Thursday 14 April 2011

Alcohol and cancer - what's the story?

While I was away, the move towards an abstinence-only alcohol policy took a couple of steps forward. I mentioned recently the importance of Scientific Temperance Instruction (STI) in bringing about Prohibition in the USA. For 40 years, STI used pseudo-science to hammer the message that there was no safe level of alcohol consumption into American school-children. This was necessary to turn the sensible advice of temperance into the fanatical doctrine of prohibitionism.

Abstinence-only has long been hampered by the knowledge that alcohol is good for the old ticker—as confirmed in a BMJ study in February—a finding that is denied by only the most blinkered anti-drink crusaders.

The new study that came out in the BMJ last week offers hope to the zero tolerance activists and will, I expect, be cited for many years to come. It claimed that:

  • alcohol is responsible for 8% of all cancers in men (UK)
  • alcohol is responsible for 3% of all cancers in women (UK)
  • former (male) drinkers have a 54% increase in cancer risk
  • alcohol caused 16,355 cancer cases in 2008 (UK)
  • risk increases by 3% for every drink (per day)

It is the last of these points that most excites the temperance lobby since it seems to support the 'no safe level' doctrine. The study does not, of course, try to calculate how many cases of heart disease are prevented by alcohol so the net cost/benefit cannot be ascertained. 

The study concludes with a paragraph on 'policy implications' which is always a bit of a red flag.

This strongly underlines the necessity to continue and to increase efforts to reduce alcohol consumption in Europe, both on the individual and the population level.

It's at times like this that I wish I didn't drink. Not because of the cancer risk but because I could better enjoy the schadenfreude as the smug non-smoking pub-goers, CAMRA members and the "drinking's different" brigade struggle to disassociate themselves from the cigarette fiends and fatties. It's no longer just about 'binge-drinkers', alcoholics and lager louts. The temperance lobby has its eye on that agreeable glass of pinot noir and they've got the science behind them.

What has been most interesting for me has been the way in which the study has been attacked by those who now see their own pleasures under threat. Much of it is very reminiscent of the weak arguments made in the 20th century against the smoking/lung cancer link. At the Telegraph, for example, Judith Potts wrote:

The story hit the headlines. However, according to Cancer Research UK’s figures published in July 2010, the total number of cancer cases in the UK is 298,000 – which covers over 200 different types of the disease. This means that 285,000 cancers are not caused by alcohol. I wonder how many of these 285,000 cases are alcohol-blamed cancers which have developed in non-drinkers?

Well, yes but this is not an argument against public health clamping down on booze. You might equally say that most cancer cases are not smoking-related. You'd be right, but some are and they are considered preventable through behaviour modification.

The Science Bit unwittingly took the traditional tobacco industry line that mere statistics do not prove anything:

Rather than establishing a causal link between alcohol consumption and cancer occurrence, the researchers conducted a study that relied on the premise that such a link exists. In other words, a causal link between alcohol and cancer is not a conclusion of the BMJ study — it is one of its assumptions.

Yes and no. Epidemiology can never prove causation. However, the study did control for a large number of confounding factors so it is not unreasonable to assume causality. Saying that an epidemiological study doesn't prove causation is asinine. You need to provide a convincing alternative explanation for the statistics or demonstrate that the statistics themselves are flawed. Otherwise, we must accept the most obvious interpretation—the one given by the researchers—is the correct one.

Instead, this was very much a “What if…?” kind of study. The basic research question was, “If alcohol consumption causes cancer, then how much cancer might it be causing?”

Actually, it wasn't. It collected the statistics like any other cohort study and derived the conclusions from the data.

...it is impossible to know for sure whether all extraneous factors have been accounted for.

It always is. This is an argument against all epidemiology and, again, was an argument used against the smoking/lung cancer link for years. Then, as now, we need to hear what these 'extraneous factors' are otherwise the hypothesis stands. I am very open to hearing criticisms of this study—I'm quite prepared to believe that much of it is wrong, exaggerated or misleading—but general criticisms of epidemiology as a whole just don't cut it.

Regular readers will know that I am no defender of junk science. This study may well be flawed and you can be sure that as more money moves towards temperance work, the quality of the research will decline and the claims will become more sensational.

Where does that leave us right now? The claim that alcohol increases cancer risk from the very first drink certainly needs to be shown in further studies before anyone outside of temperance circles takes it seriously. The claim that ex-drinkers have a 50% increase in risk that is independent of other factors also needs to be reproduced. But while there are reasons to question some of the associations (a causal link with breast cancer, in particular, has never been as well documented as if often assumed), surely we are not denying that alcohol is linked to cancer of the larynx, liver or mouth? 

This particular study may well overestimate the number of cancer cases but there is strong evidence that excessive drinking, and possibly moderate drinking, increases the risk of several forms of cancer. We've known this for a long time. Whether alcohol causes 16,000, 8,000 or even 4,000 cases of cancer, the temperance lobby will push on with its neo-prohibitionist policies. (Bear in mind that the favoured statistic is 40,000 deaths per year.)

As I've said before, a public health movement that bans smoking in all 'public' places on the basis of 3,000 hypothetical deaths from secondhand smoke—based on much weaker epidemiological evidence that the study discussed here—is not going to think twice about hammering drinkers who are allegedly responsible for tens of thousands of deaths.

The solution to the problem does not lie in quibbling over statistics, but in telling the powers that be that our lives are our own and defending personal liberty in general. As Dick Puddlecote has said time and again:

You simply cannot pick and choose which freedoms you like and which you don't. You either stand up to all of the dictatorial bullying, or you will inevitably become a target.

If the BMJ study is correct and every drink carries a risk—and that even those who give up drinking are in danger—it serves as a reminder that pretty much everything carries a risk so you might as well enjoy yourself. Whether or not the epidemiology is up to scratch, the researchers can file their 'policy implications' over a cliff while we, as grown adults who can make our own decisions, can say "Thanks for the information, I'll take my chances."




25 comments:

Anonymous said...

Funny how the health lobby keep quiet about the demonstrable risks of engaging in anal sex, because to point that out isn't politically correct.

Christopher Snowdon said...

To be fair, I think they do recommend condoms.

Christopher Snowdon said...

This comment section has gone off-topic in record time.

Carl V Phillips said...

Chris,
Great analysis. It is quite interesting to watch how some people's interpretation of science changes with the topic. I will probably post on that.

On the point about epidemiology not showing causation, I agree with this and your other points about epi. Epidemiology -- including the branch of epidemiology known as randomized clinical trials -- cannot prove/show/demonstrate causation. But to clarify this, neither can any other empirical science -- proof is only possible in an axiomatic system like mathematics. In physics, chemistry, anthropology, medicine, and public health we are stuck with empirical sciences that let us infer causation with varying degrees of confidence.

As for the ex-drinkers statistic, that is not surprising at all -- and it is completely meaningless. People who are ex-drinkers are not like ex-smokers, made up mostly of people who thought what they were doing was a bad idea (always and for anyone) and quit. Someone who completely quits the generally healthful and socially accepted behavior of drinking typically did so for a reason like their drinking was heavy and out of control (and thus they may have already damaged their body with it), they acquired some health condition that made drinking unwise or unpleasant (and thus being sick with something cause the behavior), or some other lifecycle shock that probably had various health implications (e.g., suffered a major financial setback and quit drinking to save money).

Even ex-smokers are a bit of a mess to analyze, mixing in healthy people who quit to avoid disease or social pressure and people with COPD who quit to try to mitigate their disease. But ex-drinkers (or ex-soda drinkers, ex-weightlifters, ex-vegetarians, etc.) are such a mess that any study needs to either go into great depth about why they quit and what their health was at the time of quitting, or needs to just set them aside as far too complicated to say anything honest about.

Fredrik Eich said...

Well, assuming there is some small reduced risk for vascular diseases and drinking, and if I had my anti hat on, I could always claim that non-drinkers exposed over decades to the air borne poison of alcohol increases risk of disease and thus explains part of the
the illusion of a decreased risk for drinkers. So in the case where drinking saves a 2,000 lives in drinkers but causes 1,000 deaths in non-drinkers, drinking should be restricted to private outdoor places, just as is planned for smoking, at the very least. And lets face it, there is nothing more annoying than sitting down for a meal and an OJ in a pub and having to put up with the reek of
booze on peoples breaths and the general stink of booze. I would visit pubs much more often if only these selfish people would take their drinks out side. There is no point in having booze free areas in pubs because that is like have a piss free swimming pool ... etc,ad nauseam.

Anyway Chris, is that a typo of 3% women not men?

Anonymous said...

@14 April 2011 13:05
I think the reason the prohibitionist lobby never come down on sexual deviance is because in my experience of life people like that are more than often the kind who suffocate in a gimp suit.
Know what I mean.

Anonymous said...

I haven't looked at this study, but a similar one (see link) was criticised by many people including Prof David Spiegelhalter on R4's More or Less, for excluding one quarter of the population who didn't drink at all. The authors, including Naomi Allen, also one of the authors of this recent study, declined to appear on the programme.

http://jnci.oxfordjournals.org/content/101/5/296.long

I've always assumed that if consuming large amounts of x leads to a noticeable increase in the risk of y, then consuming even a tiny amount of x leads at least to an infinitesimal increase in risk. This may be a physiological impossibility in the case of alcohol (perhaps a critical amount needs to be consumed before sufficient depth of stomach/bowel lining is penetrated), but there is little chance that more than a handful of people will care either way. As you and many others have pointed out, alcohol consumption is in decline, but to hasten this decline would require taxes and bans/restrictions, which would fail because alcohol is so easy to make at home and the ingredients, sugar and yeast, cannot realistically be taxed or banned.
JB

Anonymous said...

They got one thing right:

"Smoking, known to be closely related to alcohol consumption..."

Perhaps they could get second jobs as consultants to the pub industry.
JB

Unknown said...

At least drinkers have more fun.

"Those who abstain from alcohol are also more likely to lack social skills and have higher levels of anxiety, it was claimed.

Non-drinkers even have more mental health issues than those considered heavy drinkers, the survey found.

One reason why non-drinkers were more gloomy could be that they have few friends, the study suggests.

“We see that this group is less socially well-adjusted than other groups,” said research leader Dr Eystein Stordal, from the Norwegian University of Science and Technology."

http://pubcurmudgeon.blogspot.com/2009_08_01_archive.html

Anonymous said...

"This study may well be flawed and you can be sure that as more money moves towards temperance work, the quality of the research will decline and the claims will become more sensational."

Here we go, the denormalisation has begun. There is a measurement somewhere done that shows alcohol in an open container evaporates very quickly into the air and engulfs those in the vicinity very quickly. And there is also the argument of drunk-driving going to come into it too. This will be enough to shut down the pubs entirely and ban alcohol from restaurants in the long run. Thanks to the second-hand-smoking fear, prohibition this time will be successful coming in through the backdoor method. Plus all the drinkers will have tacitly agreed when they did riddance to the smokers based on a fraud, so there will be no arguing back when their time comes next, nor anyone to stand up for them.

Kris said...

On a general note: How do these studies control for underreporting of alcohol-use?
Contrary to tobacco, I always have the feeling respondents are somewhat more reluctant sharing information about how much alcohol they really use(if they even know this accurately).

Unknown said...

"alcohol is responsible for 8% of all cancers in men (UK)
alcohol is responsible for 3% of all cancers in men (UK)"

Is there a gender typo there?

"Saying that an epidemiological study doesn't prove causation is asinine. You need to provide a convincing alternative explanation for the statistics or demonstrate that the statistics themselves are flawed. Otherwise, we must accept the most obvious interpretation—the one given by the researchers—is the correct one."

I disagree with this. Your logic suggests that we know everything and have at our disposal all possible complications that should be looked at, but we don't. And just because we don't know of a possible factor doesn't mean it doesn't exist. Look at HPV for example, it wasn't long ago it wouldn't have even been suggested or considered, and now we know what it does. HPV is said to cause what, almost 100% of cervical cancer? And yet there were many studies linking it to something else for some time. Same thing with throat cancer - long said to be caused by drinking and smoking, no mention of HPV for ages and then bam, there's that virus again.

Statistics only provide a marker for further investigation. I gave an example in CATCH of headache tablets being strongly linked, with a dose dependent relationship, with headaches - but the tablets don't cause the headaches. However your logic of the statistics don't disagree with the hypothesis would easily suggest the tablets do cause the headaches, and hey, there's dose dependency to prove it.

Someone else in CATCH pointed out epidemiology is like a reverse logic. It's one thing to point out that smokers/drinkers/people who dye their hair/ get more cancer than those who don't, but if we then look at the percentage of smokers/drinkers/hair dyers who don't get the disease we get a better look at the bigger picture.

And drinking will be particularly difficult to study, because how often have we heard of drunken one night stands or unprotected drunken sex? Sex is highly linked with intoxication, and HPV is an STD that can cause cancer. So there is one massive missing factor from studies on drinking and cancer. Hell, we can even say it's missing from studies looking at smoking and cancer, because it's implicated in throat cancer and we have evidence of it being present in lung cancers too.

Without the causation it is all anyone's guess. Yes we can have very suggestive figures, but as HPV has shown, there can be an unconsidered but causative agent lurking around the corner just waiting to be discovered.

Christopher Snowdon said...

Fredrik/Rich,

Typo corrected, ta.

Anonymous said...

Carl Phillips mentioned that physicists also have to rely on less than axiomatic proof. But unlike the softer sciences, they do take great care with statistical significance. In particular, whereas epidemiologists place great faith in a 2 sigma result (significant at 95%), physicists tend to work at 4 or 5 sigma which equates to 99.9999% or higher.

In this article, (physicist) Lubos Motl correctly points out that to move from 2 sigma (95%) to 3 sigma (99.9%) significance you just need to double your sample size (9/4 to be precise). Moving from 3 sigma (99.9%) to 4 sigma (99.9999%) requires a sample 16/9 times larger. And moving from 4 sigma (99.9999%) to 5 sigma (99.999999%) requires a sample 25/16 times larger.

I also agree with Rich in that having found a correlation, the onus is on the proposers of a causal link to argue their case for causality. Assumption of causality is not the default position. And to argue against it should not require an alternative causal explanation.

http://motls.blogspot.com/2010/03/defending-statistical-methods.html#more

Tony

Unknown said...

In addition to my previous comment, i should say this is a great article and analysis Chris. I don't agree with your points that i said already, but i do agree with the gist of the article and the conclusion is spot on.

Unknown said...

Carl:

"Even ex-smokers are a bit of a mess to analyze, mixing in healthy people who quit to avoid disease or social pressure and people with COPD who quit to try to mitigate their disease. But ex-drinkers (or ex-soda drinkers, ex-weightlifters, ex-vegetarians, etc.) are such a mess that any study needs to either go into great depth about why they quit and what their health was at the time of quitting, or needs to just set them aside as far too complicated to say anything honest about."

You're spot on about the mess of analysing ex-drinkers and other groups, but why are smokers less messy? Certainly, a certain percentage of smokers quit in the hope of avoiding future disease, but so do drinkers - if not cancer than cirrhosis or even just alcoholism. But many smokers quit because they have developed some illness already, or become aware of some genetic illness that they are predisposed to and think smoking may increase the risk, etc etc.

Anonymous said...

One problem with the study is the fact that they studied mostly people from the age of 37 to 70.

About 2% of cancers occur below the age of 35 and about 71% of cancers occur over the age of 65.

They studied an age group where only about 30% of cancers occur and the ratios might be rather different if those over the age of 70 were more prevalent.

Gary K.

Anonymous said...

In other sciences we have theories and models. From these models, we get propositions. And then we test these propositions. It is the theory that provides the causal relations. Empirical test only allows us to say we have not found sufficient evidence to reject the propositions.

Carl V Phillips said...

Richard, I agree that there is nothing easy about analyzing ex-smokers. Most of the time that category exists in a study just to get them out of the way to avoid contaminating the cleaner categories. And maybe you are right that the notion that we can effectively study ex-smokers is as misguided as studies of ex-drinkers.

If someone is serious about analyzing ex-smokers, it is usually possible to distinguish useful strata. Someone who quit at a relatively early age and does not have a relevant diagnosis for a long time pretty clearly was not driven to quit by a heart attack or COPD. But a drinker may have quit because of sub-clinical neurological or psychological problems that could be an uncontrolled confounder.

But wait: Perhaps the smoker quit because of some health problem that will not show up in any chart, not something caused by smoking but something that just made him more worried about his health. A valid point. Without taking up too much space here on someone else's blog, I think there are a series of related points and counterpoints.

Here is the main difference, I think: Giving up smoking is a common thing to do, usually caused by general health concerns (not a problem for the analysis) but is sometimes caused by a proximate smoking-caused disease (which should be in the data and thus can be separated out). Completely giving up every drop of drinking is sufficiently uncommon, and is not legitimately motivated by a general health concern. So in cases when there is not a health problem in the data, being an ex-drinker is much more likely to have been caused by a specific health problem, or by knowledge that the drinking was out of control. That is, a smoker who quits for no apparent reason is just "doing the right thing" most of the time, while a drinker who quits entirely for no apparent reason is quite likely actually sick (it is just not in the data) or was drinking at dangerous levels and is now afraid to touch a drop.

For both drugs if someone quit just as they got a diagnosis of some disease then we should have that in any dataset that is useful for analyzing ex-users, and can stratify them accordingly. It is the annoying subjects with a disease that does not show up in the data that is the problem.

I agree that as we learn more about genetic dispositions then the ex-smoker category will be further confused, since they will then include those with knowledge of their own high risk. But for now enough ex-smokers are "normal" but for the residual damage from their history of smoking, whereas something is odd about ex-drinkers.


Tony, It is certainly the case that the evidence supports causal inference much better in some cases than others. That does not change the fact that it is still all inference, not proof. Also, it is useful to keep in mind that causal inference is not all about random sampling error, as you imply it is. is. Indeed random sampling is often the least of our challenges in population sciences. But even in physics there is still a Duhem-Quine problem - the experiments are not just coin flipping. Also, physics is an outlier among the sciences in many ways, making it a misleading starting point for trying to understand the nature of typical scientific inquiry.

Magnetic said...

As I've said before, a public health movement that bans smoking in all 'public' places on the basis of 3,000 hypothetical deaths from secondhand smoke—based on much weaker epidemiological evidence that the study discussed here—is not going to think twice about hammering drinkers who are allegedly responsible for tens of thousands of deaths.

Sorry for O/T, but it’s a circumstance worth highlighting because your point can be made another way that points back to the medical establishment instead. (in 2 parts)

Consider iatrogenesis which refers to any detrimental outcome produced by medical conduct (e.g., adverse drug reactions, medical errors, poor care of the bed-ridden resulting in infected bed sores). In America, from the very few studies that have been done, iatrogenic deaths are estimated at 750,000-1,000,000 per annum. It dwarfs the so-called tobacco “death toll” (450,000) and is approaching half of the total annual death toll in America (2,500,000). The medical establishment is by far the leading cause of preventable death and disability and associated costs. The translation for Australia is that iatrogenic deaths are estimated at 35,000-50,000 per annum compared with the tobacco “death-toll” of 15,000. The translation for the UK would be 250,000-330-000.

Further, the iatrogenic toll is far more plausible than the tobacco “toll”. The tobacco “toll” is based on lifetime use that also brings into play a veritable plethora of other factors over a lifetime. It is also argued from the population level that has very poor extrapolation to the individual level. Conversely, the iatrogenic toll is argued from the individual level and then estimated for the population level. For iatrogenesis, causation is typically demonstrable at the individual level. For example, with adverse drug reactions (these are properly prescribed, FDA approved drugs), it may take just one or a few pills to produce cardiac arrest in some patients that can be fatal or leave permanent injury. The patient is stable and within an hour of taking a pill goes into atypical catastrophic failure. The temporality of many of these associations is not over a lifetime, but involves hours or days. Of the two – the iatrogenic or tobacco tolls – it is the former that should attract very serious scrutiny. But, as will be seen, it has been the other way around, or upside-down.

Anonymous said...

Carl

"Completely giving up every drop of drinking is sufficiently uncommon and is not legitimately motivated by a general health concern."


Unless you are a very modest drinker who discovers she is pregnant and then having been on 24 hour call for so long and with so many demands on her time, never quite gets in the way of relaxing with a drink again.

Rose

Jean said...

Two points :
- 3% more chance of cancer for each extra drink per day is not that much. One extra drink per day is in itself a lot of alcohol. If I got it right.
- I'm starting to think that all these studies about products that cause cancers should be put into perspective with the increase in life expectancy. Precisely, do alcohol and tobacco are primarily responsible for the cancers you get at 50 or for those you get at 90?

Ivan D said...

3% in women is the claim for cancer overall Jean but I think that the study is claiming an alarming sounding 25% relative risk per unit for mouth and throat cancers.

Even if we accept the 25% relative risk as being true then what responsible scientists should do is give that some context in terms of your individual risk which for say mouth cancer in the UK is 1 in 20,000 for a woman. So your extra unit might increase your risk from say 1 in 20,000 to 1.25 in 20,000.

Worried?

Rather than give us the absolute risk and let us make our own choices, too many “researchers” and “experts” issue tabloid style press releases claiming big changes in relative risk. They seek not to educate and inform but to remove choice by manipulating our elected representatives into legislating in favour of a chosen cause. As a society we should be asking whether we want informed free choice with respect to lifestyle and health issues or soviet style authoritarianism under the auspices of the epidemiologists and zealots.

My personal issue with the BMJ paper, aside from it being a political pamphlet is that it contains no meaningful results and expects us to simply believe the conclusions of the authors. No scientist worth the name would publish without results.

Unknown said...

Carl, thanks for the reply, very informative and i largely agree. One thing i will add for a further reason someone may quit drinking (and this is just something i've heard anecdotally) is when the drink begins to interfere with life. Notably this doesn't happen with smoking, because aside from not being able to smoke indoors, there are no restrictions i.e. you can smoke and drive, or, using Rose's example, smoke while on call etc. So drinkers may just find themselves in situations where drinking isn't allowed and then they just fall our of habit.
Similarly, someone may have a bad experience with drink (hangover, violent vomiting etc) and quit, as one of my friends did, and, stretching anecdotal to 'i read this in a biography' but i have no doubt it's not isolated to one case, having a violent episode while drunk and realising drinking is having a negative effect.

With smoking, the only concern is the worry on illness later in life. With drinking there can be more factors that are more immediate but not related to illness, and that makes any study messy i think.

Jean Granville said...

I was refering to this :
"isk increases by 3% for every drink (per day)"

Which seems to mean that if you drink one drink a day, you get a 3% increase, and so on with every extra daily drink.

My point being that one drink a day is already relatively important. I don't drink alcohol at most of the week and I probably don't drink more than seven drinks during the week-end. Multiply that by two, you get a 6% increase, multiply it by four, you're still at 12 and really drunk from time to time.

And a 12% increase is still not much.