Estimated vaccine effectiveness (VE) can vary markedly across years and between populations, even for the same vaccine. The authors examine how heterogeneity in pre-vaccination infection risk—both within and between vaccinated and unvaccinated groups—can cause estimated VE to diverge from true vaccine protection. They emphasize that VE studies capture contrasts in risk in a particular time and place, and therefore population-level differences in prior immunity can produce counterintuitive variability in VE estimates.
The investigators analyzed the difference between estimated VE and true vaccine-mediated protection using theoretical and simulation models that represent continuous distributions of pre-vaccination infection risk. They considered three conceptual scenarios to isolate sources of bias in VE estimates related to heterogeneity in prior infection risk. Analytical and regression work was performed in R (version 4.3.3) and simulations used Julia (version 1.10.2). The authors made code and model implementations publicly available at the project repository.
One scenario examined a straightforward contrast in mean pre-vaccination risk between vaccinated and unvaccinated populations. When the two groups differ in average risk of prior infection before vaccination, estimated VE can be biased in either direction: it can appear higher or lower than the underlying true vaccine protection. The direction and magnitude of this bias depend on how the average pre-vaccination risks compare between groups, because VE estimates reflect the relative hazards faced by the two groups at the time of measurement.
A second scenario considered cases in which vaccinated and unvaccinated populations share the same mean pre-vaccination risk but have different risk distributions (for example, differing levels of variance or skew). The authors show that differences in the shape of risk distributions alone—without differences in mean—can produce similar patterns of discrepancy between estimated VE and true protection. In other words, heterogeneity in who was previously infected within each group, not only average levels, affects how VE is estimated.
The third scenario addresses the temporal dynamics when infection-derived immunity lasts longer than vaccine-mediated protection. Under that condition, annual VE estimates can vary substantially between years despite constant vaccine-induced protection. The authors report that yearly VE estimates may change by tens of percentage points across seasons, driven by the accumulation and waning of infection-derived immunity in the population rather than changes in vaccine efficacy itself.
These theoretical results underscore that VE estimates are context-specific: they quantify contrasting risk between vaccinated and unvaccinated individuals at a given time and place, not an immutable property of the vaccine. Observed variability in VE across years or populations may therefore reflect underlying heterogeneity in prior infection and the distribution of susceptibility, as well as susceptible depletion effects, rather than true changes in vaccine protection.
Interpreters of VE studies should consider the population distribution of prior infection risk and how that distribution differs between vaccinated and unvaccinated groups. Explaining variability in estimated VE will often require more granular data on pre-vaccination immunity in the study population and attention to temporal dynamics of infection-derived immunity relative to vaccine-induced protection.
The authors have made all analytical and simulation code available at the GitHub repository: https://github.com/cobeylab/heterogeneity_in_pre-vax_immunity_contributes_to_ve_variability. Analytical and regression analyses used R version 4.3.3, and the simulation model was implemented in Julia version 1.10.2.
Competing interests are disclosed: B.J.C. reported consulting for multiple vaccine companies; M.L. disclosed limited consulting for a vaccine manufacturer and paid work for other organizations on unrelated matters. Other authors reported no potential conflicts. Funding sources included the National Institute of Allergy and Infectious Diseases, the U.S. Department of Education, and Seoul National University startup funds. The manuscript was posted to medRxiv on August 31, 2026.