Test negative designs with uncertainty, sensitivity, and specificity

Test-negative designs (TNDs) can be used to estimate vaccine effectiveness by comparing the relative rates of the target disease and symptomatically similar diseases among vaccinated and unvaccinated populations. However, the diagnostic tests used to identify the target disease typically suffer from imperfect sensitivity and specificity, leading to biased vaccine effectiveness estimates. Here we present a solution to this problem via a Bayesian statistical model which can either incorporate point estimates of test sensitivity and specificity, or can jointly infer them directly from laboratory validation data. This approach enables uncertainties in the performance characteristics of the diagnostic test to be correctly propagated to estimates, avoiding both bias and false precision in vaccine effectiveness. By further incorporating individual covariates of study participants, and by allowing data streams from multiple diagnostic test types to be rigorously combined, our approach provides a flexible model for the analysis of TNDs with explicitly stated assumptions.