A practical illustration of the importance of realistic individualized treatment rules in causal inference

The effect of vigorous physical activity on mortality in the elderly is difficult to estimate using conventional approaches to causal inference that define this effect by comparing the mortality risks corresponding to hypothetical scenarios in which all subjects in the target population engage in a given level of vigorous physical activity. A causal effect defined on the basis of such a static treatment intervention can only be identified from observed data if all subjects in the target population have a positive probability of selecting each of the candidate treatment options, an assumption that is highly unrealistic in this case since subjects with serious health problems will not be able to engage in higher levels of vigorous physical activity. This problem can be addressed by focusing instead on causal effects that are defined on the basis of realistic individualized treatment rules and intention-to-treat rules that explicitly take into account the set of treatment options that are available to each subject. We present a data analysis to illustrate that estimators of static causal effects in fact tend to overestimate the beneficial impact of high levels of vigorous physical activity while corresponding estimators based on realistic individualized treatment rules and intention-to-treat rules can yield unbiased estimates. We emphasize that the problems encountered in estimating static causal effects are not restricted to the IPTW estimator, but are also observed with the $G$-computation estimator, the DR-IPTW estimator, and the targeted MLE. Our analyses based on realistic individualized treatment rules and intention-to-treat rules suggest that high levels of vigorous physical activity may confer reductions in mortality risk on the order of 15-30%, although in most cases the evidence for such an effect does not quite reach the 0.05 level of significance.

💡 Research Summary

The paper tackles a fundamental methodological problem in causal inference when estimating the effect of vigorous physical activity (VPA) on mortality among older adults. Conventional approaches define a causal effect by comparing two hypothetical static interventions: one in which every individual in the target population engages in a specified level of VPA and another in which no one does. Identification of such static effects from observational data requires the positivity (or overlap) assumption—that every subject has a non‑zero probability of receiving each treatment level. In the elderly, this assumption is implausible because many individuals with serious health conditions simply cannot perform high‑intensity exercise. Consequently, estimators that ignore this violation tend to produce biased, overly optimistic estimates of the protective effect of VPA.

To address the issue, the authors propose two alternative frameworks that explicitly incorporate the set of feasible treatment options for each participant. The first is a realistic individualized treatment rule (ITR), which maps each subject’s baseline health status, functional limitations, and other covariates to the highest level of VPA that is realistically attainable for that person. The second is an intention‑to‑treat (ITT) rule, which treats the assignment to a VPA recommendation as the exposure, regardless of whether the individual ultimately follows it. Both frameworks respect the positivity constraint by limiting the comparison to interventions that are actually possible for each individual.

The empirical illustration uses a cohort of older adults with detailed measurements of physical activity, comorbidities, and mortality outcomes. Four widely used causal estimators are applied under three analytical scenarios: (a) static treatment effect, (b) ITR‑based effect, and (c) ITT‑based effect. The estimators are: (1) inverse‑probability‑of‑treatment weighting (IPTW), (2) G‑computation, (3) doubly‑robust IPTW (DR‑IPTW), and (4) targeted maximum likelihood estimation (TMLE).

Results show a consistent pattern. When a static intervention is assumed, all four estimators suggest a large mortality reduction—often exceeding 30%—associated with high‑intensity VPA. However, this effect is driven by the positivity violation: subjects who are unable to exercise vigorously are forced into an unrealistic counterfactual, inflating the apparent benefit. By contrast, the ITR and ITT analyses yield more modest risk reductions, typically in the 15–30% range, and the confidence intervals are substantially wider. In most cases the p‑values do not reach the conventional 0.05 threshold, indicating that the evidence, while suggestive, is not definitive. Importantly, the bias is not confined to the IPTW estimator; G‑computation, DR‑IPTW, and TMLE exhibit the same over‑estimation under the static framework, confirming that the problem is structural rather than methodological.

The authors discuss several implications. First, causal inference in observational studies must respect the realistic set of treatment options for each unit; otherwise, the positivity assumption is violated and effect estimates become unreliable. Second, individualized treatment rules provide a principled way to define causal parameters that are both scientifically meaningful and statistically identifiable. Third, intention‑to‑treat analyses retain the advantages of randomised‑trial‑like comparisons while accommodating non‑compliance, making them useful when actual adherence cannot be guaranteed. Fourth, the choice of estimator (IPTW, G‑computation, DR‑IPTW, TMLE) does not rescue a misspecified causal parameter; careful definition of the target estimand is paramount.

In conclusion, the paper demonstrates that static causal effect estimators can dramatically overstate the benefits of vigorous exercise in the elderly because they ignore the fact that many older adults cannot safely engage in high‑intensity activity. By shifting the focus to realistic individualized treatment rules and intention‑to‑treat frameworks, the authors obtain unbiased—or at least less biased—estimates that suggest a moderate mortality benefit of VPA, though the statistical evidence remains borderline. The work underscores the necessity of incorporating feasibility constraints into causal models and provides a template for future studies where treatment feasibility varies across individuals.