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

Considerations must be taken when designing group-randomized trials due to the hierarchical structure of the data. Longitudinal group-randomized trials have an added layer of nesting adding more complexity to the study design. Simulation studies have been performed to compare the operating characteristics and validate statistical models for these hierarchical data structures, but many provide simulations from parametric distributions under set assumptions.

Our manuscript aims to use previous study data to compare two statistical analysis methods in group-randomized trial designs through data-driven simulations for a prospective study design. Creating simulated datasets using existing study data from a previous study allows the existing data to drive the assumptions of the models. The motivation for this simulation study was a potential concern that our proposed longitudinal mixed-effects model could have inflated type I error. We compare the empirical power and type I error rate for our proposed model against a baseline adjusted model at a single time point when modeling a continuous outcome, % weight change at 24 months. The longitudinal model includes three follow-up time points, while the other models the outcome with an adjustment for a baseline measure, weight. The empirical power of the models is calculated and compared for varying effect sizes.

Results showed that the models had comparable power for the tested effect sizes and type I error rates of 3.09% and 3.87% for the longitudinal and the baseline adjusted model, respectively.

These results show our proposed longitudinal model does not result in an inflated type I error rate and would be sufficient to use for the future trial.