Open cohort designs for cluster-randomised trials in institutional settings

Cluster-randomised trials (CRTs) are often conducted in institutions such as care homes, schools and hospitals. These institutions can be defined as 'open cohorts', as individuals join and leave the clusters over time. I propose that open-cohort designs allow recruitment of participants after cluster-randomisation and have the ability to link participants' repeated measurements. This thesis aims to develop methods surrounding open-cohort designs for parallel-group CRTs.

A scoping review was conducted, which suggested care homes, palliative care and prisons could have greater need for open-cohort designs than other institutional settings. Previous definitions of an open-cohort design were inconsistent, and it was often ambiguous whether the population, design or analysis was 'open'.

Identifying CRT designs in the scoping review was challenging. To overcome this, I developed a CRT classification system which allows trialists to examine particular design components and assess whether a design falls into one of six proposed sub-types. Timescales were highlighted as a crucial design aspect.

A simulation study was used to compare established closed-cohort and cross-sectional designs to open-cohort designs over a range of study parameters and complications inevitable in practice. Analysis models, all variations of a mixed effects model, were fixed. Results were highly dependent on the scenario and estimand of interest.

Alternative analysis models were compared in a second simulation study, including a joint model for longitudinal and survival outcomes, a cluster-weighted model, and heteroscedastic models which partition individual variance over time. Results again varied depending on the estimand and complications, but overall the heteroscedastic models exhibited more improvements in terms of bias, precision and convergence.

CRT designs can be complex with many unique features. I therefore consider throughout how to fully specify CRT designs. The thesis concludes with suggestions on how to improve reporting of CRT designs, including specific recommendations on how established frameworks could be updated.