Assessing intervention effects in a school-based nutrition intervention trial: which analytic model is most powerful?

Area of Science:

• Biostatistics
• Public Health Research
• Epidemiology

Background:

• Group-randomized trials (GRTs) with nested cohort designs are common in health research.
• Accurate statistical analysis is crucial for estimating intervention effects in such complex designs.
• Existing methods may not fully optimize power for dietary outcome studies.

Purpose of the Study:

• To compare four mixed-model analyses for group-randomized trials with nested cohorts.
• To provide intraclass correlation (ICC) estimates for dietary outcome GRTs.
• To offer formulas for calculating the benefits of covariate adjustments on intervention effect standard errors.

Main Methods:

• Comparison of four mixed-model statistical analyses.
• Application of methods to the Teens Eating for Energy and Nutrition at School (TEENS) study data.
• Development of formulas for standard error adjustments using covariates and time correlations.

Main Results:

• Mixed-model analyses of covariance demonstrated the highest statistical power in the analyzed dataset.
• Intraclass correlation (ICC) estimates were provided for dietary outcome GRTs.
• Formulas illustrated the potential reduction in standard error of the intervention effect (sigma(delta)) through covariate adjustment.

Conclusions:

• Mixed-model analyses of covariance are recommended as a powerful approach for GRTs with nested cohorts and dietary outcomes.
• The provided ICC estimates and adjustment formulas can inform the design of future GRTs.
• Researchers can use these methods to estimate a priori detectable differences and sample size requirements for various analytic options.