Skip to contents

admr: Aggregate Data Modeling in R

⚠ Deprecated: admr is no longer maintained. Please use the successor package admixr2 instead. Install with: devtools::install_github("LeidenPharmacology/admixr2")

⚠ Go to admixr2 (successor package)

admr (Aggregate Data Modeling in R) is an open-source R package designed to facilitate pharmacometric modeling using summary-level data. It enables users to work with aggregate data, such as mean observations and variance-covariance matrices, to fit pharmacokinetic and pharmacodynamic (PK/PD) models efficiently. This package implements a newly developed Expectation-Maximization (EM) algorithm to enhance computational performance and provides tools for advanced modeling applications.

Overview

The admr package provides a comprehensive framework for aggregate data modeling in pharmacometrics, offering several key advantages:

  • Efficient Parameter Estimation: Uses the Iterative Reweighting Monte Carlo (IRMC) algorithm for robust and fast parameter estimation.
  • Flexible Data Integration: Works with both individual-level and aggregate data, making it ideal for meta-analyses and literature-based modeling.
  • Advanced Modeling Features: Supports complex PK/PD models with various error structures and parameter transformations.
  • Comprehensive Diagnostics: Built-in tools for model assessment, convergence checking, and parameter stability analysis.

Key Features

  • Efficient Model Fitting: Uses iterative reweighted Monte Carlo (IRMC) for robust parameter estimation, improving speed and scalability compared to traditional Monte Carlo methods.

  • Flexible Data Formats: Supports both raw and aggregate data formats, allowing for the integration of summary-level data from diverse sources, including published literature and simulated models.

  • Comprehensive Diagnostics: Built-in tools for model assessment, convergence checking, and parameter stability analysis.

  • Meta-Analysis Support: Facilitates model-based meta-analyses by enabling the combination of summary data across studies.

  • R Integration: Fully compatible with R, leveraging popular pharmacometric modeling libraries like rxode2.

  • Open-Source: Developed for accessibility and ease of use by the pharmacometric community.

Installation

This is an R package. R is required, RStudio is recommended.

You can install the development version of admr from GitHub with:

# install.packages("devtools")
devtools::install_github("/admr")

Quick Start

Below is a complete example of how to use admr to fit a pharmacokinetic model to aggregate data:

# Load required libraries
library(admr)
library(rxode2)
library(nlmixr2)
library(dplyr)
library(tidyr)
library(mnorm)

# Load and prepare the simulated individual-level data
data(examplomycin)
examplomycin_wide <- examplomycin %>%
  filter(EVID != 101) %>%
  dplyr::select(ID, TIME, DV) %>%
  pivot_wider(names_from = TIME, values_from = DV) %>%
  dplyr::select(-c(1))

# Create aggregated data as example
examplomycin_aggregated <- examplomycin_wide %>%
  admr::meancov()

# Define RxODE model
rxModel <- RxODE({
  cp = linCmt(
    cl,           # Clearance
    v1,           # Volume of the central compartment
    v2,           # Volume of the peripheral compartment
    q,            # Inter-compartmental clearance
    ka            # Absorption rate constant
  )
})

# Define prediction function
predder <- function(time, theta_i, dose = 100) {
  n_individuals <- nrow(theta_i)
  if (is.null(n_individuals)) n_individuals <- 1
  
  ev <- eventTable(amount.units="mg", time.units="hours")
  ev$add.dosing(dose = dose, nbr.doses = 1, start.time = 0)
  ev$add.sampling(time)
  
  out <- rxSolve(rxModel, params = theta_i, events = ev, cores = 0)
  cp_matrix <- matrix(out$cp, nrow = n_individuals, ncol = length(time),
                      byrow = TRUE)
  return(cp_matrix)
}

# Create options
opts <- genopts(
  time = c(.1, .25, .5, 1, 2, 3, 5, 8, 12),
  p = list(
    beta = c(cl = 5, v1 = 10, v2 = 30, q = 10, ka = 1),
    Omega = matrix(c(0.09, 0, 0, 0, 0,
                    0, 0.09, 0, 0, 0,
                    0, 0, 0.09, 0, 0,
                    0, 0, 0, 0.09, 0,
                    0, 0, 0, 0, 0.09), nrow = 5, ncol = 5),
    Sigma_prop = 0.04
  ),
  nsim = 2500,
  n = 500,
  fo_appr = FALSE,
  omega_expansion = 1.2,
  f = predder
)

# Fit model to data
result <- fitIRMC(opts = opts, obs = examplomycin_aggregated, chains = 2)
print(result)

Documentation

The package documentation is available at https://vanhasseltlab.github.io/admr/. Key documentation sections include:

Use Cases

The admr package is particularly useful for:

  • Meta-Analysis: Combining data from multiple studies where individual-level data is not available
  • Literature-Based Modeling: Fitting models to published summary statistics
  • Simulation Studies: Evaluating model performance with aggregate data
  • Population PK/PD: Fitting complex models to summary-level data

Getting Help

Citation

If you use admr in your research, please cite it as:

citation("admr")

License

This project is licensed under the GPL-2 License - see the LICENSE file for details.