July 14, 2022

expert reaction to study on global alcohol consumption and population level risks

A study published in The Lancet looks at population-level risks of alcohol consumption by amount, geography, age, sex, and year.

Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“This is a new major publication on the global health consequences of alcohol use, from the Global Burden of Disease project. The project team last published a report on these matters in 2018 (reference 1 in the new paper), which reported on the position in the year 2016. The new report is about the position in 2020. So what’s changed?

“The most obvious change is that the 2018 publication concluded that the level of alcohol consumption that minimised health loss, at the level of the global population, was zero. That attracted some attention at the time because there had been many pieces of research concluding that there was a so-called J-shaped relationship between alcohol use and measures of mortality or other health burdens. That is, these other studies found that people who drank small amounts of alcohol had a slightly smaller risk of early death or disability than people who drank no alcohol at all. These studies still concluded that people who drank larger, or even moderate, amounts were at higher risk than non-drinkers, and there were also several studies that did not find a J-shaped relationship but instead concluded that the risk of early death or disability rose from the first drink. But, nevertheless, there was a certain amount of surprise that the 2018 Global Burden of Disease publication concluded that there were adverse health consequences, albeit small ones, on drinking any non-zero amount of alcohol.

“The new publication does not make that conclusion. At the level of the population of the whole world, the researchers now conclude that there is some evidence of a J-shaped relationship, though the amount of alcohol consumption that gives the lowest risk is very small. (More on this at the end of my comment.) But, more importantly, the researchers now say that it is not so appropriate to average out these risk measures across the whole world, so that instead they concentrate on giving separate estimates for different age groups, for men and women separately, and for different regions and countries.  That makes it difficult to summarise in a few numbers what they found – but essentially that’s because they are now saying that the picture is complicated and varies a lot from place to place and between ages.

“The reasons for these changes are actually summarised well in the linked Comment by Burton and Sheron in the Lancet, which is quoted in the press release. The data are updated from 2016 to 2020, and there have been some revisions to the estimated curves that describe the risks from some specific diseases of drinking different amounts of alcohol. But, as Burton and Sheron make clear, those are not the main reasons for the differences. Instead a new method has been used to put together the risks from different diseases and causes of death and disability, in relation to alcohol consumption, and most importantly (in my view) the position looks different because the report concentrates on figures for separate age groups, sexes and regions instead of concentrating on broad global or regional figures.

“Because, globally and for separate regions of the world, the researchers found a J-shaped relationship between alcohol consumption and risk in older age groups, a lot of what they report is concerned with two separate alcohol consumption figures. They report the ‘Theoretical minimum risk exposure level’ or TMREL, which is the daily level of alcohol consumption that produces the minimum risk of early death or disability. It’s ‘theoretical’ because it comes from statistical modelling – you can’t measure these things directly. For a given region, age group and sex, if the risk curve is J-shaped, the minimum risk will correspond to an alcohol consumption above zero. The researchers also report what they call the ‘Non-drinker equivalence’ or NDE. That’s the daily alcohol consumption that would give the same level of risk as the experience by a non-drinker. (This should also really be called ‘theoretical’ as well, because it also comes from statistical modelling.) So, where there is a J-shaped relationship, people drinking at the NDE level have (theoretically) the same risk of early death or disability as do non-drinkers; those drinking an average daily amount between zero and the NDE will have a risk below the risk experienced by non-drinkers, with those who drink at the TMREL having the minimum risk; and those drinking more than the NDE level have a higher risk than do non-drinkers.

“That’s the position when there is a J-shaped relationship – but the relationship is not J-shaped for some regions, age groups and sexes, particularly for young men. In these cases the risk starts increasing right from zero consumption, so if people in the populations in question drink any alcohol at all, their risk will be above the risk run by non-drinkers.

“It probably seems obvious why there might be differences between men and women in the risk of drinking a given amount of alcohol, since (on average) male and female bodies react rather differently to alcohol. The reason why there are age differences in the risk curves, and so in the TMREL and NDE, is also because bodies of different ages react differently on average. In particular it’s largely because, if you look at the curves describing the association between risk from specific causes or disease and alcohol consumption, most of them are not J-shaped, but a few quite clearly are – for example the curves for heart disease, strokes, and type 2 diabetes. Those are all conditions that mostly affect people in middle age or older. So the risks to health from drinking alcohol in young people generally come from causes where there isn’t a J-shaped association, so the overall risk pattern for young people won’t be J-shaped either. But for older people, a greater part of their risk comes from diseases with a J-shaped curve, so their overall risk curve is J-shaped too.

“It’s perhaps a bit less obvious why the TMREL and NDE should vary between regions of the world, for a given age group and sex. It’s essentially for the same reason as the variation with age. The prevalence of different diseases varies a lot between regions. In regions where heart disease, strokes and type 2 diabetes are particularly common, particularly in people in middle age, the J-shaped associations corresponding to those diseases will tend to dominate the overall picture, and so the TMREL and NDE will be greater than in another region where those diseases aren’t such important causes of early death.

“And in turn, it seems to be because the TMREL and NDE vary so much between age groups and regions that the researchers have chosen to give prominence to those figures, rather than the averages for the whole world.

“The researchers do not go into any detail on how their TMREL and NDE figures might be used to set recommended limits for alcohol consumption, though they do argue that the limits should take into account age and region – though not, they suggest, sex, on the grounds that, within each age group and region, the TMREL and NDE do not in fact differ very much between males and females. And the detail of how to set limits, even given all this new information, isn’t obvious anyway. As Burton and Sheron say in their comment, “The finding that no drinking minimises health loss among some groups is not synonymous with a recommendation for abstinence. But the finding that small amounts of alcohol might confer a lowered health risk is not synonymous with a recommendation for drinking either.”

“Because of the variation between regions, we do need to take some care in interpreting the figures given in the press release in the UK context, or indeed the context of other specific countries. For instance, the press release says that “For individuals over 65 years in 2020, the risks of health loss from alcohol consumption were reached after consuming a little more than three standard drinks per day.” In the terms used in the research paper, the global average NDE (Non-drinker equivalence) was a bit over three of the standard drinks used in this research per day. But that’s a global average, and the figures vary by region. For the Western Europe region, which includes the UK, the NDE daily level seems to be slightly less. (I can’t give exact figures for the over 65s taken together, since they are not given in the paper, but the numbers for narrower age groups are in Figure 3 in the paper.). But for some other regions, notably Central Asia and Eastern Europe, the NDEs for older people are considerably higher than the global averages.

“Another thing to be careful of in interpreting the numbers in the paper and the press release, is that the ‘standard drink’, referred to in this research, is bigger than a UK standard alcohol unit. The standard drink in this research is 10 grams of pure alcohol (ethanol). A UK alcohol unit is defined in volume terms (10ml of pure alcohol), and corresponds to 8 grams of pure alcohol. So a standard drink’ in the terms of this research is 1.25 UK units.

“I’d take issue with one sentence in the press release – the first one in its main body, that reads, “Young people face higher health risks from alcohol consumption than older adults, according to a new analysis published in The Lancet.” At least it needs a bit of interpretation. It’s true that the way that risk varies with alcohol consumption is different at different ages, and it’s also true that more individuals in young age groups are drinking more than the non-drinker equivalent than for older age groups. But the personal risk seems not to be higher at younger ages. This new research does not give figures for the number of deaths attributable to alcohol consumption at different ages, but the global overall picture is not hugely different from what was in the 2018 report. There, globally, it was estimated that there were (in 2016) about 150,000 deaths of people aged 15-29 attributable to alcohol, compared to about 900,000 deaths of people aged 60-74. Overall, the risk from alcohol is considerably higher for the older age group, but the way the risk varies with alcohol consumption is different at different ages.

“Just to clarify both that it is not completely pointless to refer back to the 2016 estimates, as I’ve just done, and to emphasise that the most important difference between this new report and the one published in 2018 is the concentration on results for age groups and regions rather than the overall picture, I’ll mention that the new report says “Our results are consistent with previous findings at a global level.” Also, they go into some detail in comparing the two publications, in Table 2 of the new paper. This does indeed show that the changes in the risk curves, and the update to 2020 data, would make the overall curve slightly J-shaped, but with risk below the risk level for non-drinkers only for very low daily consumption figures.”

Dr Colin Angus, Senior Research Fellow, Sheffield Alcohol Research Group, University of Sheffield, said:

“The idea behind this paper is sound – estimating how the relationship between alcohol consumption and risk varies between countries depending on their demographic composition, patterns of alcohol consumption and the prevalence of a wide range of alcohol-related health conditions is a valuable research goal and the analysis itself appears to be well-conducted. However there are a number of serious problems in the way the authors have interpreted their results. The most significant issue is in the headline interpretation that younger age groups should have lower drinking guidelines, which is not supported in any way by the study itself. The analyses presented in this study focus only on relative risk within age groups, but say nothing whatsoever about absolute risk. Younger people are, on average, much less likely to become ill or die from any cause than their older counterparts. Assessing how we should set drinking guidelines or prioritise interventions to minimise the total harm of alcohol requires an assessment of absolute, not relative risk. The GBD’s own figures suggest that there are over 14 times as many alcohol-attributable deaths in the UK among 70-74 year-olds than 20-24 year olds, which rather contradicts the assertion in this new study that we should focus on the drinking of younger age groups. There may be valid reasons to target younger drinkers for public health interventions, but those arguments are not presented in this study.”

Dr Tony Rao, Visiting Clinical Research Fellow, Institute of Psychiatry, Psychology and Neuroscience, King’s College London (IoPPN), said:

“Although the press release summarises the main findings of the study, the conclusions quoted by the senior author that “older people may benefit from drinking small amounts” is inaccurate and does not take into account the shortcomings in the way that the data was measured and interpreted. This study focussed on cardiovascular disease, which is known to be more common in older people.

“The study is an update to a similar review published using a similar but different approach to assessing the impact of alcohol in health across different countries. Although the conclusions are based on recognised ways of alcohol consumption, these are different from the GBD review published in 2018.

“The work does not fit with existing evidence, particularly for alcohol related harm in older people in the UK. The proportion of over-65s in the UK drinking alcohol over the previous 12 months has increased by 11% over the past 30 years – 4 times higher than in Western Europe. For over-55s in England, rates of alcohol related deaths and alcohol specific admissions have risen more sharply than other age group over the past 15 years. As a risk factor for Disability Adjusted Life Years (DALYs) – the measure used in the current study – alcohol had risen most sharply in people aged 55 and over the past 30 years. 

“There are major flaws in this study, two of which stand out. Firstly, there are numerous references to the lower risk of cardiovascular disease compared with non-drinkers with small amounts of alcohol consumption. The authors also point out that this disease burden is more common in older populations compared with younger ones – where alcohol-related injuries predominate. It does therefore not compare like with like, where different thresholds for harm may be present for different disorders. The second flaw is the use of relative risk as measure of outcome and interpretation. Relative risk does not give an indication of absolute risk, from which a more precise estimate of how many more people are affected by ill health from drinking can be calculated.

“The study does not adjust for confounders – factors that could affect both risk and outcome. There was no assessment of relative risk for alcohol use disorders and no assessment of mental health outcomes such as depression, anxiety and dementia. It also used self-reported alcohol consumption , which is known to vary.

“There are also 2 major limitations. Firstly, there was no “dose-response” assessment matching different levels of alcohol intake to different health outcomes. Secondly, the study did not include frequent episodic “binge” drinking, which is known to occur in up to 1 in 5 drinkers. 

“The elephant in the room with this study is the interpretation of risk based on outcomes for cardiovascular disease – particularly in older people. We know that any purported health benefits from alcohol on the heart and circulation are balanced out by the increased risk from other conditions such as cancer, liver disease and mental disorders such as depression and dementia. The study also fails to consider why using non-drinkers as a comparison group may be misleading. This group also includes those who have stopped drinking from poor health and those who choose not to drink because of poor health. If we are to simply draw the conclusion that older people should continue or start drinking small amounts because it protects against diseased affecting heart and circulation – which still remains controversial – other lifestyle changes or the use of drugs targeted at individual cardiovascular disorders seems like a less harmful way of improving health and wellbeing.”

‘Population-level risks of alcohol consumption by amount, geography, age, sex, and year: a systematic analysis for the Global Burden of Disease Study 2020’ by the GBD 2020 Alcohol Collaborators was published in The Lancet at 23:30 UK time on Thursday 14th July.

Declared interests

Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee.  My quote above is in my capacity as an independent professional statistician.”

Dr Colin Angus: “No conflicts to declare.”

Dr Tony Rao: “No conflict of interest to declare other than receiving a fee for speaking on alcohol misuse in older people at a Pharma sponsored educational event, but which did not contain any reference to the drug manufactured by that company.”