Reactions to a study published in the BMC Public Health which estimates the risk of cancer when drinking alcohol as compared with smoking.
A RoundUp accompanied this Before the Headlines.
Title, Date of Publication & Journal
‘A comparison of gender-linked population cancer risks between alcohol and tobacco: how many cigarettes are there in a bottle of wine?’ by Theresa J. Hydes et al. was published in BMC Public Health at 01:00 UK time on Thursday 28 March 2019.
Study’s main claims – and are they supported by the data
As the authors say this study does not suggest that the cancer risk of alcohol is equivalent to that of smoking. The first sentence of the discussion section is really important: they have definitely not shown that drinking a bottle of wine a week is as bad as smoking. In fact, if you take their numbers at face value, then they have shown that even ‘hazardous’ drinking is equivalent to extremely low levels of smoking.
It’s also important to note the authors’ comparison is one bottle of wine per week with five cigarettes per week (not per day). That is a very very light level of smoking.
In general it could potentially be useful to compare the cancer risks of drinking with those of smoking to aid the public understanding of the risk, as the authors suggest, but only if that’s done based on very robust number crunching – and I’m not sure that is quite the case here for the reasons given below. The “equivalent number of cigarettes” for a bottle of wine needs to be seen as only a very rough approximation.
The authors were clear that this study only looks at cancer risk, not risk of other diseases. But leaving out the effects on CVD is an important omission – I don’t think it makes sense to consider cancer in isolation when so many people die of CVD, given that a substantial body of research has shown that moderate alcohol consumption is associated with a lower risk of cardiovascular disease. Some previous evidence has suggested a J-shapes curve of CVD risk with increasing alcohol consumption, e.g. Ronksley et al 2011, BMJ, and even the paper by Corrao et al cited in this paper and used as the main source of relative risks for alcohol on cancer. So although drinking one bottle of wine a week may slightly increase cancer risk, it is possibly that those harms would be more than offset by the benefits on CVD. The wording in the press release around risk could be misinterpreted as meaning that the total risk of developing cancer is 1% in men and 1.4% in women when it actually means that this is the increase in the risk of developing cancer (the total risk is of the order of 50%), as the authors say in the abstract of the paper.
The authors have been clear to state they are not saying the cancer risk of alcohol is equivalent to that of smoking, and that is correct.
I have some concerns about how the authors came up with their figures – a lot of the data they used appears to have come from 2004 (Corrao et al 2004 was where the authors sourced their data on the relative risks of alcohol and cancer), so that is now quite old. The numbers in that 2004 paper do not match the numbers given in supplementary table 1 in this new paper despite being its source. (That 2004 paper also, incidentally, reported an association between moderate alcohol consumption and protection against cardiovascular disease.). There is a more recent meta-analysis – Bagnadi et al 2015 – which looked at alcohol consumption and cancer and I am unsure why the authors did not use data from that instead. Using slightly different numbers for the relative risks could possibly have given very different results. Ideally this would have been quantified by sensitivity analysis, but only a very crude sensitivity analysis was attempted which does not shed much light on this possibility.
The way they estimated relative risks due to cigarettes is based on certain assumptions that may or may not be true, as the risks for the very low levels of smoking considered in the paper were extrapolated from higher levels of smoking using a statistical model – so there is a lot of uncertainty around the numbers the authors have given.
There are a great many assumptions in the analysis, and a great many numbers are combined in complex calculations. Given that the input numbers themselves are not necessarily the best ones to use and have some uncertainty around them, the overall numbers must be hugely uncertain, and I think it’s an important and significant limitation of the paper that they don’t really attempt to quantify that uncertainty.
CVD = cardiovascular disease.
Any specific expertise relevant to studied paper (beyond statistical)?