Thursday 4 June 2020

Smoking and COVID-19 update

I'm finding it difficult to keep up with all the evidence on smoking and COVID-19.  I suggest you follow Phil on Twitter if you want to get up to speed. He's catalogued 117 pieces of evidence so far in this heroic thread, (UPDATE: he is now at over 3,000!)  most of which show smokers significantly under-represented among Covid patients/cases.

I still have my doubts about whether smoking/nicotine confers any protection against the disease. Simple comparisons between the proportion of Covid patients who smoke and the proportion of smokers in the general population are not the strongest evidence. The smoking rate in Britain, for example, is 15% and all the British studies find smoking prevalence to be about half of that among Covid patients. This one, for example, finds that only 6% were current smokers.

But the average age of patients in that study was 73 and smoking rates are much lower at older ages. In the UK, it is only 8 per cent among people aged 65+. This still implies a difference - and it is certainly striking that decades of smoking don't make people more likely to be hospitalised with this respiratory disease - but claims based on unadjusted figures are bound to exaggerate any protective effect.

This is why it is better to focus on the relatively small number of studies that use epidemiological methods and adjust for other factors. There are not many of them (UPDATE: there are quite a few now, and the list below is incomplete), but it has to be said that their results are fairly consistent.

Let me know if I am missing any, but I can find the following... (UPDATED 7 May 2021)

1. This recent study from Israel involved testing 114,545 people for the virus. The smoking rate was 9.8% among those who tested positive (the national rate is 19%). After adjustments, the authors found that smokers were half as likely to test positive, with an odds ratio of 0.46 (0.41-0.51). This is, as the authors conclude, an 'intriguing finding' which 'may reveal unique infection mechanisms present for COVID-19 which may be targeted to combat the disease and reduce its infection rate.' (Israel et al. 2020)

2. That is very similar to this British study which found that 'active smoking was linked with decreased odds of a positive test result' with an odds ratio of 0.49 (0.34–0.71). (de Lusignan et al. 2020)

3. This British study found that smokers were 27% less likely to test positive for COVID-19 although you have to look carefully to find the evidence because the lead author is the chairman on Action on Smoking and Health and he buries it as much as he can. The odds ratio is 0.73 (0.65-0.81). (Hopkinson et al. 2020)

4. Meanwhile this British study looked at the likelihood of dying from COVID-19 and found that smokers were slightly more at risk, or slightly less at risk, or neither, depending on how the figures were adjusted. (Williamson et al. 2020)

5. Then there is this study from the USA which looked at 3,789 US military veterans aged between 54 and 75 who were tested for COVID-19, of whom 585 tested positive. Smokers were 55% less likely to test positive, with an odds ratio of 0.45 (0.35-0.57). (Rentsch et al. 2020)

6. Finally, there is this study from a badly hit area of France where 661 people were tested for COVID-19 and smokers were 67% less likely to test positive. Odds ratio after adjusting for age: 0.23 (0.09 –0.59). (Fontanet et al. 2020)

7. Large study from Mexico: "Current smokers were 23% less likely to be diagnosed with COVID-19 compared to non-smokers." (Giannouchos et al. 2020)

8. Study from the USA finds smokers 90% less likely to test positive. Odds ratio: 0.1 (0.01-0.8). (Lan et al. 2020)

9. This study - another one from Britain, this time using the Biobank data - claims in the abstract that smokers were 'slightly more likely' to test positive. In fact, the results section show that there was no statistically significant difference. (Cho et al. 2020)

10. By contrast, this study using Biobank data shows smokers under-represented (with an implied odds ratio of 0.73 (0.6-09). (Kolin et al. 2020)

11. French study finds that daily smokers are 76% less likely to be infected with COVID-19 (after adjusting for age and sex). Odds ratio for inpatients: 0.24 (0.14-0.40). For outpatients: 0.24 (0.12-0.48).  (Miyara et al. 2020)

12. This study from Italy found that "current smokers were significantly less likely to be hospitalized for COVID-19 compared with non-smokers (Odds ratio 0.23; 95% CI, 0.11-0.48 after adjusting for age and gender). (Meini et al. 2020)

13. Interesting study of a Covid outbreak on a French Navy aircraft carrier. 76% of the crew members got the disease but smokers were 36% less likely to get it. There was also a "trend towards a lower risk among e-cigarettes users". (Paleiron et al. 2021)

14. Web-based study from Italy finds smokers were half as likely to get COVID-19 than nonsmokers. There was a dose-response relationship, with heavy smokers 62% less likely. (Prinelli et al. 2021)

15. Study of 4,137 Covid patients in South Korea finds smokers 67% less likely to be infected (0.33, CI = 0.28–0.38). (Lee et al. 2021)

16. Study of Covid hospital patients in Turkey finds no difference in disease severity between smokers and nonsmokers, but finds that the "mortality rate was significantly increased in ex-smokers (p= 0.037) and non-smokers (p= 0.001) in comparison to active smokers (p= 0.123)". (Gonca et al. 2020)

17. Large study from Germany finds, based on a seroprevalence survey, that regular smokers are 52% less likely to have had COVID-19 (0.48 (0.31-0.72)). (Gornyk et al. 2021)

18. Study of 10,614 nurses in Madrid finds smokers 77% less likely to have had COVID-19 (0.23 (0.20-0.27)). The finding is not mentioned in the abstract. (Candel et al. 2021)

19. UK study finds smokers are 27% less likely to be hospitalised with COVID-19. The authors do not mention this finding in the text (0.63 (0.44-0.88)). (Cummins et al. 2021)

This meta-analysis finds smokers to be 27% less likely to test positive, based on seven 'fair' quality studies. These include studies 1, 2, 5, 9 and 10. It takes an implied odds ratio from this study of 76 patients, of whom 2 were smokers and 1 tested positive. It also manages to derive an odds ratio from this study, but it is not obvious how this was achieved.

The meta-analysis doesn't include study 3 because it relied on self-diagnosis. There is no explanation for why studies 6, 7 and 8 are absent. Studies 7 and 8 are not in the references so perhaps the authors are unaware of them. Study 11 was published after the meta-analysis was conducted.

There is also an ecological study that compared smoking rates and COVID-19 mortality rates in 38 European countries and found an inverse relationship.


With the exception of the Goldacre study, all of these look at the likelihood of testing positive for the virus, not the probability of being hospitalised or dying from it. (Update: this is no longer true. There are several studies testing whole populations for antibodies which find that smokers are less likelyyyy to have had the disease. This makes what is written below somewhat redundant.)

There are dozens of other studies and datasets showing a low proportion of smokers among Covid hospital patients, but these studies do not specifically look at the question of smoking's effect on the virus. In many cases, the authors do not even comment on the rate of smoking, and in some of them the rate of smoking is not particularly low anyway.

If smokers are less likely to be hospitalised with the coronavirus, it could be because smoking/nicotine makes symptoms less severe or it could be because smokers are less likely to get it in the first place.

The original hypothesis put forward by Konstantinos Farsalinos lent towards the former explanation, but I tend to lean towards the latter because there are studies adjusting for other factors, crucially age and sex, which back it up. In addition, there are several other studies which do not produce odds ratios, but which strongly suggest protection against infection, such as this, thisthisthis and this.

All in all, the evidence that smoking reduces the chance of a person testing positive for the virus is methodologically stronger than the evidence implying that smoking helps combat the disease once a person is infected. From the evidence above, one might tentatively conclude that smoking reduces the risk of infection by 50 per cent.

But there is one thing that has been bugging me about this theory for weeks. A lot of the evidence showing low rates of smoking among patients comes from China where the smoking rate is 27%. That figure masks a huge gender divide, with 52% of men smoking and only 3% of women. The official figure for women is probably an underestimate as there is a certain taboo about women smoking in much of Asia, but there is nevertheless a very large gap in prevalence between the two genders.

Therefore, if smoking confers some sort of protective effect, we should expect to see men enjoying a disproportionate benefit. Do we see that in the statistics for testing and hospitalisation? Not really, no. Men tend to be slightly over-represented in Covid wards.

However, we also know that men are more likely to be hospitalised and die from COVID-19 - twice as likely in some studies. (We also know, although it won't have a big impact on these statistics, that there are more men than women in China as a horrible consequence of the one child policy.)

So the real question is whether China has a smaller proportion of men with COVID-19 than countries that have a more equal distribution of smokers across the sexes. The evidence must be out there, but I have yet to find anything conclusive.

Looking at the studies that have crossed my radar while looking at the smoking hypothesis, here's some evidence for China. I include only relatively large studies involving more than 100 patients/cases:

Large study of 1,056 COVID-19 cases in China found that 536 (50.8%) were male.

Study of 585 confirmed cases in Beijing found that 268 (45.8%) were male. No smoking history recorded.

Study of 292 patients in China found that 50.7% were male. The smoking rate among the patients was just 5.1%, although smokers were more likely to be classified as 'severe' than 'mild'.

Study of 285 patients in China found that 128 (44.9%) were male. 11% of the patients were smokers.
Study of 276 patients in the Zengdu District found that 56.2% were male. 12.8% of the total had a smoking history.

Study of 202 patients in China. 107 (51.4%) were males and 101 (48.6%) were females. 5.8% had a history of smoking.

Study of 200 patients in Wuhan found that 99 were male (49.5%). Unusually, most of them were smokers, although 'no difference on the fatality rate of COVID-19 patients was found between smokers and nonsmokers'.

Study of 143 patients in Taizhou, China found that 77 were male (53.8%). 9.8% had a history of smoking (see table - text is wrong) and smokers were no more likely to suffer severely than nonsmokers.

Study of 134 patients from 9 cities outside Wuhan found that 65 (48.5%) were male. 10.4% were smokers.

Study of 121 patients in Wuhan found that 66 (54.6%) were male. 5% were current smokers. 

Study of 101 Chinese medical staff infected with COVID-19 found that 32 (31.7%) were male. Only four were smokers. However, what proportion of Chinese medical staff are female?

I've also looked at the larger Chinese studies previously mentioned in this blog post. The numbers are:

Guan (1,099) 58% male

Shi (487) 53% male

Zhou (191) 62% male

Zhang (140) 51% male

Wan (135) 53% male

Liu (78) 50% male

From this sample of studies, the gender balance is not far off 50/50, with several studies showing a majority of women. Of particular interest are the two large studies at the top because they show cases rather than patients (ie. people who have tested positive but are not necessarily hospitalised). Taking those two sets of numbers together, a slight majority of 51% of cases were women.

How does this compare to countries where there is less of a gender gap in smoking rates? It's not easy to say, but here are the results from the largest of the studies I have come across. The figures refer to COVID-19 hospital patients unless otherwise noted.

Various countries (8,910 patients of whom 1,507 were in China) 59.6% male (5.5% were smokers)

UK (8,699 patients) 71.1% male (smoking data not included)

UK (7,346 patients) 60% male (number of smokers is unclear)

New York (5,700 patients) 60.3% male (15.6% had history of smoking)

UK (5,683 deaths) 63.1% male

New York (4,103 patients) 50.5% male (5.1% were current tobacco users) 

New York (1,999 patients - same study as above) 62.6% male (5.2% were current tobacco users)

Sweden (2,158 ICU patients on 3 June) 74% male

USA (1,482 patients) 54.5% male (smoking data not included)

Italy (441 patients) 61.9% male (4.8% were smokers)

New York (393 patients) 60.6% male (5.1% were smokers)

Switzerland (200 patients) 60% male (4.5% were smokers)

In all of these studies, the majority of patients were male. In most of them, at least 60% were male. By contrast, in the Chinese studies above, only one had a male rate above 60%.

Perhaps this is telling, but unfortunately all of these studies look at patients, not cases. Global Health 50/50 has the gender breakdown of cases for various countries, but there is no clear pattern. In some, such as Pakistan and Ecuador, there are more men. In other, such as Sweden and the Netherlands, there are more women.

I'm not sure what to make of this. I only mention it in the hope that somebody will pick up the gauntlet and find something more conclusive. But it seems to me that for the smoking hypothesis to be correct, we should be seeing disproportionately few cases among groups in which the smoking rate is high (eg. Chinese men). If such an effect is not apparent from the aggregate data then any positive effect from smoking/nicotine is either non-existent or too small to get excited about.

No comments: