Generalised pairwise comparisons in dose-optimisation oncology trials: beyond safety to multi-outcome dose selection

PURPOSE: The primary objective of dose-finding oncology trials (DFOTs) is to determine the recommended phase II dose. Whilst dose-selection has traditionally relied upon clinician-reported safety outcomes, the goal of DFOTs increasingly focusses upon the integration of safety, activity, and tolerability within decision-making, in line with modern dose-optimisation strategies. Seamless Phase I/II designs often use decision frameworks to quantify trade-offs between outcomes to guide dose selection. However, assigning numerical values to reflect such trade-offs can be difficult as clinical judgements and interpretation often vary between investigators. EXPERIMENTAL DESIGN: With stakeholders including clinical teams and patients potentially considering their own prioritisation of outcomes, generalised pairwise comparisons (GPCs), including the win ratio (WR) provide a statistical framework mirroring this clinical decision-making by evaluating treatment benefit against prioritised outcomes. Using the WR, doses are compared across pre-specified prioritised outcomes sequentially, with the optimal dose yielding the largest proportion of favourable (winning) patient-pair comparisons. This paper presents WIN-DOSE, a hierarchical, multi‑outcome win ratio-based approach for dose optimisation. We demonstrate the performance of WIN-DOSE in a two-arm randomised dose-optimisation trial incorporating dose-limiting toxicities (DLT) for safety, preliminary response for activity, and both dose intensity and patient-reported outcomes for tolerability. RESULTS: When one dose is clearly favourable, WIN-DOSE consistently identifies the optimal dose. We also demonstrate how the WR can accommodate different trade-offs between safety, activity, and tolerability, supporting transparent and clinically relevant dose selection. CONCLUSIONS: The WR can support transparent and clinically relevant patient-centric dose-selection decision-making, aligning with the broader goals of early phase DFOTs.