Optimal and efficient crossover designs under different assumptions about the carryover effects.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Experimental Design

Background:

• Crossover designs and repeated measurements designs are crucial for within-subject treatment comparisons in studies.
• These designs involve assigning treatments to subjects across multiple periods, requiring careful consideration of design parameters (subjects, periods, treatments).
• Selecting an appropriate design is complex, with many options available even for fixed design parameters.

Purpose of the Study:

• To compare different crossover designs based on their efficiency in estimating treatment differences.
• To identify designs that remain efficient across multiple statistical models, particularly concerning first-order carryover effects.
• To evaluate designs under two distinct models: one established and one novel.

Main Methods:

• The study focuses on comparing designs for a given set of design parameters (number of subjects, periods, and treatments).
• Efficiency is assessed based on criteria related to the objective of comparing treatments and estimating treatment differences.
• Designs are evaluated under two statistical models that differ in their handling of first-order carryover effects.

Main Results:

• The research compares selected designs for specific design parameters under the chosen statistical models.
• Findings highlight the performance variations of designs across different models, especially concerning carryover effects.
• The study identifies designs that exhibit efficiency under a variety of plausible modeling assumptions.

Conclusions:

• Designs that are efficient under multiple statistical models, including those accounting for carryover effects, are preferable.
• The choice of design significantly impacts the reliability of treatment comparisons, necessitating consideration of model robustness.
• This work provides insights into selecting optimal crossover designs for repeated measurements studies with varying statistical assumptions.