Gan (2013) reported that approximately 27% of 120 randomized Phase III trials with a primary end- point of OS had statistically significant outcomes. Ratain (2005) suggested that the low Phase III success rate in Oncology may stem from a low positive predictive value (PPV) in Phase II trials. We proposed two rank-based (RB) randomized Phase II designs that differentially weigh the risk of death by the type and time of disease progression and the percentage change in tumor burden. Both RB designs utilized the Wei-Lachin test (Lachin 1992); one (MI-WL) utilized multiple imputation prior to applying the Wei-Lachin test (Mogg and Mehrotra 2007), the other did not (WL). We then compared these designs with one based on progression-free survival (PFS) by simulating 2500 randomized Phase II studies from Phase III trials with known OS outcome with respect to sensitivity, specificity, and PPV.
The MI-WL test had the greatest PPV among the three methods considered. The increase relative to PFS reflected a gain in specificity, whereas the increase relative to the WL method was due to increased sensitivity. The decreased specificity of the WL method may be due to a biased missingness which is addressed by the MI-WL method.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.