Tuesday, 28 September 2021

Smoking, COVID-19 and Mendelian Randomisation

It's been interesting to watch how the Guardian has been covering the research into smoking and COVID-19. Like many media outlets, they reported the news last spring that researchers had found an inverse association between smoking status and Covid-related hospitalisations and that work was underway to see if nicotine patches would help people recover from the disease. 

Since then there have been many more relevant studies published which the Guardian has ignored, possibly because they nearly all show that smokers are less likely to be infected with the virus than non-smokers

It has only revisited the topic twice. The first was when some researchers in San Francisco did some modelling and claimed that smokers (and vapers) were more 'medically vulnerable' to COVID-19. The second was when a study was retracted because of undisclosed 'links to the tobacco industry'.

In the meantime, there have been six editions of a comprehensive meta-analysis, the last of which looked at 87 studies and concluded that smokers are 33% less likely to be infected with the coronavirus, no more likely to be hospitalised with Covid than non-smokers and no more likely to die from Covid than non-smokers. 

The Guardian is not alone in ignoring all this research. Hardly anybody in the media has touched the issue for over a year. But the Guardian is back today with a new story...

Smokers up to 80% more likely to be admitted to hospital with Covid, study says

Smokers are 60%-80% more likely to be admitted to hospital with Covid-19 and also more likely to die from the disease, data suggests. 

 A study, which pooled observational and genetic data on smoking and Covid-19 to strengthen the evidence base, contradicts research published at the start of the pandemic suggesting that smoking might help to protect against the virus. This was later retracted after it was discovered that some of the paper’s authors had financial links to the tobacco industry.

The unwary reader might assume from this that only one study has suggested that 'smoking might help to protect against the virus' and that this study was retracted. This is untrue. Dozens of studies have come to that conclusion, including in top journals such as the NEJM, Nature and the Lancet, and the one that was retracted was not published 'at the start of the pandemic'. (It is also important to note that it was not retracted because there was anything wrong with it; only that the journal has a policy of not publishing research by anyone with 'links' to tobacco.)

The new study uses Mendelian Randomisation which looks at the health outcomes of people who have certain genes. In this instance, it looks at people who have a genetic propensity to smoke. Apparently these people were more likely to be hospitalised with Covid, although the authors acknowledge that a genetic propensity to smoke may also correlate with a more general propensity to take risks with one's health such as ignoring social distancing guidelines. 

Mendelian Randomisation can be a very useful tool, but it has so far been shown to be weak at best and misleading at worst when it comes to lifestyle-related diseases. The obvious problem is that people who have genes which give them a propensity to do something do not necessarily do it. This is a particular problem when it comes to smoking because people's propensity to smoke has been severely curtailed by decades of public education, stigmatisation, high taxes and so on. 

Consequently, MR studies have found a link between smoking and lung cancer, but the association is much smaller than that found in observational epidemiology. In MR studies, it barely doubles the risk whereas smoking increases the risk by a factor of 10 or 20 in reality. If you take these studies seriously, smoking doesn't increase the risk from most cancers at all and only mildly increases the risk from a few others. (I have written about this before.)

MR research into alcohol has also been near-useless. MR studies have been cited as evidence that moderate drinking doesn't reduce the risk of heart disease (in contrast to many dozens of epidemiological studies which show that it does), but it is an inconvenient fact that MR studies don't show any risk from drinking either. An MR study published last year failed to find any statistically significant link between drinking and any disease except, bizarrely, lung cancer.

The methods are simply too blunt to find anything other than the strongest associations (smoking and lung cancer being the obvious example). Advocates of MR in lifestyle research claim that it cuts out confounding factors, such as socio-economic status, but it has a much bigger problem in that it incorrectly identifies people as smokers or drinkers when they are not. 

“The study adds to our confidence that tobacco smoking does not protect against Covid-19, as their Mendelian randomisation analyses are less susceptible to confounding than previous observational studies,” wrote Dr Anthony Laverty and Prof Christopher Millett of Imperial College London in a linked editorial published in the journal Thorax.

This is activist-driven spin (we have encountered Christopher Millett before and the editorial in question openly calls for more 'tobacco control' policies). The obvious point is that if smokers were 60-80% more likely to be admitted to hospital with Covid, we would see evidence of it in the hospital admissions data. But we don't. Time and time again, we find that smokers are less likely to be admitted to hospital with Covid. 

MR studies take a bit of work for journalists to get their heads around and the studies themselves rarely provide enough information for the reader to see what is going on. In this instance, as with the moderate drinking issue, MR seems to have been wheeled out to contradict findings that are inconvenient for the 'public health' lobby rather than to provide illumination.

The study also contains some observational epidemiology. The number of smokers in this part of the study is suspiciously small. Only 3.3% of the sample confess to smoking (the national rate is 14%). This may reflect the demographics of the kind of people who sign up for these things, or it may reflect undisclosed smoking, or both. Whatever the reason, the researchers still found that smokers were less likely to be infected with Covid, with heavy smokers being half as likely to be infected. You have to go to the supplementary tables to discover this. The authors do not report it in the text.

The idea that smokers are more likely to suffer more from COVID-19 if they get it is, of course, very plausible. Nearly all the evidence shows that former smokers are more likely to be hospitalised with Covid, presumably because they are more likely to have pre-existing health conditions.

The question is whether smokers are less likely to catch the virus in the first place. And if they are, why? If they aren't, why do antibody tests of whole populations (before vaccination campaigns began) find that smokers are less likely to have antibodies, i.e. they are less likely to have been infected? 

Confounding factors cannot be blamed for this and a study which combines the MR method of assuming somebody is a smoker based on their genetic profile with an observational element based on a sample that is clearly not representative of the public at large and contains hardly any smokers does not give us a compelling answer. Nor, if this study is correct, does it explain why all the other studies are wrong. 

No comments: