I just posted this REPLY on a comment on the Levi / Ladapo Florida study paper.
I hope it gets published.
Robert, unfortunately it was not possible to match everyone in Florida 1:1 on 7 demographics plus vaccination center and time of vaccination. This is a VERY restrictive matching condition. Did you do a computation showing this is possible to match more than 84% exactly like this? Could you post it here?
The authors simply matched as many as they could match. They posted their R code in the supplement. m.out <- matchit(formula, data = df, method = "exact", ratio = 1)
The method "exact" forces an exact match. The R code appears to be correct. Did you find an error in their code? Can you point it out please? There are not any lines in the code that would "discard" any rows which didn't fit the hypothesis of the authors. Did you find such a line(s)?
In Supplement Fig 3, the graph for NCACM was mislabeled. They reversed it. The text of the study is correct. The "cherry picked" matched group nearly exactly matched the FULL cohort. From the text, FULL cohort OR was 1.32. For the matched cohort, OR was 1.356. So the matching showed the effect was slightly larger than the raw data. I've reported the labeling error in the figures to Professor Levi. Thank you for highlighting the label switch mistake in these Figures!
As for your bigger concern about the unvaccinated, I've discussed that with Professor Levi and he didn't include the unvaccinated because the comparison would be unproductive. Unvaccinated typically die at 2x to 3x higher than those seeking vaccination. This is a well known long term HVE effect. See Atanasov for example. So including the unvaccinated would be like having a comparison between two brands of apples and comparing that with pears. The pear comparison wouldn't yield any useful information. Studies attempting to normalize this HVE bias have been uniformly unsuccessful. If you know of a study that successfully accounted for that, please post it. The differences are non-linear, so using Cox models or similar methods don't work. I just verified with ChatGPT which wrote: "As of now, no published observational study has fully neutralized the long-term Healthy Vaccinee Effect (HVE)—particularly over a 12 month period post-vaccination" So if you know of such a study, that would be extremely valuable. Thanks!