‘OH [expletive deleted],’ I said to myself when I heard about the Bangladesh mask study, the results of which were published at the beginning of last month.
It was a randomised control trial (RCT), described as the ‘gold standard’ of scientific evidence! It involved more than 340,000 people! It was conducted by academics from some of the world’s most prestigious universities: Stanford! Yale! University of California, Berkeley!
And it found that face masks reduced Covid infections by more than 9 per cent.
After refusing to wear a mask for 13 months of abject misery (avoiding public transport; shopping only for food, and only when necessary; keeping well away from cafes, and steeling myself for a confrontation with a masker on the odd occasion that I did venture out), imagine how gutted I was to see ‘conclusive proof’ that I was so wrong in my maskepticism.
Even if it didn’t demonstrate the slam-dunk prophylactic effect maskers often credit masks with, nearly 10 per cent is still a difference so big it’s impossible for maskeptics like me to hide behind.
I was horrified: what if I’d unwittingly passed on the disease during one of my (rare) trips out? If I’d caused someone to get really ill? Or die? Or if the person I’d needlessly infected had transmitted Covid to someone who died?
If I had murdered someone, I decided, the very least I could do was read the report that exposed my culpability.
For anyone who, like me, saw only the headlines about the study, https://www.poverty-action.org/publication/impact-community-masking-covid-19-cluster-randomized-trial-bangladesh, it involved more than 340,000 adults in 600 villages in Bangladesh. It was a ‘cluster’ trial in that it studied communities not individuals. Around half of the communities were provided with masks and given lots of ‘encouragement’ to wear them properly, and the rest were left to their own devices.
After the trial, participants were asked whether they had experienced any WHO-defined Covid symptoms, and any who had were invited to have their blood tested for antibodies. However, in the end, only just over a third of those who reported symptoms had their blood tested. (Some samples were too small to be tested, for others the ID bar codes were illegible.)
I’m not sure exactly what I thought I would find when I started to read the 94 pages (!) properly, but given the whole 9.3 per cent thing, at the very least something sensational like: ‘30-odd thousand non-maskers (ie nearly 10 per cent of 340,000) tested positive for Covid antibodies, but only a handful of the maskers did.’
But there wasn’t. Just lots of references and tables and pages and pages on selecting and randomising communities; strategies for promoting masking and assessing the success of that promotion; the (literally) gory details of how the blood was taken and tested, and even on ‘polling policy-makers’.
The abstract (basically the introduction) told me: ‘Adjusting for baseline covariates, the intervention reduced symptomatic seroprevalence [ie antibodies] by 9.3 per cent . . . control prevalence 0.76 per cent; treatment prevalence 0.68 per cent’.
I didn’t understand the 0.76 per cent and 0.68 per cent thing too much but both sounded tiny. But the same information cropped up again a couple of times, only while the 0.76 per cent was constant, the other figure was either 0.68 per cent or 0.69 per cent. I’m assuming it’s either a typo or something to do with rounding up or down. (In a ‘gold standard’ study, of course.)
But there was no what I would call ‘working out’, or not the sort of working out a lay person could understand, anyway; ‘intervention coefficient’? ‘0.005***’? Nope, me neither.
Before we go any further, a couple of disclaimers: I’ve seen blogs elsewhere quibbling with how the survey was conducted; I’m in no way qualified to comment on stuff like that, so I’m using only the information the researchers were confident enough about to include it in their report. Secondly, I’m not a statistician, mathematician or anything like that, so I may be completely wrong in all this – and if I am, I would love to get a response from someone who really does understand this stuff.
However, here’s how I decided they came to that 9.3 per cent (spoiler alert: it’s nowhere near as dramatic as it sounds).
There were 160,323 people in the ‘intervention’, ie masker, group and 146,783 in the ‘control’/non-masker one, but that’s not that important, seeing as it’s not the size of the ‘pool’ we’re interested in but the percentage/s of people infected.
Masking was not mandatory in any village, but the distribution and promotion of masks increased use in ‘treatment’ villages to 42 per cent, while in the ‘control’ villages it was 13 per cent.
0.69 per cent of the people in the masker group were found to have Covid antibodies in their blood, and 0.76 per cent of the people in the non-masker villages.
That’s a difference of 0.7, which, as a percentage of that 0.76 is 9.2 per cent. OK, that’s not 9.3 per cent, but considering the lack of consistency for the masker infection rate – is it 0.68 per cent or 0.69 per cent? – it’s close enough.
Was my theory borne out by reality? What happens if you apply those percentages to the groups in the study?
There are nearly 15,000 more in the masker/intervention group than in the control one, so let’s make both groups the same size, that of the non-mask/control one: 146,783 people.
0.76 per cent of them had Covid antibodies – that’s 1,116 people.
0.69 per cent of the masker group had antibodies, and in that 146,783, that means 1,013 of them, 103 fewer than the non-maskers.
Which would suggest that masks reduced infections by 103. Which is 9.229 per cent of 1116.
I reckon I must have been doing something right, as in the ‘Discussion’ a few pages later, I read: ‘The intervention led to a 9.3 per cent reduction in symptomatic SARS-CoV-2 seroprevalence (which corresponds to a 103 fewer symptomatic seropositives).’ In a footnote, the authors suggest that: ‘If we assume that non-consenting symptomatic individuals were seropositive at the same rate as consenting symptomatic individuals, the total estimated symptomatic seropositives prevented would be 258.’ Even if that assumption were correct, 258 is still a lot less scary than the tens of thousands I’d been expecting.
So, basically, that massive ‘9.3 per cent’ is not something like nine per cent of the total number of people studied, but 9.2/9.3 per cent of the less than one per cent of the people who got infected in the ‘control’ group.
What attracted less attention is that the researchers also evaluated the effectiveness of different masks. They promoted cloth masks among 54,122 people in the masker group and surgical masks among the 106,201 people in the other villages.
How many people in the cloth mask villages tested positive? Er, 0.74 per cent (ie just 0.02 per cent fewer than in the non-mask villages). Surgical masks fared better: 0.67 per cent of people in the villages where they were distributed had Covid antibodies – an ‘11.2 per cent relative reduction’ according to the report. In a group of 146,783, though, 0.67 per cent still represents just 983 people – only 133 fewer than the 0.76 per cent of the non-maskers.
Even when they compared people reporting WHO-recognised Covid symptoms (not everyone who reported symptoms agreed to have their blood tested), there was only a 1 percentage point difference between the ‘comparison’/non-mask villages and the ‘intervention’/masker ones: 8.6 per cent in the former, 7.6 per cent in the latter, although the study authors describe this as an ‘11.9 per cent relative reduction’.
But we can’t be sure how many of them, in either group, actually had Covid, rather than Covid-like symptoms: of the 160,323 in the ‘treatment’ group proper (i.e. not scaled down to the 146k of the ‘control’ group), 13,273 people reported symptoms; 5,006 of them had their blood tested and ‘only’ 1,106, or just over a fifth, of them (or 0.69% of that 160,323)actually had antibodies.
In the ‘control’/non-masker group, where 4,971 of the 13,893 people who said they experienced symptoms had their blood tested, only 1,116, or just over one in five, had antibodies.
The study does address one of my main concerns about masks, that they may actually facilitate transmission rather than reduce it; it looks like they don’t. And I can’t deny that, according to this evidence, masks do reduce transmission. But possibly by only a couple of per cent, while causing all manner of distress to lip readers, people with autism and others, and wreaking environmental havoc to boot.
In their ‘Discussion’, the authors point out that even in the ‘intervention’ villages, only 42 per cent wore masks. ‘The total impact with near-universal masking – perhaps achievable with alternative strategies or stricter enforcement – may be several times larger than our 10 per cent estimate,’ they say. Maybe they have a point: who knows what reduction might have been achieved if 80, 90, 100 per cent of people in the treatment villages had worn masks?
But let’s look at this a different way: even with just 13 per cent of people in the ‘control’ village wearing a mask, so 87 per cent not wearing one, a mere 1,116 out of 146,783 got infected during the study period. OK, so we have no idea how many of the 8,335 in that group who experienced symptoms but didn’t get their blood tested were infected too, nor how many in the ‘treatment’ group. Or how many people, in either group, caught Covid but remained asymptomatic – while still perhaps being capable of infecting other people. None of this we know.
But one thing I do know for sure (and I am keeping in reserve, just in case our government, in its wisdom, tries to foist masks on us again): the next time some masker tells me that masks reduce Covid transmission by 9 per cent I’ll be happy to put them straight . . .