Reply to Skowronski, De Serres, and Orenstein

To the Editor—We agree with Skowronski et al [1] that caution is required in conducting and interpreting studies of influenza vaccine effectiveness (IVE) using medical and public health data. Indeed, in our stratified results and discussion, we addressed many of the points the authors raised regarding IVE differences between sites, by illness severity (and thus thresholds for admission), and during peak and nonpeak weeks of influenza circulation [2]. We agree that clinician-ordered testing may bias IVE results if selection of who is tested is associated with both the risk of influenza positivity and the probability of influenza vaccination. Obviously, in the PREVENT cohort, clinicians often tested women with suspected influenza virus infection, and thus influenza positivity was high. Therefore, the risk of bias depends on the association between testing and the probability of vaccination. As we described in our article, published findings regarding this association are mixed. In a simulation study of the possible impact of selection bias on IVE estimates using the test-negative design (TND), Jackson et al found that bias from differential care seeking by patients was only meaningful (ie, reduced a true IVE of 50% by >5%) when vaccination doubled the likelihood of care seeking [3]. If we apply the same model but substitute clinician testing for patient care seeking as the action that selects patients into a study, Jackson et al’s simulation suggests that clinician testing among vaccinated versus unvaccinated pregnant women would have to differ by >2-fold in order to meaningfully bias IVE estimates. We acknowledge that the description of vaccination documentation in our article was limited, and greater detail on these methods is now published [4]. Most prospective IVE networks rely in part or entirely on self-reported vaccination status, which can increase false positive reports and thus reduce specificity of vaccine exposure measurement. In contrast, our study exclusively used vaccination documentation from medical records and registries, which can increase false negative records and thus reduce sensitivity. In a recent simulation study, Jackson et al concluded that low sensitivity presented a lesser risk of bias to IVE estimates using TND than did low specificity; records with only 40% sensitivity to true vaccination status could result in an IVE estimate of 40% when the true IVE is 50% [5]. This is consistent with our interpretation that we likely underestimated true IVE. Finally, we are concerned that a reader might misinterpret Skowronski et al’s observation that we focused on “general laboratory-submissions without standardization of the influenza testing-indication” to mean that we simply sampled all patients with clinical testing. Our study focused on real-time reverse transcription polymerase chain reaction testing among pregnant women hospitalized during weeks of local influenza circulation with a diagnosis associated with influenza in previous studies. Although prospective IVE networks play a vital role in IVE monitoring, further population-based research like PREVENT that builds on our methodological strengths and mitigates limitations is also needed to assess IVE in preventing less frequent and severe outcomes, such as influenza illness that is a primary or secondary cause of hospitalization during pregnancy. Financial support. This study was funded in part by the US Centers for Disease Control and Prevention. Potential conflicts of interest. A. N. reports grants from Pfizer, MedImmune/AstraZeneca, and Merck, outside the submitted work. N. K. reports grants from GlaxoSmithKline, Sanofi Pasteur, Pfizer, Protein Science, Merck & Co, MedImmune, Novartis (now GlaxoSmithKline), and Dynavax, outside the submitted work. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.