Using Previous Longitudinal Group-Randomized Rural Weight-Loss Study Data to Design a Prospective Rural Weight-Loss...

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Group-randomized trials (GRTs) present unique challenges due to hierarchical data structures.
• Longitudinal GRTs add further complexity with nested data layers.
• Existing simulation studies often rely on parametric assumptions, limiting generalizability.

Purpose of the Study:

• To compare the performance of a proposed longitudinal mixed-effects model against a standard baseline-adjusted model for GRTs.
• To assess the type I error rate and empirical power of the longitudinal model using data-driven simulations.
• To validate the suitability of the longitudinal model for a prospective study analyzing % weight change.

Main Methods:

• Data-driven simulations were generated using existing study data to inform model assumptions.
• A longitudinal mixed-effects model with three follow-up time points was compared to a baseline-adjusted model.
• Empirical power and type I error rates were calculated for a continuous outcome (% weight change at 24 months) across varying effect sizes.

Main Results:

• Both models demonstrated comparable empirical power across the tested effect sizes.
• The longitudinal model exhibited a type I error rate of 3.09%, while the baseline-adjusted model showed 3.87%.

Conclusions:

• The proposed longitudinal mixed-effects model does not inflate the type I error rate.
• The validated longitudinal model is suitable for use in future group-randomized trials with hierarchical and longitudinal data.