The authors argue that when vaccinated individuals change their behaviour because they believe they are protected, that change can increase their exposure to pathogens and thereby alter measured vaccine effectiveness. They propose that the effect of interest, in many circumstances, is the effect of vaccination on disease when the vaccine's influence on behaviour is blocked. By definition, this corresponds to the natural direct effect from causal inference frameworks. The paper positions natural direct effects as a conceptually relevant target for understanding what vaccines would do in the absence of behaviour change.
Although natural direct effects are conceptually useful, the authors emphasize practical complications for their estimation in vaccine studies. The review does not provide new empirical estimates but discusses methodological obstacles that arise when attempting to separate the vaccine's biological protection from behavioural pathways. These complications include identification issues inherent to mediation analysis and the need for assumptions that may be difficult to justify or verify in observational vaccine effectiveness research.
A central concern highlighted is confounding of both the exposure-outcome and mediator-outcome relations by healthcare-seeking behaviour. Individuals' propensity to seek care may influence both their likelihood of being exposed or tested and their reported behaviours after vaccination. Such patterns can create bias when estimating how much of a vaccine's effect is direct (biological protection) versus mediated through behaviour. The paper points out that these forms of confounding are particularly problematic because healthcare seeking can correlate with both mediator (behaviour) and outcome (disease detection), complicating causal interpretation.
The authors discuss possible approaches to make estimation of natural direct effects more feasible in practice. Although specific implementation details or empirical protocols are not provided in the abstract, the review highlights causal inference tools—such as potential outcomes and formal definitions of direct effects—as the conceptual foundation. The paper suggests that addressing confounding and carefully defining the mediator and counterfactual scenarios (for example, the hypothetical absence of vaccine-induced behaviour change) are necessary steps. The authors also draw attention to the need for strategies that account for healthcare-seeking patterns when attempting mediation analyses in vaccine studies.
A clear recommendation from the review is the importance of collecting data on behaviour in vaccine effectiveness studies. Without behavioural measurement, it is difficult to assess whether vaccines induce riskier behaviour and to quantify how much behaviour change contributes to observed disease outcomes. The authors argue that integrating behavioural data will support evaluation of whether, and to what extent, observed vaccine effects are mediated by behaviour rather than being purely biological.
The paper concludes that understanding the effects of vaccines when their influence on behaviour is blocked is important for interpreting vaccine impact. Estimation of such natural direct effects is subject to practical and methodological challenges, including confounding by healthcare-seeking behaviour. The authors underscore the need for thoughtful data collection on behavioural mediators and for using causal inference concepts to clarify what vaccine studies measure. Overall, the review calls for attention to post-vaccination behaviour in vaccine effectiveness research and for development of methods and data collection practices that can better separate biological protection from behavioural pathways.
Conflict of interest
The abstract includes a conflict of interest statement noting that several authors report consulting or advisory relationships with multiple pharmaceutical companies. The review otherwise focuses on conceptual and methodological issues around direct effects, behaviour, and measurement in vaccine effectiveness studies.