Causal Inference
Comparison of methodological approaches in COVID-19 vaccine effectiveness estimation using observational data Anne Huiberts* Anne Huiberts Bente Smagge Brechje de Gier Henri van Werkhoven Susana Monge Hester de Melker Susan Hahné Janneke van de Wijgert Susan van den Hof Mirjam Knol
Introduction: Target trial emulation (TTE) is an analytical framework to systematically assess and address potential biases in causal inference when using observational data. We aimed to assess vaccine effectiveness (VE) of the Omicron XBB.1.5 booster vaccination against self-reported SARS-CoV-2 infection between 2 October 2023 and 2 April 2024 using two TTE approaches.
Methods: A hypothetical target trial in which eligible participants would be randomized to booster vaccination or no booster vaccination was formulated. Two approaches were used to emulate this hypothetical trial using data of an ongoing prospective cohort study in the Netherlands (VAccine Study COvid-19; VASCO) (Figure 1). The first analysis used a single time zero at the start of the booster vaccination rollout and considered booster vaccination as a time-varying variable. The second analysis used multiple time zeros. On each day, all persons who received a booster vaccination on that day were matched 1:1 to persons who had not (yet) received a booster vaccination. Two additional approaches have been planned.
Results: 24,728 participants were eligible for vaccination and 16,295 (66%) were vaccinated. All were included in the single-time-zero analysis and 1,799 and 2,362 infections were reported during vaccinated and unvaccinated follow-up time, respectively. In the multiple-time-zero analysis, 15,138 vaccinated participants were matched to unvaccinated participants, who reported 960 and 1,518 infections, respectively. Overall VE was 39% (95% CI 35-43) using single- and 35% (31-39) using multiple-time-zero analysis. VE decreased as time since vaccination increased, being 46% (41-50), 29% (22-36) and 13% (-8-30) at 1-6, 7-12 and 13-18 weeks post-vaccination using single- and 42% (38-46), 39% (33-46) and 18% (-18-42) using multiple-time-zero analysis, respectively.
Conclusion: Results showed that VE analyses using different approaches within the TTE framework yielded rather similar VE estimates.
