Primary Submission Category: Sensitivity Analysis
On the role of the Potential Outcome Association Structure for Principal Causal Effects
Authors: Arianna Nuti, Alessandra Mattei, Bjoern Bornkamp, Tianmeng Lyu,
Presenting Author: Arianna Nuti*
In clinical trials, subgroup analyses based on biomarkers that may lie on the causal pathway between treatment and outcome can provide valuable clinical insight. Assessing causal effects among patients who respond to treatment according to their biomarker levels, namely those with a sufficient biomarker reduction, may help physicians tailor therapy early after treatment initiation. We formalize this research question using the principal stratification framework, recognized in the ICH E9(R1) addendum as an estimand strategy for addressing post-treatment variables. Within this framework, the target causal estimand is the principal average causal effect for the principal stratum of responders, defined as patients whose biomarker value under treatment would fall below a pre-specified threshold. We investigate how this principal causal effect depends on the association structure between the potential outcomes for the biomarker and the primary endpoint Assuming a multivariate normal joint distribution of all potential outcomes, we explicitly express the causal effect as a function of the association parameters. We derive a closed-form formula, showing the sensitivity to non-identifiable association parameters. Our results provide insights into the role of the association parameters and highlight how causal conclusions may depend on assumptions about them. These findings may be informative even beyond the normal setting to assess the plausibility of this type of assumptions.