by Professor Byron Sharp
short answer: The scientific evidence is that alcohol helps cause (only) two cancers: aerodigestive and liver. These are rare causes of death, so the increased chance that drinkers will die from these cancers is tiny. A very large body of evidence continues to support a causal association between moderate drinking and longer life, due largely from less risk of diabetes and less risk of dying from heart disease.
sub-title: Can Good Intentions Corrupt Science?
Recently an article by a Kiwi Professor gained global press coverage with sensational headlines along the lines of “proof that alcohol causes cancer”. This is a very significant claim which therefore deserves significant scrutiny. Unfortunately in this case I’m afraid that there has been an abuse of science. I’m very interested in the corruption of science, and science communication. This doesn’t always arise from corrupt individuals – more common causes of bias are good intentions, moral vanity, and a desire to please a dominant funding body. That’s what it looks like here: perhaps a case of seeing what you want to see in the evidence, or perhaps even deliberately misusing the evidence to achieve a social/political objective (‘the ends justifies the means’)?
The scientific evidence suggests alcohol helps cause (only) two cancers, areodigestive and liver, though these are rare and the elevated risk of death is tiny. It’s wrong and unethical to scare the public, and to misrepresent the medical evidence.
I’m going to document how Professor Jennie Connor’s article missed out important studies and misrepresented others (she even missed evidence that helps a part of her argument).
Connor’s conclusions are not based on new data, nor data analysis. Her article was published in the journal Addiction‘s “for debate” section. The article’s purpose was to examine whether it might be reasonable to consider the (weak) correlations seen in population studies as causal evidence that drinking alcohol increases your risk of cancer.
Connor says alcohol causes some cancers. This conclusion is based almost entirely on epidemiological studies that show that in surveys of populations drinkers have slightly higher rates of some specific cancers (and slightly lower rates of a few others). In doing so she places great faith in epidemiology, far more than many of her colleagues. A more sober analysis would adopt the epidemiologists’ rule of not considering risk estimates of less than a factor of 3 as indicating any causality (see Taubes 1995 ‘Epidemiology faces its limits’, Science, 269:5221, 164-69), e.g. smokers have more than 20 times the risk of lung cancer. Unlike smoking, alcohol-cancer studies seldom report risk estimates anywhere near a factor of 3.
The well known problem with epidemiology studies is that they show correlations that are often not causal – they can be completely spurious due to confounding factors. For example, population data might show that elderly people who exercise regularly tend to live longer. But elderly people who are afflicted with debilitating health issues are very likely to exercise less (many will be simply unable to undertake exercise), so really it is health that is determining both the exercise as well as the longevity. Exercise might well be doing good but the risk estimate (for not exercising) is going to be inflated.
The problem with comparing alcohol drinkers, of various levels, to lifetime non-drinkers is that alcohol drinkers tend to smoke more, and they also tend to be less likely to live in rural areas and more likely to be wealthier. Moderate drinkers might be people who tend to be adopt other healthy behaviours, while heavy drinkers may be self-medicating because of mental or physical ill-health. The ability for statistics to control for such confounding correlations is rather poor. So there are plenty of reasons for caution in treating a correlation seen in population data as causal. Here are some funny examples of real but non causal correlations.
Connor understands that confounds produce spurious correlations yet refrains from discussing any likely confounds in her assessment of whether alcohol causes cancer. Instead she goes off on a tangent discussing possible confounds for the well established link between drinking and a (substantially) lower risk of heart disease. Here she suddenly loses her faith in epidemiology, i.e. when she doesn’t like the results. For Connor to even discuss heart disease is rather odd, given that her article is about cancer. And to discuss confounds for heart protection but then to dismiss potential confounds for her claimed causal link with cancer is at best an abrogation of scientific scepticism.
In the absence of controlled experiments (which for cancer research are pretty much impossible for ethical and practical reasons) we need other evidence to support treating a correlation in population studies as causal. Connor’s article makes 3 such points:
1. First, Connor says, there is a dose-response relationship between alcohol consumption and some cancers, that is, any increase in drinking is associated with increased cancer risk. However, she then (rightly so) contradicts herself in pointing out that for most of these cancers meta-analysis shows no risk for light drinking. Connor should have noted that some meta-analyses show heightened risk only for the heaviest level of persistent drinking (which is problematic, as this group may contain alcoholics and people drinking as self-medication for health problems, a confound completely ignored by Connor (and many others)).
2. Second, Connor notes “evidence that, for some cancers, the risk associated with alcohol attenuates when drinking ceases”. Now this is what we would expect if there were a causal relationship, yet it is odd that in such population studies the correlation between drinking and cancer risk drops so slowly, taking decades before drinking cessation results in the small elevated risk subsiding to that of a non-drinker. This suggests the apparent alcohol risks have more to do with confounding factors than a direct causal relationship (i.e. lifestyle factors that slowly regress to the mean). Regardless, this evidence is of poor quality, to quote from one of the meta-analyses that Connor references “Too few studies have addressed this question, and of the studies that have, all have significant limitations” and further “the only statistically significant relationship that we observe is that drinkers who recently quit drinking have a higher risk of liver cancer than current drinkers”. It would be wrong to make much of this evidence, because the models have huge error margins, and anyway they are hardly supportive of a causal link between alcohol and cancer.
3. Third, Connor briefly discusses potential biological mechanisms for how alcohol might cause cancer. Alcohol in itself is not carcinogenic to human cells, and the epidemiology shows associations with only some cancers and not others (indeed, drinking appears to be protective against some cancers). As Connor writes: “The mechanisms by which alcohol causes cancer are not well understood, but are thought to depend upon the target organ. Pure ethanol does not act as a carcinogen in animal studies, and evidence that it causes mutations directly in humans is weak”. A possible mechanism is that bacteria in the mouth and digestive tract convert alcohol to acetaldehyde, which is a carcinogen. As Connor notes “Stronger associations and more susceptibility at low doses is seen for the cancers where alcohol and [hence] acetaldehyde come into direct contact with the tissues”. She doesn’t mention that this potential bio-mechanism is supported by the evidence that people whose genes mean they are slow to break down acetaldehyde (somewhat common in Asia, very rare for West Europeans), have substantially higher rates of oral/throat (aerodigestive) cancers. Other potential biological mechanisms, such as alcohol increasing estrogen levels, remain very speculative and face a number of difficulties, such as explaining why women who drink alcohol have lower, not higher, rates of endometrial cancer (Je et al 2014, Sun et al 2010).
So for aerodigestive cancers, we see the highest correlations in epidemiology data (though nowhere like the magnitude for smoking), and we have a plausible mechanism (the culprit being acetaldehyde). Plus we can add the evidence that Connor missed – there are much higher rates of aerodigestive cancer among people who do not have the genes to quickly metabolise acetaldehyde (Seitz and Becker 2007). This is by far the best case for making a causal connection between drinking alcohol and a cancer. Though it must be noted that aerodigestive cancers are rare (eg a tiny 0.3% of US deaths are from oral cancers, and most of these are caused by smoking, age and genes) – even though most of the population drinks alcohol.
For the other cancers Connor mentions, the evidence of causality is extremely poor. In contrast, the evidence is vastly more convincing that drinking alcohol moderately reduces heart disease, diabetes, dementia, and overall mortality. In addition to meta-analyses of many population studies, we have plausible biological mechanisms, supported by lab studies, animal experiments, and hospital trials. Meta-analysis of dozens of experiments shows administering alcohol to subjects leads to rapid positive changes in biomarkers for heart health (Brien et al 2011). We also see changes in drinking linked to changes in rates of heart disease, diabetes and overall mortality, e.g. people who increase their drinking lower their risk of heart disease. In sum, this is what a plausible case for causality looks like.
So apart from aerodigestive and perhaps liver cancer, Connor is wrong, the evidence does not support the hypothesis that drinking alcohol causes cancer. And even for oral/throat cancer, the risk may be confined to smokers and people with particular ALDH enzyme mutations.
Alcohol does not increase risk of dying from breast cancer
But that’s if we stop with Connor’s superficial analysis. There is more evidence to consider if we are to form a proper judgement about causality. Read on…
Connor concluded that drinking probably was causal for 7 cancers. This has a nice ring about it…. like the 7 Deadly Sins. Now, Connor used the old medical convention of labeling cancer according to where the tumour occurs. It’s possible to talk of hundreds if not thousands of different cancers, but more reasonably it’s not seven. It’s actually (1) oral/throat/oesophagus (aerodigestive cancers) which we have already discussed, (2) liver, (3) colorectal, and (4) breast cancer. It’s this last one that drives most public health forecasts that reductions in drinking will reduce cancer deaths. This is because breast cancer is the most common potentially deadly cancer for women. That said, less than 3% of female deaths in the USA are from breast cancer. 90% of breast cancer diagnoses turn out not to be fatal, due to curative treatment but also because many breast cancer diagnoses are for non-fatal cancer – ‘over-diagnosis’ is a very real problem.
Breast cancer is a prime example of why Connor should have discussed evidence on confounding factors, and should have distinguished between diagnosis and mortality statistics. Meta-analysis shows no link between drinking and breast cancer mortality (Thun et al 1997, Fuchs et al 1995) while studies of those diagnosed with breast cancer also show no increased chance of death, nor with recurrence for drinkers (Gou et al 2013). A new study from the Women’s Health Initiative (Lowry 2016) again shows no link between mortality and drinking before or after breast cancer diagnosis. A smaller US study published in the same year showed the same thing, women who drank (at all levels from 1-36 standard drinks per week) before their breast cancer diagnosis had no higher risk of dying from the cancer than the non-drinkers (Din et al 2016). Similarly a large study with long follow-up of women with breast cancer (Newcomb et al 2013) showed breast cancer patients had better chances of survival if they were regular drinkers before diagnosis. If they altered their drinking after diagnosis this did not alter their chance of dying from breast cancer. But an increase in drinking was associated with an overall increase in life expectancy (largely due to substantially fewer heart disease deaths among those who increased their alcohol consumption). This is strong causal evidence that alcohol prevents heart disease, and it seriously conflicts with the idea that alcohol causes breast cancer.
Similarly while animal experiments show that alcohol consumption results in less heart disease, and longer life overall, they do not show a link between alcohol consumption and breast cancer (see Hackney et al 1992, Singletary 1997). Alcohol actually reduced the risk of breast cancer metastasis in mice (Vorderstrasse et al 2012).
The (weak) correlation seen in population studies between drinking and breast cancer diagnosis is then probably due to a confounding factor, as so often can happen in population studies. In this case it is probably simply that drinkers are more likely to screen for breast cancer (shown in Mu and Mukamal 2016) – screening definitely increases diagnosis. Women from lower socio-economic superbs are less likely to drink and less likely to be screened for cancers (see here). Land et al 2014, which screened all their subjects (i.e. controlling for screening incidence), showed no link between drinking and breast cancer – indeed drinking was associated with slightly less risk of diagnosis (of both breast and colon cancer).
The modest degree of increased risk of colorectal cancer for alcohol drinkers may also be spurious or exaggerated for the same reason – drinkers screen more, which results in more diagnoses. Thun et al 1997 shows exactly the same low chance of developing colorectal cancer irrespective of drinking, for men and women.
As far as a potential biological mechanism, some fecal bacteria have been shown (in lab studies) to convert alcohol to acetaldehyde (Jokelainen et al 1994). But then other bacteria have been shown to break down acetaldehyde (Nosova et al 2000) – the flora of the human gut are complex and not well researched. Acetaldehyde has been shown to exist in the colons of rats but the level was not affected by feeding the rats alcohol (Seitz et al 1990). Research is needed to see if drinking really can increase acetaldehyde levels in the colon. In humans alcohol is absorbed in the stomach and small intestine which makes it harder to explain how alcohol might reach the colon where it might be converted to acetaldehyde by bacteria. Whereas the risk of oral cancer is much higher among people without the genes to produce some acetaldehyde processing enzymes, this does not appear to be the case for colorectal cancer (Tiemersma et al 2003) (though the link between drinking and colorectal cancer is highest in Asian studies while barely statistically significant in Europe/America/Australia (Fedirko et al 2011)). So a causal link between drinking alcohol and colorectal cancer remains speculative, and if there is a link, it is not strong – drinkers still get colorectal cancer at much the same rate as non/rare drinkers.
That leaves us only with liver cancer remaining to be discussed. This is a rare but deadly cancer, causing 1% of deaths in the USA, though most cases are due to viral infection (hepatitis), then obesity, diabetes, other disease, and genetics. We perhaps again have the plausible biological mechanism of acetaldehyde but only at high levels of drinking as the liver is very efficient and fast at breaking down acetaldehyde. It’s more plausible that alcohol, through its effect on liver disease, leads to higher risk of liver cancer. However very few drinkers develop liver disease, so the degree of absolute increase in risk in absolute terms is tiny and probably only for long-term alcoholics or those unlucky enough to have liver damage from hepatitis or other causes.
Alcohol is a surprisingly benign hepatotoxin (i.e. the liver is remarkably good at dealing with it). In mice and rat laboratory studies it isn’t possible to induce cirrhosis from alcohol alone, and this may be the case for humans as well.
Drinkers live longer. Do they die more from cancer?
So it’s plausible that alcohol contributes to two cancers, both rather rare, and its influence can’t be great. If this conclusion is correct then we would not expect drinkers to die much more often from cancer, and this is exactly what is observed in longitudinal population studies. Meta-analyses of mortality studies report cancer death is barely higher among drinkers than occasional/zero drinkers, and confined to the cohort who admit to consumption of more than 50g of alcohol per day (see Jin et al 2012, table 2) which is very probably an under-estimate of actual consumption. In Thun et al 1997 (a study of death among middle class, middle aged and elderly Americans) the heaviest drinkers showed higher rates of death from alcoholism and injury, as expected; they were also more likely to die from aerodigestive and liver cancer, again as we now expect, though none of these deaths were common; as expected, they were less likely than non/light drinkers to die from heart disease, stroke and other circulatory disease (each of which were major causes of death); there was no consistent relationship between rates of drinking and rates of death from colorectal or breast cancer, as we now should expect. So the evidence fits together – drinking alcohol might ever so slightly increase the risk of dying from aero-digestive or liver cancer, that’s all.
Finally, it is worth noting that there is one clear way in which drinking alcohol (at least to non-alcoholic levels) increases your chance of getting cancer, and that is through its effect on living longer. For most cancers, age is the dominant ’cause’ (e.g. half of all colorectal cancers occur in people aged over 70 years old, half of breast cancers are for women aged over 62). Drinkers live longer, largely due to reduced levels of heart disease (also diabetes, and dementia). Perhaps it’s also due to drinkers being more social or some other beneficial behaviours – but whatever the mechanisms, drinkers live longer and will therefore will survive long enough to get more cancer – because everyone has to die from something and the incidence of the vast majority of cancers increases substantially with age. This fact alone is enough to produce an association, but not a direct causal one, between drinking and cancer.
Thanks to these people for helping improve my earlier drafts:
Dr Nick Danenberg, Ehrenberg-Bass Institute for Marketing Science, University of South Australia.
Ian Olver, Professor of Translational Cancer Research, Director Sansom Institute for Health Research, University of South Australia.
Philippa Martyr, Communications and Research Officer, North Metropolitan Health Service Mental Health; Adjunct Senior Research Fellow, Psychiatry and Clinical Neurosciences, University of Western Australia.
Wiemer Snijders, consultant, The Commercial Works, Netherlands.
Malcolm Wright, Professor of Marketing, Massey University, New Zealand.
Copies were emailed to Prof Jennie Connor at her Otago University email address on 8th and 24th of August, inviting her to correct any errors or omissions. She has not yet replied.
Bagnardi et al (2015) “Alcohol consumption and site-specific cancer risk: a comprehensive dose–response meta-analysis” in British Journal of Cancer 112, 580–593 | doi: 10.1038/bjc.2014.579.
Brien, Susan et al (2011) “Effect of alcohol consumption on biological markers associated with risk of coronary heart disease: systematic review and meta-analysis of international studies”, in British Medical Journal, BMJ 2011;342:bmj.d636
Connor, Jennie (2016) “Alcohol consumption as a cause of cancer”, in Addiction, “For Debate” section. doi:10.1111/add.13477
Din N; Allen IE; Satariano WA; Demb J; Braithwaite D. (2016) “Alcohol consumption and mortality after breast cancer diagnosis: the Health and Functioning in Women study”, Breast Disease Vol 36, No 2-3, 2016, pp77-89.
Fedirko, Tramacere, Bagnardi, Rota, Scotti, Islami, Negri, Straif, Romieu, La Vecchia, Boffetta & Jenab (2011) “Alcohol drinking and colorectal cancer risk: an overall and dose–response meta-analysis of published studies” in Annals of Oncology 22: 1958–1972, doi:10.1093/annonc/mdq653
Gou, Xie, Yang, Liu, Zhang, Li, and He (2013) “Alcohol Consumption and Breast Cancer Survival: A Meta- analysis of Cohort Studies”, in Asian Pacific Journal of Cancer Prevention, Vol 14, No 8, pages 4785-4790. DOI:http://dx.doi.org/10.7314/APJCP.2013.14.8.4785
Je, DeVivo and Giovannucci (2015) “Long-term alcohol intake and risk of endometrial cancer in the Nurses’ Health Study, 1980–2010”, in British Journal of Cancer (2014) 111, 186–194 | doi: 10.1038/bjc.2014.257
Jin, Cai, Guo, Zhu, Li, Yu, Zhang and Chen (2012) “Alcohol drinking and all cancer mortality: a meta-analysis” in Annals of Oncology 24: 807–816, 2013 doi:10.1093/annonc/mds508
Jokelainen, Roine, Vaananen, Farkkila, Salaspuro (1994) “Invitro acetaldehyde formation by human colonic bacteria”, in Gut, 35: 1271-1274
Hackney, John F. and Robert W. Engelman & Robert A. Good (1992) “Ethanol calories do not enhance breast cancer in isocalorically fed C3H/Ou mice”, in Nutrition and Cancer, 18:3, 245-253, DOI: 10.1080/01635589209514225
Land, Liu, Wickerham, Constantino, and Ganz (2014) “Cigarette smoking, physical activity, and alcohol consumption as predictors of cancer incidence among women at high risk of breast cancer in the NSABP P-1 Trial”, in Cancer Epidemiology Biomarkers and Prevention, 23(5): 823–832. doi:10.1158/1055-9965.EPI-13-1105-T
Lowry, Sarah J. and Kris Kapphahn, Rowan Chlebowski, and Christopher I. Li (2016) “Alcohol Use and Breast Cancer Survival among Participants in the Women’s Health Initiative”, in Cancer Epidemiology, Biomarkers and Prevention, 25(8); 1268–73. doi: 10.1158/1055-9965.EPI-16-0151
Mu L, and KJ Mukamal (2016) “Alcohol consumption and rates of cancer screening: Is cancer risk overestimated?”, in Cancer Causes Control, Feb, 27(2):281-9. doi: 0.1007/s10552-015-0692-3.
Newcomb Polly A. and Ellen Kampman, Amy Trentham-Dietz, Kathleen M. Egan, Linda J. Titus, John A. Baron, John M. Hampton, Michael N. Passarelli, and Walter C. Willett (2013) “Alcohol Consumption Before and After Breast Cancer Diagnosis: Associations With Survival From Breast Cancer, Cardiovascular Disease, and Other Causes” in Journal of Clinical Oncology, DOI: 10.1200/JCO.2012.46.5765
Nosova, H. Jousimies-Somer, K. Jokelainen, R. Heine, M. Salaspuro (2000) “ACETALDEHYDE PRODUCTION AND METABOLISM BY HUMAN INDIGENOUS AND PROBIOTIC LACTOBACILLUS AND BIFIDOBACTERIUM STRAINS”, in Alcohol and Alcoholism, DOI: http://dx.doi.org/10.1093/alcalc/35.6.561 561-568
Seitz, HK, and Becker P. Alcohol Metabolism and Cancer Risk. Alcohol Research & Health. 2007;30(1):38-47
Singletary, Keith (1997) “Ethanol and Experimental Breast Cancer: A Review”, in Alcoholism: Clinical and Experimental Research, Vol. 21, No. 2. pp 334-339.
Sun Q, Xu L, Bo Zhou, Wang Y, Jing Y, Wang B (2011) “Alcohol consumption and the risk of endometrial cancer: a meta-analysis”, Asia Pacific Journal of Clinical Nutrition, 20(1): 125–133.
Taubes, Gary (1995) “Epidemiology faces its limits”, in Science, 269:5221, 164-69.
Thun, Petro, Lopez, Monaco, Henley, Heath and Doll (1997) “ALCOHOL CONSUMPTION AND MORTALITY AMONG MIDDLE-AGED AND ELDERLY U.S. ADULTS”, in The New England Journal of Medicine, Vol.337, No.27.
Vorderstrasse, B.A., Wang, T., Myers, A.K., Katherine L. Wyrick and Gary G. Meadows (2012) “Alcohol consumption suppresses mammary tumor metastasis in a syngeneic tumor transplantation model”, in Breast Cancer Research and Treatment 136: 729. doi:10.1007/s10549-012-2275-2