The C33D design, with or without the Storer (2001) modification, has no real statistical basis, and more efficient alternatives have been sought. Efficiency here can be thought of as achieving the right MTD and with as few patients as possible. However, the design is easy to implement and requires little (statistical) manipulation - only keeping a count of the number of patients experiencing DLT at each dose tested. However, published studies appear to suggest that many variations from the basic C33D occur in practice. Indeed, Smith, Lee, Kantarjian et al. (1996) comment, following a review of Phase I studies conducted at the MD Anderson Cancer Center, Houston, USA, that: 'investigators sometimes entered cohorts of patients at a dose intermediate between two previously tested levels'. This clearly makes designing Phase I trials somewhat problematic but perhaps unavoidable since critically ill patients are often involved. Nevertheless, such difficulties imply that the results need to be interpreted with due caution and carefully reviewed before taking the next step in the development process.
O'Quigley, Pepe and Fisher (1990) and O'Quigley (2001) have proposed the continual reassessment method (CRM) as an alternative to C33D. This design recruits the first patient to a dose closer to the centre of the range of pre-specified doses than the dMINIMUM of C33D. Essentially, if DLT is observed in this first patient then the next patient (Patient 2) is given the dose below dSTART, whereas if no DLT is observed he or she receives the dose above dSTART. Once this second patient receives the corresponding dose, and presence or absence of DLT is observed, the subsequent dose to utilise (which may be below, at or above the dose last used) is determined. However, at any stage of this process, the results from all individual patients so far recruited are utilised to provide the basis for the choice of the dose to be tested in the next patient recruited.
Was this article helpful?