The Danish face mask study

Posted on

The Swedish approach to the COVID-19 crisis stands out compared to the rest of the developed world in several ways. One of these is the extremely low percentage of the population wearing face masks.

This behavior has been consistently egged on by the Public Health Agency of Sweden (FHM), who throughout the pandemic have refused to recommend the use of face masks other than in medical care. This puts them at odds with the WHO and with the global scientific community at large. There is some opposition to FHM’s policy also in Sweden, but I doubt that even a report last week from the Nobel Prize awarding Royal Swedish Academy of Sciences (recommending the use of face masks) will have any effect on the FHM.

Due to the lack of transparent decision making at FHM, it remains unclear exactly why they take their unusual stance on face masks, although in my previous blog post here at Statisticians React to the News I offered some speculations. When the Swedish news channel TV4 requested the scientific basis for their stance, the FHM offered a list of 37 scientific articles, and when TV4 went on to ask British expert Melinda Mills to go over it she found that the set of articles pointed overwhelmingly in the direction of face masks having a substantial effect. Astonishingly, in a recent press conference, Karin Tegmark Wisell at the FHM hinted at a fairness aspect: a face mask recommendation would induce costs that would be felt more strongly by low-income households.

Be all that as it may, the results of a Danish study (published in the Annals of Internal Medicine) on the effects of wearing a face mask were announced last week. Given the news coverage the study received, it is likely to feed into and reinforce the negative attitude towards face masks that dominates in Sweden. Among the headlines in leading Swedish news media were “Danish study: wearing face mask has limited utility” (Dagens Nyheter), “Face mask ineffective according to new study” (Expressen), and “New study: no clear effect for the wearer of face mask” (the public service television channel SVT). Further afield, readers of the New York Times were told that “the study’s conclusion flies in the face of other research suggesting that masks do protect the wearer” and, in a statement attributed to the study’s lead author, Dr Henning Bundgaard at the University of Copenhagen, that the study “gives an indication of how much you gain from wearing a mask: not a lot”.

So does the study give evidence in support of the claim that – in the words of the Expressen article – “face masks have no particular effect on the spread of infection”? The short answer is no, and there are mainly two reasons for this.

First, the study only addresses the protection that the face mask offers to the wearer, whereas it says nothing at all about the main purpose of face masks, which is to protect other people in case the wearer happens to carry the infection. This point is made in some of the media reporting, but far from all, and even when the point is made it easily gets lost amidst all the anti-mask rhetoric.

Second, let’s look at the numbers. A randomized controlled study was carried out, in which half of the participants (none of whom had been infected when the study began) were given instructions to wear a mask, and the other half were assigned to the control group and given no such instructions. Among participants assigned masks, 2392 completed the study, and of these, 42 had been infected by the end of the study. In the control group, 2470 completed the study, 53 of whom were infected. This leads in the obvious way (and as reported in the article) to the point estimate that wearing mask in public reduces the risk of infection by 18%. As it happens, this is not a statistically significant deviation from 0, but it is not a small amount. If an entire population switches to wearing masks, it corresponds to reducing the infection’s reproduction number R from (say) 1.2 to 1.0 – which is enough to stabilize the prevalence of the infection in the population as opposed to doubling it within a time frame of less than a month. With such a point estimate, it is plain misinformation to say that the study shows face masks to be of “limited utility” or have “no particular effect”.

A confidence interval is reported for the relative risk when wearing a face mask, ranging from 0.53 to 1.23. It is true that this interval covers the value 1 (no effect), which is in fact what it means for the result to fail to reach statistical significance. But it is also true that the confidence ranges almost all the way down to a risk reduction by half. This is not what statistical evidence for “limited utility” of face masks looks like. What the lack of statistical significance says here is not that masks have “limited utility,” but that the study turned out to be too small to say much on its own. (I don’t want to blame the research team for poor design of experiment, however, as they were not in a position to know how the infection would spread in Denmark during their trial. With more spread, their study would have gained more statistical power.) When all is said and done, and an appropriate meta-analysis carried out and survey papers written, I believe that the most likely verdict on the the Danish study will be that it provides one more little piece of evidence in favor of the overall conclusion that wearing face masks in public has substantial effect on the pandemic.

So who is to blame for this media debacle – have Bundgaard and his coauthors gone wrong, or are they innocent victims of overenthusiastic media coverage? I do not know, but the Bundgaard quote in the New York Times does not look good for him. Another bad sign for him and his team is that three weeks prior to last week’s announcement and publication of the paper, there was a news article in Svenska Dagbladet about the study, and how it had been rejected by three different top-tier journals (The LancetNew England Journal of Medicine, and the Journal of the American Medical Association), along with speculations about how the rejections might be unfair and caused by the journals being biased against certain study outcomes. This is not the kind of thing to bring to the press if you handle media relations responsibly.


All posts are written by authors in their personal capacity and in no way represent the view of the organisations, universities, governments, or agencies where they are employed or with which they are associated, or the views of the International Statistical Institute (ISI).

Leave a Reply

Your email address will not be published. Required fields are marked *