Friday 17 April 2020

Smoking and COVID-19 - the Glantz/Patanavanich meta-analysis

With the FDA admitting that there is no evidence that vaping exacerbates COVID-19 and the Norwegian Institute of Public Health taking smoking off its list of risk factors for the disease, I've been expecting a reaction from Stanton Glantz. I imagine it has been all hands to the pumps in San Francisco this week as the legacy 'public health' establishment plans its counter-attack.

It arrived last night when Glantz and a Thai colleague, Roengrudee Patanavanich, published a meta-analysis of sorts as a pre-print claiming that smoking more than doubles the risk of COVID-19 progressing in people who have the virus. It includes a telling sentence in the Methods section:

We will widely disseminate our findings working with the media and interested clinical and public health groups. 

I bet you will, Stan.

You can tell it has been produced in a hurry. The results section consists of two short paragraphs and the conclusion is just two sentences long.

It looks at twelve studies, ten from China and one each from South Korea and the USA. The large study from New York which found no association between smoking and COVID progression - and found that smokers are significantly less likely to be hospitalised with the disease - is conspicuous by its absence.

The headline claim is...

A total of 9,025 COVID-19 patients included in our meta-analysis, 878 of whom (9.7%) experienced disease progression and 495 with a history of smoking (5.5%). A total of 88 patients with a history of smoking (17.8%) experienced disease progression, compared with 9.3% of never smoking patients.

The meta-analysis showed an association between smoking and COVID-19 progression (OR 2.25, 95% CI 1.49-3.39, p=0.001)

It also claims that...

There was not significant heterogeneity among the studies 

You could have fooled me. The odds ratios in the studies range from 0.11 to 12.19, ie. a 89% reduction in risk to a twelve-fold increase! Clearly, some of these results are very wrong. Bunging them all together in a meta-analysis is not going to get you anywhere. Garbage in, garbage out.

As you can see from this graphic, only three of the studies (supposedly) produced statistically significant results. One of them is Liu et al., which only included five smokers and is the source of the wild claim that smoking increases the risk of coronavirus progression fourteen-fold. The others are Guan et al. and this webpage from the Centers for Disease Control. Glantz gives both of these a big weighting and large odds ratios, but neither of the figures he attributes to them come from the studies themselves.

Guan et al. showed some raw data for smokers and ex-smokers in Chinese coronavirus wards, but the authors did not comment on smoking as a risk factor, nor did they produce an odds ratio. Both current and former smokers are under-represented in their study (compared with smoking prevalence in China), but they do appear somewhat more likely to develop severe symptoms. The increased risk is more like 50%, however, whereas Glantz claims the study shows a trebling of risk.

If the Guan study implies some risk of COVID progression from smoking, the same cannot be said of the CDC study which Glantz gives the greatest weighting to. This has hospitalisation data for 6,354 COVID cases in the USA. Most of the cases in the meta-analysis therefore come from this one data set.

It found that only 1.3% of the COVID cases were current smokers, a striking statistic in itself. Of the 88 smokers for whom the CDC has hospitalisation data, 27 were hospitalised (30.7%) and five ended up in ICU (5.7%).

2.3% of the cases were former smokers. Of the 178 ex-smokers for whom the study has hospitalisation data, 78 were hospitalised (43.8%) and 33 ended up in ICU (18.5%).

The CDC doesn't give a specific figure for nonsmokers, but the overwhelming majority of cases had never smoked (or so they said) and the overall rate of hospitalisation was 34%. The overall proportion of cases that ended up in an intensive care unit (ICU) was 5.8%.

Former smokers therefore did worse than average, presumably because of underlying health conditions, and current smokers did better than average. Both groups were heavily under-represented among the cases as a whole: 14 per cent of Americans are current smokers and an even larger proportion are ex-smokers.

A major caveat is that we don't have hospitalisation status for three quarters of the total cases recorded by the CDC, and the CDC says itself that...

...for some underlying health conditions and risk factors, including neurologic disorders, chronic liver disease, being a current smoker, and pregnancy, few severe outcomes were reported; therefore, conclusions cannot be drawn about the risk for severe COVID-19 among persons in these groups.

Glantz pays no attention to that advice. Instead, he lumps ex-smokers in with current smokers and produces a relative risk of 2.60 (1.82-3.73). Again, this figure does not come from the CDC and it does not reflect the data in the CDC study which shows current smokers to be less likely to be hospitalised and slightly less likely to enter ICU. Glantz's odds ratio of 2.60 - ie. 160% greater risk - is due to the outcomes seen in ex-smokers, not smokers.

Glantz's gives so much weight to the imaginary results from these two studies that it hardly matters about the rest. Nevertheless, there are some bizarre interpretations of sketchy evidence. For example, the study from South Korea (Kim et al.) involved 28 COVID patients, of whom five were smokers. Six of the 28 patients were given supplementary oxygen in hospital, of whom two were smokers. From these crumbs of evidence, Glantz portrays the study as showing that smokers are more than three times likely to become critically ill if they contract COVID-19.

Dong et al. provides detailed case reports on eleven COVID patients in China. Only one patient had a history of smoking. The authors do not attempt any statistical analysis and it is impossible to derive any relative risks from the doctors' notes. Glantz nevertheless attempts to do, somehow conjuring up an odds ratio of 1.15.

In Zhang et al.'s study of 140 COVID patients in Wuhan, only two of the patients (1.4%) were current smokers. The smoking rate in China is 27 per cent and the authors comment on how few smokers appear in their sample, saying that 'the exact underlying causes of the lower incidence of COVID-19 in current smokers are still unknown'. For Glantz's purposes, the only thing that matters is that both of the smokers had severe symptoms, from which he derives an odds ratio of 3.04 (0.73-12.69) which, once again, does not feature in the study itself.

Despite only having two smokers in the sample, Zhang et al.'s study is given a relatively heavy weighting of 6.74% by Glantz, thereby amplifying its purported findings in the meta-analysis. By contrast, Huang et al.'s study published in the Lancet had three smokers, none of whom developed severe symptoms. Glantz gives that study a weighting of just 1.74%.

If you read the twelve studies, you will struggle to find much evidence for the counter-intuitive theory that smoking reduces the risk of serious COVID-19 complications, but nor is there good evidence that it increases the risk. There is a suggestion of increased risk in Guan et al. - and, at a push, in Shi et al. - but the authors of the studies do not draw that conclusion and it seems to be more of an issue for former smokers than current smokers (perhaps because they are older). Contrary to Glantz's claims, the CDC data do not support the hypothesis that smoking increases the risk of coronavirus complications, and the agency explicitly cautions against drawing such a conclusion.

The New York study of 4,103 patients is the best evidence yet that current smoking is irrelevant to the current viral pandemic, but that isn't in the meta-analysis. All the other studies have far too few smokers in them to draw any conclusions either way.

And it is the lack of smokers coming to the attention of health authorities with COVID-19 that is the most interesting thing. It is why Dr Farsilinos started researching this in the first place. Since most of the studies are based on hospital data, you have to be hospitalised to be studied - and smokers are strangely under-represented in coronavirus wards. Current smoking rates in all the studies in Glantz's meta-analysis are much lower than would be expected: 10 per cent in Yang et al., 6.7% in Wan et al., 3.9% in Mo et al., 7% in Huang et al., 9% in Dong et al., 1.9% in Guan et al., 6% in Zhou et al., 1.4% in Zhang et al. and in 6.4% in Liu et al. In Shi et al., only 8.2% of cases had any smoking history.

The highest smoking rate recorded in any of the studies is 18.5% in Kim et al.'s study from South Korea and that is still well below the national smoking rate. In the USA, the figure from the CDC is 1.3%. In New York, it is 6.7%.

Under-reporting of smoking status by patients is always an issue and it no doubt plays a part here, but is hard to believe that under-reporting alone can explain such remarkably low figures. If there is something about nicotine that helps people fend off the virus, or stop it progressing, it would be useful for the world to know. (It could also help explain why current smokers seem to do better than former smokers.)

There may be nothing in the theory, but there seem to be plausible biological mechanisms for it and serious people are giving it serious consideration. Stan Glantz is not a serious person and it was inevitable that he would respond to the suggestion that tobacco might be anything less than the root of all evil with yet another shabby piece of junk science.

No comments: