Conference Agenda

The ICH E9 (R1) addendum on estimands, coupled with recent advancements in causal inference, has prompted a shift towards using model-free treatment effect estimands that are more closely aligned with the underlying scientific question. This represents a departure from traditional, model-dependent approaches where the statistical model often overshadows the inquiry itself. While this shift is a positive development, it has unintentionally led to the prioritization of an estimand's ability to perfectly answer the key scientific question over its practical learnability from data under plausible assumptions. We illustrate this by scrutinizing assumptions in the recent clinical trials literature on principal stratum estimands, demonstrating that some popular assumptions are not only implausible but often inevitably violated. We advocate for a more balanced approach to estimand formulation, one that carefully considers both the scientific relevance and the practical feasibility of estimation under realistic conditions.

The adoption of the estimand framework within the clinical trial community, extending beyond data science, has made significant progress. The extensive educational efforts to generalize the understanding and application of causal concepts and the estimand framework are now facilitating more effective and meaningful cross-functional conversations. In healthcare companies, these efforts also guide the allocation of limited resources dedicated to biostatistical innovations towards the most impactful.

Drawing on experience from early and late-stage clinical development in Alzheimer’s disease and other neurological conditions, this presentation will review real-life examples of how this paradigm shift has impacted the design, conduct, analysis and interpretation of clinical trials. We will look into future areas of development and opportunities for advancing the field.

In phase III oncology trials, the analysis of overall survival (OS) is often complicated by unidirectional treatment crossover, where patients randomized to control are permitted to switch to the experimental treatment upon disease progression. While a treatment policy estimand is routinely adopted as the primary analysis in this setting, hypothetical estimands have gained considerable attention as supplemental analyses; most relevant are estimands that contrast the “experimental treatment” with the hypothetical regime of “control treatment without the option to crossover after progression”. Rank preserving structural failure time models (RPSFTM) and inverse probability weighting (IPW) methods remain popular for estimating such estimands, but they all rely on unverifiable assumptions and have their limitations.

In this talk, we illustrate the utility of multi-state models for constructing and estimating marginal, causally interpretable OS estimands in the hypothetical scenario that crossover is not permitted. Specifically, following the idea of Gran et al. (2015), we define hypothetical treatment regimes by artificially manipulating transition intensities in the observed multi-state process and estimating OS probabilities by g-computation under some common causal assumptions. We report simulation results to demonstrate the performance of the proposed approach in realistic clinical settings and present a roadmap for its implementation in R. We discuss potential extensions to other relevant hypothetical treatment regimes. The talk will conclude with remarks on the robustness of modeling assumptions required by the method.