The effect of cluster randomization on sample size in prevention research

Area of Science:

• Health Services Research
• Clinical Trial Design
• Biostatistics

Background:

• Investigates cluster randomization in primary care intervention trials.
• Examines the cluster effect on preventive maneuver performance among physician groups.
• Discusses the intracluster correlation coefficient's role in sample size determination for cluster-randomized trials.

Purpose of the Study:

• To assess the impact of cluster randomization on sample size requirements in primary care.
• To evaluate the intracluster correlation coefficients (ICCs) for different outcome measures.
• To inform the design of randomized controlled trials (RCTs) with clustered randomization.

Main Methods:

• Cross-sectional study of 46 practices and 106 physicians in Southern Ontario.
• Data collected via questionnaires and chart audits (100 charts/practice).
• Analysis of variance used to calculate ICCs for 'up-to-datedness' and 'inappropriateness' indices for 13 preventive maneuvers.

Main Results:

• Mean 'up-to-datedness' score was 53.5%; mean 'inappropriateness' score was 21.5%.
• ICCs varied significantly by outcome measure, from 0.005 (blood pressure) to 0.66 (chest radiographs for smokers).
• Required sample sizes ranged from 20 to 42 physicians per group, depending on the outcome measure.

Conclusions:

• Cluster randomization and physician-level analysis significantly affect sample size needs.
• Higher ICCs indicate greater physician interdependence within clusters, increasing required sample size.
• Researchers must select outcome measures carefully and adjust sample sizes for cluster-randomized trials where randomization and analysis units differ.