Estimands for combination therapy oncology trials with competing risks and their integration with Project SignifiCanT
Year: 2026
Type: article
Abstract: The evaluation of combination therapies in oncology increasingly demands estimand frameworks that reflect component-specific treatment strategies, complex intercurrent events (ICEs), and competing or semi-competing risk structures. Traditional hazard-based analyses often lack causal interpretability in these settings, particularly when selective discontinuation, immune-mediated toxicity, or death modifies subsequent exposure and influences event pathways. To address these limitations, we propose a unified causal-estimand framework specifically tailored for combination oncology trials. Grounded in potential-outcomes theory and aligned with ICH E9 (R1), the framework (i) defines regimen-level, component-specific, and pathway-specific estimands; (ii) formally maps these estimands to ICE strategies - treatment policy, hypothetical, composite, principal stratum, and while-on-treatment; (iii) evaluates the compatibility of commonly used statistical approaches, including cause-specific hazards, subdistribution hazards, and multi-state models, with these estimands; and (iv) integrates these elements with the regulatory objectives of the FDA's Project SignifiCanT, with particular emphasis on mechanistic interpretability, benefit-risk assessment, and transparent justification of analytical choices. By expressing estimands as contrasts of counterfactual risk or state-occupation functions - rather than relying solely on model-based hazard ratios - the framework enhances clinical relevance, supports attribution of component-specific effects, and facilitates cross-trial comparability. This work provides a principled foundation for designing, analysing, and reporting combination oncology trials, thereby advancing both methodological rigour and regulatory clarity.
Institutions: Newcastle University, Pfizer (United States), University of Dundee
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7221810641