Type I error in sample size re-estimations based on observed treatment difference.

摘要: Sample size re-estimation based on an observed difference can ensure adequate power and potentially save a large amount of time resources in clinical trials. One the concerns for such approach is that it may inflate type I error. However, possible inflation has not been mathematically quantified. In this paper mathematical mechanism explored two-sample normal tests. A (conditional) error function data derived. This only provides quantification but also gives mechanisms due to sample re-estimation. Theoretically, their decision rules (certain upper lower bounds), people calculate exactly visualize changes Computer simulations are performed results. If there no bounds adjustment, evident. proper adjusting used, be well controlled. some cases even reduced. The trade-off give up ‘unrealistic power’. We investigated several scenarios which change different. Our show similar results apply other distributions. Copyright © 2001 John Wiley & Sons, Ltd.