Quantifying clustering by GP practice in individually randomised trials in primary care

Problem

In randomised controlled trials, the assumption of independence of individual observations is fundamental to the design, analysis and interpretation of studies. However, in individually randomised trials in primary care, this assumption can be violated because patients are naturally clustered within GP practices. Not only may individuals have similar background and socio-economic characteristics, the effectiveness of some GPs at delivering an intervention may vary and the definition of “usual care” may vary between GPs or practices. Clustering has implications for both the analysis and the design of trials, in particular sample size calculations. Our study aimed to quantify the amount of clustering present in previous individually randomised trials in primary care to provide empirical estimates of clustering that can be used in the design of future studies.

Approach

Clustering can be quantified by intra-cluster correlation (ICC), a measure of the similarity between individuals within a cluster with respect to a particular outcome. We reviewed 20 trials undertaken by the Department of Primary Care at the University of Southampton over the last ten years. We calculated the ICC for the primary and secondary outcomes in each trial at the practice level and determined whether ignoring practice-level clustering still gave valid inferences. We carried out analyses both adjusting for and not adjusting for baseline covariates. Where multiple studies collected the same outcome measure, the median ICC was calculated for the outcome.

Findings

17 trials contributed usable data to the study. The mean cluster size was 20 patients per practice. The median ICC was 0.012 (IQR 0.001, 0.03). The median ICC for symptom severity was 0.02 (IQR 0.01 to 0.07) and for reconsultation with new or worsening symptoms if was 0.01 (IQR 0.00, 0.07). For HADS anxiety the ICC was 0.04 (IQR 0.02, 0.05) and for HADS depression if was 0.02 (IQR 0.00, 0.05). The median ICC for EQ5D-3L was 0.01 (IQR 0.01, 0.04) In 16/17 (94%) trials, accounting for clustering did not change the result of the primary analysis. In the one trial where the result did change, the published analysis had accounted for clustering.

Consequences

There seems to be evidence of clustering at the practice level in individually randomised trials in primary care. The level of clustering found here is in line with that found in a review of cluster randomised trials in primary care by Adams et al. (2004), which gave a median ICC of 0.010 with IQR 0 to 0.03. This may have implications for future sample size calculations, suggesting that a small inflation of sample size to take into account the naturally clustered nature of the data may be necessary in order to ensure adequate power. However, this represents evidence only from trials undertaken within one centre and we aim in future work to extend this to other UK primary care trials.