Example Liver Pathology Data for GEE Analysis

Simulated dataset of liver biopsies from dogs with multiple samples per subject. This dataset is designed to demonstrate Generalized Estimating Equations (GEE) analysis for correlated/clustered data common in pathology studies.

Usage

gee_liver_data

Format

A data frame with 135 observations and 10 variables:

dog_id

Factor. Unique identifier for each dog (cluster ID)

sample_number

Integer. Sequential sample number for each dog (1-4)

time_point

Integer. Time sequence for longitudinal analysis

age

Numeric. Dog's age in years

breed

Factor. Dog breed (Labrador, German Shepherd, Poodle, Mixed)

sample_method

Factor. Biopsy method (Fine Needle, Core Biopsy, Surgical)

pathologist_experience

Numeric. Years of pathologist experience (5, 10, 15, 20)

fibrosis_score

Numeric. Liver fibrosis score (0-6 scale, continuous)

diagnosis

Factor. Binary diagnosis (Negative, Positive)

cell_count

Integer. Inflammatory cell count (Poisson distributed)

Source

Simulated data generated using data-raw/gee_example_data.R

Details

Study Design:

• 50 dogs with 2-4 liver samples each (unbalanced design)

• Samples from the same dog are correlated

• Represents typical veterinary pathology study with multiple biopsies

Clinical Context: In veterinary pathology, multiple liver samples are often taken from the same animal to assess disease distribution and severity. These samples are not independent, requiring GEE or mixed models for proper statistical analysis.

Correlation Structure:

• Within-dog correlation: Samples from the same dog share baseline characteristics

• Exchangeable structure: Recommended for this data (samples equally correlated)

• AR(1) structure: Can be used if time_point represents sequential sampling

Example Analyses:

1. Binary Outcome (Diagnosis):

   
   # Predicting diagnosis from fibrosis score and age
   geemodel(
     data = gee_liver_data,
     outcome = 'diagnosis',
     predictors = c('fibrosis_score', 'age', 'sample_method'),
     cluster_id = 'dog_id',
     family = 'binomial',
     corstr = 'exchangeable',
     robust_se = TRUE
   )

2. Count Outcome (Cell Count):

   
   # Modeling inflammatory cell count
   geemodel(
     data = gee_liver_data,
     outcome = 'cell_count',
     predictors = c('fibrosis_score', 'breed'),
     cluster_id = 'dog_id',
     family = 'poisson',
     corstr = 'exchangeable',
     robust_se = TRUE
   )

3. Continuous Outcome (Fibrosis Score):

   
   # Predicting fibrosis from age and sample method
   geemodel(
     data = gee_liver_data,
     outcome = 'fibrosis_score',
     predictors = c('age', 'sample_method', 'pathologist_experience'),
     cluster_id = 'dog_id',
     family = 'gaussian',
     corstr = 'exchangeable',
     robust_se = TRUE
   )

Why GEE is Needed:

• Standard regression assumes independent observations

• Multiple samples per dog violate independence assumption

• GEE accounts for within-dog correlation

• Provides valid standard errors and inference

Data Generation: Data was simulated with realistic correlations:

• Dogs with higher age tend to have higher fibrosis scores

• Samples from the same dog share baseline disease severity

• Diagnosis probability increases with fibrosis score and age

• Sample method affects diagnostic accuracy

See also

• geemodel() for GEE analysis function

• geepack::geeglm() for the underlying GEE implementation

Examples

# \donttest{
# Load data
data(gee_liver_data, package = "ClinicoPath")

# View structure
str(gee_liver_data)
#> 'data.frame':	125 obs. of  10 variables:
#>  $ dog_id                : Factor w/ 50 levels "Dog_001","Dog_002",..: 1 1 2 2 3 3 4 4 5 5 ...
#>  $ sample_number         : int  1 2 1 2 1 2 1 2 1 2 ...
#>  $ time_point            : int  1 2 1 2 1 2 1 2 1 2 ...
#>  $ age                   : num  8.7 8.7 3.6 3.6 0.7 0.7 5.2 5.2 7 7 ...
#>  $ breed                 : Factor w/ 4 levels "German Shepherd",..: 2 2 3 3 2 2 3 3 2 2 ...
#>  $ sample_method         : Factor w/ 3 levels "Core Biopsy",..: 1 1 1 2 1 2 1 3 2 1 ...
#>  $ pathologist_experience: num  20 15 15 5 10 5 15 10 10 15 ...
#>  $ fibrosis_score        : num  3.6 3.9 3.5 2.3 0.1 0.9 0.2 0.4 1 0 ...
#>  $ diagnosis             : Factor w/ 2 levels "Negative","Positive": 1 1 2 1 2 1 1 1 1 1 ...
#>  $ cell_count            : int  8 15 12 6 2 3 7 5 4 4 ...

# Check cluster sizes
table(table(gee_liver_data$dog_id))
#> 
#>  2  3  4 
#> 30 15  5 

# Summary statistics by diagnosis
aggregate(fibrosis_score ~ diagnosis, data = gee_liver_data,
          FUN = function(x) c(mean = mean(x, na.rm = TRUE),
                              sd = sd(x, na.rm = TRUE)))
#>   diagnosis fibrosis_score.mean fibrosis_score.sd
#> 1  Negative            1.535211          1.065444
#> 2  Positive            2.255769          1.112745
# }