Read on Twitter
Today I'll be live tweeting (Chu et al. June 1, 2020), the best systematic literature review on the effects of masks, distance, and eye protection that I am aware of: https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)31142-9/fulltext
A summary of their findings was included in a short writeup by MacIntyre and Wang https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)31183-1/fulltext
This is the best evidence I am aware of. It's worth revisiting in light of papers that are trying to use much worse evidence, e.g. observational data inferring masks must be unreasonable effective because weak protest outbreak, or this abortive PNAS paper https://twitter.com/RexDouglass/status/1278088077529935873
I see this as THE question for the U.S. covid-19 response because we have two political communities, one arguing masks don't work and so they can congregate without them, and one arguing they work supernaturally and so they can congregate with them. Neither are reasonable beliefs
The topline from this review is that masks, when worn correctly, in clinical settings, work. Not supernaturally well, but well enough that they should be mandatory. Especially when we're not doing any other thing we need to do to bring Rt down. I'll unpack that below.
The paper is a meta-review and so the vast differences across research designs and how they tried to aggregate across them creates difficulties in summarizing and evaluating the research design. I'm going to try then to focus in specifically on one outcome and one treatment.
That outcome is transmission, from one confirmed or probably COVID-19, SARS, or MERS case to another confirmed or probable negative person.
They found no randomized trials which is bad news. Most studies also focused on bundles of different PPE together, it was rarely just masks or no masks. 30 Studies specifically focused on face masks and respirators. Their universe were health-care workers, patients, or both.
Here's their main summary paragraph which I'll start to unpack
And here's their main summary table behind the conclusions in that paragraph
Everything is either measured in or has been converted to relative risk (RR). That is 1 is identical risk with or without masks, greater than 1 masks are more dangerous, less than 1 masks are less dangerous. Confidence intervals in this setting mean see if they overlap 1 or >1.
That risk is of a transmission 'event' which means they've got some groups that they're monitoring over some period of time and counting how many in the treatment or the control group get sick in that window. One thing to note is a lot of these groups are super small.
Also the event is rare. In 6 of these studies there were no transmission at all in either the control or the treatment group. In others it's just a couple. In 10,170 patients only 709 transmissions happened 7%. This is good for humanity, bad for science. Rare events are HARD.
The big claim to fame then is this line, 163/3,686 (4.4%) transmissions in the treated group and 546/6,484 (8.4%) in the untreated group. So when people say masks work, the mean they maybe half your risk. They don't mean they eliminate your risk.
These results are also conflating masks and respirators, and clinical and non-clinical settings, so let's break those out. Here are the non-clinical settings. Those are straightforward, none of them have respirators. Now it's only 37/244 (15%) vs 101/481 (20%).
So when people say masks work, they mean you, a civilian, with a regular surgical mask or less, who keep taking it off, touching your face, and licking the people around you, it's cutting your risk by a quarter, from super bad 1 in 5 to just bad 1.5 in 10.
They go through a bunch of statistical adjustments to try to aggregate across all of these small and slightly different studies into some larger answer. Their assessment of certainty I agree with is that this is a substantively important but statistically noisy result
In Sum: Wear masks because there's weak evidence that it saves lives and your moral and intellectual responsibility is to try. Don't exaggerate their effectiveness as an excuse to be irresponsible. And don't exaggerate the strength of the evidence to demean your political enemies
If you're an at risk group and can get hold of an N95 the math works out much more in your favor. N95 performs much better than no mask, single layer mask, or surgical mask. The only problem is most of our evidence is from clinical settings, by people who know how to wear them.
The evidence on n95 vs lesser masks (in clinical settings) is also better because we can ethically randomize those treatments. They include studies like this one which found reductions of closer to 75% in risk. https://pubmed.ncbi.nlm.nih.gov/23413265/
I'll conclude by briefly touching on the eye-wear result which is comparable to the above. No randomization, small studies, noisy effects, but seems to help. My family personally has been doing it in close quarters settings but it's another thing that 'helps' not 'solves' covid.
If this was helpful, this is the 5th in a series of COVID19 papers I've done public review on and I'll be collecting all of them at this thread here: https://twitter.com/RexDouglass/status/1278115752747253760
