Calibrated phase II clinical trials in oncology.

This paper proposes the use of calibrated designs in phase II oncological clinical trials and evaluates their statistical properties in terms of power recovery and cost. A calibrated phase II design for a new cancer treatment for a specific tumour, e.g. colo-rectal, consists of random allocation of patients to receive either the investigational treatment or a standard treatment known to have activity at a certain level in phase II trials (e.g. 5 FU, expected response proportion = 0.20). Patients assigned to the standard treatment form the calibration group. The calibration group is not a control group in the traditional sense and one does not conduct a formal efficacy comparison between the investigational treatment group and the calibration group. Instead, one uses the calibration group to evaluate whether the sample of patients who receive the investigational treatment has the capability of showing a response. If the data do not support the hypothesis that the expected response proportion prevails in the calibration group, one declares the investigational group results suspect and recommends a second trial. Assuming acceptable results of the second trial, we use binomial calculations to find the effect of the calibration design on power recovery and relative cost. We show that when an unrepresentative sample occurs, calibration designs generally recover 90 per cent or more of nominal power at a cost of three to fivefold increase in sample size. We recommend for calibrated phase II trials a 'master protocol' approach in which several investigational treatment arms share one concurrent calibration group.