4 ms·

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. T

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.

timr 3y ago
> 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.
- roywiggins 3y ago
  > 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.
  - timr 3y ago
    > 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!
    - roywiggins 3y ago
      > 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?

timr 3y ago

> 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.

roywiggins 3y ago
> 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.
- timr 3y ago
  > 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!
  - roywiggins 3y ago
    > 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?