ABSTRACT

In some dose-finding trials, there are several groups of patients, and the goal is to estimate a maximally tolerated dose (MTD) within each group. These groups may be defined by the patients’ degree of impairment at baseline [9, 16], amount of pre-treatment [17], or genetic characteristics [6]. For example, Ramanathan et al. [16] enrolled 89 patients with varying solid tumors to develop dosing guidelines for the administration of imatinib in patients with liver dysfunction. Prior to dosing, patients were stratified into none, mild, moderate, or severe liver dysfunction at baseline, according to serum total bilirubin and AST. A similar classification is used by LoRusso et al. [9]. Kim et al. [6] define three groups of patients according to the number of defective alleles, either 0, 1, or 2. In each of these cases, parallel phase I studies were conducted within each group, but they did not account for the expectation that the MTD would be lower in the more severely impaired patients at baseline, or in the subset of patients with a greater number of defective alleles. In these cases, even with an efficient design, with the sample sizes typically seen in phase I trials, ignoring the orderings among the groups can lead to reversals in the MTD estimates, meaning that the estimated MTDs in the groups can contradict what is known clinically. For example, the parallel designs might recommend a greater dose level as the MTD in the most severely impaired group compared to a less severely impaired group. Even in cases where the ordering is not known, running parallel studies can be inefficient compared to a design that uses a model to pool information from all patients across all groups in order to estimate the dose–toxicity relationship.