Design and analysis issues for crossover designs in phase I clinical studies

Area of Science:

• Clinical pharmacology
• Biostatistics
• Experimental design

Background:

• Assessing new drug efficacy requires efficient and safe clinical trial designs.
• Traditional designs like parallel and crossover studies have limitations regarding safety and data analysis.
• Latin square designs, while efficient, may not meet essential safety criteria.

Purpose of the Study:

• To compare the efficiency of various clinical trial designs, including parallel, two-period crossover, three-period crossover, and Latin square designs.
• To evaluate the suitability of these designs for early-phase drug efficacy assessment while ensuring subject safety.
• To address analytical challenges in crossover designs, such as nonconstant variances and carryover effects.

Main Methods:

• Employed a mixed analysis of variance model to compare variances of estimators across different designs.
• Investigated generalized Box-Cox transformations to handle issues like nonconstant variances within the mixed model.
• Utilized simulation studies to assess the sensitivity of the analysis to first-order carryover effects.

Main Results:

• Proposed three-period crossover designs demonstrated slightly lower efficiency than Latin square designs.
• Latin square designs were found to be incapable of satisfying necessary safety conditions for subjects.
• Analysis revealed that results from models ignoring carryover effects are only reliable when carryover is minimal.

Conclusions:

• Three-period crossover designs present a practical and safe alternative for early-phase drug efficacy studies.
• Generalized Box-Cox transformations can effectively address data analysis challenges in mixed models for crossover designs.
• The presence of carryover effects significantly impacts the reliability of study results, necessitating careful consideration in the analysis phase.