Natural direct effects of vaccines and post-vaccination behaviour

Knowledge of the protection afforded by vaccines might, in some circumstances, modify a vaccinated individual’s behaviour, potentially increasing exposure to pathogens and hindering effectiveness. Although vaccine studies typically do not explicitly account for this possibility in their analyses, we argue that natural direct effects might represent appropriate causal estimands when an objective is to quantify the effect of vaccination on disease while blocking its influence on behaviour. There are, however, complications of a practical nature for the estimation of natural direct effects in this context. Here, we discuss some of these issues, including exposure-outcome and mediator-outcome confounding by healthcare seeking behaviour, and possible approaches to facilitate estimates of these effects. This work highlights the importance of data collection on behaviour, of assessing whether vaccination induces riskier behaviour, and of understanding the potential effects of interventions on vaccination that could turn off vaccine’s influence on behaviour.

💡 Research Summary

The paper addresses a subtle but important source of bias in vaccine effectiveness studies: the possibility that knowledge of being vaccinated changes an individual’s behavior, thereby altering exposure to the pathogen and confounding the measured effect of the vaccine. The authors argue that the conventional total effect—comparing disease risk between vaccinated and unvaccinated groups—mixes two distinct causal pathways. One pathway is the direct biological protection conferred by the vaccine (V → Y), and the other is an indirect pathway where vaccination influences post‑vaccination behavior (the mediator M), which in turn affects disease risk (V → M → Y). When the research objective is to isolate the pure immunological benefit of the vaccine, the appropriate causal estimand is the natural direct effect (NDE), defined as the effect of vaccination on disease risk while holding the mediator (behavior) at the level it would have taken without vaccination.

The authors construct a causal diagram that highlights two major sources of confounding. First, exposure‑outcome confounding arises when factors such as health consciousness or health‑care seeking behavior affect both the likelihood of being vaccinated and the risk of infection. Second, mediator‑outcome confounding occurs when the same or other unmeasured variables influence both the post‑vaccination behavior and disease risk. Health‑care seeking behavior is a particularly problematic confounder because it can simultaneously affect vaccination status, the mediator (e.g., willingness to attend crowded places), and the outcome (through early testing or treatment). Ignoring these confounders leads to biased NDE estimates.

To address these challenges, the paper proposes a multi‑layered methodological framework. At the design stage, researchers should collect detailed, time‑ordered data on behavior (e.g., mobility, mask use, social contacts) and on health‑care utilization (e.g., clinic visits, testing frequency). Such data can be gathered via surveys, mobile‑phone tracking, or linkage to electronic health records. In the analysis stage, the authors recommend using structural equation modeling or weighted mediation analysis to adjust for observed confounders and to estimate the NDE. When a suitable instrumental variable (IV) is available—such as a region‑level vaccination education campaign that influences behavior independently of infection risk—IV methods can help control for unobserved mediator‑outcome confounding. Sensitivity analyses (e.g., Rosenbaum bounds) are essential to assess how robust the NDE estimate is to residual, unmeasured confounding.

The paper illustrates the approach with two case studies. For seasonal influenza vaccines, the authors find that risk‑compensating behavior is minimal; consequently, the NDE and total effect are similar. In contrast, for COVID‑19 vaccines, substantial risk compensation (e.g., increased travel, reduced mask wearing) is observed. Empirical estimates show that while the NDE remains high—reflecting strong immunological protection—the total effect is attenuated because behavior offsets part of the benefit. This empirical contrast underscores why separating the direct effect from the mediated effect is crucial for accurate policy evaluation.

Policy implications are emphasized. When evaluating vaccination programs, decision‑makers should report both the NDE (the vaccine’s intrinsic protective value) and the total effect (the real‑world impact that includes behavioral responses). Complementary interventions—such as continued mask mandates, public information campaigns, or targeted behavioral nudges—can be designed to “turn off” the vaccine’s influence on risky behavior, thereby aligning the observed total effect more closely with the NDE. Moreover, systematic collection of behavioral and health‑care seeking data should become a standard component of vaccine trials and observational studies.

In conclusion, the authors make a compelling case that natural direct effects provide a theoretically sound and practically relevant estimand for vaccine effectiveness research when behavior change is a plausible mediator. By outlining concrete data‑collection strategies, analytic techniques, and sensitivity‑analysis tools, the paper offers a roadmap for researchers to obtain unbiased estimates of the vaccine’s pure biological benefit while simultaneously accounting for the complex interplay between vaccination, behavior, and disease risk. This framework can improve the scientific rigor of vaccine evaluations and help policymakers design more effective, behavior‑aware immunization programs.