Trisha Greenhalgh Profile picture
Feb 26 95 tweets 26 min read Twitter logo Read on Twitter
New mask thread. Addresses
- Why so much controversy?
- Airborne transmission
- RCTs of masks - strengths/weaknesses, community, healthcare settings
- Meta-analyses - gold standard or lazy lumping?
- Non-RCT evidence - why is it needed, what does it show?
2.Before I start, here’s the OLD mask thread from July 21, which ran to >100 tweets. Some of that is now rather dated but a lot still holds. I’ll paste the best old tweets into this new thread.
First, ‘masks’ and ‘masking’ are very broad terms. They cover different technologies (cloth, medical, respirator), wearers (public, patient, healthcare worker), settings (low v high risk) and requirements (mandated v free choice). ImageImageImage
Because of this multiplicity, statements like “masks [don't] work” may be referring to very different interventions in different populations for different activities at different times, not to mention different diseases. The literature, in short, is a mess.
Second, masking is a cultural practice as well as a public health intervention. It has symbolic meaning. Pirates and burglars wear masks. To some, masking means you’re a ‘pansy’, ‘sheep’ etc. Masking offends libertarians.…
There’s strong social pressure to mask to contribute to the greater good (“my mask protects you”). People may feel pushed into doing something which a) feels awkward, b) symbolises an allegiance they don’t identify with and c) they feel is ineffective. Image
Third, masking (whatever its benefits) has downsides. It’s (depending on the design) uncomfortable. It can interfere with communication (especially if hard of hearing). If you work long shifts, it gets hot & sweaty behind the mask.…
In short, masking is something many of us feel strongly about. It’s not even possible to separate “the science” of masks from “the politics”, since science does not sit in some protected bubble away from the rest of society. Read on!
Understanding why & to what extent masks work requires knowledge of how the SARS-CoV-2 virus is transmitted. With colleagues, I published a summary of evidence that it spreads *predominantly* through the air. Thread + paper here
I responded to critics of that paper here (plus see more in thread below).
For those who want depth and detail, here’s Chia Wang and team’s stunning review of the evidence for airborne transmission in Science (one of the very top scientific journals).…
In sum, it's *settled science* that SARS-CoV-2 is transmitted via the air. Settled, that is, in all branches of science *except* key groups of doctors, including some in powerful positions at the World Health Organisation. I return to that paradox below.
Medical/surgical (~same thing) masks are designed to stop spatter (e.g. your dentist dribbling into your mouth). Most aren’t made to a specific quality standard. Respirators are designed to filter out airborne particles. They ARE made to a standard.
If SARS-CoV-2 were spread mainly by droplets, we’d need to focus on handwashing and med/surg masks which stop wet drops. If mainly via air, we need to focus on air quality and use the kind of face coverings that stop airborne particles (ie. respirators)
All masks stop some airborne particles (even if the holes are bigger than the particles, fibres are criss-crossed and electrostatic forces kick in). But N95 (FFP2) respirators stop >95% of particles of SARS-CoV-2 size, and N99s (FFP3s) stop >99%.
A year ago, we wrote this article recommending that everyone upgrades their mask from either medical/surgical to respirator.…
Whilst respirators protect best, ANY mask is better than none. Efficiency at reducing transmission of an artificial respiratory aerosol:
N95 respirator 99%
Medical mask 59%
Bandana (single) 47% (double) 60%
Face shield 2%…
The above, a laboratory study of artificial aerosols, should not be taken as direct evidence that these masks would be this effective/ineffective in preventing real-world SARS-CoV-2 transmission. But they *add to the mix of evidence*. More on that later
Face shields are only 2% effective cos they’re designed to stop spatter NOT filter air. Medical masks don’t fit snugly, so instead of breathing filtered air, you breathe what leaks in round the sides. This is why a doubled-over bandana works pretty well!
While I believe that cloth & medical masks have an important place (better than nothing), some scholars (Lisa Brosseau is one) feel that the benefits of respirators are so much greater, we should junk all lesser kinds of masking.…
Remember, respirators (and masks) only work if they fit you snugly. You don’t need a professional fitting any more than you need to have your underwear professionally fitted, but do check for gaps.
RCTs of masks have continued to generate more heat than light. I’m going to reproduce what I said 2 years ago, which still holds (even more so now, since what has happened since could have been predicted from my earlier thread).
23. RCTs
In the name of evidence-based medicine (EBM), the West got off on the wrong foot. We became obsessed with the holy grail of a definitive randomised controlled trial (RCT) that would quantify both the benefits and the harms of masks, just as you would for a drug.
24. RCTs
If you were raised in the EBM tradition, where “rigorous RCTs” are mother’s milk, it’s not easy to get your head round why this was a bad way to approach the problem. Looks like Prof G has lost it, dropped her standards, joined the dark side etc. Bear with me.
25. RCTs
A RCT is a controlled experiment. Since people are randomly allocated to one or other group (‘arm’), any confounding variables are distributed evenly between the arms so they all cancel out (so long as the study is large enough and allocation is truly random).
26. RCTs
Random allocation means that differences between the arms of a RCT are highly likely to be due to the intervention (in this case, masks) and not to confounders. But it doesn’t follow that a RCT is better, for any scientific question, than a non-RCT design. Why not?
27. RCTs
Many reasons. Drugs are (arguably) a simple intervention, but masks are a highly complex one. There are two key questions: do they protect the wearer from other people’s germs – and do they protect other people from the wearer’s germs (‘source control’)?
28. RCTs
The RCT design can’t (usually) cope with this. It’s easy to design a study where the main outcome is infection in wearers, but how would a RCT of source control work? Answer: do an enormous study to randomise entire towns to either masking or not masking!
29. RCTs
Impressively, @jabaluk and team did just this. They randomised 600 villages in rural Bangladesh (over 300,000 people) to getting free masks (‘interventions’) or not (‘controls’). For 2 months, they observed mask-wearing and counted covid-19 cases in all villages.
30. RCTs
The Abaluk RCT was published in Science (really top-notch academic journal). It showed that in the intervention villages, a) mask wearing was three times as high as in the control villages, and b) incidence of covid-19 was lower.…
31. RCTs
Despite the large size of the Bangladesh RCT, differences in covid-19 incidence were relatively modest and only just reached statistical significance when measured by blood test. Differences were greater where surgical rather than cloth masks were used.
32. RCTs
Does this mean “masks don’t work” in the community? NO! Looks like even in free-living conditions (people can choose not to wear), they have a modest but important effect—reducing transmission by ~10%. Let me return to some of the arguments I put out two years ago.
For policy decisions, we’re not just interested in whether masking reduces community incidence of covid-19 in a short intervention period (say, a month or 2). We’re interested in how masking impacts on the *exponential spread* of an accelerating pandemic.
Take the number 1, double it, then keep going. 1 becomes 2, then 4, etc. After 10 doubles, you get 512. After 10 more doubles, 262144. Now instead of doubling, multiply by 1.9 (a tiny reduction in growth rate). After 20 cycles, the total is only 104127.
=> if masks reduce transmission by a TINY bit (too tiny to be statistically significant in a short RCT), population benefits are still HUGE. If instead of doubling every 9 days, covid rates increased by 1.9, after 180 days total cases would be down by 60%.
These two issues—the near-impossibility of using RCTs to test hypotheses about source control and over-reliance on “statistically significant effects” within a short intervention period—is why a RCT of masks is *highly likely to mislead us*.
Let’s look at the other large community-based RCT of masks in covid-19. The *big Danish mask trial* became one of the most talked-about RCTs in the pandemic. Its main finding was negative (no significant impact of masks compared to no masks).…
38. RCTs
But the DANMASK RCT was *fatally flawed* as
and I argued here. I’ll summarise in the next tweet.…
39. RCTs
DANMASK RCT flaws: no CONSORT statement, no ethical approval, inappropriate setting (there was almost no Covid circulating at the time!), under-powered sample, wrong primary outcome, wrong intervention period, inaccurate test, misinterpretation of own findings.
40. RCTs
There *never will be* RCT evidence to definitively answer the question “do masks work in community settings?” because RCTs can’t generate dynamic evidence to understand complex social interventions in diverse and changing contexts. Answer will always be “to some extent”
The question of how to protect HCWs is politically contentious, since it involves the sensitive topic of occupationally-acquired disease and the risk (in rare but well-described cases) of a HCW dying from covid-19 caught at work.
42. RCTs in HCWs
Broadly speaking, HCW unions want high-grade protection for their members; employers want to spend as little as possible on personal protective equipment (PPE). So RCTs comparing medical masks (cheap) with respirators (more expensive) have special significance.
43. RCTs in HCWs
There’s been only one published peer-reviewed RCT of masking in HCWs during covid-19. Loeb et al did a 5-country study which showed no statistically significant difference between medical masks and respirators:
44. RCTs in HCWs
The Loeb study has been heavily criticised. I have two major concerns. First, according to the original study protocol, HCWs were asked to wear the mask/respirator “when providing routine care to patients with COVID-19 or suspected COVID-19”.
45. RCTs in HCWs
Because SARS-CoV-2 is airborne (see tweets 9-20), removing masks when not in the close vicinity of known or suspected covid patients is illogical. See why we need to understand the mode of transmission to interpret the studies?
46. RCTs in HCWs
Even though the words quoted in tweet 42 are taken directly from their participant information sheets, Loeb and colleagues say that participants were “expected” to wear their mask or respirator all the time when at work.
47. RCTs in HCWs
There is confusion about which participants wore what and when in the Loeb RCT. This may not be entirely the authors’ fault, as the study was occurring in a changing policy context (universal masking of HCWs came and went in some sites as the trial progressed).
48. RCTs in HCWs
My other concern with the Loeb RCT is that results are heavily skewed by data from Egypt, the largest national subsample. There was a big wave of Omicron in Egypt at the time, so HCWs in both arms could well have caught the virus outside work.
49. RCTs in HCWs
Even if Egyptian HCWs wore their protection all the time at work, if they then went home and played with their infected kids, they’d be as likely to catch covid as anyone else. Egypt was the only country where *more* infections occurred in the respirator arm!
50. RCTs in HCWs
Loeb et al are currently responding to these and other methodological criticisms. ONE explanation for their findings is that respirators are no more effective than medical masks. Another is that a benefit exists but this trial failed to demonstrate it.
You may have seen the headlines in the last couple of weeks, that a ‘Cochrane review’ has demonstrated the lack of RCT evidence for efficacy of masks. Here’s the review so you know what I’m criticising. I don’t endorse it.…
Despite widespread claims that this new meta-analysis (actually, an update of an old meta-analysis) has definitively demonstrated that “masks don’t work” in reducing covid-19 transmission, it doesn’t show anything of the sort. Let me explain.
As @KelseyTuoc points out, of 78 studies in in this Cochrane review, only 6 (& only 2 on masks) were conducted during the covid-19 pandemic. All the rest were old RCTs in flu & other (less contagious) outbreaks in non-pandemic times.…
As well as combining covid studies with non-covid studies, Jefferson et al made another elementary lumping error. They combined RCTs in which HCWs had worn protection *some of the time* with RCTs in which they wore them *all the time*.
As we argue in this article, a meta-analysis that combines studies of different designs asking different questions will tend to produce un-nuanced and often negative findings. Mixing the paints in the paintbox gives you dull brown.…
Here, @JenniferNuzzo (a fan of Cochrane revs) points out additional subtleties that cast doubt on the masks-don’t-work conclusion (which is not in any case what the review's authors concluded—at least, not in the review itself).
Here, Jason Abaluk @jabaluk of the impressive Bangladeshi mask RCT argues that the updated Jefferson review made the same mistakes they had made in previous versions.
Here, Satoshi Akima @ToshiAkima explains why Cochrane reviews should not be blithely swallowed as an assumed ‘gold standard’.
Satoshi argues that whilst RCTs may provide *predictive* evidence (intervention A is more/less likely to produce outcome X than intervention B), they don’t provide *mechanistic* evidence (how does intervention A work?). I’ll come back to this.
Here, @danwalker9999, citing @BlakeMMurdoch, accuses the authors of the Jefferson Cochrane review of “amazing sloppiness” (failing to access and incorporate raw data from the Abaluk trial that was in the public domain).
Have I persuaded you that Cochrane reviews aren’t a magic higher form of science that automatically knocks all other evidence out of the park? They look at a narrow slice of evidence and can be done well or badly. Where’s the quality control? Good question.
Some (though not all) Cochrane authors believe that the RCT is such a superior form of evidence that we don’t even have to look at any other kinds of evidence (unless our search for RCTs turns up none). Let me explain why this is a dangerously naïve assumption
Bradford Hill (famous for doing the 1st ever RCT) emphasised that we can’t rely solely on probabilistic evidence from RCTs. We also need mechanistic evidence (see tweet 58). We explain in this paper on EBM+:…
When Jefferson et al ignore ALL evidence except RCTs and also use synthesis methods that *make no sense* when you take the *mechanism of transmission* into account (i.e. lumping continuous masking RCTs with intermittent masking RCTs), I have a problem.
Mechanistic evidence (which Jefferson et al conspicuously ignore) means evidence – from basic science laboratory experiments and real-world case studies for example – that helps us understand how the virus spreads and how covering the face *could* work.
My original mask thread (see tweet 2) presented lots more mechanistic evidence that whilst *some* transmission of SARS-CoV-2 might conceivably happen by droplets, *most* transmission happens by airborne viruses which enter the lungs when we breathe in.
Airborne transmission happens *mainly* when you’re very close to someone who is exhaling the virus. But airborne viruses *float in the air* so they spread throughout an indoor space. Rules like “wear your mask/respirator within 2 metres” don’t make sense
Airborne transmission of SARS-CoV-2 explains why protection that *blocks airborne particles* (i.e. respirators) is in theory better than protection that blocks droplets, and why we’d expect a well-designed RCT of respirators vs masks to favor the former.
Just because mechanistic evidence suggests respirators will outperform medical masks doesn’t mean we should ignore RCTs that show no advantage. But it does mean we should *design those RCTs properly* so that an advantage, if it exists, would show up.
Because RCTs and meta-analyses were poorly designed (overlooking mechanism of transmission), many produced negative findings. Some researchers (+ press and public) have conflated *absence of evidence of effect* with *evidence of lack of effect*.
I’ve argued from philosophical perspective that flawed mental models assuming a ‘droplet’ mode of spread led scientists to greatly under-estimate the importance of masking. (NB I’m NOT saying it’s a bad idea to wash hands) (…
For an unforgivably long time, WHO and others denied that SARS-CoV-2 was airborne (see tweets 9-20). We wrote a paper on why this error happened (TL;DR: powerful doctors had entrenched mental models & reputations to lose).
These flawed mental models led to flawed policy narratives, diverting us into an *obsessive concern with handwashing and squirty sanitisers* and thinking that medical masks (designed to block droplets) will do for HCWs.…
Let’s return to the question of community masking. Notwithstanding @JABALOK’s impressive RCT, it remains the case that this kind of study is difficult & expensive to do & hard to interpret. What other evidence is there? I get asked this question a lot.
Countries that introduced mandated masking within 30 days of the first case (mostly Asian) had *dramatically* fewer Covid-19 cases than those that delayed beyond 100 days (mostly Western).…Image
Was this association or causation? Early on, we didn’t know (though, importantly, nobody in these Asian countries seemed to come to harm from wearing a mask). Since that early study, there have been numerous studies of the impact of mask mandates.
Various systematic reviews of mask mandates have been published. This from 2021 showed a statistically significant effect of mask mandates (same effect size as handwashing, with physical distancing having an even greater effect)…
Yanni Li et al did a meta-analysis of case-control studies. Acknowledging the limitations of such designs, they found masks reduced SARS-CoV-2 infections by (on average) ~70% in healthcare workers and ~60% in lay people.… 35/
What’s a case-control study? Find people who caught Covid; match them with people who were similar but didn’t catch Covid; ask whether each was wearing a mask. Relatively weak design (bc unmeasured confounders), but not as weak as a badly-designed RCT
This review of observational studies of various public health measures in the pandemic found that 6 of 7 studies of mask mandates showed a statistically significant effect. Studies rated higher quality found greater effects.…
The Talic review (tweet 77) highlighted this study by Liu et al, who looked at (inter alia) introduction of mask mandates in different US states. After correcting for confounders, mask mandates reduced transmission by 29%.…
A more recent study of observational data across US states by Ahlers et al showed that (after complex statistical measures to correct for confounders) mask mandates were associated with significantly reduced transmission rate…
Two more studies of mask mandates across US states, again showing the localities with mask mandates had lower transmission whilst the mandates were in place, and lower hospitalisations from covid-19……
And here’s Miriam Van Dyke’s study of more granular data on mask mandates across the state of Kansas, USA, in 2020. Again, localities that implemented mask mandates had lower transmission after confounders were controlled for.…
In another recent study, Daniel Cooper et al try to unpack whether the *association* between mask mandates and lower covid-19 cases across US states from 2020 is *causal*. They conclude that it is.…
86. AOB
We wrote a summary review of masks and masking here (now somewhat out of date but includes a lot of useful papers). [need break - will resume thread after tea]…
87. AOB
The rest of this thread links to additional key evidence on the mask/masking debate. This from @jljcolorado on why people wrongly assumed for so long that respiratory illnesses are droplet-spread.
88. AOB
A landmark paper from Adriaan Bax’s team showing that breathing, speaking and singing are key activities that transmit SARS-CoV-2…
89. AOB
The Bagheri et al modelling study: what if one person masks? What if both people mask?
90. AOB
TIL = total inward leakage (vertical axis). Jargon for "air getting in round the sides of your badly-fitting mask". High TIL is bad. Aim for PURPLE: wear a FFP2 & use a bit of tape to close the gaps. GREEN is an untaped surgical mask - useless!… Image
91. AOB
Here’s a nice theory-based review of the psychology of mask-wearing (or mask-refusing). Talks about 3 main constructs: autonomy, psychological relatedness, and competence.…
92. AOB
Thanks several people for pointing out I forgot to add the before-and-after Cambridge study showing HCW covid infections went right down after N95s introduced…
93. AOB
We don't have to wear masks everywhere, for ever. We should adjust our masking behaviour to reflect the prevailing risk of transmission.…
94. PS
Another thread, from Tomas Pueyo, pointing out serious flaws with the Jefferson Cochrane review which is being (wrongly) interpreted as proving that masks don't work.
95. PS
Another relevant thread from @dontwantadothis pointing out the differences between RCTs of behavioural interventions such as masking and the more conventional RCTs of drug v placebo.

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Trisha Greenhalgh

Trisha Greenhalgh Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @trishgreenhalgh

Mar 1
Our paper on orthostatic tachycardia (fast heart rate when you stand up) after covid-19 is now #OpenAccess in @bmj_latest. Summary 🧵 1/
#longcovid @AnaBelenEspino2 @HarshaMaster3 @RogersNat70…
We found that fast heart rate on standing was common and sometimes very disabling in people whose recovery from covid-19 was prolonged. 2/
'Orthostatic tachycardia' means the heart rate goes up (sometimes very dramatically) on standing up, but the blood pressure doesn't drop (the latter is another condition called orthostatic hypotension). 3/
Read 22 tweets
Feb 26
Here’s my new mask thread from late Feb 2023:

Other key threads (which I’ve written in the past) are linked in this short pinned thread:
Read 4 tweets
Jan 21
A thread on primary health care 101 (#PHC101). Do you even know what primary [health] care is? Do you care about it? Most people, if they were honest, would probably answer ‘not really’ to both. Read on if interested. Mute this thread if not.
Primary health care (PHC) has 4 defining features:
-First-contact (i.e. you don’t need a referral)
-Accessible (you can get an appointment, whoever you are*)
-Relationship-based (“my GP”)
-Undifferentiated (any illness, any problem—‘generalist’ rather than ‘specialist’)
* Yeah I know—PHC in UK is not currently accessible to everyone. This is because there is a chronic lack of resources and staff. I’ll return to this later.

For now, let’s focus on PHC at its best, and ask “what *becomes possible* when a country has a strong PHC system?”
Read 35 tweets
Jan 8
An adequately funded system of universally-accessible primary health care is associated with
- better overall health
- lower overall healthcare costs
- fewer inequalities between rich and poor
- lower mortality

Which of these benefits do you have a problem with, @wesstreeting?
Good primary health care is characterised by
- accessibility
- therapeutic relationships
- continuity of care
- multidisciplinary teamwork
- coordination

A vote-winning policy for LABOUR would be to STRENGTHEN PRIMARY CARE.

Happy to help you write this one, @wesstreeting.
Read 6 tweets
Jan 4
MASKS are back (=> they’re trending on social media and I’m getting hate mail). A short thread. See my pinned tweet for a longer thread with lots of peer-reviewed articles.🧵
Masking is a deeply symbolic practice. In some social groups, a mask is a muzzle, a ‘face diaper’ or a sign that you’ve given in to state control. Criminals, pirates, political protesters and others with something to hide wear masks. Masked people are suspicious.
In other social groups, masking during a pandemic means that you care about the common good. My mask protects ME from inhaling airborne virus—and it also protects YOU because I may be infected but not know it. Even if I’m healthy, others may be vulnerable.
Read 16 tweets
Oct 2, 2022
We’re on the nth wave of covid-19 and all the old anti-mask arguments are doing the rounds. “No robust evidence”. YES THERE IS – see my pinned tweet – and btw the I-only-accept-RCTs trolls don’t have a monopoly on the word ‘robust’. 1/
Do we have to wear masks everywhere, forever, for all time? NO!!! We need to get used to ASSESSING THE RISK and adapting our behaviour accordingly. If the incidence of covid is low, risk is low (but not zero). But if it’s high, risks are high. 2/
I just called up a pal who sounded TERRIBLE. Rotten symptoms. They said “I think I might have covid again”. Maybe this new bout could have been prevented if they’d assessed the risk of various activities in the light of high local incidence last week. 3/
Read 10 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!


0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy


3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!