Some problems related to the design and analysis of clinical trials

The evaluation of new cancer treatments requires demonstration of their efficacy in clinical trials. Many nonrandomized clinical series have suggested benefit of a new treatment compared with historical controls or with controls from other institutions that received standard treatment. Often, however, the new treatment is not found to be beneficial when evaluated in a subsequent randomized trial. The interpretation of clinical trials has been confounded by many problems, several of which were described or summarized in the writings of Raymond S. Bush: These include (a) invalid comparison of nonrandomized controls because of patient selection or stage migration, (b) inappropriate endpoints, and in particular the failure to evaluate quality of life and complications of treatment, (c) inadequate power to detect clinically meaningful differences in randomized trials, and (d) multiple comparisons and retrospective analysis used to make inappropriate conclusions from randomized trials. The present paper reviews briefly some of these methodological issues that may lead to inappropriate interpretation of clinical trials, with particular reference to trials of combined modality therapy.