Abstract
With Phase III failure rates of 50%, better ways of predicting late-stage success are needed. One concept that has been used is “assurance.” Rather than conventional power calculations hypothesizing a known effect of a drug, assurance provides an expected power calculation based on some prior distribution for the treatment effect. It therefore has appeal in Phase III planning and decision making, especially when the prior is based on Phase II data. However, assurance has counterintuitive properties that can serve to confuse and concern the nonstatistician. Appreciation of these properties is helpful to ensure an informed use of assurance in strategic drug development.
Notes
a Phase III: 508 events to provide 90% power to test the hypothesis θTRUE = 0.75 at the 0.025 one-sided α-level.
b Phase II: Observed HR =0.75 on 70 events.