Sensitivity of methods for analyzing continuous outcome from stratified cluster randomized trials - an empirical...

Area of Science:

• Statistics
• Public Health
• Clinical Trials

Background:

• Little attention has been given to the sensitivity of statistical methods in cluster randomized trials (CRTs), particularly for stratified designs with continuous outcomes.
• This study addresses the need for empirical examination of analytical methods in stratified CRTs.

Purpose of the Study:

• To empirically examine the sensitivity of five statistical methods for analyzing continuous outcomes in a stratified cluster randomized trial.
• To compare the efficacy of the Classroom Communication Resource (CCR) intervention versus usual care on peer attitudes towards children who stutter among grade 7 students.

Main Methods:

• Five statistical methods were used: linear regression, mixed-effects model, GEE with exchangeable correlation structure, cluster-level linear regression, and meta-regression.
• The analysis compared methods with and without adjustment for stratification.
• The primary outcome measured was the Stuttering Resource Outcomes Measure.

Main Results:

• The direction of estimated differences was consistent across all methods, except for meta-regression.
• Adjusting for stratification resulted in narrower 95% confidence intervals for the treatment effect.
• All methods yielded similar overall conclusions, with minor variations in effect estimates and confidence interval widths.

Conclusions:

• Statistical method choice and stratification adjustment can influence results in stratified CRTs.
• The findings provide empirical evidence on the performance of different analytical approaches for continuous outcomes in stratified CRTs.
• The study supports the use of stratification adjustment for more precise estimates in such trial designs.