Infectious Disease
Examining the potential for bias in influenza, COVID-19, and respiratory syncytial virus test-negative vaccine effectiveness designs due to inclusion of vaccine preventable disease controls Margaret K. Doll* Margaret K. Doll Margaret K. Doll Jana L. Hirschtick
Background: The test-negative case-control design (TND) is used to estimate influenza, COVID-19, and respiratory syncytial virus (RSV) vaccine effectiveness (VE). In these studies, participants with a respiratory illness test positive (cases) or negative (controls) for the disease of interest. Research has shown that bias may arise from inclusion of controls with another vaccine-preventable disease (VPD), since vaccination behaviors across VPDs are associated. If uncontrolled, the magnitude of this bias depends, in part, on the prevalence of VPD controls; therefore, we estimated the prevalence of VPD controls eligible for inclusion in TND VE studies.
Methods: Using 2019-24 data from the largest US Midwestern nonprofit healthcare system, we identified hospitalizations for respiratory illnesses via ICD-10 criteria used in prior TND VE studies. To examine the potential for bias in influenza (2019-24), COVID-19 (2020-24), and RSV (2023-24) TND VE studies, we calculated the prevalence of alternate VPD etiologies among TND control participants. VPDs were identified via ICD-10 codes for Streptococcus pneumoniae, Haemophilus influenzae type B, influenza, SARS-CoV-2, and RSV (2023-24), with the disease of interest excluded. Based on VPD prevalence, we classified the potential for VE bias as no/low (<10%), moderate (10-<25%), or high (≥25%).
Results: Between 2019-24, we identified 796,149 hospitalizations due to respiratory illness. Of these, 188,274 (23.6%) had a VPD etiology. Figure 1 presents VPD prevalence among potential TND VE controls by season and disease of interest. The percentage of VPD controls varied from 0.4% to 31.5%, representing a range of no/low to high potential VE bias.
Conclusions: We demonstrate that VPD controls may represent up to nearly one-third of TND VE controls, depending on season and disease of interest. Researchers should consider whether alternate VPD controls are included in their TND VE studies, and the implications for potential bias.
