Overview
tern.gee provides an interface for generalized estimating equations (GEE) within the tern framework to produce commonly used tables (using rtables and graphs. It builds on the R-package geepack for the actual GEE calculations.
When to use this package
If you would like to use the tern framework for tabulation and graphs, then this package is ideal for your GEE fits. However if you use another reporting framework then it will be better to directly use geepack and perform the tabulation and plots differently.
Main Features
- Fitting of GEE models to continuous longitudinal data collected over several time points (called visits) and optionally treatment arms.
- Tabulation of least square means per visit and treatment arm.
- Tabulation of the covariance matrix estimate.
Installation
tern.gee is available on CRAN and you can install the latest released version with:
install.packages("tern.gee")or you can install the latest development version directly from GitHub by running the following:
if (!require("remotes")) install.packages("remotes")
remotes::install_github("insightsengineering/tern.gee")Note that it is recommended you create and use a GITHUB_PAT if installing from GitHub.
Getting started
You can get started by trying out the example:
library(tern.gee)
fev_data$FEV1_BINARY <- as.integer(fev_data$FEV1 > 30)
fev_counts <- fev_data %>%
dplyr::select(USUBJID, ARMCD) %>%
unique()
gee_fit <- fit_gee(
vars = list(
response = "FEV1_BINARY",
covariates = c("RACE", "SEX"),
arm = "ARMCD",
id = "USUBJID",
visit = "AVISIT"
),
data = fev_data
)
lsmeans_df <- lsmeans(gee_fit, data = fev_data)
basic_table(show_colcounts = TRUE) %>%
split_cols_by("ARMCD", ref_group = "PBO") %>%
summarize_gee_logistic() %>%
build_table(lsmeans_df, alt_counts_df = fev_counts)This specifies a GEE with the FEV1_BINARY outcome and the RACE and SEX covariates for subjects identified by USUBJID and treatment arm ARMCD observed over time points identified by AVISIT in the fev_data data set. By default, logistic regression is used and an unstructured covariance matrix is assumed. The least square means assume equal weights for factor combinations.
For more information on how GEE models and their rtables tables are created see the introduction vignette.
