Causal Inference
Estimating vaccine effectiveness in observational studies: a matching alternative Emily Wu* Emily Wu Elizabeth Rogawski-McQuade Mats Stensrud Razieh Nabi David Benkeser
Matching is a popular approach for estimating vaccine effectiveness in observational studies wherein vaccinated individuals are matched to unvaccinated individuals on key covariates. However, matching can obscure the causal estimand of interest and yield inefficient estimators thereof. In this work, we critically examine the estimand targeted by a matching-based analysis and propose a more general class of causal effectiveness estimands. We further propose G-computation style estimators of these estimands that are shown via simulation to enjoy significantly improved efficiency relative to matching-based estimators. The method is illustrated using a study of the effectiveness of the COVID-19 vaccine in children aged 5-11 years during the 2021-2022 school year. In this application, the proposed method produced point estimates of waning vaccine effectiveness comparable to those obtained through matching-based estimators, but with significantly narrower confidence intervals (Figure 1).
