Pierre
POUPIN

In Cluster Randomized Trials, units of randomization are groups of individuals, also called clusters, rather than individuals themselves. Such designs involve considering correlations between individuals of a same cluster. Trials involving older adults are sometimes carried out in nursing homes (NH) that are therefore the units of randomization. Residents leaving in NH are often frail and a third of them deceased in the first year after their admission. When trials are planned in such setting the problem of discontinuation due to death must be added to the problem of correlation. The problem of the management of discontinuation has 2 levels:
- The choice of the optimal design (« closed cohort », « open cohort », « continuous recruitment with short exposure »), depending on the intervention and on the outcome,
- If a closed-cohort is the selected design, the statistical analysis strategy to estimate the intervention effect without bias.
This PhD aims to contribute to answering those 2 problems.