A scoping review identified additional considerations for defining estimands in cluster randomised trials.

Abstract

Objective: An estimand is a clear description of the treatment effect a study aims to quantify. The ICH E9(R1) addendum lists five attributes that should be described when defining an estimand. However, the addendum was primarily developed for individually randomised trials. Cluster randomised trials (CRTs), in which groups of individuals are randomised, have additional considerations for defining estimands, such as the population of clusters and how individuals and clusters are weighted. We aimed to identify a list of additional items that may need to be considered when defining estimands in CRTs.

Study design and setting: We conducted a systematic search of multiple databases as well as the authors' personal libraries to identify articles that described an aspect of an estimand definition for CRTs which was not explicitly covered by one of the five attributes listed in the ICH E9 (R1) addendum. From this, we generated a list of items that may require consideration when defining estimands for CRTs beyond the five attributes listed in the ICH E9(R1) addendum.

Results: From 46 eligible articles, we identified 8 items that may need to be considered when defining estimands in CRTs: (i) population of clusters; (ii) population of individuals under selection bias; (iii) exposure time of individuals/clusters on treatment; (iv) how individuals and clusters are weighted (e.g. individual-average vs. cluster-average); (v) whether summary measures are marginal or cluster-specific; (vi) strategies used to handle cluster-level intercurrent events; (vii) how interference/spillover is handled; and (viii) how individuals who leave or change clusters are handled.

Conclusion: This review has identified additional items that may need to be considered when defining estimands for CRTs. Study investigators undertaking CRTs should consider these items when defining estimands for their trials, to ensure estimands are unambiguous and relevant for end-users such as clinicians, patients, and policy makers.