Thread by @VikiLovesFACS

Just out! A new meta-analysis of 81,349 ppl receiving #CovidVaccine during #pregnancy compared to 255,346 unvaccinated. Vaccination...

👉🏻 Protects against infection

👉🏻 Lowers chance of stillbirth by 27%

👉🏻 Lowers chance of babies needing NICU by 12%

jamanetwork.com/journals/jamap…

I’ve been asked for some additional context around whether the 27% reduction in stillbirth reported by this paper is likely to be an overestimate…

🧵

https://twitter.com/hicksyalex/status/1577536665240870912

This is particularly worth considering because a similar meta-analysis found that the rate of stillbirth was reduced by 15% among those vaccinated in pregnancy - rather less than the 27% reported here… 2/

nature.com/articles/s4146…

As I’ve mentioned before, people who take medical advice and get vaccinated in pregnancy are also likely to be taking care of their health in other ways which might improve outcomes, and this might make vaccination look better than it really is. 3/

https://twitter.com/vikilovesfacs/status/1585547966676439040

Most studies try to take this into account by controlling or matching on factors that we know are correlated with or indicative of healthcare seeking behaviour, such as socioeconomic status or whether a person has got their flu vaccine. 4/

This gives us what is called an adjusted odds ratio, or aOR. What Alex is asking here is why the stillbirth rate in this meta-analysis was calculated using the unadjusted odds ratio, or OR, for one of the biggest studies. Won’t that inflate the apparent effect? 5/

Yes, and that’s probably why this meta-analysis finds a larger protective effect than the previous one.

So why have they used this approach? Two reasons… 6/

First, not all the studies included in the meta-analysis have an aOR, and this one instead gives an adjusted Hazard Ratio. aHR cannot be converted into aOR so the authors have taken the raw data and calculated an OR from it. But the paper contains insufficient data to adjust. 7/

2. They pre-specified that this would be their approach. So even if - as Alex suggested to me in DM - they decided to change their statistical approach after the fact to take account of this, this would not be ethical… 8/

crd.york.ac.uk/prospero/displ…

… since the whole point of a systematic review and meta-analysis is that we publish what we will do before we start. This prevents cherry picking of data, or fiddling analyses if we don’t like what we find. 9/

I hope you found this trip into the bowels of SR and MA helpful! 10/10