Sample size can be a key design feature that not only affects the probability of a trial＇s success but also determines the duration and feasibility of a trial. If an investigational drug is expected to be effective a...Sample size can be a key design feature that not only affects the probability of a trial＇s success but also determines the duration and feasibility of a trial. If an investigational drug is expected to be effective and address unmet medical needs of an orphan disease, where the accrual period may require many years with a large sample size to detect a minimal clinically relevant treatment effect, a minimum sample size may be set to maintain nominal power. In limited situations such as this, there may be a need for flexibility in the initial and final sample sizes; thus, it is useful to consider the utility of adaptive sample size designs that use sample size re-estimation or group sequential design. In this paper, we propose a new adaptive performance measure to consider the utility of an adaptive sample size design in a trial simulation. Considering that previously proposed sample size re-estimation methods do not take into account errors in estimation based on interim results, we propose Bayesian sample size re-estimation criteria that take into account prior information on treatment effect, and then, we assess its operating characteristics in a simulation study. We also present a review example of sample size re-estimation mainly based on published paper and review report in Pharmaceuticals and Medical Devices Agency （PMDA）.展开更多
Efron(1997)considered several approximations of p-values for simultaneous hypothesis testing.An extension of his approaches is considered here to approximate various probabilities of correlated events.Compared with mu...Efron(1997)considered several approximations of p-values for simultaneous hypothesis testing.An extension of his approaches is considered here to approximate various probabilities of correlated events.Compared with multiple-integrations,our proposed method,the parallelogram formulas,based on a one-dimensional integral,not only substantially reduces the computational complexity but also maintains good accuracy.Applications of the proposed method to genetic association studies and group sequential analysis are investigated in detail.Numerical results including real data analysis and simulation studies demonstrate that the proposed method performs well.展开更多
文摘Sample size can be a key design feature that not only affects the probability of a trial＇s success but also determines the duration and feasibility of a trial. If an investigational drug is expected to be effective and address unmet medical needs of an orphan disease, where the accrual period may require many years with a large sample size to detect a minimal clinically relevant treatment effect, a minimum sample size may be set to maintain nominal power. In limited situations such as this, there may be a need for flexibility in the initial and final sample sizes; thus, it is useful to consider the utility of adaptive sample size designs that use sample size re-estimation or group sequential design. In this paper, we propose a new adaptive performance measure to consider the utility of an adaptive sample size design in a trial simulation. Considering that previously proposed sample size re-estimation methods do not take into account errors in estimation based on interim results, we propose Bayesian sample size re-estimation criteria that take into account prior information on treatment effect, and then, we assess its operating characteristics in a simulation study. We also present a review example of sample size re-estimation mainly based on published paper and review report in Pharmaceuticals and Medical Devices Agency （PMDA）.
基金supported by the Intramural Program of NIHsupported in part by National Natural Science Foundation of China(Grant No.10901155)supportedin part by NIH(Grant No.EY014478).
文摘Efron(1997)considered several approximations of p-values for simultaneous hypothesis testing.An extension of his approaches is considered here to approximate various probabilities of correlated events.Compared with multiple-integrations,our proposed method,the parallelogram formulas,based on a one-dimensional integral,not only substantially reduces the computational complexity but also maintains good accuracy.Applications of the proposed method to genetic association studies and group sequential analysis are investigated in detail.Numerical results including real data analysis and simulation studies demonstrate that the proposed method performs well.
