Magnitude of potential biases in COVID-19 vaccine effectiveness studies due to differential health care seeking following home testing: implications for test-negative design studies

Observational studies that use the test-negative design (TND) approach for estimating vaccine effectiveness (VE) aim to reduce bias associated with differential health care seeking (HS) between vaccinated and unvaccinated individuals.1,2 For COVID-19 VE studies, COVID-19 vaccination status is compared between symptomatic patients who test positive for SARS-CoV-2 infection at health care facilities, and patients with similar symptoms who test negative for SARS-CoV-2. The design assumes that HS is independent of infection status among vaccinated and unvaccinated individuals. With widespread availability of at-home rapid diagnostic tests (RDTs) for SARS-CoV-2 infection, the effect of self-testing prior to HS has the potential to introduce selection bias in TND studies.3 We constructed a decision tree model that simulates a TND study of COVID-19 VE to estimate the magnitude of potential bias in COVID-19 VE studies due to differential HS after self-testing. We explored the potential bias in observed VE estimates (⁠|$\widehat{VE})$| using a directed acyclic graph (DAG) (Appendix S1 and Figure S1). The DAG illustrates the causal relationship between COVID-19 vaccination (Vcov), COVID-19-like illness (Icov), and other potential determinants of Vcov and Icov and selection into the TND study, such as use of at-home RDT, at-home RDT result (positive or negative), patients’ sociodemographic characteristics, and health care seeking (HS) behavior.