Development of a consensus extension of the estimands framework for cluster randomised trials (CRT-estimands):...

Estimands are increasingly used in randomised trials to clarify research objectives. The ICH E9(R1) addendum sets out five attributes necessary to describe a well-defined estimand. However, the addendum was primarily developed for individually randomised trials. There is growing recognition that estimand descriptions for cluster randomised trials, where groups of individuals are randomised, may require specification of additional considerations. We conducted a Delphi study to assess stakeholder views on additional items for inclusion in a consensus extension of the ICH E9(R1) for cluster randomised trials.

We invited experts in estimands and cluster randomised trials to participate in a modified Delphi process to identify critical items for describing estimands in cluster randomised trials. The research team generated an initial list of eight items and definitions. Across three Delphi rounds, panellists scored items, suggested additional items, and provided open-ended rationales for responses. The consensus threshold was set as ≥70% of respondents rating an attribute as "essential" (i.e., score of ≥7 on a 9-point Likert scale) and <15% of respondents rating the item as "not important" (i.e., a score of ≤3).

Seventy-three (52%) invited individuals participated in Round 1. Response rates were 85% in Round 2 and 95% in Round 3. Panellists included largely statisticians (62, 85%) and clinical trialists (18, 25%). After Round 1, one additional item was added for Round 2 inclusion. After Round 3, five items met consensus criteria: how individuals and clusters are weighted, population of clusters, exposure time of clusters and individuals to the intervention, whether treatment effects are marginal or cluster-specific, and handling of cluster-level intercurrent events.

This Delphi identified expert consensus around the importance of several key items for defining estimands in cluster randomised trials. These results can inform the development of consensus guidance outlining the set of attributes to describe when defining estimands for cluster randomised trials.