Quantifying the impact of pathogen interactions on vaccine effectiveness estimates in the test-negative design

The test-negative design (TND) has been widely adopted for real-world vaccine effectiveness (VE) assessment, assuming that control selection is independent of vaccination status. However, pathogen interactions may contradict this assumption, posing challenges to VE evaluations. We developed a theoretical framework to quantify the pathogen-dependence-related biases in VE estimates. The framework was further empirically evaluated using data from 66,117 pediatric patients tested for multiple respiratory pathogens during 2023-2024, comparing influenza VE estimates using standard influenza-negative controls with those using specific non-influenza pathogen positive or negative controls. The theoretical framework indicated that the TND may overestimate VE even without pathogen dependence. Pathogen dependence further modified influenza VE estimates in a pathogen-specific direction, i.e., positive pathogen dependence with influenza tended to overestimate influenza VE, whereas negative pathogen dependence biased VE estimates downward. However, the magnitude of dependence-related bias was generally smaller than the bias attributable to the baseline model-specific discrepancy. The use of all-pathogen-negative controls minimized estimation bias. These findings suggest that pathogen dependence and control selection should be considered when using TND for VE estimation.