Friday, 16 December 2022

Take no one's word for it

I take it as a slight badge of honour that an article I wrote in the early days of Covid was flagged as fake news on Facebook. I wrote a follow up to the offending article for Spiked two weeks ago titled 'The great Covid and fags cover up'. Regular readers will know what I'm talking about. If you don't, hit the link.

An organisation called Logically is Facebook's official fact-checking partner.

When Logically rates a piece of content as false, Facebook will significantly reduce its distribution so that fewer people see it, apply a warning label to let people know that the content has been rated false, and notify people who try to share it.

They have now rated my new Spiked article as 'misleading' which is a rung above 'false' in their ladder of truthiness. Why is it misleading? According to the accompanying blog post...

The article includes links to Snowdon's blog, where he has compiled a list of studies (including those published by reputed medical journals such as Lancet and Nature), and claims that an overwhelming “87 percent” of the studies acknowledge that smokers are typically less likely to catch the virus and be hospitalized.

On reviewing the exhaustive list of studies that Snowdon refers to, Logically found that several of them are pre-prints rather than peer-reviewed published papers.

You don't need to dig too deeply to see that. I explicitly say a study is a pre-print in the few cases where that is the case. There are only four of them, two of which use the same UK Biobank data. Excluding them makes no difference to the overall picture. The vast majority of studies have found smokers at less risk of getting COVID-19 and hardly any of them have found the opposite.
The author himself says that he has not done a "systematic review" of all the studies that have been listed.
Quite true. It is a compilation of every relevant study that crosses my radar, although I have some expert help.

In his article, Snowdon argued that search results for COVID-19 and tobacco/smoking lead to WHO briefs and articles in publications such as the British Medical Journal (BMJ). He claims that these studies merely state that smokers are at greater risk without addressing the actual findings of the studies that suggest smokers had a reduced risk of contracting COVID-19.

Indeed I do. Anyone who does a casual search would come away thinking that there was little to no evidence of a protective effect. Worryingly, this is all the fact-checker seems to have done. He focuses on those very same out-of-date and inaccurate websites that come up at the top of the listings - in particular, a short opinion piece by Jonathan Grigg (or 'John Griggs', as the fact-checker calls him):

In an article published in The Lancet on August 16, 2022, Professor John Griggs, a respiratory and environmental medicine specialist, acknowledged that there were studies in the early stages of the pandemic that appeared to provide evidence that smoking might protect against COVID-19. However, he also noted that during those early stages of the pandemic, "scientific journals rightly responded to the SARS-CoV-2 pandemic by more rapidly publishing their COVID-19 research”.

Yes, but the point is that these studies kept coming out long after the early days of the pandemic and have been remarkably consistent. 

Prof. Griggs also cited a more detailed article published in The Lancet, which examined epidemiological studies of tobacco use and COVID-19, and found that the studies have produced conflicting results, meaning further research is needed before one can claim that smoking affects susceptibility to the virus. The article nonetheless concludes that "the evidence indicates that smokers are at greater risk of poor outcomes from COVID-19, including hospital admission and progression to severe disease than are non-smokers."

This is a reference to an article by the renowned anti-smoking academic Neal Benowitz and his colleagues. It looked at 51 relevant studies, none published after August 2021, and found the evidence to be 'conflicting'.
Firstly, given what the fact-checker says about systematic reviews above, let's bear in mind that Grigg notes that Benowitz's article is...
...not a formal systematic review and based largely on searches up to August 2021
So how does this collection of studies differ from mine, apart from being half the size (and also smaller than the meta-analysis I mentioned in the article but which the fact-checker does not address)?
We identified conflicting evidence for the effects of cigarette smoking on the incidence of SARS-CoV-2 infection (appendix pp 3–6). Specifically, we identified 12 multivariable and eight univariate or frequency-reporting studies that found a positive effect of cigarette smoking on SARS-CoV-2 infection rates, and 14 multivariable and 12 univariate or frequency-reporting studies that found a negative effect.

That does indeed sound like a highly conflicting literature, but is it? Dip into the appendix and you will see that many of the studies are either irrelevant or don't say what Benowitz claims. The criteria for my list is that the study must be an epidemiological study showing the relative risk for SARS-CoV-2 infection among current smokers. Benowitz reckons he found twelve studies using multivariate models which show that smokers are more likely to get COVID-19, but he is clearly using far more generous criteria because they include the following... 
Li et al. - an ecological study about vaping, not smoking.  
Yoshikawa et al. - an ecological looking at smoking rates in Japanese districts. 
Liu et al. - another weak ecological study.

Wang et al. - a study of hospitalisations, not infections.

Gaiha et al. is a new one to me. It found, rather implausibly, that smokers were more likely to get COVID-19 but only if they vaped as well. Past use of e-cigarettes - but not current use - was strongly associated with COVID-19 risk. Make of that what you will.

McQueenie et al. is another study using Biobank data (two pre-prints using this data are in my list). It found that current/former smokers (the study didn't distinguish between the two) were 26% more likely to test positive. 

Colaneri et al. clearly states that 'being a current smoker was negatively associated' with COVID-19 infection. Presumably, Benowitz's excuse for including in his list was that it found that former smokers (like other nonsmokers) were more likely to get it.
Chen et al., which is in my list, didn't find an association between smoking or vaping and Covid infection but did find an association between vaping and smoking (i.e. dual use). 
Didikoglu et al. is another Biobank study. This one found that people whose mothers smoked in pregnancy were more likely to test positive. It is unclear from the study whether the same was true of people who smoked themselves.
Ho et al. is yet another Biobank study. This one found that current/former smokers were 45% more likely to test positive for SARS-CoV-2 (1.45 (95% CI 1.19 to 1.79)).
How many times was the Biobank data dredged during the pandemic?! Prats-Uribe et al. is yet another study using this dataset. This one separated current smokers from former smokers and found that current smokers were 42% more likely to test positive (1.42 (95% CI, 1.21–1.66)).
Of these twelve studies, only two fit my criteria and one is already on it so I've amended the list to include Prats-Urbine et al.
Benowitz also claims to have four studies using univariate models which find smokers to be at greater risk of infection. They are...

Albiges et al. didn't look at infections, it looked at mortality risk. 

Li et al. is the same study mentioned above; an ecological study about vaping, not smoking.

Dev et al. found an association between smoking and infection but this disappeared after adjustment. I'm happy to include this study in the list as supporting the null hypothesis.

Similarly, Mostafa et al. found associations with both former and current smoking which disappeared after adjustment for confounding factors. I'll add this one as a null finding too.

Unless we think epidemiological findings are more robust before they are adjusted for exogenous variables, none of these studies support Benowitz's argument. And none of the 'frequency-reporting studies' are relevant at all (I won't list them but feel free to check them yourself).

Benowitz also cites five studies which, he says, support the null hypothesis, i.e. they find no association either way. They are...

Liu et al. did not look at infection although it did find that smoking was not associated with Covid mortality risk.

Zhu et al. looked at hospitalisations, not infections. 

Dadgari et al. is a study from Iran which does indeed support the null hypothesis. I'll add it to the list.

Dayem et al. is a study from England which doesn't give relative risks for smoking but does say that 'current smoking status appeared to have a protective effect in our cohort after adjusting for comorbidities, as has been observed by others'.

Finally, Benowitz lists 14 studies using multivariate analyysis which show a protective effect from smoking, plus six using univariate analysis. Some of them are on my list already and a few of them don't meet my criteria, but there are eight studies that I've missed. They are...

Ghinai et al. is a study of homeless people in Chicago which found that smokers were less likely to be infected (0.71; 95% CI, 0.60-0.85).

Gu et al. is a US study which found: 'Being a current smoker (self-reported in the latest EHR encounter) was associated with a reduced chance of having positive test results (OR, 0.31 [95% CI, 0.20-0.48]; P < .001).'

Vila-Córcoles et al. is a study from Spain which found smokers 57% less likely to be infected (0.43 (0.25-0.74)).

Fernandez-Fuertes et al., is a study from Spain involving HIV patients. It found that 'active tobacco smoking was the only factor independently associated with lower risk of SARS-Cov-2 infection [Incidence rate ratio: 0.29 (95% CI 0.16–0.55)'.

Green et al. is a study from Israel involving patients with bronchial asthma which found: 'A significantly higher proportion of smokers was observed in the COVID-19–negative group than in the COVID-19–positive group (4734 [13.45%] vs 103 [4.55%]; P < .001).'

Gu et al. is a study from Michigan which found that smokers were much less likely to test positive for COVID-19 (0.31 (0.20-0.48)).

Lombardi et al. is a seroprevalence survey of Italian healthcare workers which found smokers were 59% less likely to test positive (0.41 (0.27-0.61)).

Holuka et al. is a study from Luxembourg which found smokers half as likely to test positive for COVID-19 (0.50 (0.30–0.83)).

These eight, plus the four studies that reached different conclusions, mean that we have an extra twelve studies to add. So let's get rid of the four pre-prints, two of which use the Biobank data. If they were going to be published in journals, they probably would have been by now. I've also taken a couple of studies out of the list, like this one, which are interesting but don't quite fit my criteria. I kept the Mendelian Randomisation studies even though they don't really fit the criteria because they're virtually the only studies that find an increased infection risk from smoking.

The new list is here. It has a grand total of 105 epidemiological studies looking at the relative risk of SARS-CoV-2 infection among current smokers. 

Of the 105 studies, 90 find a statistically significant reduction in risk of infection from SARS-C0V-2 among current smokers.

Eight of them find no statistically significant association either way (although theyy are nearly all below 1.0).

Four of them find an increase in risk.

Three of them found mixed results.

I mean, I guess you could say that these studies are 'conflicting' but there's a pretty clear tendency in one direction, isn't there?

The fact-checker then goes to the next link in the search results and finds the WHO... 

The WHO, in an updated 2020 review, which excluded the non-peer-reviewed reports and pre-prints, found that smoking is associated with increased severity of disease and death in hospitalized COVID-19 patients. However, it could not quantify the risk to smokers of infection by SARS-CoV-2 or hospitalization for COVID-19. The brief said that more population studies are needed to address these questions

Even if we ignore the fact that the WHO is corrupt and incompetent, the important thing to note is that this was a '2020 review'. More population studies have been conducted in the last two and half years and they overwhelmingly point to a protective effect. Perhaps this explains why the WHO haven't updated their webpage?

The fact-check ends with the thoughts of some random doctor in India... 

Logically also contacted Dr. Satyanarayana Mysore, head of the pulmonology department at Manipal Hospitals in Bengaluru, to get more clarity on the matter. Dr Mysore dismissed the claim that smokers are less likely to catch COVID-19, and said that there are major limitations in the studies associating smoking with with reduced risk of getting COVID-19. He noted that WHO has listed smoking and chronic obstructive pulmonary disease (COPD) as major risk factors for COVID-19. He added that smoking does not protect anyone from COVID-19 and has the potential to harm.

I've never heard of this guy and I doubt you have either. He could be as reliable as Aseem Malhotra for all I know. What are the 'major limitations in the studies'? Are they any more serious than in other epidemiological studies? The fact-check asserts that the early studies in this area were 'riddled with various problems' and had 'several biases' but it doesn't say what they are. It doesn't include a single link, even to the article it is fact-checking.

There are some very good fact-checking websites out there. The good ones lay out the facts, link to the data and explain why the claim is right or wrong. They don't appeal to authority or expect you to take their word for it just because they've declared themselves to be fact-checkers. Nullius in verba and all that. 
But I suppose if you are only accountable to Facebook, you can say what you like.

No comments: