Bronner P. Gonçalves, Piero L. Olliaro, Sheena G. Sullivan, Benjamin J. Cowling
7 min
Abstract
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.
Alex: Wait, so vaccinated folks in the example ramp up risky contacts more than unvaccinated. But how do you get data on those "what if" behaviors?
Sam: To estimate it, they rely on a formula that weights observed disease rates among unvaccinated people by their behavior levels, then applies vaccinated rates at those same behaviors—adjusting for shared factors like age or work environment that influence actions. This works under assumptions like no hidden biases linking vaccination status to outcomes beyond measured traits. The paper notes these are standard but tricky in practice, especially since healthcare-seeking might twist behavior reporting.
Alex: Right, so it's not fixing behavior to one level for all, like forcing everyone to act cautious. It lets behavior vary naturally, just without the vaccine nudging it riskier.
Sam: Exactly—that's the difference from controlled effects, where you'd clamp behavior uniform across groups. Here, behavior follows its natural causes, like job demands, but minus vaccination's push toward risks. Studies from COVID showed vaccinated people did cut back less on gatherings or masks.
Alex: Okay, so collecting those contact diaries sounds key—but what practical hurdles come up when trying to measure behavior accurately in real studies?
Sam: One issue is that behavior might not be measured perfectly. People report things like social contacts or mask use, but they could forget details or report inaccurately. When that happens, estimates of the natural direct effect tend to shift toward the overall effect that includes behavior changes. The paper points to methods that handle this kind of measurement error. Another challenge is capturing all relevant behaviors—some studies track just one or two, but infection risk might depend on many factors working together.
Alex: Huh. So sloppy reporting on how often someone meets friends blurs the line between biology and actions?
Alex: Right—and tying back to confounders like healthcare-seeking, how do they suggest getting data on that without perfect records?
Sam: They recommend databases with long histories, using pre-vaccination clues like past vaccinations for other diseases, number of tests, or screening participation—these predict who seeks care more often. Table 2 in the paper contrasts this with behavior data: healthcare markers come from registries before the study starts, while behaviors need ongoing tracking like diaries after. Shared factors like age or income affect both, so adjusting is crucial.
Alex: So pre-shot habits flag the doctor-goers, letting you correct biases in who reports infections or changes contacts. But if healthcare-seeking shifts after vaccination, it complicates things further.
Sam: Yes. The paper discusses alternatives like sensitivity checks or bounds that don't assume it stays fixed.
Alex: You mentioned fallback strategies—what about when direct behavior data like diaries isn't available? How do they suggest checking for those riskier actions indirectly?
Sam: One approach uses infections from another germ, unrelated to the vaccine—like catching a different respiratory bug. Since the same habits raise risk for multiple germs, seeing more of these other infections among vaccinated people hints that vaccination nudged behaviors riskier. The paper's diagrams show this works under simple rules, like no vaccine protection against that second germ, after adjusting for healthcare habits.
Alex: Huh... so a spike in colds among the vaccinated could flag they're meeting more people, even without asking them directly?
Sam: Yes. It flags the vaccine-to-behavior path specifically when behaviors overlap for both germs. But if some habits only matter for one germ, the signal weakens.
Alex: Right, that sidesteps needing diaries. But with contagious diseases, doesn't one person's shot affect others around them, complicating the picture?
Sam: That's interference—one person's vaccination status influences neighbors' infection risk through spread. The paper extends the natural direct effect idea to groups, comparing outcomes where someone gets vaccinated but keeps their unvaccinated-level behavior, while holding others' statuses fixed. Identifying it demands advanced tools like network simulations.
Alex: Huh... so for policies, targeting that behavioral nudge could close the gap between lab promise and street reality. Trials dodge some mess by blinding, so no one knows and acts reckless.
Sam: Precisely—blinding keeps behaviors steady, unlike real life where knowledge sparks changes, as COVID studies showed with less masking post-shot. The natural direct effect targets what protection would look like without that nudge, guiding policies to curb risks, like messages reinforcing caution.
Alex: So ultimately, separating these threads clarifies why real-world numbers lag trials, pushing for smarter data and nudges. Valuable for next outbreaks. That's a solid wrap on sorting vaccine biology from our reactions to it. Thanks, Sam—and thanks for listening to this ResearchPod episode.