Abstract
The primary goal of a two-stage Phase I/II trial is to identify the optimal dose for the following large-scale Phase III trial. Conventional Phase I designs focus on the maximum tolerated dose (MTD), assuming that efficacy increases monotonically with dose. However, this assumption may not hold for novel therapies such as immunotherapies and targeted agents, where efficacy can plateau or decline at higher doses. In such cases, the optimal biological dose (OBD)-finding designs are more appropriate. In practice, several recommended Phase II doses (RP2Ds) are typically selected after Phase I for further evaluation in Phase II, where dose optimization designs assess toxicity and efficacy outcomes to support "go/no-go" decisions. The final recommended Phase III dose (RP3D) is then chosen from the RP2Ds meeting these criteria. We compared fifteen Bayesian model-assisted two-stage designs, combining five designs (BOIN, TITE-BOIN, BF-BOIN, BOIN12, and TITE-BOIN12) for dose finding with three designs (TS, BOP2, and TOP) for dose optimization, through extensive simulations under fixed and random scenarios, with and without the existence of the OBD. Our results show that BOIN12 and TITE-BOIN12 achieve better overdose control and RP2D selection, while BOP2 and TOP consistently outperform TS in RP3D selection.