Yeah, but...it's also subject to sample bias. Especially in a place like east Africa, having enough resources (not just money -- time, transport, etc.) to get a crazy expensive new vaccine could easily have co-variates that affect all-cause mortality.
It's great to report it, but you definitely want to see the cause-specific numbers, too.
They used two sets of controls: 1, an entire set of communities where the vaccination wasn't offered at all, and 2, children who were not eligible due to age. They weren't just comparing children who got the vaccine with children who didn't, they were comparing children who were eligible for the vaccine and those who weren't, and that couldn't have been so easily confounded:
"79 areas where the RTS,S vaccine was administered and 79 comparator areas where it was not available"
"Researchers then compared the death rates of babies whose age made them eligible to receive three doses of the vaccine with those of young children who were not age-eligible for three doses, in both RTS,S areas and unvaccinated areas."
This is obviously not foolproof, but it seems like a substantial bulwark against obvious confounders.
If you can show that 1) the mortality rate in the vaccinated communities was lower in the age-eligible children than not-age-eligible[0] and 2) this effect was absent in unvaccinated communities that seems like it goes quite far towards proving that it's real and not an artifact. It probably even underestimates how protective it is at an individual level.
(communities which had the vaccine offered also had fewer cases of severe malaria recorded in their hospitals, so there's a straightforward reason to think it does something)
[1] okay probably age-ineligible children have a different mortality rate regardless. You'd be taking the difference between those two groups of children and comparing it to the same difference in communities without the vaccination offered. This difference-of-difference would tell you whether the vaccine is doing anything. It wouldn't matter if only the richer children received the vaccine if the communities have basically the same distribution of wealth since you're comparing differences between age cohorts, not between vaccinated and unvaccinated.
> This is obviously not foolproof, but it seems like a substantial bulwark against obvious confounders.
I mean...they had controls? That's good. But otherwise, there's no way you can tell from what is written. Even if the controls are perfect (which I don't grant without more detail), the implementation of the controls can still allow for confounding.
> they were comparing children who were eligible for the vaccine and those who weren't, and that couldn't have been easily confounded
You can trivially get confounding based on that. Just pick the kids who are healthier, and call them "eligible". Or pick the ones who are older (which they did, as you note), and voila...infinite time bias! Kids who make it to age N are more likely to survive to age N+1 than kids who make it to age N-1 (I mostly disregarded the between-group differences based on age, for this exact reason. It's too easily confounded.)
More commonly, bias of this form sneaks into a study. Particularly in a place like sub-saharan Africa, the set of people who are even willing to engage with you, mysterious doctor-magician, are of a fundamentally different nature than the ones you never see. They probably do all sorts of things that make them a little bit healthier, on average.
It's hard to correct for that, and it's a real problem when the primary metric is "community survey of all-cause mortality". For example: are the community surveyors also magical stranger doctor-magicians, or are they just regular people? It matters.
> > This is obviously not foolproof, but it seems like a substantial bulwark against obvious confounders.
> I mean...they had controls? That's good. But otherwise, there's no way you can tell from what is written. Even if the controls are perfect (which I don't grant without more detail), the implementation of the controls can still allow for confounding.
> > they were comparing children who were eligible for the vaccine and those who weren't, and that couldn't have been easily confounded
> You can trivially get confounding based on that. Just pick the kids who are healthier, and call them "eligible". Or pick the ones who are older (which they did, as you note), and voila...infinite time bias! Kids who make it to age N are more likely to survive to age N+1 than kids who make it to age N-1 (I mostly disregarded the between-group differences based on age, for this exact reason. It's too easily confounded.)
They did use two control groups, that's the whole point so you compare the difference between the age groups for the area where they vaccinated and the area where they didn't. That reduces confounding factors, e.g. based on area.
> More commonly, bias of this form sneaks into a study. Particularly in a place like sub-saharan Africa, the set of people who are even willing to engage with you, mysterious doctor-magician, are of a fundamentally different nature than the ones you never see. They probably do all sorts of things that make them a little bit healthier, on average.
And you base that assertion on what? Your prejudice ("mysterious doctor-magician", do you think about what you're implying here?!). In reality based on the studies I have read people in underdeveloped nations are significantly more likely to engage with health professionals and less likely to believe in anti vax or other anti science prkpaganda across all classes than in developed nations. Also one should note that in developed nations the effect is the other way around, anti vax sentiments are strongest (and therefore less likely to get vaccinated) in richer classes, who tend to generally have healthier lifestyles.
> It's hard to correct for that, and it's a real problem when the primary metric is "community survey of all-cause mortality". For example: are the community surveyors also magical stranger doctor-magicians, or are they just regular people? It matters.
It's much easier to look at all cause mortality than cause specific mortality, because you include more confounding factors. It's the much better study.
Lots and lots of prior research, as well as direct experience talking to people who run these kinds of experiments. It's practically the #1 most common theme you will hear from anyone who has run a public health campaign in a third-world country.
Just for example [1]: "From the onset, Northern Nigeria presented an extreme challenge. The transmission of polio in Northern Nigeria was due to complex health, economic and social issues such as poor demand for and access to health services, low immunization coverage, few available skilled health workers, extreme poverty, low literacy, and community resistance to immunization and government services. Other factors such as the safety of the vaccine, religious factors, and community distrust of government health systems played a major role in increasing transmission. This led to a reemergence of polio in Nigeria, especially in the Northern states. Even in areas where polio immunization was not controversial, failure to engage parents and discuss why a fully vaccinated child may develop polio disease, for instance, reinforced and increased parents’ negative perceptions of the polio program."
> "mysterious doctor-magician", do you think about what you're implying here?!
I'm not implying anything. I'm saying it explicitly. I'm certainly exaggerating for effect, but I'm saying it explicitly: lots of people in poor countries are fearful of medical professionals.
I don't know why that's surprising -- it's true right here in the USA, as well, and one of the reasons why certain ethnic groups have disproportionately bad medical outcomes.
> less likely to believe in anti vax or other anti science prkpaganda across all classes
Oh, stop. Nobody in this discussion is "anti science" -- I have a doctorate, in a biological science. Nor am I "anti-vax".
It's helpful if you don't characterize people who critically analyze research with an entire class of fictional villains. Because that actually is what scientists do.
Just a quick clarification, I was not accusing you of being anti-science or anti-vax. I was using that as a short-hand for people the who don't want to get vaccinated, I'm sorry if I gave you the impression that I'm talking about you.
Regarding the "reaching only people based on certain educational background" I think choosing a citation about northern Nigeria is quite selective. The assertion that people in Africa are more vaccination skeptical seems to be a gross overgeneralisation and is vaccination acceptance rates vary greatly between countries (not surprising as this is the same in the developed world as well).
> I'm not implying anything. I'm saying it explicitly. I'm certainly exaggerating for effect, but I'm saying it explicitly: lots of people in poor countries are fearful of medical professionals.
Well your choice of language certainly makes an association to stereotypes of "superstitious primitives"
> I don't know why that's surprising -- it's true right here in the USA, as well, and one of the reasons why certain ethnic groups have disproportionately bad medical outcomes.
Yes some ethnic groups, would these somehow be more likely to engage with the medical professionals that engage with the control groups, or go to the hospitals while being opposed to the "mysterious doctor-magicians"? Also the modern "health-suspicious" population in the USA (and other developed nations) is primarily composed of well off, well educated socio-economic backgrounds, e.g. just look at where recent measles outbreaks happened.
> Just pick the kids who are healthier, and call them "eligible".
Sure, yeah. Or whatever. But there's no evidence for that in the article, it says it was done based on age, and then matched between comparable communities. You'd have to not only mess with the eligibility, but only do so in the vaccinated communities. Because they compared eligible and ineligible children in the unvaccinated communities, too. And again, the cohorts were split apart by age. Maybe a bunch of unhealthy children didn't get the vaccine for that reason, but they'd be included in the age-cohort anyway.
Of course they could maliciously juice the study but the "what if richer, healthier children were the ones that got the vaccine" just doesn't seem a reasonable criticism at least as described. It seems like a perfectly good design to avoid being confounded that sort of thing.
> You'd have to not only mess with the eligibility, but only do so in the vaccinated communities. Because they compared eligible and ineligible children in the unvaccinated communities, too.
Yes, I get that. I'm not suggesting malfeasance here [1]. I'm just saying controls are hard, and these problems pop up in the best studies.
The difference between the clinical trials and this was that the clinical trials were an actual RCT, and this is an observational study. Observational studies almost always have confounding issues.
[1] I do think the immortal time bias is real, however. Whether or not the bias was consistent between groups is a separate question, but I almost don't really care. The fact that they're reporting that older children survive a bit longer than younger children, and not mentioning this issue, is sketchy to me. They either don't understand the problem (bad), or are exaggerating (typical, but still bad), or they're hiding something (really bad).
Honestly stuff like this just makes me exhausted for the state of medical science. You spent a crapload of money on this. Immortal time bias is confounding 101. We know how to avoid it. Do the damned RCT!
> Kids who make it to age N are more likely to survive to age N+1 than kids who make it to age N-1
Look, maybe I'm just giving them credit because this is filtered through journalism, but isn't that the point of the control communities? You can subtract out this bias using the control community. It's all down to picking comparable controls, obviously. If I had to point to a place you could screw up it would be picking the wrong control communities. Ideally you'd probably pair communities and then assign them at random to get the vaccines or not.
I'm not objecting to the idea that there could be confounders, just that it's probably not sampling bias along the lines of "richer and healthier children probably got the vaccine." If all you're saying is, "it's not an RCT" then... yeah, it's not?
The question is how hard is to pick good control communities. During the pandemic I've read a lot of preprints about cures for covid-19 (like Ivermectin) and many of them used a similar aproach. It looks very hard. I've seen too many bad results with this kind of controls. Double blind randomized controled trial or it didn't happen.
We already did the RCT and saw that the incidence of severe malaria was reduced.
Now we've done an actual rollout, and have seen in observational data a bigger reduction in all-cause mortality than we expected. It's relatively high quality observational data, but of course the risk of confounds is larger than the RCT.
It's great to report it, but you definitely want to see the cause-specific numbers, too.