Building firm foundations for therapy development.

The phase I trial in the oncology treatment development paradigm is supremely challenged on the one hand by being required to identify a satisfactory dose/regimen that will be carried forward in pilot and definitive efficacy evaluations and on the other hand to minimize patient's exposure to ineffective doses. The result of this compromise is often scant data on which to base the entire subsequent program of evaluation, the failure of which at any juncture could lead to abandoning further development of what might be a promising treatment strategy. At times, it can seem that we are, as the parable warns against, attempting to build a sturdy structure with a foundation of sand. In these necessarily small experiments, after the puta-tive maximum tolerated dose (MTD) is identified, sometimes an additional group of patients, known as the dose expansion cohort (DEC), is enrolled. As Boonstra and colleagues point out, as a prelude to their comprehensive simulation investigation (1), the practice is becoming increasingly common, although specific elements related to the utility of this cohort (sample size justification , for example) are often absent (2–4). Furthermore, the role that the expansion cohort may play in estimating (or redefining) the MTD is often not clearly specified. We surmise that the main motivation for the DEC is simply because the most commonly used design (3+3) results in a maximum of only six patients treated at the terminal dose, which is inherently unappealing. In their simulation studies, the traditional 3+3 approach and two variations on the continual reassessment method (CRM), a progressive dose-response curve model-based approach, are compared. Each was evaluated with and without a subsequent DEC, and the DEC was incorporated with the dose-escalation portion into re-estimation of the MTD. The investigators make several recommendations, supported by the observations and at odds with current phase I trial practice. First, with respect to design choice, the 3+3 approach remains ubiquitous despite recognized shortcomings, perhaps owing to its simplicity (one needn't even interact with a statistician during trial conduct, an attractive feature). Wider adoption of alternatives may be because of real and perceived complexity in implementation, as well as the assumption that the 3+3 design performs serviceably. Boonstra's study illustrates again that too often this design does not arrive at the true MTD. The CRM naturally incorporates information from all patients (not just the terminal dose tier) in dose escalation decisions and MTD estimation and has better …