The primary goal of phase I studies in oncology is to determine the MTD (Maximum Tolerated Dose) for a
drug. This MTD is determined with respect to an accepted risk (usually 33%) to see a limiting toxity for patients. In this
paper we propose a new mathematical model to determine the MTD. An important feature of this model is that the
limiting toxicity can be formulated as a combination of several basic graded toxicities such as hematologic or
neurological. Another feature is the possibility to take into account several patient covariates to individualize the
determination of the MTD. The model is a bayesian model where some prior information has been considered. The model
is expected to work better than traditional empirical schemes for determining the MTD because it uses at every step all the
available information on patients, and adds some major improvements as compared with existing CRM strategies because
it uses whole data made available, including low-grades toxicities. Finally the model has been validated with a
retrospective data set on 17 patients from a phase I study on paclitaxel in pediatric oncology. Calculated MTDs for each
patient were found to be markedly different than the doses actually given following a traditionnal dose-escalation methodology.
Results suggest that our new model provides a better and safer way to drive dose-escalation in phase-I trials as
compared with traditionnal schemes.
Keywords: MTD, DLT, CRM, Toxicity grades, Skewed Logistic, MCMC methods, Prior distributions
Rights & PermissionsPrintExport