The Impact of Enrollment Strategies on the Performance of Time-To-Event Dose-Escalation Designs for Phase I Drug Combination Oncology Trials.

Dose-escalation trials make sequential dosing decisions for cohorts of patients based on accumulating toxicity data, often in the form of dose-limiting toxicities (DLTs). Challenges arise due to the necessity of complete follow-up for accurate dosing decisions. There are many settings where patients may become available to be enrolled before there is complete DLT information from the patients already in the trial. In radiotherapy trials, for example, the DLT assessment window is far longer than traditional chemotherapies and waiting for complete DLT information can result in impractically long trial durations. This issue can be exacerbated in escalation trials of dose-combinations, where the space to be explored and the sample sizes are generally larger. The time-to-event (TITE) approach, which utilizes partial follow-up data, has been effective in reducing trial lengths while ensuring accuracy and safety. Whether TITE methods are beneficial depends critically on the relationship between the DLT assessment window length and patient inter-arrival time, characterized by the assessment-to-interarrival (A/I) ratio. However, the impact of the rate of accrual into the trial and the size of each dosing cohort have not been carefully considered because it is widely assumed that accrual rates will be slow and the A/I ratio will be small. We use comprehensive simulation studies based on a dose-combination case-study and show that accrual rates corresponding to larger A/I ratios can result in poor accuracy and safety, while a dosing cohort size of one can additionally induce unfavorable operating characteristics. To address this issue, we propose a modification to the TITE design where the trial is paused if there are a certain number of non-completers. We demonstrate that this simple modified approach is robust across the range of A/I ratios investigated and has non-inferior operating characteristics to a non-TITE design while providing meaningful reductions in trial duration.