Medicine, poison, poison, poison, ……

As of the 23rd May 2022 this website is archived and will receive no further updates. was produced by the Winton programme for the public understanding of risk based in the Statistical Laboratory in the University of Cambridge. The aim was to help improve the way that uncertainty and risk are discussed in society, and show how probability and statistics can be both useful and entertaining.

Many of the animations were produced using Flash and will no longer work.

Yesterday the Chief Medical Officer announced new guidelines for alcohol consumption. The Summary says,

The proposed guidelines and the expert group report that underpins them, have been developed on the basis of the following principles:

  • People have a right to accurate information and clear advice about alcohol and its health risks.
  • There is a responsibility on Government to ensure this information is provided for citizens in an open way, so they can make informed choices.

This admirable aim of treating the public as adults capable of making up their own minds seems in stark contrast with the authoritarian tone of publicity from the Department of Health, which baldly says “Men should not drink more than 14 units of alcohol each week, the same as women”.

Neither the press release nor the Summary of the guidelines provide any information on the effects of higher consumption, but Tables 10 and 11 of the full expert report provides the estimates plotted below.


Estimated lifetime risk from regular daily consumption of alcohol, as given in Expert Report: the half-bottle indicates roughly what 4.5 units means. The 'Guideline' is for 14 units over a week rather than 2 units a day, and the curves are slightly different if the 14 units are not evenly spread over the week

This, I feel, really does provide important information:

  • At one unit a day (half a standard glass of wine), they estimate a tiny overall benefit for men and a slightly larger, but still small, overall benefit for women. But the Department of Health appears very eager to avoid the message of “a small glass of something for medicinal purposes”.
  • However, after the first ‘medicinal’ unit, each subsequent one steadily raises the risk.
  • The ‘low-risk’ threshold is at a very low level, where the small benefit in heart disease and stroke has been overwhelmed by cancer and other harms.
  • There is a dramatically increased risk for higher consumption, with women having a steeper gradient than men.

A simple summary of these curves might be that units are medicine, poison, poison, poison,..... . As two units is a standard glass of wine, which seems to leave you roughly quits, maybe best only to have a mouthful. If you can stick to that.

Popular response to the announcement seems to have been, predictably, extremely varied. Although only 24% of men and 18% of women report drinking regularly more than the guidelines, many were vocal in the condemnation of the ‘nanny-state’. Perhaps if they had been given clear information as to the potential consequences of their habit, they might not so easily have been so dismissive, although it's limited what can be put over in a few soundbites.

Changing individual behaviour and the surrounding culture is an immensely complex and subtle issue. But I think there is an irreconcilable tension between providing information with the aim of enabling people to make an informed choice, and telling people what they should do. Rather than trying to do both in a single announcement, perhaps future public health communications should clearly separate these two processes.

Explaining a 1% lifetime risk

I got a surprising amount of coverage due entirely to submitting a quote to the Science Media Centre.

“These guidelines define ‘low-risk’ drinking as giving you less than a 1% chance of dying from an alcohol-related condition. So should we feel OK about risks of this level? An hour of TV watching a day, or a bacon sandwich a couple of times a week, is more dangerous to your long-term health. In contrast, an average driver faces much less than this lifetime risk from a car accident. It all seems to come down to what pleasure you get from moderate drinking.”

My comparators were chosen as daily activities that could be linked to all-cause mortality 'hazard ratios' of around 1.05, that is a 5% increased risk of dying each year. I used two measures:

  • One hour's TV a day: a large and reputable recent UK study found that 2 hours TV watching each day was independently associated with a hazard-ratio of 1.09 – I scaled this down to 1 hour giving a hazard ratio of around 1.05
  • A couple of bacon sandwiches a week: another study from EPIC estimated a hazard ratio of 1.19 from daily consumption of 50g of processed meat a day: a couple of bacon sandwiches a week is around 15g a day, so might be associated with a hazard ratio of around 1.05.

These hazard ratios cannot be directly compared with the metric of “lifetime risk”. But Figures 12 and 13 in the Expert Report give hazard ratios associated with alcohol consumption for different age-groups, and it is clear that for the guideline consumption the hazard ratios for older age-groups (where the main impact will occur) are well under 1.05. I openly acknowledge some fairly heroic interpolation of risk to low exposures of TV watching and bacon sandwiches, but I feel that these are reasonable indicators of the potential harms of sedentary behaviour and harmful diet.

For driving, I used the current GB estimate of 2 car-occupant deaths per billion miles – this means that a a 1% chance of death would require around 5,000,000 miles, or 100,000 miles a year over 50 years, a vast quantity of driving. Alternatively, we can simply observe that around 1 in 300 deaths is on the roads, as 1700 people were killed in Great Britain last year out of around 550,000 deaths. So transport accidents now give a lifetime risk of well under 1 in 100.

The main point I wanted to make was that people should be provided with information so they can make an informed choice. But maybe 'medicine, poison, poison, poison, poison..." is all people need to know.


Is it not true that the evidence comes from observational studies rather than randomised controlled trials? And is it not also true that cancer is a complex disease with many types and many causes. Would it not, therefore, be fair to criticise the evidence this advice is based upon as not having proven alcohol as the cause and that other confounding factors may be influential? For example, evidence is emerging of high sugar and carbohydrate consumption being linked to a number of cancers as well as metabolic syndrome. So, if sweet alcoholic drinks are consumed the risks may be higher than dry drinks. Could this be why women, who in my experience prefer sweeter drinks, have a higher risk? In the LCHF/Paleo lifestyles only red wine, whisky and other low-sugar drinks are advised.

Remember that the figure shows the risk per units of alcohol. Women might prefer sweeter drinks, but tend to be in general lighter (in weight), so that the dose consumed (i.e., alcohol unit) is distributed over less body weight, hence more pronounced effect.

If these figures as shown on David's graph, taken from Tables 10 and 11, are actually observed values, this is an extremely important guide as to safe alcohol consumption. However, the figures in Tables 10 and 11 give point estimates of lifetime risk for 14 groups (including both sexes and excluding the teetotal nulls) each to four significant figures, an extraordinary level of precision. Moreover, if you draw the figures in the tables up at a large scale, these very precise point estimates also lie exactly on two very smooth curves from 7 units upwards. Given the precision and the smoothness of the curves these cannot possibly be the means of observed risks by group. They must be the calculated outputs from a global fitted model, reported to an inappropriate level of precision, or the result of a later smoothing procedure. If (as one hopes) the numbers of observations from groups with higher levels of consumption are smaller, one would expect raw data to produce more erratic means and larger errors of estimate as the consumption rises. Given the importance of the issue, could David enquire or could someone familiar with the Sheffield work explain the figures in these tables and supply the actual observed point estimates of the means by group and the errors of estimate for each group for each sex, so they can be related to the results in David's earlier 2015 post on alcohol risk and give members or the public a better guide as to what is actually known about the increase in risk??

Presumably the 'medicine' effect of alcohol consumption on non-alcohol related causes of death automatically increases the likelihood of dying from an alcohol related cause. Surely the reasonable measure - especially at such low risk levels - ought to be overall increased annualised risk of death. The method described in the paper implies that the graphs are produced by a mathematical model, which in term seems to be based on epidemiological measurements (rather than, say, taking account of genomic differences).

There's a good critique of the underlying research here -

mtimc makes a good point. Unless some remarkable progress occurs in medical science we will all die from some cause or combination of causes. Apart from the issue that some causes may result in a more unpleasant death than others the critical estimate we require is the projected impact on our life expectancy.