Symbolic two-step method compared with single-step methods to model the center-mean outcome in cluster randomized...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Data Analysis

Background:

• Cluster randomized trials (CRTs) often employ single-step analysis methods.
• A novel symbolic two-step method has been developed within the symbolic data analysis framework.
• This method uniquely adjusts for patient-level factors in estimating and testing center-level effects on both average outcomes and their variation.

Purpose of the Study:

• To evaluate the performance of the symbolic two-step method compared to single-step methods in challenging settings.
• To determine when the symbolic two-step method is preferable to single-step analyses for CRTs.
• To explore the utility of modeling within-center outcome variation.

Main Methods:

• A simulation study compared the symbolic two-step method with single-step multilevel linear models (allowing for heterogeneous variances and not).
• Both methods were applied to analyze data from a cluster randomized trial.
• The analysis focused on estimating and testing center-level effects on average outcomes and outcome variation.

Main Results:

• Single-step models showed increased statistical power for center-level factors when patient and center-level factors were uncorrelated.
• The symbolic two-step method demonstrated superior power in the presence of correlation between patient and center-level factors.
• The two-step method identified a factor predicting within-center variance, a capability lacking in single-step methods.

Conclusions:

• Single-step methods are recommended only under strict assumptions of no correlation and no effect on center-outcome variation.
• The symbolic two-step method is recommended when patient and center-level factors are correlated or when analyzing center-outcome variation.
• Identifying factors influencing center-outcome variation can inform practice changes to reduce variability.