Historical controls for metastatic pancreatic cancer: benchmarks for planning and analyzing single-arm phase II trials
Philip A Philip, Kari Chansky, Michael LeBlanc, Lawrence Rubinstein, Lesley Seymour, S Percy Ivy, Steven R Alberts, Paul J Catalano, John Crowley, Philip A Philip, Kari Chansky, Michael LeBlanc, Lawrence Rubinstein, Lesley Seymour, S Percy Ivy, Steven R Alberts, Paul J Catalano, John Crowley
Abstract
We compiled and analyzed a database of cooperative group trials in advanced pancreatic cancer to develop historical benchmarks for overall survival (OS) and progression-free survival (PFS). Such benchmarks are essential for evaluating new therapies in a single-arm setting. The analysis included patients with untreated metastatic pancreatic cancer receiving regimens that included gemcitabine, between 1995 and 2005. Prognostic baseline factors were selected by their significance in Cox regression analysis. Outlier trial arms were identified by comparing individual 6-month OS and PFS rates against the entire group. The dataset selected for the generation of OS and PFS benchmarks was then tested for intertrial arm variability using a logistic-normal model with the selected baseline prognostic factors as fixed effects and the individual trial arm as a random effect. A total of 1,132 cases from eight trials qualified. Performance status and sex were independently significant for OS, and performance status was prognostic for PFS. Outcomes for one trial (NCCTG-034A) were significantly different from the other trial arms. When this trial was excluded, the remaining trial arms were homogeneous for OS and PFS outcomes after adjusting for performance status and sex. Benchmark values for 6-month OS and PFS are reported along with a method for using these values in future study design and analysis. The benchmark survival values were generated from a dataset that was homogeneous between trials. The benchmarks can be used to enable single-arm phase II trials using a gemcitabine platform, especially under certain circumstances. Such circumstances might be when a randomized control arm is not practically feasible, an early signal of activity of an experimental agent is being explored such as in expansion cohorts of phase I studies, and in patients who are not candidates for combination cytotoxic therapy.
©2014 American Association for Cancer Research.
Figures
Source: PubMed