Balancing sample size and accuracy of dose selection in phase 1 oncology trials – designs tiered by cohort size and maximum number of patients treated per dose

The objective of phase 1 oncology dose‑escalation trials is to identify the dose with a toxicity rate that matches or is the closest to the target dose‑limiting toxicity (DLT). There are mainly two categories of design methods for phase 1 trials: the rule-based methods and model-based/assisted methods. All literature has focused on the designs, mostly on how the various model-based/assisted methods are better than the 3 + 3 design, which is the most used rule-based design. The accuracy of dose selection is perhaps the most important operating characteristics (OC). Our investigation shows that the accuracy is more consequentially affected by the maximum number of patients treated per dose (MPTPD) than by the type of designs. Simulation examples show that the model-based BOIN and CRM design has lower accuracy for dose selection than the 3 + 3 design with the same cohort size and the same MPTPD. Larger MPTPD is associated with larger sample sizes. Simulations show that selection accuracy improves with increasing MPTPD (hence sample sizes). Our findings suggest that the design of phase 1 trials should include consideration of the balance between MPTPD, sample size and expectation of selection accuracy. Rule-based designs that are tiered by cohort sizes and MPTPD are proposed. These designs perform similarly with the BOIN and CRM designs with identical cohort sizes and MPTPD, but can be used without any software.