Abstract:
|
Since the introduction of the continual reassessment method in 1990, model-based adaptive dose-escalation approaches have evolved to address initial challenges with this approach including Bayesian Logistic Regression Model and alternative algorithmic designs, such as the modified toxicity probability interval method. With oncology drug development heavily focused on combinations, approaches continue to be broadened into the combination setting to study multi-drug therapies and to incorporate historical and emerging data. While models or algorithms have advanced, the endpoint on which these designs are applied remains focused on dose-limiting toxicity (DLT) within a defined time window (usually one or two cycles of treatment) with a goal to establish a maximum tolerated dose (MTD). In the setting of many new cancer treatments, the observation of DLT may not occur and “dose-finding” will rely on more than the primary “dose escalation” design. We reflect on the learnings and provide areas to consider when integrating historical and emerging data and examples of decisions using non-DLT data. Data borrowing in the context of efficacy signal evaluation will be discussed as well.
|