Identification and Estimation of Causal Effects with Outcomes Truncated by Death

Linbo Wang, Xiao‐Hua Zhou, Thomas S. Richardson

Abstract: It is common in medical studies that the outcome of interest is truncated by death, meaning that a subject has died before the outcome could be measured. In this case, restricted analysis among survivors may be subject to selection bias. Hence, it is of interest to estimate the survivor average causal effect, defined as the average causal effect among the subgroup consisting of subjects who would survive under either exposure. In this paper, we consider the identification and estimation problems of the survivor average causal effect. We propose to use a substitution variable in place of the latent membership in the always-survivor group. The identification conditions required for a substitution variable are conceptually similar to conditions for a conditional instrumental variable, and may apply to both randomized and observational studies. We show that the survivor average causal effect is identifiable with use of such a substitution variable, and propose novel model parameterizations for estimation of the survivor average causal effect under our identification assumptions. Our approaches are illustrated via simulation studies and a data analysis.

The Challenge of Truncation by Death

In clinical trials and observational studies, researchers often aim to measure non-mortality outcomes, such as quality of life or cognitive decline. However, if a participant dies before the follow-up, their outcome is undefined—it isn't just "missing," it is truncated.

A common pitfall in research is comparing only the survivors in the treatment group to the survivors in the control group. This approach frequently leads to selection bias. For example, if a new drug keeps the sickest patients alive, the "survivor group" in the treatment arm will appear sicker on average than the control group, potentially masking the drug's true benefits. To overcome this, researchers focus on the Survivor Average Causal Effect (SACE): the effect of a treatment specifically on individuals who would have survived regardless of their treatment assignment.

Defining Survival Strata

To identify the SACE, we must categorize individuals into latent groups based on their potential survival under different treatment conditions. These groups, known as principal strata, are defined in the table below.

Table 1: Definitions of survival types based on potential survival status under treatment (S(1)) and control (S(0)).

The "always-survivor" group is the only subgroup for whom a causal effect is well-defined, as their outcome exists under both treatment and control scenarios. The challenge lies in the fact that we can never directly observe which individuals belong to this group.

Methodology: The Substitution Variable Framework

The authors introduce a novel framework using a substitution variable ($A$). This is a baseline covariate that is highly predictive of a person's survival type but has no direct effect on the final outcome ($Y$) except through the treatment and survival status.

This relationship is formalized in the Directed Acyclic Graph (DAG) below, which illustrates the causal paths between baseline covariates ($X$), the substitution variable ($A$), treatment ($Z$), survival ($S$), and the outcome ($Y$).

Figure 1: Causal diagram showing the role of the substitution variable (A) in identifying effects on outcomes (Y) truncated by death (S).

By leveraging the Exclusion Restriction (the assumption that $A$ does not directly affect $Y$) and Substitution Relevance, the authors demonstrate that the SACE can be identified even when traditional propensity score methods fail. This framework is particularly robust because it allows for the relaxation of common assumptions like "Monotonicity" (the idea that treatment never causes death).

Key Results and Clinical Implications

The proposed methods were validated through extensive simulations and a re-analysis of a Southwest Oncology Group (SWOG) trial comparing treatments for metastatic prostate cancer.

  • Simulation Consistency: In scenarios with heavy confounding, the authors' "Prop-ER" (Exclusion Restriction) and "Prop-NI" (No Interaction) methods remained accurate, while naive methods showed significant bias.
  • Re-evaluating Clinical Trials: Applying this method to the SWOG data revealed that the previously reported significant benefit of docetaxel on quality of life was likely overestimated. Once the latent "always-survivor" structure was accounted for, the effect became much smaller and statistically insignificant. This highlights how failing to account for truncation can lead to misleading clinical conclusions.

Conclusion

This research provides a powerful tool for medical researchers to disentangle treatment effects from survival-related selection bias. By using substitution variables to identify the always-survivor group, we can obtain more accurate estimates of how treatments impact the quality of life for those who survive. Future work will aim to generalize these findings to continuous treatment exposures and further refine sensitivity analysis techniques.

Summary updated

Abstract

It is common in medical studies that the outcome of interest is truncated by death, meaning that a subject has died before the outcome could be measured. In this case, restricted analysis among survivors may be subject to selection bias. Hence, it is of interest to estimate the survivor average causal effect, defined as the average causal effect among the subgroup consisting of subjects who would survive under either exposure. In this paper, we consider the identification and estimation problems of the survivor average causal effect. We propose to use a substitution variable in place of the latent membership in the always-survivor group. The identification conditions required for a substitution variable are conceptually similar to conditions for a conditional instrumental variable, and may apply to both randomized and observational studies. We show that the survivor average causal effect is identifiable with use of such a substitution variable, and propose novel model parameterizations for estimation of the survivor average causal effect under our identification assumptions. Our approaches are illustrated via simulation studies and a data analysis.
BibTeX
@inproceedings{wang2017identification,
  author = {Linbo Wang and Xiao‐Hua Zhou and Thomas S. Richardson},
  title = {{Identification and Estimation of Causal Effects with Outcomes Truncated by Death}},
  booktitle = {Biometrika},
  year = {2017},
  doi = {10.1093/biomet/asx034},
}
Figure