Conceptual issues in the analysis of cost data within cluster randomized trials.

Cluster randomized controlled trials (RCTs) are increasingly used in economic evaluations of social, educational and health care interventions. Methodological research has, therefore, been spread across several disciplines, with the result that it has taken many years for guidelines on good statistical practice in the design and analysis of such trials to become easily accessible to health service researchers. These guidelines remain incomplete, however, because they do not take account of issues specific to the analysis of cost data. In particular, they fail to recognize that the calculation of confidence intervals around costs needed to inform health care priority setting raises unique methodological issues. If poorly designed trials are to be avoided in future (including those by the authors), then collaboration between triallists and health economists is required. This paper sets out a framework that should facilitate such collaboration and draws attention to problems that must be addressed quickly in the design of cluster-based economic evaluations.