Skip to main content

Beer in fridge

Harmful alcohol consumption is known to cause several diseases and many deaths. A study published in the Lancet this August has presented the latest figures, pertaining to the year 2016.1 The authors estimate that alcohol answers for approximately 7% of all male and 2% of all female deaths, as well as 6% of male and 2% of female disability-adjusted life year losses. Moreover, respective estimates were presented for all countries over several years. But these estimates seem to be too high.

This study has attracted plenty of criticism already – including a scathing blog post by David Spiegelhalter. He pointed out that the absolute risks, not reported in the paper, were small, and based on a rather opaque statistical model. I agree. But there is more to take issue with. I shall start with the question of overestimation of alcohol-related harms, then discuss several unsupported claims made by the authors.

Death estimates

Let’s start with the alcohol-related deaths. Two methods provide the data needed to add up the total. The first one is to add up all underlying cause-of-death diagnoses with mention of alcohol in the mortality statistics, e.g. alcoholic liver disease or alcoholic cardiomyopathy. One source of overestimation is that these deaths are assumed to be 100% alcohol-related. It goes without saying that alcohol is a necessary cause for these cases. But it seldom is a sufficient cause. It is known that other risk factors play a role. For example, the risk of alcoholic liver disease is increased by obesity and proton-pump inhibitor medication.

For those diagnoses in mortality statistics with no mention of alcohol, another method is needed. First, relative risks are derived from published studies. These are then applied to population categories grouped by the level of alcohol consumption to calculate the number of deaths that are due to alcohol. A source of overestimation exists because of the unreliability of self-reports on alcohol intake.

Self-reports on alcohol intake are known to underestimate consumption approximately by 50-70% on average. The error varies between individuals, and some people may even overestimate their actual consumption. The Lancet study extrapolated the population group consumption data so that the total approximated the respective country per capita alcohol figures in the statistics. This was used as an input to attributable risk calculations. However, similar “correction” was not made for the relative risk estimates. Thus, alcohol-related burden was overestimated.

This is perhaps best explained by a fictional example, shown in Table 1. It assumes that actual risk of death increases linearly by alcohol intake, that self-reported alcohol intake is always one-third of the actual one, and that population size is 1000 at all consumption levels. The “correction” results in gross overestimation of actual number of deaths.

TABLE 1 Overestimation of deaths (row f) due to correcting for under-reporting of drinks (row e) but not for risk of death (row c), assuming linear increase of deaths by number of drinks.

a. Population size 1000 1000 1000 1000 1000 1000
b. Reported no. of drinks per day 1 2 3 4 5 6
c. Deaths as % of population 3 6 9 12 15 18
d. Actual no. of deaths 30 60 90 120 150 180
e. Actual no. of drinks per day (corrected) 3 6 9 12 15 18
f. “Corrected” no. of deaths 90 180 270 360 450 540


Disability-adjusted life year estimates

A major aim of the Lancet study was to estimate the global health burden due to alcohol. The indicator of this burden is disability-adjusted life year (DALY). It is the sum of years of life lost and years lived with disability. One DALY equals one lost year of healthy life. The calculation is complex and contains several assumptions.2,3 There are two components. One is the number of potential life years lost due to premature mortality. To get this, one has to decide on an imaginary maximal life span, before which a death is considered to be premature. This span can vary from country to country and be different for men and women. To these lost life years you then add the product of disease duration (from onset to cure or death) multiplied by a disability weight. Weights vary from zero to one. For example, the weight for a severe disease may be 0.8, meaning that 80% of the time with the disease is considered to be time with a disability. The weights have been determined by a small group of experts.

Because deaths are a major component in the DALYs, overestimation of the DALY burden is also to be expected. I compared the present and earlier DALY estimates for the year 2005. The earlier figures were extracted from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) database in 2014.4 To my knowledge, the earlier database figures were uncorrected at that time. The earlier database figures showed the alcohol-related DALY burden to be 1.11 million in France in 2005. The Lancet study, however, showed 1.56 million DALYs for the same country in the same year. An increase of 450,000 DALYs is substantial.

Data from most countries, excepting highly-developed OECD countries, are likely to be inaccurate. For example, according to the WHO Global Health Observatory data repository, the percentage of heavy episodic alcohol drinkers (HED) in Turkey in 2010 was 0.2%, but that of alcohol dependence was 0.7%.5 HED was defined as drinking at least 60 grams or more of pure alcohol on at least one occasion in the past 30 days. Do we have to believe that the majority of alcoholics in Turkey (0.5% of the population) are moderate drinkers?

Since the data can be poor for many countries, the results of the Lancet study may have been based to some extent on guesstimates, imputation of missing data points and extrapolations. Extrapolation models may be needed when essential data are not available. In these cases, data of neighbouring regions or other time periods may be used.3

DALYs have been criticized for many weaknesses, including being patently invalid, unreliable and not comparable between and even within countries,6 having flawed assumptions and value judgements,7 failing to give little emphasis to the poor, low social class, and low treatment potential (Mont 2007).8 The disease component is blind to socio-economic differences between and within countries.

Relative risks

In the relative risk estimates from past observational epidemiologic studies, the focus was on self-reported alcohol consumption. This is an estimate of recent intake, usually in the last week or month. It is supposed to indicate long-term average alcohol intake, over all days whether drinks were consumed or not. Drinking rhythm, pattern, and speed were not studied, although they play a major role.

The Lancet study claimed that: “We found that the risk of all-cause mortality … rises with increasing levels of consumption …” Surprisingly, no data, charts or tables to support this claim were presented. An even more general claim than the above was that “the level of consumption that minimises health loss [emphasis added] is zero”.

It would have been important to see evidence on the risk of all-cause mortality since it is probably the most appropriate overall health-related outcome. Death is not subject to diagnostic errors, but the underlying causes of death and clinical diagnoses are. All-cause mortality is a relevant indicator of life expectancy, comprehensive (covering all diagnoses), and is easily understood. As Spiegelhalter has noted, in a very large recent study, all-cause mortality was lower among drinkers than among never-drinkers up to the level of 300g per week (Wood et al. 2018, eFigure 10, appendix).9 This is a far cry from zero.

The zero-level conclusion in the Lancet study was reached by logistic regression. Findings from a statistical model should not be interpreted as fact if data are sparse. This might be the case with respect to the outcomes of light drinkers. To avoid false conclusions, it would be better to compare light drinkers with never-drinkers than to fit a continuous curve, which may fit closely with the data at hand but does not allow for the detection of thresholds and has little generalizability.

Policy implications

The final sentence in the Lancet conclusion reads: “Policies that focus on reducing population-level consumption will be most effective in reducing the health loss from alcohol use.” No data, analysis or references were presented to support this forecast.

These policies rely on increasing the price, reducing availability and controlling marketing of alcoholic beverages. A composite indicator of these policies did not associate with alcohol-related DALYs in OECD countries in 2005. In regression analysis, the policy indicator, alcohol consumption and excise tax rate on alcohol were not significantly associated with DALYs.4

Moreover, a recent review has criticized the value of taxes as a cost-effective intervention to address alcohol-related harms (Nelson and McNall 2016).10 It thus seems that the above policies are not as effective as commonly believed.

To sum up, there is a lot of hot air in the Lancet study.

About the author

Kari Poikolainen is an adjunct professor in public health in the Department of Public Health, University of Helsinki, Finland. His main research interests are the causes and consequences of alcohol intake and other addictive substances.


  1. GBD 2016 Alcohol Collaborators. (2018) Alcohol use and burden for 195 countries and territories, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet, published online Aug 23. ^
  2. Homedes, N. (1996) The disability-adjusted life year (DALY), definition, measurement and potential use. Human Capital Development Working Papers, HCDWP 68. ^
  3. Devleesschauwer, B., Havelaar, A.H., Maertens de Noordhout, C., Haagsma, J.A., Praet, N., Dorny, P,, Duchateau, L., Torgerson, P.R., Van Oyen, H., Speybroeck, N. (2014) Calculating disability-adjusted life years to quantify burden of disease. International Journal of Public Health, 59(3), 565-9. ^
  4. Poikolainen, K. (2016) The Weakness of Stern Alcohol Control Policies. Alcohol and Alcoholism, 51(1), 93-97. ^
  5. Poikolainen, K. (2017) Does the Tail Wag the Dog? Abstainers, Alcohol Dependence, Heavy Episodic Drinkers and Total Alcohol Consumption. Alcohol and Alcoholism, 52(1), 80-83. ^
  6. Banerji, D. (2002) Report of the WHO Commission on Macroeconomics and Health: a critique. International Journal of Health Services, 32(4), 733-54. ^
  7. Anand, S., and Hanson, K. (1997) Disability-adjusted life years: a critical review. Journal of Health Economics 16(6), 685-702. ^
  8. Mont, D. (2007) Measuring health and disability. Lancet, 369(9573), 1658-63. ^
  9. Wood, A.M., Kaptoge, S., Butterworth, A.S. et al. (2018) Risk thresholds for alcohol consumption: combined analysis of individual-participant data for 599 912 current drinkers in 83 prospective studies. Lancet, 391(10129), 1513-1523. Please see eFigure 10, Supplementary appendix. ^
  10. Nelson, J.P., McNall, A.D. (2016) Alcohol prices, taxes, and alcohol-related harms: A critical review of natural experiments in alcohol policy for nine countries. Health Policy, 120(3):264-72. ^

Significance Magazine