The test-negative design for estimating influenza vaccine effectiveness

Abstract Objective The test-negative design has emerged in recent years as the preferred method for estimating influenza vaccine effectiveness (VE) in observational studies. However, the methodologic basis of this design has not been formally developed. Methods In this paper we develop the rationale and underlying assumptions of the test-negative study. Under the test-negative design for influenza VE, study subjects are all persons who seek care for an acute respiratory illness (ARI). All subjects are tested for influenza infection. Influenza VE is estimated from the ratio of the odds of vaccination among subjects testing positive for influenza to the odds of vaccination among subjects testing negative. Results With the assumptions that (a) the distribution of non-influenza causes of ARI does not vary by influenza vaccination status, and (b) VE does not vary by health care-seeking behavior, the VE estimate from the sample can generalized to the full source population that gave rise to the study sample. Based on our derivation of this design, we show that test-negative studies of influenza VE can produce biased VE estimates if they include persons seeking care for ARI when influenza is not circulating or do not adjust for calendar time. Conclusions The test-negative design is less susceptible to bias due to misclassification of infection and to confounding by health care-seeking behavior, relative to traditional case-control or cohort studies. The cost of the test-negative design is the additional, difficult-to-test assumptions that incidence of non-influenza respiratory infections is similar between vaccinated and unvaccinated groups within any stratum of care-seeking behavior, and that influenza VE does not vary across care-seeking strata.