Comparison of analytical methods for cluster randomised trials: an example from a primary care setting

Area of Science:

• Health Services Research
• Primary Care Medicine
• Biostatistics

Background:

• Cluster randomisation is prevalent for evaluating interventions in primary care.
• A study in Scotland randomised 66 general practices to test guidelines and fast-track services for urological conditions.
• Small referral numbers per practice presented analytical challenges.

Purpose of the Study:

• To evaluate different statistical approaches for analysing cluster randomised trials in primary care.
• To investigate the impact of analytical methods on continuous and dichotomous outcomes.
• To highlight potential weaknesses in cluster trial data from primary to secondary care referrals.

Main Methods:

• Applied three analytical approaches: adjusted standard tests, cluster level analysis, and advanced statistical models (random effects, GEE).
• Investigated intervention effects on both continuous and dichotomous outcomes.
• Compared results from different analytical methods, considering cluster size and weighting.

Main Results:

• Conventional tests yielded spuriously low P values due to unaddressed clustering.
• Cluster level analysis of dichotomous outcomes differed significantly when observations were not weighted by cluster size.
• Consistent results were observed for continuous outcomes across methods.

Conclusions:

• Cluster randomised trials in primary care, especially with referral-based recruitment, face challenges like non-contributing clusters and variable cluster sizes.
• Analytical approach significantly impacts conclusions for dichotomous outcomes, emphasizing the need for weighted cluster level analysis.
• Advanced statistical methods and adjusted standard tests provide consistent results for continuous outcomes.