Statistical model
VE was estimated as (1 – hazard ratio)*100. Hazard ratios comparing
vaccinated and unvaccinated were estimated using Cox proportional
hazards models with vaccination as a time-varying exposure; the
vaccination status of some individuals changed over time from
unvaccinated to vaccinated, and therefore the same participant could
contribute person-time to both exposure categories. Study time was used
as the underlying time scale in the Cox regression model. Participants
were considered fully vaccinated 14 days after they received their
second vaccine dose.
We calculated unadjusted and adjusted VE estimates. We categorized
hospitals into two groups based on geographical location; three
hospitals in central Baku were considered “central,” while four
hospitals located on the outskirts of Baku were considered
“peripheral.” Both unadjusted and adjusted VE estimates included
hospital group and prior infection as fixed effects. We assessed other
potential confounders (e.g., month, age, sex, occupation, hands-on care,
smoking, household size, any chronic condition, BMI) if they changed the
VE estimate by more than 5% using step-wise backward selection.
We defined previous infection as a PCR-confirmed infection prior to
enrolment documented in either the Etabib database or the MoH/Mandatory
Health insurance database. Participants who were unvaccinated at
enrolment and had PCR-confirmed SARS-CoV-2 infection prior to enrolment
began to contribute person-time on the date they became eligible for
vaccination (six months after their last positive test). Participants
who were vaccinated at enrolment and who had PCR-confirmed COVID-19
infection prior to enrolment were included in the analysis at the time
point they were considered “at risk” of reinfection, which we defined
as 90 days after their most recent positive PCR test.
Participants contributed person-time from enrolment, or, for individuals
with prior PCR-confirmed SARS-CoV-2 infection, from the start of time at
risk, until the earliest of outcome or exit from the study. Person-time
ended at whichever came first of the following outcomes: 1) the day of
the first SARS-CoV-2 infection, 2) the day of receipt of a second
COVID-19 vaccination if it occurred before the recommended interval
between the first and second dose 3) the day of receipt of a third
vaccine dose, or 4) the day of the last weekly questionnaire before
complete loss to follow-up, withdrawal from the study, transfer or
retirement from their hospital of employment, death, or censor date for
the analysis period (1 December 2021). Participants were also censored
from the primary analysis upon receiving a dose of any COVID-19 vaccine
other than CoronaVac.