Comparison of two methods for seamless phase II/III Clinical trials using early outcome for treatment selection

Abstract

Seamless phase II/III trials have been advocated following the US Food and Drug Administration’s Critical Path Initiative lunched in 2004. Statistical methods have been proposed to address the treatment selection (phase II portion) and final test (phase III portion) issues. We compare two popular methods of Shun, Lan, and Soo (2008) and Friede et al. (2011) in terms of the familywise error rate and power as a function of interim analysis information time and the correlation between the short-term surrogate endpoint, on which the treatment selection is based, and the long-term primary endpoint, on which the final test is focused. The numerical study shows that there is generally a slight gain of power by the method of Shun et al. owing to being less conservative, compared to Friede et al. regardless when the interim selection is performed, as long as the two endpoints are not perfectly correlated. For the case of perfectly correlated endpoints, the gain is shown for the scenario when the patients in the unselected treatment arm is no longer followed up, where the gain for Shun et al. is around 1%∼2%.