From Adaptive Design to Modern Protocol Design for Drug Development: Part II. Success Probabilities and Effect Estimates
- PDF / 9,736,888 Bytes
- 10 Pages / 612 x 792 pts (letter) Page_size
- 75 Downloads / 163 Views
333
From Adaptive Design to Modern Protocol Design for Drug Development: Part 11. Success Probabilities and Effect Estimates for Phase 3 Development Programs
Frank l r e t z Hannover Medical School, Hannover. Germany Sre-Jane '"g Johns Hopkins University, Baltimore, Maryland
In this article we investigate success probabilities and efect estimates for several standard phase 3 programs consisting of two pivotal studies, each with placebo and either one or two doses. We investigate the probability of success for the entire phase 3 program as well as the resulting bias and mean square error of the effect estimates. We also investigate an alternative development stratea, where the first of the two
Key Words Bias; Clinical trial simulation; Power; Treatment selection
Correspondence Address Frank Bretz, PhD. Institute for Biometry, Hannover Medical School, CarlNeuberg-Str. I , 30625 Hannover, Germany (email: bretz.frank@ mh-hannover.de). The professional views presented are those ofthe authors and not necessarily those oftheir affiliated organizations.
INTRODUCTION According to international guidelines, demonstrating evidence of efficacy for a medical product considered for market release generally requires at least two adequate and well-controlled, so-called pivotal, trials in a confirmatory phase 3 development program, where each pivotal trial is convincing on its own (1,2). Using frequentist methods, the success probability of a single study is closely related to the power of detecting a treatment effect. While power is easily defined in the context of univariate hypothesis testing, there are different ways of defining power in trials with multiple objectives (3). In general the choice of power definition is driven by choosing power measures tailored to the study objectives. On a program level, however, the power accounts for clinical evidence of the same treatment in both confirmatory trials, which will impact the program's overall success probability. On the other hand, estimates of treatment effects based on the pivotal studies may be included in a drug label after gaining market access. A good understanding of their properties is essential. Bauer et al. (4) were among the first to investigate thoroughly selection, reporting, and admission bias under a variety of scenarios Dnq Lnfomtion Journal. Vd. 44.pp. 3 3 3 - 3 4 2 , 2 0 1 0 -
pivotal trials uses an adaptive trial design with treatment selection at interim. The dose selection at the interim analysis of that adaptive trial triggers a second confirmatory, nonadaptive trial comparing the selected dose@)with placebo and othenvisefollowing the same protocd as the adaptive trial. After describing key considerations, we report the main results of extensive simulation studies.
for both adaptive and nonadaptive trials. From their findings one concludes that selection and bias are inherently tied together in a complex way that depends on a multitude of different factors. Motivated by the two development paradigms described in part I of this article (this issue),
Data Loading...
