Global tests for combination drug studies in factorial trials

Area of Science:

• Biostatistics
• Pharmacometrics
• Clinical Trial Design

Background:

• Existing statistical tests by Hung, Chi, and Lipicky assess if drug combinations are more effective than individual components.
• These tests traditionally require normally distributed data and complete factorial designs.
• Many real-world outcomes, like disease response rates, follow a binomial distribution, and studies may have missing data.

Purpose of the Study:

• To extend the applicability of Hung, Chi, and Lipicky's alpha-level tests to non-normally distributed data (e.g., binomial outcomes).
• To adapt these tests for incomplete factorial designs where certain dose combinations are not studied.
• To investigate the impact of excluding study cells on the statistical power of these extended tests.

Main Methods:

• The study mathematically extends existing alpha-level tests to accommodate outcome variables with variance dependent on the mean.
• It adapts the methodology for scenarios involving incomplete factorial designs.
• Power performance is evaluated through simulations or theoretical analysis considering the exclusion of specific experimental cells.

Main Results:

• The proposed extensions allow for the analysis of drug combination efficacy with binomial or other non-normal data.
• The methodology is validated for incomplete factorial designs, increasing its practical utility.
• Excluding cells from study demonstrably impacts the power of these statistical tests, with the extent of the impact quantified.

Conclusions:

• The enhanced statistical tests provide a robust framework for evaluating drug combination therapies with common data types and designs.
• Researchers must consider the consequences of incomplete data when interpreting results from these tests.
• This work improves the statistical toolkit for pharmacometricians and clinical trial designers.