Detection of a treatment effect when not all experimental subjects will respond to treatment

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Methods

Background:

• Non-responders in clinical trials can reduce statistical power by inflating sample variance.
• Standard parametric tests may be less effective when treatment response is inconsistent across subjects.
• Existing statistical methods may not adequately account for the presence of non-responders.

Purpose of the Study:

• To introduce and evaluate a new Fisher's type randomization test designed for situations with potential non-responders.
• To compare the sensitivity of the new randomization test against traditional parametric tests, specifically Student's t-test.
• To demonstrate the applicability of the new test using real-world data from drug abuse and virus titre studies.

Main Methods:

• Development of a Fisher's type randomization test without a direct parametric analogue.
• Evaluation of the new test's performance using Monte Carlo simulations.
• Application of the test to analyze drug abuse data and virus titre data.

Main Results:

• The Fisher's type randomization test showed increased sensitivity to treatment effects compared to Student's t-test in most trial applications.
• The new test is particularly useful when not all subjects are expected to respond to treatment.
• Monte Carlo evaluations confirmed the test's statistical properties.

Conclusions:

• The proposed Fisher's type randomization test is a valuable tool for analyzing data where non-response is a concern.
• This method enhances statistical power and sensitivity in clinical trials with variable subject response.
• The test provides a more robust alternative to standard parametric tests in specific research contexts.