Using Benchmarks Based on Historical Survival Rates for Screening New Therapies for Stage IV Melanoma Patients

Abstract

For the great majority of patients diagnosed with stage IV melanoma, prognosis is poor and death comes quickly. After three decades of clinical trials in metastatic melanoma, systemic therapy has failed to alter the natural history of this disease. Counterbalancing this is the promise of new therapeutic agents developed through a deepening understanding of the molecular circuitry and immunobiology of melanoma. The article by Korn et al in this issue of the Journal of Clinical Oncology presents an alternative framework for evaluating phase II trials in melanoma with the aim of making fewer errors in identifying candidate therapies worthy of phase III testing. There are two aspects of this approach that differ from standard phase II trial designs. First, 1-year overall survival (OS) rates and 6-month progression-free survival (PFS) rates are proposed as outcomes rather than tumor response. These end points also have been explored in trials of patients with glioblastoma multiforme, where the disease is similarly aggressive and therapies are similarly inactive. The 6-month PFS rate has been adopted for use as a clinical end point for phase II trials conducted in the brain tumor cooperative groups. Second, a trial-specific benchmark, calculated retrospectively at the end of the trial, is used in a decision rule to move an agent forward. This proposed benchmark is tailored for a specific trial. It depends on historical rates within the prognostic classes defined by Korn et al and on the numbers of patients actually accrued in each prognostic class. Korn et al took advantage of the lack of progress in therapeutic development in melanoma to pool data from 2,100 patients from 70 arms of 42 Southwest Oncology Group trials done between 1975 and 2005 who were treated with a variety of agents and regimens, some of which appeared sufficiently promising to proceed to phase III trials. However, no phase III trial in melanoma has ever demonstrated an improvement in OS, including those with objective response rates so high that some investigators doubted the equipoise of trials that randomized patients to the historical but inactive standard therapy, dacarbazine. The authors’ meta-analysis based on 1,278 patients provided an estimate of the 1-year OS rate (25.5%) and 1-year OS rates for 24 prognostic classes (ranging from 5.5% to 63.8%) defined by four statistically significant independent prognostic factors: three patient factors (performance status, presence of visceral metastasis, sex) and one trial factor (exclusion of patients with brain metastasis). Their analysis confirmed the findings of the 2001 analysis by the American Joint Committee on Cancer staging committee that site of metastatic disease is an important prognostic factor for patients with metastatic melanoma. However, the most important prognostic factor that they identified, performance status, is one that all oncologists recognize but which the current American Joint Committee on Cancer staging system does not take into account. The strength of the association between performance status and both OS and PFS clearly justifies its inclusion as an important prognostic factor for stage IV melanoma patients. To facilitate future use of the information from the authors’ prognostic model presented in Table 3, the number of patients within each prognostic class should be made publicly available in addition to the OS rates. These data are essential for the computation of the 1-year OS rate when one or more of the prognostic factors are not available. In addition, these data can be used to compare the characteristics of patients at specific sites versus those patients included in the meta-analysis. In the proposed framework, the decision to advance a therapeutic intervention to a phase III trial depends on the difference between a trial’s benchmark (based on historical 1-year OS) and the observed 1-year OS rate. The standard parameter of interest, the true 1-year OS rate for (all) stage IV patients, is replaced with one that is tailored, providing more flexibility to reflect site-specific patient characteristics or inclusion/exclusion criteria for the specific therapeutic trial. This is an elegant and important use of a prognostic model. Its use with a decision rule defined for a specific clinical trial balances tailoring (to a specific patient population) with generalizability (survival rates within prognostic classes from a meta-analysis). Sample size is a key element of the design of all phase II trials. The authors suggest three approaches. The first uses standard procedures and computes sample sizes based on a site-specific, historical 1-year OS rate; the second seems arbitrary and uses 35% for the 1-year OS rate; and the third is done within the context of survival analysis using a historically derived survival curve (in Appendix A). Oddly, the first two approaches do not use any information from the authors’ metaanalysis or any trial-specific features such as trial eligibility criteria. In JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L VOLUME 26 NUMBER 4 FEBRUARY 1 2008