19 November 2020 — Sebastion Rushworth MD
A few months back I wrote an article about the state of the evidence on face masks. At that point, there were no good studies looking at the effectiveness of face masks in preventing the spread of covid-19 specifically, but there was a systematic review that looked at all randomized trials that had been done on face masks for the prevention of respiratory infections more generally. That review found that surgical face masks reduced the probability of getting a respiratory infection by around 4% in absolute terms (17% in relative terms).
My conclusion was that, considering how infectious covid is, face masks were unlikely to have an effect on the spread of the virus on a population level, although I thought it made sense to use them in hospitals and nursing homes, where you want to do everything possible to minimize the risk of spread to people who are at high risk of a severe outcome.
Now we finally have a randomized controlled trial that has looked at the ability of face masks to protect wearers against covid-19. It was published in the Annals of Internal Medicine. The study was carried out in Denmark, and was funded by a charitable foundation that is connected with a company that owns supermarkets (I’m not sure whether that means they wanted the study to be a success or a failure, or just wanted to know the truth).
In order to be included in the study, participants had to be over the age of 18 and and they had to spend at least three hours per day outside the home. People were not allowed to take part in the study if they had current or prior symptoms that could indicate covid-19 infection, or a previously confirmed diagnos of covid-19. All potential participants had an antibody test performed at the beginning of the study, and if it was positive, then they were excluded from taking part. Participants were recruited through adverts in media and through direct contact with companies and other organizations.
In total 6,024 people were recruited in to the study, and of these 4,862 (81%) followed through to the end. That is a nice big number, which should be able to detect a meaningful difference, if there is one. The average age of the participants was 47 years. Half the participants were randomized to wear a face masks at all times when outside the home, and half were randomized not to. For obvious reasons, this study was unblinded, since it’s hard to create a situation where people are unaware of whether they’re wearing masks or not.
Participants in the intervention group were given 50 disposable surgical masks. This actually increases the probability of the study showing a meaningful effect compared with the reality in most countries where masks are currently being used in public. Why?
Because in most real world situations, people are wearing (and repeatedly re-wearing) non-disposable cloth masks, which are likely much less effective than disposable surgical masks that are only used once. In the systematic review I wrote about in my previous article, the little data there was on cloth masks suggested that they were completely ineffective.
Participants were followed for one month, and at the end of the month an antibody test and a PCR test for covid were carried out. If participants had symptoms suggestive of covid at any point during the month, a PCR test was also performed at that time point. All participants received written and video instructions on how to use the face masks properly. If outside the home for more than eight hours at a time, they were instructed to change to a new mask, so that a single mask was never used for longer than eight hours.
Both an intention-to-treat and a per-protocol analysis was done of the results. What that means is that they looked at what the results were, both if all participants involved in the study were included (intention-to-treat), and if only participants who reported wearing the masks as instructed a high proportion of the time were included in the analysis (per-protocol).
In general, it is considered good form to do an intention-to-treat analysis, and bad form to do a per-protocol analysis. The reason for this is that a per-protocol analysis will tend to make the results seem better than they are in the real world (in the real world, not everyone does as they’re told – annoying, right?!). In this case, however, I think it’s reasonable to also do a per-protocol analysis, because we want to know what effect, if any, masks have when used as instructed.
So, what were the results?
We’ll start with the intention-to-treat analysis. In the face mask group, 1,8% developed covid-19 over the course of the study. In the control group, 2,1% developed covid-19. That is a 0,3% difference in favor of face masks, but it is not even close to being statistically significant.
Ok, let’s look instead at the per-protocol analysis, which in practice means that the 7% of participants who often didn’t follow the mask wearing instructions properly are excluded from the analysis. In the face mask group, 1,8% developed covid-19, and in the control group, 2,1% developed covid-19. So, interestingly, the result was the same regardless of whether you look only at those who wore the masks as intended, or look at everyone, including those who didn’t follow the instructions. This in itself suggests that mask wearing doesn’t make a big difference, since the results don’t change when you only look at people who have been good at wearing their masks as intended.
As an interesting aside, the researchers didn’t just look at covid, they also looked at 11 other respiratory viruses. In the face mask group, 0,5% tested positive for one or more other respiratory viruses. In the control group, 0,6% tested positive. That is a 0,1% difference, and again, it was nowhere close to being statistically significant.
What can we conclude from this?
Wearing face masks when out in public does not meaningfully decrease the probability that the mask wearer with get covid-19. It’s possible that there is a small reduction in risk, but if there is, it is so small that it was undetectable in a study where almost 5,000 people were followed for a month.
It is worth nothing here that the effect seen in studies is usually better than the effect seen in reality. The reason for this is that study participants usually try harder than people who aren’t part of a study, and they get better instruction. In this case, they also had better masks than most people are using at present in reality, and changed to new masks on a regular basis. So, if no meaningful difference was seen in this study, then I think it’s safe to say that no meaningful difference exists in reality.
One thing that is good about this study is that it is the first randomized controlled trial that comes close to mimicking the present reality in many countries, where people are wearing face masks in public, but not at home.
One interesting result of the study was that 52 people in the face mask group and 39 people in the control group reported another individual in the home having covid-19 during the course of the study. Yet of those, only 3 actually developed covid. People sharing a home with someone with covid were really no more likely to get covid than people who weren’t. This suggests that most covid infections happen outside the home, and is in itself something that would be an interesting avenue for further research. It also suggests that most people with covid are not themselves very infectious, giving support to the hypothesis that most infections happen through a small group of highly infectious ”super spreaders”.
The main thing lacking with this study is that it only looked at risk to the person wearing the face mask. It says nothing about the risk that the person wearing the mask will infect another person. That is an equally important parameter, and at present there are no high quality studies looking in to it, so before we can truly say that masks fill no function, we need another large randomized controlled trial that looks at the ability of face masks to prevent the mask wearer infecting other people.