Fig. 1: Directed acyclic graph illustrating selection bias in test-negative studies by health-care-seeking behaviour. | npj Vaccines

Fig. 1: Directed acyclic graph illustrating selection bias in test-negative studies by health-care-seeking behaviour.

From: The need for a clinical case definition in test-negative design studies estimating vaccine effectiveness

Fig. 1

In (a), health-care-seeking behaviour HS confounds the relationship between vaccination V, SARS-CoV-2 infection status SI (e.g., by influencing engagement in risk behaviours), and COVID-19 status C19 (e.g., because of other healthy behaviours that modify disease severity). Only patients who are tested for SARS-CoV-2 are selected into the study S = 1. An individual’s health-care-seeking behaviour HS and COVID-19 status C19 influence whether they present for care, are tested and selected into the study S = 1, resulting in collider bias. In (b), the test-negative design by restricting participants to those who present to sentinel sites and meet particular clinical criteria HS = 1, the collider bias introduced by S = 1 is blocked enabling unbiased estimation of the V-C19 effect.

Back to article page