Cluster randomized trials with a small number of clusters: which analyses should be used?

Cluster randomized trials with a small number of clusters: which analyses should be used?