How reliable are the predictions for the outcome of standard therapy in randomized phase III studies

Estimation of expectedness: How reliable are the predictions for the outcome of standard therapy in randomized phase III studies (RP3) in epithelial ovarian cancer (EOC)?

Meeting:
2011 ASCO Annual Meeting

Citation:
J Clin Oncol 29: 2011 (suppl; abstr 5036)

Author(s): V. Castonguay, I. Diaz-Padilla, L. Wang, A. M. Oza; Princess Margaret Hospital, Toronto, ON, Canada; Department of Biostatistics, Princess Margaret Hospital, Toronto, ON, Canada

Abstract:

Background: The reliability and anticipated outcome of the standard control (C) arm is an important parameter in the design of RP3s, to calculate sample size, power and study duration. Changing patterns in treatment of EOC patients or in the composition of the study population can lead to discrepancies between expected and actual outcome of the C arm. We compared the expected versus actual outcome of the C arm in RP3s in EOC. Methods: Superiority design RP3s of EOC published or in abstract form from 1/2000 to 12/2010 were reviewed. Trials where the expected outcome of the primary endpoint for the C arm was not explicitly stated and those where the result was not reported were excluded. A trial C estimate was judged as accurate if the outcome of the C arm was 0.75 to 1.25 times predicted. Results: 50 RP3s in EOC were found from 2000 to 2010, 20 were excluded (4 non-inferiority design, 13 expected outcome not reported, 3 no mature data). Thirteen of the 30 trials had survival as a primary endpoint, and 17 had a progression endpoint. Of trials with survival endpoints, 5/13 accurately predicted outcome and 8/13 underestimated outcome. In 8 trials underestimation was such that the C arm outperformed the hypothesized outcome of the experimental arm (actual/expected (A/E) range 1 to 2.65). For trials with progression endpoints, 10/17 were accurate, 4 underestimated outcome, and 3 overestimated it (A/E range 0.52 to 1.57). Studies with survival endpoints underestimated outcome significantly more than those with a progression endpoint (Wilcoxon two-sample test p = 0.0072).

Conclusions: The estimation of expected outcome of standard arm in randomized studies is frequently inaccurate. Our analysis shows that overall survival has often been underestimated in recent EOC trials. This has implications for the design of future trials and in interpreting previously published data. Because of this variation, some trials might have been underpowered to show the magnitude of benefit they were designed for. Potential for variation of outcome with standard therapy needs to be factored into the statistical design and sample size calculations for RP3s.