From: Adaptive designs in clinical trials: why use them, and how to run and report them
Statistical quantity | Fixed-design RCT property | Issue with adaptive design | Potential solution |
---|---|---|---|
Effect estimate | Unbiased: on average (across many trials) the effect estimate will have the same mean as the true value | Estimated treatment effect using naive methods can be biased, with an incorrect mean value | Use adjusted estimators that eliminate or reduce bias; use simulation to explore the extent of bias |
Confidence interval | Correct coverage: 95% CIs will on average contain the true effect 95% of the time | CIs computed in the traditional way can have incorrect coverage | Use improved CIs that have correct or closer to correct coverage levels; use simulation to explore the actual coverage |
p value | Well-calibrated: the nominal significance level used is equal to the type I error rate actually achieved | p values calculated in the traditional way may not be well-calibrated, i.e. could be conservative or anti-conservative | Use p values that have correct theoretical calibration; use simulation to explore the actual type I error rate of a design |