Sometimes a trial contains several centres that stop participating in the trial after recruiting only a few patients. Since little information is available for measuring effects at such centres, a strategy which is often adopted is to combine such centres into one centre in the analysis. This has most use when a fixed effects model is used, because the centre-treatment interaction can then be assessed more effectively. However, in a mixed model it usually makes little difference whether such centres are combined together or fitted separately.
Was this article helpful?