Ping Yu Liu
The Southwest Oncology Group is a national consortium of institutions and physicians with the primary objective of evaluating new cancer treatments. The Statistical Center is actively involved in all stages of the clinical evaluation. This includes protocol design; patient registration; data collection, quality control, and management; study monitoring; result analysis; and report writing. My particular research interests are in the area of design and analysis of clinical trials with survival endpoints.
For phase III comparisons, the frequent use of the standard two-arm randomized clinical trial is due in part to its simplicity of design and interpretation. When more than two arms are included, the number of possible comparisons increases dramatically and the chance of erroneous conclusions multiplies. When faced with several promising new treatments, a pilot study using selection designs can first be employed to reduce the number of treatments compared in the ultimate phase III study. Patients are randomized to the different new treatments in a selection design. Without performing hypothesis testing or controlling for the false positive rate, the treatment with the best outcome observed at the study completion is selected for further testing against a standard treatment in a simple, two-arm phase III study. Sample sizes for the selection designs are chosen so that the probability of making the correct selection is high IF a true superior treatment exists among those tested. These sample sizes are typically much smaller than those of phase III designs and the complication of a K-arm phase III study is avoided. Unfortunately, this approach can lead to wrong impressions if for any reason the definitive phase III study is not carried out and the results of the selection study are treated as conclusive. It is especially dangerous if a control arm is included in the selection stage. In this case the pilot study becomes a phase III study with no protection for false positive findings. Efforts are underway to quantify the false positive rates in this setting.
When there are compelling reasons for a K-arm phase III study, we have addressed the general case where no systematic relationships among the treatments are assumed to exist. The approach is to start with global testing, i.e. test of equality of all arms; pairwise tests follow only if the global test is significant (the Least Significant Difference, or LSD, approach). Sample size tables are established and the probability that the LSD approach reaches the correct conclusions is investigated for a variety of situations.
Sometimes trials are designed with specific relationships hypothesized among the arms. For example, when treatments under testing consist of increasing doses of the same agent or successively added agents to a regimen, designs with order restrictions among the treatment arms are often employed. We originally proposed a class of ordered logrank tests for testing stochastically ordered alternatives. Compared to procedures basedon subjective treatment scoring and isotonic regression, the ordered logrank tests have the advantage of objectivity (i.e. subjective scoring of the samples is not needed), easy adaptability from the standard logrank test, and limiting standard normal distribution.
However, the power of the ordered logrank tests varies substantially with respect to the possible alternative configurations. Investigation of a class of modified ordered logrank tests has recently been completed. The modified tests retain the advantages of the ordered logrank tests but have more robust power properties against possible alternatives. Sample size requirements for the LSD approach are established for the modified logrank tests.
Meta-analysis of phase II cooperative group trials in metastatic stage IV melanoma to determine progression-free and overall survival benchmarks for future phase II trials.. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 26(4):527-34.. 2008.