Sunday 15 April 2018

Spin the bottle

I've not blogged or written about health for some time, since the publication of my book "The Life of Riley", but the coverage generated by the research on safe levels of drinking, published this week (link here)  produced some wild headlines across the press, warning of the dangers of demon drink. My interest in health, combined with my knowledge of the workings of the press, and some rudimentary graduate level stats knowledge, were all piqued by this.

And yet the analysis done actually seems to me to give moderate (and even slightly heavier than moderate) levels of drinking a pretty reasonable bill of health, and overall was much in line with lots of previous research (much of which I referenced in my book), which most people vaguely interested will recall as suggesting moderate drinking was not only safe, but seemed to be beneficial for health.

How is this possible?

Well the report itself was a meta analysis of a number of previous studies, merging 600,000 respondents over many years into a giant database for analysis. A good approach, and one which normally removes the wrinkles and oddities found in smaller pieces of research.

Where this report differs from many previous meta-analyses is the base population it was measuring and comparing. Most previous research has looked at the population as a whole, both drinkers and non-drinkers, in order to compare rates of mortality (i.e. how long we’ve got left, on average, given our age, sex, lifestyles etc.)

This analysis however only compared mortality rates of all current drinkers, using those who drank, but drank the least, as the base line (so giving them an index of 1). These people drank between 0-50 grams of alcohol per week. A standard bottle of wine has around 80 grams, and you can get 4 decent, medium sized glasses from a bottle - so this control group were drinking under 3 medium sized glasses of wine per week ( I’m just going to use glasses of wine for this analysis, but the comparisons hold true for beer etc, just with different volumes).

This base line group were then compared to folk who drank 50-100 grams, 100-150 grams, 150-250 grams, 250-350 grams and 350 grams+ each week (although the groups were split differently in the text, tables and graphs, for no obvious reason, which is unhelpful when trying to analyse and critique).

However, in the report's text the authors have merged the two lighter drinking groups (0-50 grams and 50-100 grams) into just one 0-100 grams group as there appeared to be little difference in mortality rates between them (although see below for the raw data) - so this larger base are those people who only drink up to 5-6 glasses of wine per week, with an average of around 3 glasses. These are deemed the least at risk and provide the base level of risk used to compare other groups to. For reference lets call this group “light drinkers 1 & 2”

Those in the group that drink 100-200 grams (so up to 10 glasses of wine per week or 2 1/2 bottles) were slightly more at risk, and estimates were that a 40 year old who drank at this level would lose around 6 months of life compared to the light drinkers. But this is over the remainder of his/her life, which might reasonably be expected to be 45-50 years - so maybe a 1% reduction overall. This hardly seems worth worrying about in my view. 200 grams per week represents 25 units in UK drinking jargon, which is of course way over the current suggested limit of 14 units per week for both men and women, despite this apparently relatively trivial level of increased risk. Lets call these people “moderate drinkers 1 & 2” as the data splits them into two groups.

Those drinking 200-350 grams had a worse life expectancy. In the middle of this band at around 275 grams you’d be drinking 3 1/2 bottles of wine a week (or half a bottle = 2 medium glasses a night), but at age 40, if you continued to drink at this level, the suggestion is you would lose 1-2 years of life compared to the lightest drinkers. This is a 3% reduction and, doing a bit of maths, I think this is where that “each drink takes 30 minutes off your life” headline reported in the press comes from (11 extra drinks per week at 30 minutes lost each drink over 50 years knocks off 1.6 years). Let’s call this group “regular drinkers”.

Those drinking over 350 grams, heavy drinkers, who are nudging a bottle of wine a night, every night (or more) had a 4-5 year loss of life in comparison - so around a 10% reduction. That sort of reduction ought not to be dismissed, and no one sensible could be shocked to discover that knocking back that much booze every night for your entire adult life might carry some significant elevated risk of illness or disease.

But, the confounding factor in this research is the absence of those who simply don’t drink - not former drinkers who may have given up for health reasons, but those who simply don’t like the taste, or who don’t imbibe for religious or other non health related factors.

These people also lived with much higher mortality risk than our super-light drinkers - a risk pretty comparable with those in the second highest drinking group, our “regular drinkers”, those drinking up to 3 1/2 bottles a week (or roughly half a bottle per night).

I understand why the researchers did some analysis excluding non-drinkers, as they do seem to be different in their overall make up. As the researchers stated “...  However, we observed notable differences in baseline characteristics between never drinkers and current drinkers (eg, in relation to sex, ethnicity, smoking, and diabetes status.....”.  However, to then only make comparisons of death rates, and years lost amongst the drinking sub-group, without also highlighting the flip side of years lost by not drinking at all, seems slightly disingenuous, knowing how the press are likely to end up reporting the paper. The data on non-drinkers is only to be found in the appendices - and no reporter, rushing for a deadline, is going to pore over the minutiae of the report in this fashion. In this sense I think the tone of the main summary, and the subsequent reporting, could be considered somewhat misleading for the layman.

In fact, all levels of drinking up to around 1/2 a bottle per night (so both light and moderate) seemed to provide much lower risk than either not drinking at all or being a regular or heavy drinker.

You’d have to say, looking at the data, that non-drinkers are doing themselves a huge disservice by their restraint, and arguably should be holding their noses two or three times a week and knocking a glass back! But no one in the public health sphere could or would ever say this of course.

In reading the report, I was interested to note the amount of “heavy lifting” being done by the weighting applied to each group. This is where the researchers apply greater weight to, say, female respondents if there are fewer than the number expected within a group, or upweight or downright particular ethnicities or age groups, so as to produce a revised sample more in line with the population as a whole (or balance for smokers, or the obese etc.). This is a perfectly sensible thing to do, but as a man who has been a professional user of weighted statistics for 30 years, with his livelihood dependent upon their accuracy (RAJAR anyone?) I can tell you weighting can produce some strange effects.

Just to show how much effect the weighting had, here is a simple table of the raw data on how many people had died in each group when they were all recontacted, around 8 years later.

Non Drinkers (53,851) 10.7%
V Light Drinkers (177,956) 8.1%
Light drinkers (128,094) 6.0%
Moderate 1 (94,653) 6.0%
Moderate 2 (94,760) 5.9%
Regular (52,020) 6.7%
Heavy drinkers (52,429) 8.4%

Just looking at the raw data, its those at the top end of the moderate banding who were least likely to die, and in fact the very light drinkers had one of the worst mortality rates! And lo and behold it is non-drinkers who were dying off the quickest - much quicker even than heavy drinkers. This is just the raw data though, and I fully accept it needs to be weighted, but intuitively you can see the weighting has had some dramatic effects.

One piece of weighting they didn’t do, and which I’d have liked to see, is a stratification by weight. I’m a pretty tall, lean guy, but I weigh nearly 100 Kg. A small, lean woman could easily weigh only 50Kg. We can’t both surely have the same tolerance for alcohol though, as it must, as a drug, be in some way dose-dependent. Therefore sorting mortality rates by alcohol consumed per Kg of body weight might have produced some interesting findings (the researchers must have had the data, as they had BMI numbers). This is one reason why having no difference in guideline limits between men and women in the UK seems so odd, given different levels of typical body mass (we are alone in the western world in doing this).

Apart from ignoring non-drinkers in their presentation of relative risk and levels of mortality, another concern I have with any alcohol related studies is a point acknowledged by the researchers themselves in this one, who state

 “........Nevertheless, our study has some potential limitations. Self-reported alcohol consumption data are prone to bias and are challenging to harmonise across studies conducted over different time periods that used varying instruments and methods to record such data.....”

This is of course a scientific way of saying that people habitually lie about their levels of drinking or simply can’t recall it accurately and on average are prone to significantly under report. In fact numerous comparisons of actual sales data and reported levels of drinking show significant gaps between what respondents were actually drinking and what they were letting on to. The ONS itself, in its latest alcohol drinking habits report here states “It is likely that the data reported here underestimate drinking levels to some extent. Social surveys consistently produce estimates of alcohol consumption that are lower than the levels indicated by alcohol sales data. This is likely to be because people either consciously or unconsciously underestimate their alcohol consumption.” 

Far be it for me to be ultra critical or cynical about the data in this most recent analysis, but of the 600,000 drinkers in the sample, fully 300,000 of them claimed to drink on average just 3 glasses of wine (or the equivalent) each week. Really? Half of all drinkers only drink 3 glasses a week? I must be mixing with the wrong crowd.

I doubt the heavy drinkers are capable of much under-reporting given how much they are already admitting to drinking (and there are only a few of them any way, less than 10% of the total), so it must be the light, moderate and regular drinkers who are under-reporting, and it seems to me all of the “drinks per week” discussed above are some way below the reality for most people.

The final query I’d have is why choose age 40 as the baseline point to reference years lost etc?

Clearly one reference point is better than 2 or 3 - but I doubt today's millennials will care much what happens to 40 year olds, and 40 year olds are, I suspect, by and large too busy worrying about jobs, kids, mortgages etc to care too much about what will happen to them when they are 80 or 85. It’s curious too, because the average age of the respondents in the surveys merged for this analysis was 57, so there was some more heavy statistical lifting involved in then “back dating” the results to produce the age 40 estimates. It’s also odd because this late 50s/early 60s cohort is the group that public health officials tell us they are most worried about.

The ONS has a great life expectancy model on its website here, and this forecasts that a typical 57 year old will live for around another 30 years (women 31, men 28).

Using this figure of 30 years longer to live at age 57 as a benchmark, and using 30 minutes lost for every drink per week for each excess drink consumed over the light drinkers, this is what happens when you crunch the data:


Non Light Mod 1 Mod 2 regular heavy all
Number in sample 53851 306050 94653 94760 52020 52429 653763
age 58 57.2 57.4 57.2 56.6 56.4 57.2
mean units per week
6 16 25 38 69
drinks per week (@ 2.5 units per drink) 3 6 10 15 28
excess over light drinkers 0 4 8 13 25
minutes lost per year -19500 0 -5850 -11700 -19500 -39000
months lost over 30 years -13.5 0.0 -4.1 -8.1 -13.5 -27.1
Years lost compared to light drinkers -1.1 0.0 -0.3 -0.7 -1.1 -2.3
Predicted mean age of death 86.4 87.5 87.2 86.8 86.4 85.2 87.0
Years lost compared to average (mean)  -0.6 0.5 0.2 -0.2 -0.6 -1.7
non-drinkers given same excess as regular drinkers as all-cause mortality rates are virtually identical

So the light drinkers get an extra six months over the average, whilst some moderate drinkers lose a couple of months, regular drinkers are 7-8 months off the average, and even heavy drinkers are under two years below average. Obviously if damage has already been done by prior excess consumption, then that will affect mortality going forward - but most folk looking at this study will probably shrug their shoulders about past indiscretions, saying "what's done is done"

Like many of my friends I am in my late 50s. If our likely life expectancy is another 30 years or so, then the light and moderate drinkers in the group appear fine, and even for the regular drinkers, if every night spent drinking half a bottle of wine lops 7 months or so off the average life expectancy compared to our cohort, I suspect many people would think that’s a trade off worth making.