Fulfillment of the uncertainty principle in cancer clinical trials.

6003 Background: For randomized controlled trials to be ethical, the relative benefits of the study treatments must be uncertain. Therefore, the prior probability that either treatment is actually better should not be too high. Concerns about violation of this uncertainty constraint adversely affect accrual and other aspects of trial conduct. METHODS: To estimate the prior probability that an adult cancer trial will favor the experimental or control treatment, we studied trials conducted by two cooperative groups that closed between 1981 and 1995. We abstracted design features from original protocols. We coded study results using statisticians' reports, published articles, or abstracts. The unit of analysis was the randomization. Using protocol-specified primary endpoints, we categorised each randomization as favoring control, showing no significant difference, or favoring the experimental treatment. We also calculated an effect size for each randomization. RESULTS: 107 randomizations were eligible. Results were available for 103 (96%), from 93 trials. Of these, three (3%) favored the control treatment, 70 (68%) found no statistically significant difference between the treatments, and 30 (29%) favored the experimental treatment. The mean effect size was 1.19 (95% confidence interval 1.12-1.27), reflecting an advantage for the experimental treatment. CONCLUSIONS: We observed a measurable average improvement in disease control associated with assignment to the experimental rather than the control arm of an adult cancer cooperative-group trial. Nevertheless, the small magnitude of the advantage and the heterogeneity of the results suggest that as a group these trials satisfy the uncertainty principle. No significant financial relationships to disclose.

Duke Scholars

Author Stephen L. George Biostatistics & Bioinformatics, Division of Biostatistics