Skip to contents

Fits Generalized Estimating Equation (GEE) models for analyzing correlated/clustered data. GEE provides population-averaged (marginal) estimates for studies with repeated measures, longitudinal data, multi-site studies, or clustered observations such as multiple samples per subject in pathology studies.

Details

GEE is essential for pathology studies where observations are correlated within clusters:

  • Multiple biopsies from the same patient

  • Bilateral organs (paired kidneys, eyes)

  • Repeated measures over time

  • Multi-site or multi-center studies

  • Hierarchical/nested data structures

Key Advantages of GEE:

  • Provides valid inference even if correlation structure is misspecified (with robust SE)

  • Does not require distributional assumptions for within-cluster correlation

  • Population-averaged interpretation (vs. subject-specific in mixed models)

  • Handles unbalanced designs (different cluster sizes)

Working Correlation Structures:

  • Exchangeable: Constant correlation between all pairs within cluster (most common)

  • AR(1): Autoregressive - correlation decays with time lag (for longitudinal data)

  • Unstructured: Estimates all pairwise correlations (requires many observations)

  • Independence: No correlation (equivalent to GLM)

Model Selection: QIC (Quasi-likelihood Information Criterion) is used for comparing models with different correlation structures or predictor sets. Lower QIC indicates better model fit.

Super classes

jmvcore::Analysis -> ClinicoPath::geemodelBase -> geemodelClass