Background: Test-negative designs (TND) are derived from case-control designs where cases are patients with a positive test for a specific disease at a healthcare facility and controls are patients with a negative test at the same facility for the same symptoms. TND has been used to assess influenza vaccine effectiveness (IVE) and is now considered for COVID-19 vaccine effectiveness (CVE) assessment.
Objectives: To evaluate the applicability and challenges of the TND to assess CVE with respect to internal validity, external validity and efficiency.
Methods: A targeted literature review of TND methods applied to IVE was performed and a summary of the strengths and limitations for CVE considering current knowledge about COVID-19 disease and vaccination was prepared.
Results: COVID-19 symptomatology is heterogeneous, raising the question of the target population for case assessment. Targeting hospitalized patients would allow the assessment of CVE on prevention of severe illness but a TND in the community setting would evaluate prevention of milder illness. The CVE objective will vary according to each vaccine’s target population. Laboratory tests to confirm COVID-19 are numerous with variable performance. In the absence of a gold standard, misclassification of cases and controls is possible, warranting a consistent and reliable diagnostic approach to limit classification bias. In addition, finding more cases than controls (which would limit statistical power) may become a risk if the frequency of non-COVID respiratory infection (e.g. influenza) is lower than expected due to widespread protective measures. Exposure depends on evolving local public health decisions (e.g., priority access, product availability, vaccination organization and schedule). Tracking these policies is key for the design and results interpretation, e.g. imbalance between exposed and unexposed groups could stem from massive vaccination in priority populations and could affect the sample size and CVE precision estimate. Exposure category definition will need to account for alternative vaccination schedules (e.g. lag between doses, no booster). Collecting vaccination data (brand, dates of 1st and 2nd injection) should be done with extra scrutiny to avoid exposure misclassification. Emergency mass vaccination might complicate data collection from medical sources, e.g. vaccination centers with no access to medical records.
Conclusions: TND is a valuable approach given the massive testing applied worldwide but results need to be interpreted with caution due to challenges specific to COVID-19.
