Abstract:
|
When designing the dose escalation trials in oncology, it is recognized that the dose limiting toxicity (DLT) events may have different onset time frames. Focusing only on the events that are likely to occur within a short time period may result in missing some important late-onset events in the dose escalation consideration and thus under-estimating the overall toxicity level of each dose. On the other hand, suspending patient recruitment for a longer DLT observation window delays the dose finding stage of the drug development, especially when numerous dose steps are evaluated. Many existing dose escalation methods have been extended to accommodate the study designs with late-onset DLT events, with which the dose for a new patient or cohort can be recommended even before all existing patients finish their entire DLT observation window. However, these methods are not designed to integrate multiple categories of DLTs defined with different observation windows given their likely onset time frames. We propose to use a time-to-event Bayesian piecewise proportional hazard (TITE-BPPH) model to handle the problem. We provide prior distribution specifications and the overall study design with the dose escalation rules derived based on the model inference. Simulation results are presented to demonstrate the operating characteristics of the method and compare it to the classical 3+3 with DLT status of all patients fully resolved before recruiting a new cohort as a benchmark, and also to an escalation with overdose control proportional hazard (EWOC-PH) method assuming constant hazard. The two model-based approaches reduce the trial duration and usually identify the maximum tolerated dose more accurately than the 3+3. Also the TITE-BPPH method exhibits more consistent performance than the EWOC-PH method among different compositions of the early- and late-onset events and compares favorably with the later method in several scenarios.
|