Related Experiment Videos
Some fundamental issues with non-inferiority testing in active controlled trials
H M James Hung1, Sue-Jane Wang, Yi Tsong
1Division of Biometrics I, OB/OPaSS/CDER/FDA, Rockville, MD 20857, USA. hung@cder.fda.gov
Statistics in Medicine
|January 10, 2003
Summary
In active controlled non-inferiority trials, preserving the active control effect is key. Standard methods struggle with unknown control effects and estimated margins, unlike preservation tests which offer exact type I error control under strong assumptions.
Area of Science:
- Biostatistics
- Clinical Trial Design
Background:
- Active controlled non-inferiority trials lack placebo arms, complicating objective definition.
- Demonstrating preservation of active control effect is a common, yet complex, goal.
Purpose of the Study:
- To critically evaluate statistical approaches for assessing effect preservation in active controlled non-inferiority trials.
- To identify limitations of classical confidence interval methods and preservation tests.
Main Methods:
- Discussion of fundamental statistical issues in active controlled non-inferiority trials.
- Analysis of type I error control in confidence interval approaches with estimated margins.
- Examination of preservation tests and their underlying assumptions.
Main Results:
- Classical confidence interval approaches fail to achieve exact type I error control due to varying sample sizes and nuisance parameters.
- Preservation tests can achieve exact type I error rates but rely on strong, potentially unverifiable assumptions like constancy.
Conclusions:
- Both confidence interval and preservation test methods face challenges when the constancy condition (active control effect consistency across populations) is violated.
- Careful consideration of statistical assumptions and limitations is crucial for interpreting results from active controlled non-inferiority trials.