Generalized Estimating Equations

Generalized Estimating Equations for analyzing correlated/clustered data. Handles repeated measures, longitudinal data, multi-site studies with marginal model approach. Essential for pathology studies with multiple samples per subject (e.g., bilateral organs, multiple biopsies per patient).

Usage

geemodel(
  data,
  outcome,
  predictors,
  cluster_id,
  time_var,
  family = "gaussian",
  corstr = "exchangeable",
  robust_se = TRUE,
  conf_level = 0.95,
  qic = TRUE,
  posthoc = FALSE,
  posthoc_adjust = "holm",
  diagnostics = TRUE,
  correlation_plot = FALSE
)

Arguments

data

the data as a data frame

outcome

outcome variable (dependent variable); can be continuous, binary, or count

predictors

predictor variables (independent variables) for the model

cluster_id

variable identifying clusters (e.g., patient ID, dog ID). Observations with the same ID are treated as correlated.

time_var

for longitudinal data, specifies time ordering within clusters. Used with AR(1) correlation structure.

family

distribution family and link function for the outcome variable. Gaussian for continuous, Binomial for binary, Poisson for counts.

corstr

Working correlation structure within clusters: - Exchangeable: constant correlation between all pairs (recommended starting point) - AR(1): correlation decays with time lag (for longitudinal data) - Unstructured: estimates all pairwise correlations (requires many observations) - Independence: no correlation (equivalent to GLM)

robust_se

sandwich estimator for standard errors (recommended). Provides valid inference even if correlation structure is misspecified.

conf_level

confidence level for confidence intervals (default 95 percent)

qic

Quasi-likelihood under Independence Model Criterion. Lower QIC indicates better model fit. Use for comparing models.

posthoc

compare levels of categorical predictors using estimated marginal means

posthoc_adjust

multiple testing correction for post-hoc comparisons

diagnostics

display model diagnostics including number of clusters, observations per cluster

correlation_plot

visualize the estimated within-cluster correlation structure

Value

A results object containing:

results$instructions					a html
results$modelInfo					a table
results$coefficientsTable					a table
results$qicTable					a table
results$posthocTable					a table
results$diagnosticsTable					a table
results$correlationPlot					an image
results$methodologyNote					a html
results$interpretationNote					a html

Tables can be converted to data frames with asDF or as.data.frame. For example:

results$modelInfo$asDF

as.data.frame(results$modelInfo)

Details

Key Features:

• Handles binary, continuous, and count outcomes

• Multiple correlation structures (exchangeable, AR-1, unstructured)

• Robust standard errors (sandwich estimator)

• Post-hoc pairwise comparisons

• Model selection with QIC

Clinical Applications:

• Multiple samples per patient

• Bilateral organs (eyes, kidneys)

• Repeated measures over time

• Multi-site studies

• Clustered randomized trials

Examples

# \donttest{
# Example: Multiple liver samples per dog
geemodel(
  data = liver_data,
  outcome = 'diagnosis',
  predictors = c('sample_method', 'fibrosis_score'),
  cluster_id = 'dog_id',
  family = 'binomial',
  corstr = 'exchangeable',
  robust_se = TRUE
)
#> Error in geemodel(data = liver_data, outcome = "diagnosis", predictors = c("sample_method",     "fibrosis_score"), cluster_id = "dog_id", family = "binomial",     corstr = "exchangeable", robust_se = TRUE): argument "time_var" is missing, with no default
# }