The continual reassessment method for phase I clinical trial design: is it always more efficient than the 3+3?
Session type: Poster / e-Poster / Silent Theatre session
It is becoming widely suggested that the continual reassessment method (CRM) is superior to the 3+3 for phase I clinical trials for reaching the maximum tolerated dose (MTD) more efficiently and accurately. The literature in this area is growing, but the majority of comparisons consider a high number of dose levels. In many phase I trials, the number of dose levels may only be 3, and it is unclear whether the CRM is beneficial in this situation.
Simulations were performed to compare the 3+3 and CRM, using 3 dose levels, maximum sample size=18 and upper acceptable limit of toxicity=1/3 (all in line with a 3+3 trial of 3 dose levels). Three different priors and cohort sizes of 2 and 3 were investigated for the CRM, and 10000 trials were simulated under six different “true” scenarios of dose limiting toxicity rates.
Simulations show that the 3+3 and CRM are largely comparable, and no benefit of the CRM is observed in this setting (brief summary of data in Table 1). This is because the accuracy of the CRM relies heavily on the prior, and where this is not close to the true scenario, the CRM can perform poorly.
Table 1. Average probability of choosing the correct MTD across all true scenarios
From these results, we can argue that for a trial considering 3 dose levels, where the prior rates of toxicities are unclear, the CRM is not more efficient unless the sample size is increased. Therefore, the growing view that the CRM is superior to the 3+3 across the board is not always accurate.