DOSET：源自一项首次人体 I 期 CAR-T 淋巴瘤试验的疗效与毒性同步评估剂量优化

Traditional phase I designs aim to identify the maximum tolerated dose, assuming higher doses increase efficacy, which may not hold in chimeric antigen receptor (CAR)-T cell therapy. We present DOSET (Dose Optimization with Simultaneous Efficacy and Toxicity) evaluation, a novel adaptive design jointly modeling safety and preliminary efficacy to identify the recommended dose, minimizing patient exposure to subtherapeutic or toxic doses, informed by a phase I CAR T lymphoma trial with a small sample.

DOSET is a two-part Bayesian adaptive design. Part 1 employs the continual reassessment method (CRM) with an initial 2 + 2 + 2 escalation across three dose levels to identify tolerable doses, allowing early stopping for excessive toxicity. Part 2 randomly assigns patients across selected doses, with ongoing CRM updates allowing dose adjustments as needed. Activity is evaluated concurrently with a Bayesian interim futility framework. Final dose selection is based on joint tolerability and activity. Simulations compared dose selection accuracy, patient allocation, and safety with the Bayesian optimal interval phase I/II design (BOIN12).

DOSET effectively selects active and tolerable doses, allocates patients efficiently, and reliably stops when all doses are excessively toxic. Compared with BOIN12, it improves dose selection without compromising safety, maintains or enhances early stopping for futile or toxic doses, and provides greater flexibility.

DOSET integrates dose escalation and optimization in a seamless framework, leveraging safety and efficacy data. Despite small-sample sizes, it offers an efficient and flexible approach to improve early-phase CAR T cell trials and applies broadly to non-monotonic dose-response settings, including targeted therapies and immunotherapies.