A couple of days ago, an “innovative study” by a research team from the University Hospital Southampton NHS Foundation Trust, Bangor University and University of Southampton presented a fresh attempt to express risk association by comparing apples with pears, or, in this specific case, by comparing the risk of smoking cigarettes with that of drinking a bottle of wine.
With such a catchy title, media attention was guaranteed – even if one might have hoped for a little more caution and diligence in scientific news reporting after last year’s two (in-)famous episodes when initial, sensationalist headlines about two Lancet publications on alcohol and health were later found to be badly wanting in accuracy and detail.
So, looking back at the fundamentals, what do we know about the risks of alcohol consumption? Taken from an all-cause mortality risk perspective, the answers are reassuring: consuming a little more than a bottle of wine (up to 100g of pure alcohol) per week has been shown to maximize one’s life expectancy. Moreover, if abstainers are included as a reference group, we learn from such an analysis that abstainers face about the same all-cause mortality risk than drinkers consuming around 3 bottles of wine (around 250g-300g of pure alcohol) per week. This fact was re-confirmed by the Lancet publication from April last year (by Wood and his colleagues in 2018) – even if initial headlines suggested the exact opposite.
How strong can the evidence in the latest study by Theresa Hydes’ research team be considered? To the question of “how many cigarettes are there in a bottle of wine”, they answer that the increase in absolute cancer risk of drinking one bottle of wine per week could be seen as “roughly equivalent to five cigarettes per week (4.7) for men […] and ten cigarettes (9.5) for women”. In other words, 0.7 cigarettes per day for men and 1.4 cigarettes per day for women.
To arrive at this conclusion, the researchers had to take recourse to a specific methodological tweak: since there is no evidence stating risk associations for very low levels of smoking, – or so the authors claim – they had to extrapolate these risk associations from high levels of smoking by assuming and using a log transformation.
In addition, for the alcohol-cancer risk association, the authors chose a selective approach, considering only one, rather dated, source (Corrao in 2004), while at the same time ignoring more recent publications such as the more comprehensive meta-analyses by Bagnadi (2015) or Yoon-Jung Choi (2018) consisting of more than 60 cohort studies.
This is surprising since Yoon-Jung Choi found that consuming up to 15g of pure alcohol per day (roughly a bottle of wine per week) ‘was not associated with the incidence of most cancers […but] was significantly associated with a decreased incidence of both female and male lung cancer’ (a finding one would deem relevant when looking at the particular comparison with smoking…). The list of cancer types considered by Theresa Hydes and her team are the same as the ones considered by Yoon-Jung Choi in 2018. However, the latter concluded that there was no significant association between light drinking and the incidence of the listed cancers (e.g., oesophageal, stomach or liver cancers).
When a meta-analysis (like Jin in 2013) takes an even broader scope and looks at all-cancer mortality risks (including non-alcohol-related cancer types and those types where light drinking may have a protective effect) the findings are even more different, showing that people drinking up to 12.5g pure alcohol per day seem to have a lower all-cancer mortality risk.
The questions that remain when discussing the reliability of the study’s findings thus are plentiful and not insignificant: why didn’t the authors include more findings from more recent meta-analyses? And how can it be they write about ‘absolute’ cancer risk when ignoring several cancer types in the analysis?
To the research team’s credit, Theresa Hydes freely admits that ‘this study is not saying that drinking alcohol in moderation is in any way equivalent to smoking’. Logically, such a statement makes about as much sense as telling somebody not to think about their ears. Try as you might, dear reader, but when reading these lines you WILL think about your ears! It’s no different with the study: try as you might, but while discussing it your brain will start comparing wine drinking and smoking (even though there is no such thing as equivalence, as the authors themselves admit).
Seen from this angle, the media echo and debate about the study bear the features of a re-framing and neuro-linguistic programming exercise, rather than that of an objective, fact-based scientific discussion. To the researchers you are almost inclined to say: si tacuisses, inquisitor mansisses…
