Generalizing the information content for stepped wedge designs: A marginal modeling approach.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Epidemiological Study Design

Background:

• Stepped wedge trials are increasingly used due to staggered implementation needs.
• Complete designs require data collection across all periods, which can be burdensome.
• Incomplete stepped wedge designs offer flexibility by reducing data collection demands.

Purpose of the Study:

• To extend the information content metric for discrete outcomes in stepped wedge trials.
• To analyze the impact of omitting data elements (cells, sequences, periods) in incomplete designs.
• To evaluate the preservation of information content patterns under various conditions.

Main Methods:

• Developed information content expressions for discrete outcomes under marginal models with general link and variance functions.
• Investigated the impact of omitting data elements on information content.
• Utilized numerical studies to assess patterns under canonical and variance-stabilizing link functions.

Main Results:

• Information content patterns for cells approximate centrosymmetry for discrete outcomes, especially with variance-stabilizing link functions.
• Patterns for sequences and periods are more sensitive to underlying secular trends.
• Omission of data elements can lead to deviations from centrosymmetry, particularly for sequences and periods.

Conclusions:

• Incomplete stepped wedge designs are viable for discrete outcomes, but data omission impacts information content patterns.
• The choice of link function influences the preservation of centrosymmetric patterns.
• Researchers must consider the sensitivity of sequence and period information content to secular trends when designing incomplete trials.