Vaccinated and unvaccinated were matched on the following variables in the following ways:
• Age – in bins of two consecutive years.
• Sex (male, female) – exact matching.
• Place of residence – exact matching, at the level of neighborhood or small town.
• Population Sector (general Jewish, Arab, ultra-orthodox Jewish) – exact matching.
• Influenza vaccines in the past 5 years (0, 1-2, 3-4, 5) – exact matching.
• Pregnancy (yes, no) – exact matching.
• Count of pre-existing conditions considered by the CDC as risk criteria (0,1,2,3,4+) –
exact matching.
We illustrate the matching process with an example: A 76 year old ultra-orthodox Jewish male
from a specific neighborhood, who received 4 influenza vaccines in the last five years and has 2
comorbidities that are known risk factors for severe COVID-19, could only be matched with a
similar ultra-orthodox Jewish male from the same neighborhood, aged 76-77 years, with 3-4
influenza vaccine count, and 2 comorbidities.
Selection of variables for matching was done in two stages. First, a set of variables was chosen
based on domain expertise. This set was then gradually trimmed while ensuring that
exchangeability was maintained. Exchangeability was evaluated using vaccine effectiveness for
COVID-19 symptomatic infection in the first 14 days after the 1st dose, which was close to 0 in the RCT1.
Matching was performed using a “rolling cohort” design. Starting from the beginning of the
vaccination campaign, each day all newly vaccinated patients who complied with the study’s
inclusion and exclusion criteria were candidates for matching, including those who had
previously been included as unvaccinated controls. They were matched with unexposed
(unvaccinated) controls that complied with the inclusion and exclusion criteria and were not
previously recruited. If at a later date an unexposed control was vaccinated, he or she was
censored from the unexposed group together with his or her exposed match, and a new matched
unvaccinated control was sought so that this newly vaccinated individual could be included as a
vaccinated person.
https://www.nejm.org/doi/suppl/10.1056/ ... pendix.pdf
