Research Project
6622262
The randomized clinical trial (RCT) is arguably the linchpin of the drug development process, and the results of an RCT are almost always analyzed using some form of statistical hypothesis test. The most frequently used hypothesis tests assume a population model for statistical inference, even though a randomization model is more consistent with real-world characteristics of RCTs. Approximate p- values returned by population-model tests can, under certain circumstances, be misleading, resulting in effective drugs being declared ineffective, or ineffective drugs being declared effective. To support analysis of RCTs using the appropriate randomization model, sophisticated software for conducting randomization-based permutation tests is needed. Ongoing advances in computing technology have created a favorable climate for widespread use of such software. The goal of this research is to develop flexible and robust software for carrying out randomization-based permutation tests for single- or multi- clinic RCTs. A subset of this functionality has been successfully implemented in a Phase I pre-prototype ("RTAnalyzer"). Phase II seeks to build a full-scale prototype capable of handling a wide variety of trial designs, including designs using adaptive randomization. The Phase II project includes collaborations with two experts in the field of permutation testing: Dr. William Rosenberger and Dr. Bonnie LaFleur. PROPOSED COMMERCIAL APPLICATION: Software that can use general permutation tests to analyze clinical trial data would have clear commercial value to clinical research organizations in academia, Government, and the pharmaceutical, biotechnology, and medical device industries. Key applications are the analysis of clinical trials with unusual randomization schemes, trials with unusual patterns of treatment response, and trials where standard distributional assumptions are invalid.
