Don't worry, drinking is still good for you

The temperance lobby hates the fact that drinkers live longer than teetotallers. Their problem is that there is a mountain of evidence showing that non-drinkers are significantly more likely to suffer from cardiovascular disease, in particular, and have lower life expectancies. Richard Doll, the doyen of public health epidemiologists, found lower rates of mortality amongst moderate drinkers in 1994. He showed that the relationship between alcohol consumption and mortality has a J-curve, with risk higher for teetotallers, lower for moderate drinkers and then a rise for heavier drinkers.

Doll et al. 1994

Doll was not the first to show this relationship and he certainly wasn't the last. In 2011, for example, a large meta-analysis published in the BMJ showed lower rates of death from heart disease amongst moderate drinkers. A previous large meta-analysis in JAMA Internal Medicine found the same for overall mortality. I could list many more.

The standard argument against this is that some people give up drinking when they become unwell (the sick quitter hypothesis) and so a simple comparison between drinkers and non-drinkers is invalid. Epidemiologists have been aware of this potential pitfall for many years and they have tested the hypothesis by excluding ex-drinkers and studying only 'never-drinkers'. And guess what? They found that the protective effect of alcohol still exists. Richard Doll showed this in 2004. Rimm and Moats (2007) showed this. Holahan et al. (2010) showed this. See here for a summary. The zombie argument has been rebutted. Teetotallers die younger, on average, than moderate drinkers. Get over it.

When the nineteenth century temperance crusader Mary Hanchett Hunt began her quack campaign to portray alcohol as deadly in any quantity she only had to overturn a general belief that booze was healthy. Her modern intellectual decedents have a tougher job in that they need to overturn decades of strong epidemiological evidence. Today, the job began in earnest in the British Medical Journal, leading to headlines such as this, from The Times...

Alcohol has no health benefits after all

Alcohol has virtually no discernible health benefits and claims that a glass of wine is good for the heart have been exaggerated, a study has concluded.

Other headlines such as 'Alcohol "has no health benefits" and previous research flawed, study finds' (The Australian) and 'Aussie scientists bust the myth that a couple of drinks a day is good for you' (Daily Mail) will have gladdened the hearts of 'public health' folk around the world.

So too would the editorial that appears in the journal, written by Mike Daube, previously a director of ASH in the UK and now an opera-banning jackboot of all trades in Australia. In a brief op-ed, Daube uses the word 'industry' no fewer than 13 times and claims that the days of people believing that alcohol is good for you are over.

These apparent [health] benefits are now evaporating, helped along by an important contribution in this week’s issue... 

What conclusions should we draw from this emerging evidence, including Knott and colleagues’ new study? Firstly, in health as elsewhere, if something looks too good to be true, it should be treated with great caution. Secondly, health professionals should discourage suggestions that even low level alcohol use protects against cardiovascular disease and brings mortality benefits. Thirdly, health advice should come from health authorities, not from the alcohol industry, and, finally, the alcohol industry and its organisations should remove misleading references to health benefits from their information materials... 

The real mortality benefits will come from determined action at the political level, not outdated advice and wishful thinking.

So what does the study actually show? The authors pull data out of the Health Survey for England and compare mortality rates amongst non-drinkers and never-drinkers to drinkers of various different amounts of alcohol.

Here are the results for men and women compared to non-drinkers. They are split between those aged 50-64 and those aged 65 and over. They are also split between two models. Model 2 is perhaps the most important as it controls for various relevant factors whereas Model 1 only controls for age. (Click to engorge.)

As you can see, for both sexes the relative risk is below 1.0 for drinkers of any amount of alcohol, ie. their mortality risk of lower than that of non-drinkers. This reduction in risk is statistically significant in almost every case.

We then look at the results for men and women compared to people who have never drunk alcohol:

Here, again, the vast majority of the relative risks for drinkers are below 1.0, indicating a reduction in risk compared to the never-drinkers. Fewer of these findings are statistically significant, but that is the inevitable result of there being fewer people in the samples. This is particularly true of the men, amongst whom there are only 95 never-drinkers in the 65+ group, compared to 307 non-drinkers. (Amongst the women there is a more statistically useful 353 never-drinkers aged 65+).

So how does this debunk the theory that moderate drinking reduces all-cause mortality risk? Quite simply, it doesn't. Even if we only use never-drinkers as the reference point, all the indicators show a reduction in risk in the region of 0.5-0.9, ie. a 10 to 50 per cent reduction. This is not trivial. We are not talking about a reduction in risk from one rare disease, but in overall mortality. This study shows, for example, that a male never-drinker is twice as likely to die between the age of 50 and 64 than a man who drinks 15 to 20 units a week (or, rather, who says he drinks that amount—the quantities are self-reported and he probably drinks significantly more). This is kind of a big deal. Imagine that instead of 'never-drinkers' we were talking about 'people who eat trans fats' or 'people who live with a smoker'. The 'public health' response to such findings would be a little different, don't you think?

The claim being made in the newspapers today—inspired by Daube's editorial and a deeply misleading BMJ press release—is that only women over the age of 65 benefit from moderate alcohol consumption. Even if this was true, it would hardly be trivial. Women make up half the population and the vast majority of deaths occur in people over the age of 65. However, it is not true. There is no plausible reason why alcohol would be protective for a woman's body but not a man's, and the data in this study give us little reason to think it does.

The claim is based on the fact that most of the risk reductions in the latter two tables are not statistically significant, except for women aged 65 and over. But there is a simple reason for this which some cynical people would call a trick. A relatively small sample has been taken and then split into different age groups, sexes and consumption levels to create dozens of even smaller samples. This, combined with the fact that there are relatively few never-drinkers to use as a reference, makes it very difficult to generate statistically significant results from any individual group.

The small sizes of the subsets mean that relative risk would, in many cases, have to fall below 0.5 before the results reached significance. No one believes that moderate alcohol consumption has an effect of that magnitude. However, there is plenty of evidence showing that risk falls by around 10-40 per cent and all the findings from this study are in line with that, albeit lacking the numbers to achieve significance. In only one group—women aged 65 and over—are there sufficient numbers to allow a risk reduction of this size to achieve statistical significance and, sure enough, statistically significant reductions in mortality risk are shown for every group of these women drinking between 1 and 20 units a week.

Dividing your sample into so many groups that statistical significance becomes practically impossible is a way of setting yourself up to fail. Failing to find a significant reduction in risk for one gender in a small subset that you have created does not mean that there is no reduction in risk. It does not mean that larger studies which have found a reduction in risk are wrong and you are right.  

Virtually every single relative risk is below 1.0 in this study. If you combined the age groups, the reduction in mortality would reach significance. If you combined the genders, it would reach significance. If you combined the various different drinking levels and simply compared those who drank moderately with those who never drank, it would reach significance.

No reasonable person could conclude from this study that 'Alcohol has no health benefits after all', as Britain's erstwhile newspaper of record claims. On the contrary, it shows much the same as all the other evidence on this subject: that moderate drinkers live longer than both non-drinkers and never-drinkers.


UPDATE

David Spiegelhalter makes similar criticisms of the study here. 

Share

Tweet