Severity-dependent test-seeking behaviors and test-negative designs: impact on estimated vaccine effectiveness and utility of analytic and design choices

Avnika B Amin,Matt D T Hitchings,Otavio T Ranzani,Jason R Andrews,Derek A T Cummings,Albert I Ko,Julio Croda,Natalie E Dean

DOI: https://doi.org/10.1093/aje/kwae303

2024-08-27

Abstract:Test-negative designs are increasingly used to evaluate vaccine effectiveness because of desirable properties like reduced confounding due to healthcare-seeking behaviors and lower cost compared to other study designs. An individual's decision to seek care often depends on their disease severity, with severe disease more likely to be captured than mild disease. As many vaccines likely attenuate disease severity, this phenomenon generally results in an upward-biased estimate of vaccine effectiveness against symptomatic disease. To address the resulting bias, analytic solutions like adjusting for or matching on severity have been suggested. In this paper, we examine the performance of the test-negative design under different vaccine effects on disease severity and the utility of adjusting or matching on severity. We further consider the implications of studies that focus only on milder disease by restricting recruitment to outpatient settings. Through an analytic framework and simulations accompanied by a real-world example, we demonstrate that, when vaccination attenuates disease severity, the magnitude of bias is influenced by the degree of under-ascertainment of mild disease relative to severe disease. When vaccination does not attenuate disease severity, bias is not present. We further show that analytic fixes negligibly impact bias and that outpatient-only studies frequently produce downward-biased estimates.