Time-to-Event Bayesian Optimal Interval Design to Accelerate Phase I Trials

Abstract

Late-onset toxicity is common for novel molecularly targeted agents and immunotherapy. It causes major logistic difficulty for existing adaptive phase I trial designs, which require the observance of toxicity early enough to apply dose-escalation rules for new patients. The same logistic difficulty arises when the accrual is rapid. We propose the time-to-event Bayesian optimal interval (TITE-BOIN) design to accelerate phase I trials by allowing for real-time dose assignment decisions for new patients while some enrolled patients' toxicity data are still pending. Similar to the rolling six design, the TITE-BOIN dose-escalation/deescalation rule can be tabulated before the trial begins, making it transparent and simple to implement, but is more flexible in choosing the target dose-limiting toxicity (DLT) rate and has higher accuracy to identify the MTD. Compared with the more complicated model-based time-to-event continuous reassessment method (TITE-CRM), the TITE-BOIN has comparable accuracy to identify the MTD but is simpler to implement with substantially better overdose control. As the TITE-CRM is more aggressive in dose escalation, it is less likely to underdose patients. When there are no pending data, the TITE-BOIN seamlessly reduces to the BOIN design. Numerical studies show that the TITE-BOIN design supports continuous accrual without sacrificing patient safety or the accuracy of identifying the MTD, and therefore has great potential to accelerate early-phase drug development. Clin Cancer Res; 24(20); 4921-30. ©2018 AACR.

©2018 American Association for Cancer Research.

Figures

Figure 1

Hypothetical phase I clinical trial using the TITE-BOIN design. Patients are treated in cohort sizes of 3, and the number above the “×” indicates the time when DLT occurs.

Figure 2

Comparative performances of the time-to-event Bayesian optimal interval (TITE-BOIN), rolling six (R6), and time-to-event continual reassessment method (TITE-CRM) designs with respect to that of the 3+3 design. (a and b) Relative percentages of correct selection (PCS) of the MTD; (c and d) Relative percentages of patients assigned to the MTD. The target DLT rate in scenarios 1-8 is 0.2, while that in scenarios 9-16 is 0.3. The accrual rate is 2 patients/month. A higher value is better.

Figure 3

Comparative performances of the time-to-event Bayesian optimal interval (TITE-BOIN), rolling six (R6), and time-to-event continual reassessment method (TITE-CRM) designs with respect to that of the 3+3 design. (a and b) Relative percentages of trials selecting the dose above the MTD; (c and d) Relative percentages of patients assigned to the doses above the MTD; (e and f) Relative percentages of patients assigned to doses below the MTD. The target DLT rate in scenarios 1-8 is 0.2, while that in scenarios 9-16 is 0.3. The accrual rate is 2 patients/month. A lower value is better.

Figure 4

Comparative performances of the time-to-event Bayesian optimal interval (TITE-BOIN), rolling six (R6), and time-to-event continual reassessment method (TITE-CRM) designs with respect to that of the 3+3 design. (a and b) Relative percentages of “regretful” trials; (c and d) Average trial durations in months. The target DLT rate in scenarios 1-8 is 0.2, while that in scenarios 9-16 is 0.3. The accrual rate is 2 patients/month. A lower value is better.