New combination regimens evaluated in phase II cancer clinical trials often show promising results compared to the standard therapy for a disease system. Selection of patients with a better prognosis can be a prominent factor for this optimism. For most disease systems, prognostic variables that are related to the outcome are available and are called risk factors. Patients are classified into risk categories depending on the number of risk factors they possess. The patient distribution is defined as the proportion of patients falling into each of these risk categories. Typically, the patient distribution observed for a phase II study differs from the standard therapy reports so that the outcomes are not comparable. A randomized trial is the ultimate step for establishing the efficacy of a new treatment. In order to determine whether a regimen should progress to a phase III trial, we suggest adjusting the standard therapy outcome for the effect of the observed phase II patient distribution. If the endpoint of interest is tumour response proportion, a weighted average utilizing the standard therapy response proportions and the phase II patient distribution would provide an estimate of the adjusted standard therapy response proportion. Confirmatory phase II trials often attempt to estimate median survival in addition to response proportion, since this is the primary endpoint for most phase III cancer studies. Because data are censored, we propose an adjustment method based on the bootstrap resampling technique. We illustrate the problem of disparate patient selection with data from melanoma studies and demonstrate the usefulness of the proposed adjustment method with data from bladder cancer studies. A simulation study indicates that the magnitude of the adjustment is heavily dependent on the degree of separation of the risk categories. SAS code is available on a website (http://lib.stat.cmu.edu) for easy implementation.
ASJC Scopus subject areas
- Statistics and Probability