Abstract
Skepticism about the assumption of no unmeasured confounding, also known as exchangeability, is often warranted in making causal inferences from observational data; because exchangeability hinges on an investigator's ability to accurately measure covariates that capture all potential sources of confounding. In practice, the most one can hope for is that covariate measurements are at best proxies of the true underlying confounding mechanism operating in a given observational study. In this paper, we consider the framework of proximal causal inference introduced by Miao et al. (2018); Tchetgen Tchetgen et al. (2020), which while explicitly acknowledging covariate measurements as imperfect proxies of confounding mechanisms, offers an opportunity to learn about causal effects in settings where exchangeability on the basis of measured covariates fails. We make a number of contributions to proximal inference including (i) an alternative set of conditions for nonparametric proximal identification of the average treatment effect; (ii) general semiparametric theory for proximal estimation of the average treatment effect including efficiency bounds for key semiparametric models of interest; (iii) a characterization of proximal doubly robust and locally efficient estimators of the average treatment effect. Moreover, we provide analogous identification and efficiency results for the average treatment effect on the treated. Our approach is illustrated via simulation studies and a data application on evaluating the effectiveness of right heart catheterization in the intensive care unit of critically ill patients.
Introduction
Causal inference from observational data traditionally relies on the "exchangeability" assumption—the belief that all variables influencing both treatment and outcome have been measured. However, in real-world scenarios, unmeasured confounders are almost always present, leading to biased results. While "negative controls" are often used to detect such bias, correcting for it remains a significant challenge.
This paper advances the proximal causal inference framework, which treats certain observed variables as "proxies" for hidden confounders. By moving beyond restrictive linear assumptions, the authors develop a robust semiparametric theory and a novel doubly robust estimator, providing a more reliable toolkit for analyzing complex observational datasets.
The Proximal Framework
The core of proximal causal inference lies in identifying two types of proxy variables that "bridge" the gap created by unmeasured confounding ($U$):
- Treatment Proxies ($Z$): Variables associated with the treatment ($A$) but having no direct effect on the outcome ($Y$), except through the hidden confounder.
- Outcome Proxies ($W$): Variables associated with the outcome ($Y$) but having no direct effect on the treatment ($A$), except through the hidden confounder.
As shown in Figure 1, these proxies allow researchers to account for various confounding structures, including scenarios where multiple hidden factors coexist. By leveraging these proxies, the framework identifies causal effects even when exchangeability fails.
Semiparametric Theory and Estimation
The authors introduce Confounding Bridge Functions to relate the proxies to the causal effect without needing to model the hidden confounder directly. This involves solving Fredholm integral equations of the first kind.
A key contribution is the development of the Proximal Doubly Robust (PDR) estimator ($\hat{\psi}_{PDR}$). This estimator is derived from the Efficient Influence Function (EIF), which defines the theoretical precision limit (semiparametric efficiency bound) for this setting. The PDR estimator is "doubly robust" because it remains consistent if either the outcome bridge model or the treatment bridge model is correctly specified.
Figure 2 illustrates a more complex DAG where multiple latent confounders ($U_1, U_2, U_3$) interact. The proposed theory accounts for such dependencies, ensuring that the estimator achieves local efficiency—using the data as efficiently as possible when both models are correct.
Empirical Results
The methodology was validated through both simulations and a real-world application:
- Simulations: In tests where one bridge model was intentionally misspecified, the PDR estimator maintained low bias (e.g., absolute bias of 0.1), whereas non-robust versions and standard estimators ignoring hidden confounding showed severe bias (up to 64.0).
- SUPPORT Study: The authors re-evaluated the effectiveness of Right Heart Catheterization (RHC) in critically ill patients. While standard methods often suggest a certain level of risk, the proximal estimators indicated a potentially more harmful effect on 30-day survival, suggesting that previous findings might have been influenced by unmeasured confounding.
Conclusion and Future Directions
Proximal causal inference offers a powerful alternative when exchangeability is untenable. However, its success depends on the valid categorization of proxies. Future work aims to address the numerical instability of solving "ill-posed" integral equations and to develop automated methods for proxy selection from high-dimensional data.
Summary updated