Abstract:

AIM: To propose a new conditional power (CP) algorithm and apply it to sample size re-estimation (SSR). METHODS: A temporary sample size was estimated based on the interim observed results, and the ratio of the current sample size to this estimated sample size was used as the weight, referred to as the estimated information time weighting (EIW). CP ($ {cp}_{\text{EIW}} $) was calculated using this weighting, and sample size re-estimation is performed. Monte Carlo simulation techniques were used to compare the accuracy of this method with three commonly used CP calculation methods: $ {cp}_{\text{D}} $ method (initial assumed parameter method), $ {cp}_{\text{I}} $ method (interim observation result method), and $ {cp}_{\text{OPW}} $ method (optimal hybrid weight method). Additionally, the statistical performance of sample size re-estimation based on four different CP algorithms was compared. RESULTS: Simulation results for CP accuracy showed that the accuracy of $ {cp}_{\text{EIW}} $ and $ {cp}_{\text{OPW}} $ were close, and both outperform the commonly used $ {cp}_{\text{D}} $ and $ {cp}_{\text{I}} $. For sample size re-estimation based on different CP algorithms, all four methods were able to control type Ⅰ error, with $ {cp}_{\text{EIW}} $ showing the highest conditional power and the best average performance score (APS). CONCLUSION: Among the four CP algorithms, $ {cp}_{\text{EIW}} $ exhibits good accuracy and provides better statistical performance for sample size re-estimation. It is recommended for application.

Key words: clinical trials, adaptive design, conditional power, sample size re-estimation, estimated information time weighting