摘要：We group approaches to modeling correlated binary data accordingto data recorded cross-sectionally as opposed to data recorded longitudinally;according to models that are population-averaged as opposed tosubject-speci c; and according to data with time-dependent covariates asopposed to time-independent covariates. Standard logistic regression modelsare appropriate for cross-sectional data. However, for longitudinal data,methods such as generalized estimating equations (GEE) and generalizedmethod of moments (GMM) are commonly used to t population-averagedmodels, while random-e ects models such as generalized linear mixed models(GLMM) are used to t subject-speci c models. Some of these methodsaccount for time-dependence in covariates while others do not. This paperaddressed these approaches with an illustration using a Medicare datasetas it relates to rehospitalization. In particular, we compared results fromstandard logistic models, GEE models, GMM models, and random-e ectsmodels by analyzing a binary outcome for four successive hospitalizations.We found that these procedures address di erently the correlation amongresponses and the feedback from response to covariate. We found marginalGMM logistic regression models to be more appropriate when covariates areclassi ed as time-dependent in comparison to GEE models. We also foundconditional random-intercept models with time-dependent covariates decomposedinto components to be more appropriate when time-dependent covariatesare present in comparison to ordinary random-e ects models. We usedthe SAS procedures GLIMMIX, NLMIXED, IML, GENMOD, and LOGISTICto analyze the illustrative dataset, as well as unique programs writtenusing the R language.
