摘要:The application of linear mixed models or generalized linear mixedmodels to large databases in which the level 2 units (hospitals) have a widevariety of characteristics is a problem frequently encountered in studies ofmedical quality. Accurate estimation of model parameters and standarderrors requires accounting for the grouping of outcomes within hospitals.Including the hospitals as random eect in the model is a common methodof doing so. However in a large, diverse population, the required assump-tions are not satised, which can lead to inconsistent and biased parameterestimates. One solution is to use cluster analysis with clustering variablesdistinct from the model covariates to group the hospitals into smaller, morehomogeneous groups. The analysis can then be carried out within thesegroups. We illustrate this analysis using an example of a study of hemoglobinA1c control among diabetic patients in a national database of United StatesDepartment of Veterans' Aairs (VA) hospitals.
关键词:Cluster analysis; logistic regression; random eects; SAS; NLMIXED.