Hierarchical Composite Endpoints for Dose Optimization in Early Phase Oncology Trials

The FDA-initiated Project Optimus has emphasized the need for dose optimization (DO) in oncology drug development, moving beyond the traditional maximum tolerated dose paradigm. Evaluating the balance of benefit and risk is central to DO. In this paper, we propose a novel benefit-risk assessment framework for DO in early-phase oncology trials, utilizing hierarchical composite endpoints (HCEs). The HCE approach enables integration of multiple endpoints within a clinically prioritized hierarchy. Generalized pairwise comparison is then used to compare patient outcomes across candidate dose levels in DO cohorts, providing interpretable metrics to guide dose selection. Our method addresses key practical challenges of existing methods by enabling straightforward, qualitative clinical inputs and integrating various DO-related evidence. The framework also supports uncertainty quantification and sensitivity analysis to ensure robust assessment. Additionally, we extend the HCE method to incorporate pharmacokinetic data through a probabilistic index model, allowing drug exposure to support dose selection. Extensive simulation studies demonstrate that the proposed method achieves superior performance in selecting optimal doses compared to utility-score-based methods, particularly in scenarios requiring careful balancing of efficacy and safety. Application to real-world trial data for DO in HER-2 positive nonsmall cell lung cancer further illustrates the method's practical utility and robustness.