Behavioral carry-over effect and power consideration in crossover trials.

Area of Science:

• Clinical Trial Design
• Biostatistics
• Epidemiology

Background:

• Crossover trials offer efficiency but are vulnerable to carry-over effects.
• Washout periods may be insufficient or unethical, especially in comparative effectiveness research with behavioral carry-over.
• Potential outcomes framework is used to analyze crossover designs with and without carry-over effects.

Purpose of the Study:

• To investigate the impact of carry-over effects on crossover trial estimators.
• To develop and evaluate methods for covariate adjustment in crossover trials.
• To determine conditions where crossover designs maintain power and avoid type I error inflation.

Main Methods:

• Utilized a potential outcomes framework to model crossover trials.
• Analyzed the impact of carry-over effects on treatment effect estimation and statistical power.
• Developed and applied covariate adjustment techniques.
• Evaluated methods using data from the MTN-034/REACH study.

Main Results:

• When a carry-over effect exists and meets a sign condition, the basic estimator underestimates the true treatment effect.
• Carry-over effects do not inflate type I error for one-sided tests but reduce statistical power.
• Conditions were derived for crossover designs to maintain power and avoid type I error inflation compared to parallel-group designs.
• Covariate adjustment methods were shown to be effective.

Conclusions:

• Crossover designs require careful consideration of carry-over effects, particularly behavioral ones.
• The developed methods and covariate adjustments enhance the reliability of crossover trials.
• The findings provide guidance for optimizing crossover trial design in various research settings.