Utility‐Based Dose Optimization Approaches for Multiple‐Dose Randomized Trial Designs Accounting for Multiple Endpoints

The initiation of dose optimization has driven a paradigm shift in oncology clinical trials to determine the optimal biological dose (OBD). Early-phase trials with randomized doses can facilitate additional investigation of the identified OBD in targeted populations by incorporating safety, efficacy, and biomarker data. To support dose comparison in such settings, we propose to extend the utility score-based approach (U-MET) to account for multiple endpoints and doses. The utility-based dose optimization approach for multiple-dose randomized trial designs accounting for multiple (≤ 3) endpoints and doses (U-MET-m) extends the U-MET, using a utility score to account for multiple endpoints jointly (e.g., toxicity-efficacy trade-off). When there are > 3 endpoints, assigning weights jointly is quite complicated; therefore, we suggest an alternative approach with CUI-MET (clinical utility index dose optimization approach for multiple-dose randomized trial designs), which uses a utility index to account for multiple endpoints marginally. We demonstrate the relationship between U-MET-m and CUI-MET to offer a guide in weight selection for U-MET-m when there are up to three endpoints. U-MET-m and (extended) CUI-MET use Bayesian inference within a hypothesis framework to compare utility metrics across doses to identify the OBD. Here we describe simulation studies and present examples to compare both the U-MET-m and CUI-MET designs with the empirical design. The U-MET-m design and CUI-MET were shown to have satisfactory operating characteristics for selecting the OBD. Based on these findings, we recommend the U-MET-m with ≤ 3 endpoints and CUI-MET with > 3 endpoints as the primary dose comparison approach to select the OBD.