The effect of missing data on design efficiency in repeated cross-sectional multi-period two-arm parallel cluster...

Area of Science:

• Biostatistics
• Clinical Trial Design
• Longitudinal Studies

Background:

• Cluster randomized trials (CRTs) offer design efficiencies but can be impacted by missing data.
• Multi-period longitudinal designs in CRTs introduce risks of intermittent missing observations and participant dropout.
• Understanding the impact of missing data on CRT efficiency is crucial for robust study planning.

Purpose of the Study:

• To evaluate the impact of missing data on the efficiency of multi-period cluster randomized trials.
• To quantify the loss of efficiency due to planned missing data and unplanned dropout.
• To provide guidance on selecting optimal design parameters for longitudinal CRTs.

Main Methods:

• Utilized a multilevel model with a decaying correlation structure to analyze longitudinal CRT data.
• Assessed efficiency by measuring the variance of the treatment effect estimator.
• Modeled dropout using the Weibull survival function and analyzed various weekly measurement schemes.

Main Results:

• Efficiency increases with more subjects per day and more weeks when no data is lost.
• Loss of efficiency from measuring on fewer days is greatest with few subjects and few weeks.
• Dropout significantly reduces efficiency, particularly when it occurs early and with many weeks.

Conclusions:

• Multi-period designs can enhance CRT efficiency, but careful consideration of missing data is essential.
• The study quantifies efficiency losses, aiding researchers in optimizing longitudinal CRT designs.
• An interactive R Shiny app is available to assist in comparing design options and selecting the most efficient approach.