Quantifying biases in test-negative studies of vaccine effectiveness

摘要: The test-negative design has become the standard approach for assessing real-world performance of vaccines against influenza, with increasing applications in studies other infectious disease interventions. Vaccine effectiveness is measured from exposure odds ratio (OR) vaccination among individuals seeking treatment acute respiratory illness who receive a laboratory test influenza infection. This argued to provide natural correction differential healthcare-seeking behavior vaccinated and unvaccinated persons. However, relation OR true vaccine not well established. We derive under circumstances consistent studies. recovers direct effect when two are met: (1) that uncorrelated or susceptibility infection, (2) confers "all-or-nothing" protection (whereby certain have no while others perfectly protected). Biased size estimates, potentially including sign bias, arise if either condition unmet. Such bias may generate time-varying estimates suggestive waning absence protection. Troublingly, fail correct persons unless stringent clinical criteria upheld enrollment testing. Our findings demonstrate need reassess how data interpreted policy decisions conventionally based on causal inferences.