Cluster sampling in hospital surveillance

Surveillance of important events in hospitals often naturally leads to sampling of cluster of individuals. It is essential that estimators of the proportions of interest and their associated standard errors be consistent with the sampling techniques employed. Assuming that a simple random sample of individuals was completed, when in reality clusters of individuals were sampled, can result in misleading conclusions. Three appropriate estimators are considered and their standard errors compared. If the larger the size of a cluster, such as a ward, floor or hospital, the smaller the number of subjects afflicted with the characteristic, then the unbiased estimator P should be employed. If the larger clusters have more subjects with the characteristic of interest, then either PR or Pppz will provide correct and relatively precise estimates.