In 1995, a report entitled “Sensible Drinking” was published by the UK Department of Health. In this report, the recommended maximum alcohol intakes were controversially raised to 4 units per day for men and 3 units per day for women. This equates to 28 and 21 units per week for men and women respectively but the expert group chose to set daily maxima to discourage binge drinking i.e. drinking a lot of units in a single day even if the weekly maximum is not breached. During pregnancy women were advised not to drink more than one or two units once or twice a week. This report even implied that elderly non-drinkers at risk of heart disease might consider consuming a little alcohol for health reasons.
In 2016, The UK’s Chief Medical Officers reduced the recommended maximum intakes to no more than 14 units per week for men and women and suggested that, to be on the safe side, women should avoid alcohol altogether when pregnant. This report implies that regular consumption of alcohol at almost any level is associated with some small long-term risk but at the recommended levels the risk is low. Anyone drinking a pint of cider or premium beer each night or having a large glass of wine with their evening meal each night would breach these new guidelines.
In this post, I will try to put these recommendations into the context of current levels of alcohol consumption and give my brief assessment of the harms and benefits of consuming alcohol. I will discuss how apparently independent expert committees, albeit 2 decades apart, can come up with such different conclusions and recommendations about something that has been investigated intensively for many decades. These alcohol reports and recommendations also illustrate the difficulties of trying to make detailed assessment of risks and quantitative consumption recommendations based upon largely imprecise observational measurements. In a previous post , I used a quote from two senior scientists at the US National Institute of Statistical Sciences:
“Any claim coming from observational studies is most likely to be wrong”.
What are units of alcohol?
The technical definition of a UK unit of alcohol is 10ml or 8g (two standard teaspoons) of pure alcohol. This translates into:
- A 25ml measure of spirits
- 75ml of “standard” (13%) wine
- About half a pint of “standard” (4%) beer.
This means that a unit is actually small in comparison to the portion sizes often used, especially by those drinking at home. The 25ml of spirits is the measure that has historically been used in pubs but some outlets now use larger measures and someone fixing a gin and tonic or whisky and water at home could easily use three times this amount. Few people would regard 75ml (a tenth of a bottle) as a normal glass of wine; 6 glasses per bottle i.e. 125ml per glass would be more usual and a “large glass of wine” might be 250ml. A couple of glasses of wine per night sounds OK but this would almost certainly breach the guidelines and in practice would probably be nearer 24 units per week and maybe close to double this. Essentially the same arguments apply to spirits drunk at home; most cider and premium beer has more than 4% alcohol.
How much do British adults drink?
The Health Survey for England (HSE) is an annual assessment of health-related measures and behaviors in a representative sample of the English population (similar surveys are conducted in the other home nations). The survey asks participants about their use of alcohol and tries to estimate their usual weekly intake and their maximum consumption on a single day in the previous week. Table 1 shows recent results from the HSE.
Table 1 Percentage of English men and women in various alcohol consumption categories
Alcohol units/week Men Women
Non-drinker (%) 17 22
<14 units 53 62
14-21 units 10 7
21-35 units 12 6
35-50 units 4 2
This survey also collects data about the maximum number of units consumed on a single day:
- 37% of men admitted consuming more than 4 units on a single day in the previous week and 22% more than 8 units.
- 27% of women consumed more than 3 units in a day and 12% more than 6 units.
Alcohol consumption tends to rise with age until late middle age and then declines as shown in table 2.
Table 2 Average alcohol units consumed per week in English adults by age and sex
16-24 25-34 35-44 45-54 55-64 65-74 75+
Men (units/week) 12.8 13.7 16.9 17.0 19.5 17.3 12.9
Women 8.0 7.8 9.0 11.8 10.2 8.2 6.5
In the National diet and Nutrition Survey (NDNS), a representative sample of the UK population record their food and drink intake in a food diary as they consume it. I have converted the data collected in the latest NDNS survey to estimated average alcohol consumption in units per week:
- All men aged 19-64 – 14.7 units/week
- Men aged 19-64 excluding non-drinkers – 27.7 units/week
- All women aged 19-64 years – 7.1 units/week
- Women aged 19-64 years excluding non-drinkers – 15.3 units/week.
People earning over £40,000 are more likely to be frequent drinkers and to binge on their heaviest days. Young people (16-24 years) are the age group most likely not to drink but when they do drink, their drinking on their heaviest day tends to be higher than other age groups (see).
There is a large body of evidence that when using self reporting methods people substantially under report their calorie intake. When the calorie intake recorded in NDNS food diaries was compared to that suggested by a high technology analytical method (doubly-labelled water) then the intake indicated by the objective method was 50% higher than that indicated by their food diary and almost every individual under reported in their diary. It is almost certain that similar and substantial under reporting also occurs with alcohol consumption surveys. Alcohol intakes suggested by national surveys only accounts for 40-60% of alcohol sales. Under reporting is likely to be higher in surveys like the HSE which use recall of past consumption as their method of assessing intake.
This means that real intakes are likely to be higher than those recorded in consumption surveys. This has major implications for epidemiological studies that relate reported alcohol intake to disease and mortality risk. If there is substantial under reporting this will make it appear as if any harmful effects of alcohol occur at a lower dose than is actually being consumed. Any attempt to correct either calorie or alcohol intake for under recording introduces another source of uncertainty and in the case of alcohol means that it may little better than an educated guess.
The economic impacts of alcohol
This is the title of a fact sheet produced by the Institute of Alcohol Studies.They suggest that a figure of around £21billion has been widely used as the estimated annual costs of alcohol-related problems in England and Wales. This is roughly divided into:
- £3.5 billion cost to the NHS of alcohol related health problems
- £11billion as the cost of alcohol-related crime
- £7billion as the cost of lost productivity caused by alcohol.
These cost estimates do not include the private costs to the individual drinker and their family, like the suffering caused by alcohol-related illness and the loss of income to the individual or family. Neither do they include intangible costs like the fear of crime or the impact of premature death. They suggested that the average cost of alcohol to society in high-income countries amounts to around 2.5% of GDP. For England and Wales the estimated cost was substantially below this international average but higher in Scotland at 3.4% of GDP. This may explain why, in May 2018, the devolved Scottish government finally introduced a minimum 50p per unit price for selling alcohol. This rule is intended to impact mainly on the sale of cheap strong beer and cider and cheap spirits and fortified wines; it is not an extra alcohol tax because the extra revenue is retained by the retailer. It has been suggested by Scottish government sources that this measure will result in 58 less alcohol-related deaths in its first year and 1300 fewer hospital admissions. Legal wrangling delayed the introduction of this measure for six years after it was first passed by the Scottish parliament.
According to self-reports in the HSE, 70% of English men and 84% of women are within the current guidelines which are at their lowest ever level. Most of those who exceed these guidelines may be taking a risk with their long-term health but are not responsible for acute hospital admissions for alcohol poisoning or the wider societal problems like alcohol-related crime; alcohol consumption is highest in the relatively affluent sections of the population.
Despite its title (The economic impacts of alcohol), this Institute of Alcohol Studies factsheet is almost entirely focused upon the negative economic impacts of alcohol upon society and essentially ignores any economic benefits of alcohol consumption that partially compensate for the negative effects:
- Alcohol duty for the past year raised around £11 billion for the UK exchequer
- The Scotch whisky industry was one of Scotland and the UK’s biggest net exporters of manufactured products at £3.7 billion net and it generates around 40,000 UK jobs
- 600,000 people work in British pubs
- Exports of French wine and spirits amount to 12.9 billion euros and 600,000 French jobs are provided by their production, including around 140,000 growers.
Moderate alcohol use also has intangible benefits like the pleasure gained from drinking and the social and cultural roles of alcoholic drinks. Most cultures from around the world have found some means of producing alcohol. Guidelines can help people to make decisions about their behavior and the risks they are prepared to take but they can also cause guilt and fear to be unnecessarily associated with a normal pleasurable activity. Unrealistic guidelines may be counter productive and encourage people to ignore them completely.
This is not meant to down play the harm caused by inappropriate or excessive alcohol consumption but to try to give a more balanced view of the positive and negative economic and social impacts. The purely economic case against moderate alcohol consumption is not clear although there is an overwhelming case for trying to minimise inappropriate (e.g. drinking and driving) and excessive alcohol use.
The so-called “J-curve” of alcohol and mortality
For several decades there was a general belief that alcohol consumption at low levels reduced the risk of death, primarily by reducing the risk of heart disease. If alcohol intake was estimated in a large cohort of people and related to subsequent risk of death then the risk of dying was lower in those consuming a little alcohol compared to complete abstainers. As alcohol intake increased further so mortality risk increased and exceeded that in both the abstainers and low drinkers. The exact point at which alcohol intake turned from apparent benefit to harm varied from cohort to cohort depending upon things like the age, sex and nationality of the cohort. An example of such a J-curve is shown in figure 1; this was produced by Augusto Di Castelnuovo and his colleagues in 2006 (more than a decade after the 1995 Sensible Drinking report) and published in the highly rated journal Archives of Internal Medicine. This data was derived from a meta-analysis of 34 studies involving over a million subjects and almost 95,000 deaths. The green lines are confidence intervals which approximate to the potential maximum and minimum curves for each sex; the true line is thus believed to almost certainly lie somewhere within these limits.
Figure 1 An alcohol – mortality J-curve
Curves such as this underpinned the recommendations made in the 1995 Sensible Drinking report. This is not a selected “one off” paper but the consensus reached by many dozens of studies spread over several decades. All observational studies of this type involve a substantial amount of correction for other variables but up until recently it was generally thought that this was a real effect and not an artefact caused by the effects of other confounding variables. The following quote from the 1995 report addresses this point:
Recent studies have made it clear that the beneficial effects are not artefacts (phenomena produced by the analytical process) but strongly indicate a direct causal relationship.
One early and persistent criticism of this type of study was that the abstaining group would include many past heavy drinkers and even alcoholics or people who had given up or abstain from using alcohol because of ill-health – so-called sick quitters. The 1995 panel considered that many later studies had allowed for this possibility and they considered it could not be a major explanation of the apparent cardio-protective effect of alcohol and that they were:
Confident that the basic protective effect for CHD by alcohol is scientifically valid.
This widespread confidence in the cardio-protective effects of alcohol consumption led to whole fields of research into possible mechanisms by which alcohol might exert its cardio-protective effect. One highly favored explanation was that alcohol increased the good cholesterol (HDL) in the blood and this is known to be cardio-protective. This HDL mechanism was the most favored mechanism in the 1995 report. The beneficial effects of alcohol and more specifically chemicals like resveratrol, polyphenols and other antioxidants in red wine were proposed as being responsible for the so-called French paradox. Recorded mortality from heart disease in France appeared to be lower than might be predicted from the saturated fat and cholesterol intakes of French people. Resveratrol in particular gained a reputation as a major health-giving panacea even though the amounts present in red wine are so small as to be highly unlikely to confer any significant cardio-protective effect (see Dipak Das case-study )
The new 2016 Chief Medical Officers’ report would have had more data to consider than the 1995 report; a figure of 40 meta-analyses is mentioned in the 2016 report. Meta-analyses are supposed to be the consensus findings of all eligible studies published by the time it is conducted. One might thus expect all meta-analyses that address the same question at about the same time to come up with similar conclusions. This is often not the case because of different, relatively subjective, decisions made by different groups of authors when collecting and analysing their data (see post ).The newer report would have had more data to consider, but this would have used essentially the same methods as the earlier report and probably any new meta-analyses would have included much of the same data used in 1995 and in the 2006 meta-analysis used to construct figure 1. There was more data available in 2016 but what really new evidence was there to bring about such a shift of tone?
The mass use of statins and better treatment of heart disease would be factors that might affect any studies looking at risk for heart disease but researchers would have been aware of this and able to allow for it.
A Committee on Carcinogenicity report in 2015 provided a detailed analysis of the cancer risks associated with drinking alcohol. They concluded that alcohol consumption increased the risk of oropharyngeal, oesophageal, colon, liver, breast and pancreatic cancers. The concluded that this risk started at low rates of alcohol consumption of only 1.5 units per day (11/week) and that the risk increased with increasing dose and only slowly diminished in the years after cessation of alcohol drinking.
The link between alcohol and cancer is not a new one and increased deaths from cancer, liver disease and trauma have long been accepted as responsible for the rising mortality at higher alcohol doses in figure 1. The 1995 committee also accepted that there was almost certainly a causal and dose-dependent link between alcohol consumption and certain cancers. Heavy drinking of over 60 units/week caused a 3 to 15 fold increase in risk depending upon the site. They felt that at consumption levels of over 35 units/week the evidence of a link was convincing. They also felt that they could not exclude the possibility that at intakes below 18 units per week there might be some small increase in relative risk but that the number of cases attributed to low levels of alcohol would be very small.
The 1995 report concludes that half of deaths due to cirrhosis of the liver were alcohol-related but that the risk of cirrhosis for light to moderate drinkers was remote. They acknowledge that drink driving greatly increased the risk of a (fatal) accident and that one in seven road deaths involved a driver over the legal alcohol limit. Around half of pedestrians deaths in those aged 16-60 years also involved people who were above the legal alcohol limit for drivers.
The 1995 and 2016 conclusions with respect to these risks associated with alcohol do not explain to me why the recommended limits, especially for men, should be so different in the two reports. Where they do clearly differ materially, is in their views about the likely protective effect of alcohol for heart disease and thus the J-curve of mortality that it seemingly causes. Many cohort studies and amalgamations of cohort studies (meta-analyses) have yielded an apparent cardio-protective effect of alcohol and a mortality J-curve so it cannot be a simple mistake or aberrant finding. If it is now to be disregarded then this must be due to a change in the way the data is manipulated and interpreted.
An alcohol cohort study involves estimating alcohol consumption in a large sample of people, relating it to subsequent mortality and then correcting for likely confounding variables like smoking, obesity, diet and social class. Any residual extra or reduced risk is then attributed to an effect of alcohol. It has already been noted that the estimation of alcohol intake is usually a relatively crude process that is highly likely to underestimate the intake and thus potentially overestimate the risk of any given amount of alcohol. It would be very difficult to reliably correct for this probable underestimation in a large study. Likewise the measure of confounding variables is also often quite crude and so no matter how sophisticated the mathematics of the correction process it is subject to considerable degree of uncertainty. The decision about what potential confounders to try to correct for is also a matter of judgement and precedent. This process is at the core of many epidemiological studies and so it is not surprising that cohort studies that yield relatively small differences in relative risk (say 20%) are regarded sceptically by many people and why cohort studies addressing the same question may yield conflicting results. The alcohol J-curve has been reported in most cohort studies and meta-analyses.
It would be impossible for a single non-specialist to re-evaluate all of the data examined by the expert groups responsible for the 1995 and 2016 guidelines but I will just look at a couple of recent reports that may illustrate why these two expert groups produced such different recommendations.
Tim Stockwell and his international colleagues produced a meta-analysis in 2016 that specifically addressed the question of whether moderate drinkers have a reduced mortality risk. They did a meta-analysis of 87 studies involving almost 4 million subjects and 367,000 deaths and they initially produced the classic J-curve. Former drinkers had an elevated risk. When they adjusted for “abstainer biases” and “quality related characteristics” then this essentially removed the apparent benefit for low consumers. They concluded that low-volume alcohol consumption has no benefit when compared to lifetime abstainers or occasional drinkers.
Craig Knott and his colleagues writing in the BMJ in 2015, used data from the Health Survey for England to explore the association between alcohol consumption and mortality. They linked data from 10 annual cohorts of this survey to the national mortality register i.e. they identified which subjects had died and the listed cause. They used 4 categories of subjects: men or women aged 50-64 years or over 65 years. Once again they found that the crude data suggested a protective effect of alcohol across these age and sex groups. However when they adjusted for a range of personal, socioeconomic and lifestyle factors the apparent benefits were reduced. After the exclusion of former drinkers the effects were reduced still further and were only consistent in older women. They then did further analyses where they compared groups to people classified as occasional drinkers and any benefits of alcohol seemed to disappear.
Both of these groups of authors are essentially arguing that any apparent protective effect of alcohol is an artefact caused by previous generations of authors not fully correcting their results for confounding variables, study quality and the distorting effects of ex-drinkers within the non-drinking category. I am generally unconvinced when cohort studies and meta-analyses produce evidence of a small increase or decrease in relative risk of a disease or mortality when using a relatively crude measure of the key variable and a highly complex system of adjustment for confounding variables that are often only very crudely categorised. I am thus sceptical about this new analysis even though the technical mathematical modelling processes are outside my expertise; the quality of the end product of mathematical modelling and statistical analyses is dependent upon the quality of the original data and the assumptions. I think that this change of mind by “the experts”, despite both groups having huge volumes of data to work with, illustrates just how unreliable observational epidemiological methods are when looking for small protective or causal effects of crudely assessed dietary and lifestyle factors.
Summing up the alcohol debate
- Excessive use of alcohol has major adverse social and economic consequences.
- Most drinkers are moderate in their habits and although this may carry some small risk for their long-term health, they are not responsible for most of these economic and social costs.
- The adverse economic impacts of alcohol are substantially offset by major positive economic effects of the production and sale of alcoholic drinks.
- Alcohol affects judgment, co-ordination and mood so there are times when alcohol use is inappropriate:
- When at work, especially where there are health and safety implications; this also applied to many sporting and leisure activities (e,g, swimming, riding and climbing)
- When driving any vehicle; intoxicated pedestrians are also at much greater accident risk.
- Regular drinking and substantial single intakes should be avoided during pregnancy.
- Alcohol is acutely toxic at very high doses and a hangover is a warning that the person has drunk enough to cause some damage.
- Regular long-term consumption of substantial amounts of alcohol or regular alcohol binges increase the risk of cancer and liver disease.
- The apparent protective effect of alcohol on heart disease risk and total mortality found in scores of studies may be an artefact or it may be overstated.
- Even if the J-curve is an artefact, it provides some reassurance that any harmful effects of moderate alcohol use are insufficient to wipe it out and thus are likely to be small.
We need to be sceptical of observational studies that suggest that some lifestyle or dietary factor slightly increases or decreases the risk of a disease or mortality.
“Any claim coming from observational studies is most likely to be wrong”.
Some of the major successes of epidemiology involve quite readily assessable factors (e.g. smoking habits, asbestos exposure and infant sleeping position) and large, several fold, increases in risk.
Conflicts of interest
I have never received any financial award from any alcohol-related organisation. This does not mean that I am totally unbiased (is anyone?). I have always enjoyed using alcohol and many of the life events that I look back upon with pleasure have involved alcohol-use by at least some of the participants. My current drinking is usually modest on a week when I have no social engagements (around ten units per week) but would rise significantly above this on socially active weeks. My consumption in earlier years was significantly higher than this and at times in my youth well above any recommended levels.