I'll use data from a huge study in the UK, where they compared the general population (17 million people) with people who died of/with #COVID19 (Pre-print here: shorturl.at/hqtH6). Data on #smoking are from GP files prior to disease, so very little missing (4%). 2/
Among the general pop with smoking data, ~18% were current smokers. Among #COVID19 dead, ~7% were current smokers. Wow, smoking must be really protective, as smokers among the dead are less than half than expected! Not quite. 3/
The authors had sex/age data. Adjusting for age & sex, current smokers were 25% MORE likely to die. This is a strong association in the exact opposite direction. I'm not saying that this 25% figure is the truth, but certainly more likely than comparing unadjusted %. 4/
Smoking data come from GP files. If we collected data directly at the hospital, % of current smokers would likely be lower because some may have recently quit (health reasons) or because nobody asked the patients (smoking info in hosp files is often missing in 'good' times). 5/
Assuming a very optimistic (during a pandemic) 10% underreporting of current smoking (which would be analysed as never smoking), we would have had ~6% current smokers. Former smokers among dead were 62%. Compare these numbers to the (in)famous French preprint. 6/
Miyara et al. (qeios.com/read/WPP19W.3) reported 6.2% current smoking and 61% former smoking among inpatients. Admittedly, smoking prevalence is higher in FRA vs. UK, but study included health prof etc, so maybe not much higher in the source population. 7/
Nevertheless, I hope this illustrates that even big differences in #smoking prevalence between patients/dead and general pop don't necessarily mean lower risk among smokers. Smokers may prove to be at lower risk, but this is definitely not the way to show this. 8/
Finally, underreporting can be a major or minor issue depending on the context. In most studies so far there didn't seem to be a checklist for the physicians to complete. If they had made a note about smoking in the file, fine. If not, hard to know what was the case. 9/
