortality, as done by Di Castelnuovo et al9 for example, it becomes clear that while the nadir of the J-shaped curve relating alcohol consumption to mortality is indeed in the range of about one-half of a typical drink/day, the relative risk of all-cause mortality remains lower than that of abstainers up to a reported average of several drinks/day. In other words, the number of deaths would be expected to be lower among drinkers than among abstainers at all levels of alcohol intake until the “reversion point” is reached; in most studies, that point is in the range of 2 to 5 drinks/day, which is considerably different from the implications given by the authors in this paper.
It is interesting that the authors of the present paper state within the text that when various possible rates of abstinence in the population were tested, “Theoretically, optimal results were achieved when there were zero non-drinkers in the population, which resulted in 4,160 chronic disease deaths averted or delayed compared with 2006 mortality rates.” They add that such a situation (where everyone in the population consumed alcohol) would presumably increase the number of predicted deaths from cancer and liver cirrhosis, but that such an increase would be more than offset by averting cardiovascular disease deaths. These results are not reflected in the authors’ conclusions of the paper, which focus entirely on recommendations for reducing alcohol consumption.
Calculation of “optimal” level of alcohol consumption should be specific for age groups: While the analyses in the present paper were based on a comparison of disease outcomes according to non-drinking or to varying levels of intake at specific ages, the conclusions are given as a single figure for the entire population. Since almost all of the so-called “protective” effects of alcohol relate to the diseases of ageing (especially coronary heart disease, ischemic stroke, diabetes, and dementia), it is of little value to give a single number of drinks/week or drinks/day without taking age into account. Mixing data on drinking by the young (who frequently drink excessive amounts in binges) with those of the effects of regular moderate drinking in older people is not justified.
Calculation of “optimal” level of alcohol consumption should be gender-specific: Essentially all scientific studies have shown that women tend to show greater evidence of harm from smaller amounts of alcohol than do men, and the benefits are usually shown at lower levels of consumption among women. The authors recommend that the public health target should be to reduce median alcohol consumption to half a unit per day for both men and women. This suggestion ignores the considerable differences between men and women in the effects of alcohol on mortality risks from cardiovascular disease, liver disease, and most other chronic diseases.
Effects of underreporting of alcohol intake: In the present paper the authors mention the issue of underreporting of alcohol intake, but do not include the uncertainty that underreporting would have on their results. Drinkers in general tend to underestimate the reported quantity and frequency of their alcohol consumption. Comparison studies in the alcohol literature have suggested that self-reported alcohol consumption accounts for only 40–60% of alcoholic beverages sold as measured by sales and tax data.10
Development of guidelines for drinking: Of further concern by Forum members is that the authors of the present paper use only their statistical calculations as a basis for recommendations to the public. As has been discussed by Harding and Stockley,11 “The objective of guidelines is to influence and change behavior among target populations. It follows, therefore, that several factors then become relevant: behavior that is thought to be in need of change, the culture and mindset of the target populations, and the kind of message that is likely to be effective. There are some tensions between advice intended only to reduce the prevalence of misuse and that which also seeks to reflect the evidence on the beneficial health effects of moderate consumption. Providing messages about moderate drinking is the domain of policy makers, governments, and health experts.”11 Smallwood has pointed out that in translating scientific evidence into policy, a balance must be achieved between the desire to reduce alcohol misuse and the importance of reflecting accurately any beneficial health effects of drinking.12
1. Duffy JC. Alcohol consumption and all-cause mortality. Int J Epidemiol 1995;24:100-105.
2. White IR. The level of alcohol consumption at which all-cause mortality is least. J Clin Epidemiol 1999;52:967–975.
3. Britton A, McPherson K. Mortality in England and Wales attributable to current alcohol consumption. J Epidemiol Community Health 2001;55:383–388.
4. Britton A, Nolte E, White IR, Grønbæk M, Powles J, Cavallo F, McPherson K. A comparison of the alcohol-attributable mortality in four European countries. European Journal of Epidemiology 2003;18:643–651.
5. World Health Organization. Global Health Risks: Mortality and Burden of Disease Attributable to Selected Major Risks. Geneva: World Health Organization, 2009.
6. WCRF/AICR. World Cancer Research Fund and American Institute for Cancer Research. Food, Nutrition, Physical Activity, and the Prevention of Cancer: a Global Perspective. Washington, DC: AICR, 2007.
7. Rehm J, Taylor B, Mohapatra S, et al. Alcohol as a risk factor for liver cirrhosis: a systematic review and meta-analysis. Drug Alcohol Rev 2010;29:437-445.
8. Allen NE, Beral V, Casabonne D, Kan SW, Reeves GK, Brown A, Green J, on behalf of the Million Women Study Collaborators. Moderate alcohol intake and cancer incidence in women. J Natl Cancer Inst 2009;101:296–305.
9. Di Castelnuovo A, Costanzo S, Bagnardi V, Donati MD, Iacoviello L, de Gaetano G. Alcohol dosing and total mortality in men and women. An updated meta-analysis of 34 prospective studies. Arch Intern Med 2006;166:2437-2445.
10. Midanik, L. The validity of self-reported alcohol consumption and alcohol problems: a literature review. Br J Addict 1982;77:357-382).
11. Harding R, Stockley CS. Communicating through government agencies. Ann Epidemiol. 2007;17(Suppl):S98-S102.
12. Smallwood R. Communicating with the public: dilemmas of a chief medical officer. Ann Epidemiol. 2007;17(Suppl):S103-S107.
Scientists from Australia and Oxford University have carried out a complex analysis in an attempt to determine the “optimal” level of alcohol consumption that is associated with the lowest rates of chronic disease in the UK. They conclude that the intake of about one-half of a typical drink per day would result in the healthiest outcomes, and the authors conclude that the recommended alcohol intake for the UK should be reduced from the current advised level of drinking.
There were a number of concerns by Forum members about the paper. It is based on an extremely complex statistical model that requires many assumptions, most of which cannot be validated. The parameters chosen to use in such a model determine the results, and a number of instances were identified where the values used in this analysis do not reflect current scientific knowledge. Further, the conclusions of the authors are based exclusively on the lowest point of the “J-shaped” curve for alcohol and disease, and disregard the findings that the risk of death, in comparison with non-drinkers, remains lower for drinkers of alcohol of up to several drinks per day.
There were other aspects of the paper that were of particular concern: (1) focusing only on the average amount of alcohol consumed, while the pattern of drinking (regular moderate versus binge drinking) has the greatest effect on health outcomes; (2) giving a single recommended level of alcohol intake irrespective of age; the greatest risks and lowest expected benefits of alcohol intake are among the young, whereas most of the putative health benefits relate to the diseases of ageing; (3) giving a single value for both men and women, since women are known to react (both adversely and beneficially) to lower levels of alcohol than do men; (4) the use of their estimated value alone for making recommendations for the population; guidelines should be based on a large number of social and behavioral factors, not just on the results of one scientific study.
The level of alcohol that is most likely to be associated with the lowest risk of adverse health outcomes and the most likely health benefits varies markedly among individuals. Middle-aged men and post-menopausal women are most likely to demonstrate enhanced health (e.g., lower risk of cardiovascular diseases, diabetes, dementia) from moderate drinking. For all ages, binge drinking is associated with predominantly adverse outcomes. In general, women should drink less than men. While the analyses presented in this paper are of scientific interest, they alone do not support changes in the current population recommendations for alcohol consumption.
* * *
Comments on this paper have been provided by the following members of the International Scientific Forum on Alcohol Research:
Erik Skovenborg, MD, Scandinavian Medical Alcohol Board, Practitioner, Aarhus, Denmark
Yuqing Zhang, MD, DSc, Epidemiology, Boston University School of Medicine, Boston, MA, USA
Giovanni de Gaetano, MD, PhD, Research Laboratories, Catholic University, Campobasso, Italy
Andrew L. Waterhouse, PhD, Marvin Sands Professor, Department of Viticulture and Enology, University of California, Davis; Davis, CA, USA
Ulrich Keil, MD, PhD, Institute of Epidemiology and Social Medicine, University of Münster, Münster, Germany
Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA
Fulvio Ursini, MD, Dept. of Biological Chemistry, University of Padova, Padova, Italy
Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway
Francesco Orlandi, MD, Dept. of Gastroenterology, Università degli Studi di Ancona. Italy
Gordon Troup, MSc, DSc, School of Physics, Monash University, Victoria, Australia
R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA