ABSTRACT
Despite much research on the topic, little work has been done comparing the use of methods to control for confounding in the estimation of COVID-19 vaccine effectiveness in routinely collected medical record data. We conducted a trial emulation study to replicate the ChAdOx1 (Oxford/AstraZeneca) and BNT162b2 (BioNTech/Pfizer) COVID-19 phase 3 efficacy studies. We conducted a cohort study including individuals aged 75+ from UK CPRD AURUM (N = 916,128) in early 2021. Three different methods were assessed: Overlap weighting, inverse probability treatment weighting, and propensity score matching. All three methods successfully replicated the findings from both phase 3 trials, and overlap weighting performed best in terms of confounding, systematic error, and precision. Despite lack of trial data beyond 3 weeks, we found that even 1 dose of BNT162b2 was effective against SARS-CoV-2 infection for up to 12 weeks before a second dose was administered. These results support the UK Joint Committee on Vaccination and Immunisation modelling and related UK vaccination strategies implemented in early 2021.