Causal inference in population health research evaluates the effects of hypothetical interventions and requires a clearly defined estimand, yet most studies leave its key components implicit. Author et al. (Am J Epidemiol. 2026;xxxxx) used the target trial emulation framework to make the causal estimand explicit and estimate effects of hypothetical loneliness interventions on memory function. Using their study as a motivating example, this commentary examines four components of a causal estimand: (1) the definition of the "treatment," (2) the choice of comparator, (3) the intervention strategy, and (4) the summary measure for population health impact. I discuss how complex social exposures like loneliness raise challenges for the consistency assumption through multiple versions of treatment; how using the natural course as a comparator, rather than a fixed counterfactual level, better reflects population-level impact; how sustained and stochastic interventions expand the space of scientific questions beyond deterministic elimination of exposure; and how estimands can move beyond population averages to address whether interventions narrow or widen health disparities. By carefully specifying each component of the estimand, investigators can move observational research toward more consequentialist evidence that directly informs what interventions would improve population health and mitigate health disparities.
Koichiro Shiba (Thu,) studied this question.