Sigh – I’m sort of even pissed off that I feel the need to comment 🙂
So what’s wrong with this?
Some things in pretty random order:
This study doesn’t even seek to answer the question of whether alcohol is good or bad for you!!!, as implied by the BBC report
Yeah, that’s right. Despite what the article might be suggesting.
It’s about population health, which is quite different. It is about what “burden” alcohol might put on the health of a country/society.
So, for example, if alcohol use increases the risk I might cause someone else to be hurt in a car accident, that counts towards the statistics.
And all the statistics are worked out against this objective.
There’s a lot of violence
As with almost all the similar studies I have seen, a serious part of the negative statistics comes from an increased correlation with the results of violence:- car accidents, interpersonal violence, self-harm and increase in communicable disease.
OK. Fair enough. But this is not what people understand about “health problems” I would suggest.
What is missing?
A hard thing to do with this sort of research is to work out how the whole thing might be flawed, as opposed to examining the details of the research. In this case, the big question to ask is what is missing; and it turns out there might be quite a lot.
If the authors can take such care on the input side to include tourism and even home brewing in the consumption of alcohol figures, surely they can take care to include all the stuff on the output side?
How about other illnesses that might be affected by alcohol? Maybe ME/CFS/PVF? Does peoples’ experience of MS change? And are there other physical illnesses that should be included? But the big one…
What about mental health?
At first sight the authors look like they have tried to characterise all possible adverse or positive health-related issues that might be affected by alcohol.
It has to be that mental health has the possibility of being an important factor in the increase or reduction in the “burden” of alcohol? Mental health problems can be utterly debilitating and disabling, and surely should contribute to the “disability-adjusted life- years (DALYs)”.
For example I find no mention of “depression” or “anxiety” in the entire paper. These illnesses have enormous cost to society, and any increase or reduction in them might even swamp all the other considerations. I would love to know similar research to this around these illnesses.
It is enormously sad for me that such a paper can be peer-reviewed and published without reference to mental health; are we still in a world where the only “real” illnesses are physical? Yes, the paper mentions “self-harm”, but that just makes it worse: the pain of mental illness is seen through the lens of people hurting themselves physically, rather than the pain they are feeling.
It’s not a big number
I can’t see it in the Lancet paper, but the BBC article says:
“They found that out of 100,000 non-drinkers, 914 would develop an alcohol-related health problem such as cancer or suffer an injury.
But an extra four people would be affected if they drank one alcoholic drink a day.”
(Edit: David Spiegelhalter’s excellent Medium article tells me that it is The Lancet press office that published these figures.)
Let’s just ponder that for a moment.
Almost 1% of non-drinkers will have a problem. That’s perhaps quite a lot. Or maybe it isn’t – I don’t know; people do get cancer, beaten up or fall.
Now, if all those 100,000 people decided to drink one unit a day, an extra 0.004% of them will have a problem. That is definitely not quite a lot.
And note that it includes violence (see above).
Another way of looking at it is relative risk, which the paper is hot on.
My (ignorant!) calculations suggest for these figures it is a Relative Risk of 1.0044, which actually corresponds pretty much to what I can see in the graphs.
This might be a significant figure when aggregated for a nation (although actually I think not, especially given the error bars (see below)), but is definitely not the sort of level at which I suggest people should be making life choices!
Correlation is not causation
Maybe I have missed something in the paper and appendices, but I can’t find out where they tackle this for some things.
They seem to say they account for possible confounding stuff in the area of the abstainer because they are ill. But do they for everything?
A statistical correlation between alcohol use and self-harm, for example, seems particularly difficult. Does self-harm cause increased alcohol use? Probably not, but then that is as likely as increased alcohol use causing self-harm, in the absence of other information. I don’t see any evidence that the authors tried to filter for confounding variables around these.
And remember that violence is a serious part of many of the statistics, so this is potentially significant.
It’s not really a “new global study” – it’s a meta-study of existing research, and doesn’t contain a single new measured value
I can’t even find out whether they have discarded any of the studies as being unreliable. If not, this is the worst sort of meta-study. A proper Cochrane Collaboration review would have discarded the majority of the studies they found as not meeting the criteria for inclusion.
As I said, I can find no evidence that they discarded a single study as being unsatisfactory.
This may well be “big data” at its worst.
It’s a “Global” study – but the data is very location-dependent
Tuberculosis is a significant number in many of the graphs, and so presumably is a significant influence on the headline numbers. I have to question whether this means that any inference can made at more local level, given the variation.
India (and many African countries in particular) has a TB level orders of magnitude greater than the countries that might be trying to use these statistics to direct health care.
UK: 0.01%, US: 0.003%, India: 0.211%, China: 0.064% (UK gov figures)
They even extrapolate from the USA to the rest of the world
RTAs (Road Traffic Accidents): They say they can only get statistics for the effect of alcohol use on RTAs for the USA, and so “Because of data availability, we assumed that locations outside the USA would follow a similar pattern to what we estimated with FARS” (US Fatality Analysis Reporting System)”
Really? You might have tried to be honest, but can we assume that Saudi Arabia, Japan or Turkey have anything like the same patterns, no matter how hard you try? And the USA has an RTA death rate more than three times the UK’s – can the patterns really be assumed to be similar?
“car crash involving alcohol” becomes “drunk driver”
Deciding on whether alcohol (or speed) is a genuinely contributing factor in an RTA is notoriously difficult. The FARS data is simply reporting whether alcohol was “involved”, not whether it was a contributing factor in the death. Although the authors may have allowed for this, it is hard to tell – are there confounding variables they are ignoring?
And as far as I can see (Appendix 1, Section VIII), they simply use “Driver BAC >= 0.01” as the criteria. And the fact their terminology moves smoothly from “alcohol involvement” (which of course it might not actually be) to “drunk driver” in the figures’ captions deeply undermines their claim to objectivity, in my eyes.
There are huge error bars in almost all the figures
I think the only reason that the authors can even begin to say anything with any confidence is that they have so many studies. In effect, however, they can only do this if they consider each of the studies as separate “experiments”. But can you? Many of these papers will be using similar, possibly flawed experimental methods, and who knows?, some of them may even be re-using the same data!
This is particularly true at the low end of alcohol consumption, which is where all the attention is likely to be focussed.
The consumption scale is too coarse
Looking at the graphs, it seems that many of them only do curve fitting at integral number of units per day, and others only half. This really doesn’t give me confidence that the graph fitting they have done at the lower end is valid. There could easily be more J-curves hiding in there, or certainly other interesting things.
Figure 3 from the main paper is really strange on this. “Ischaemic heart disease” (both) are very angular, and clearly only fitting curve points at integral values. Whereas “Diabetes” (male) has a point at 0.5 which gives a very strange angle.
It doesn’t actually address costs
If this is for forming public policy (which apparently it is), then it should actually be about costs, rather than “alcohol-attributable deaths and disability-adjusted life- years (DALYs)”
Perhaps the cost of different problems is different? I know that treatment for diabetes and its related problems, such as the need for limb amputation, can be very complex and expensive to both health and social care. Given that alcohol use seem to be beneficial in reducing diabetes (good news, especially for women – possibly up to five units per day is still better than none!), maybe the reduced cost is sufficient to outweigh all the other problems. Who knows? The study makes no attempt to address this issue.
I am guessing/hoping the authors have tried very hard to be honest, and would be able to satisfy all my concerns if they were asked. And indeed they discuss these and other limitations. But at the moment I am left with considerable unease as to the real value of these statistics, and concern they will be used to drive public policy decisions based on unreliable data.
And I have to repeat that the lack of a mental health angle is deeply worrying, We have all heard people justifying drinking on mental health grounds! Surely we need the science on this, so that assertions on the benefits of alcohol to mental health can be challenged or supported? I am left worrying that the reasons that this is missing, and that I can’t easily separate out the violence from other issues is because the authors sense or even know that the results would be other than they would like.
Mind you, I think it is quite a good paper – well-written, and it enabled me to see a lot of what they were doing!