All Times EDT
Keywords: cohort size, phase I dose-escalation, 3+3, CRM, BOIN
The cohort size of phase 1 clinical trials and thus the timing of the interim decisions are typically prespecified in the trial protocol. During trial implementation, however, the cohort size often deviates from the planned one, which shifts the schedule of the interims. Despite its pervasiveness in phase 1 trials, the association of cohort size deviation with the operating characteristics of these trials is not clear. In this cross-sectional simulation study, a review was conducted of 102 phase 1 dose-escalation trials published between January 2017 and May 2018 in 3 peer-reviewed journals (Journal of Clinical Oncology, Clinical Cancer Research, and Cancer). After exclusion of studies that did not report the cohort size, 45 trials remained for analysis. Based on the analysis results, a simulation study was performed to systematically investigate the association of cohort size deviation with the operating characteristics of the trials. Of the 45 reviewed trials, 10 (22.2%) adhered strictly to the planned cohort size. The simulation study showed that when cohort size deviation was random, it had little association with the performance of novel model-based and model-assisted designs. When the cohort size deviation was informative and made based on the observed data on toxicity (eg, if dose-limiting toxicity was observed, the size of the next or current cohort was reduced or expanded), the variation of the design performance increased. The range of the change in the percentage of correct selection was -3.7 to 1.3 percentage points for the continual reassessment method and -4.5 to 0 percentage points for the Bayesian optimal interval design. The findings suggest that when novel phase 1 clinical trial designs are used, some cohort size deviation is acceptable and has little association with the performance of the designs. These deviations may be used by expert investigators to properly interpret the data, ensure safety, and leverage flexibility in the protocol.