Skip to contents

admiral

ADaM in R Asset Library

Purpose

To provide an open source, modularized toolbox that enables the pharmaceutical programming community to develop ADaM datasets in R.

Installation

The package is available from CRAN and can be installed with:

install.packages("admiral")

To install the development version of the package from GitHub run:

# install.packages("devtools")
devtools::install_github("pharmaverse/admiral")

Cheat Sheet

Release Schedule

The {admiral} family has several downstream and upstream dependencies and so releases are done in two Phases:

Release Schedule Phase 1- Date and Packages Phase 2- Date and Packages
Q4-2024 December 2nd December 9th
{pharmaversesdtm} {admiralonco}
{admiraldev} {admiralophtha}
{admiral} {admiralvaccine}
Q2-2025 June 2nd June 10th
{pharmaversesdtm} {admiralonco}
{admiraldev} {admiralophtha}
{admiral} {admiralvaccine}

Main Goal

Provide users with an open source, modularized toolbox with which to create ADaM datasets in R. As opposed to a “run 1 line and an ADaM appears” black-box solution or an attempt to automate ADaM.

One of the key aspects of admiral is its development by the users for the users. It gives an entry point for all to collaborate, co-create and contribute to a harmonized approach of developing ADaMs in R across the pharmaceutical industry.

Scope

To set expectations: It is not our target that admiral will ever provide all possible solutions for all ADaM datasets outside of study specific needs. It depends on the user’s collaboration and contribution to help grow over time to an asset library that is robust, easy to use and has an across-industry focus. We do not see a coverage of 100% of all ADaM derivations as ever achievable—ADaM is endless.

We will provide:

  • A toolbox of re-usable functions and utilities to create ADaM datasets using R scripts in a modular manner (an “opinionated” design strategy).
  • Pharmaceutical communities and companies are encouraged to contribute to admiral following the provided programming strategy and modular approach
  • Functions that are comprehensively documented and tested, including example calls—these are all listed in the Reference section.
  • Vignettes on how to create ADSL, BDS and OCCDS datasets, including example scripts.
  • Vignettes for ADaM dataset specific functionality (i.e. dictionary coding, date imputation, SMQs …).

The {admiral} Family of Packages

There are three types of packages in the admiral family:

  • Core package—one package containing all core functions required to create ADaMs, usable by any company (i.e. general derivations, utility functions and checks for ADSL, OCCDS and BDS).
  • TA (Therapeutic Area) package extensions—one package per TA with functions that are specific to algorithms and requirements for that particular TA (e.g. {admiralonco}).
  • Company package extensions—specific needs and plug-ins for the company, such as access to metadata (e.g. {admiralroche} or {admiralgsk}).

Related data packages include:

Both these packages are developed by the admiral team, but can used across the pharmaverse as common, open-source test SDTM or ADaM data.

The following packages are also useful when working with ADaM datasets:

  • {metacore} and {metatools}—these enable users to manipulate and work with dataset metadata.
  • {xportr}—this provides functionality to get xpt files ready for transport.

Admiral Manifesto

For admiral and all extension packages, we prioritize providing our users with a simple to adopt toolkit that enables them to produce readable and easily constructible ADaM programs. The following explains our philosophy, which we try to adhere to across the admiral family of packages. There isn’t always a clear single, straightforward rule, but there are guiding principles we adhere to for admiral. This manifesto helps show the considerations of our developers when making decisions.

We have four design principles to achieve the main goal:

Usability

All admiral functions should be easy to use.

  • Documentation is an absolute priority. Each function reference page should cover the purpose, descriptions of each argument with permitted values, the expected input and output, with clear real-life examples—so that users don’t need to dig through code to find answers.
  • Vignettes that complement the functional documentation to help users see how best the functions can be applied to achieve ADaM requirements.
  • Functions should be written and structured in a way that users are able to read, re-use or extend them for study specific purposes if needed (see Readability below).

Simplicity

All admiral functions have a clear purpose.

  • We try not to ever design single functions that could achieve numerous very different derivations. For example if you as a user pick up a function with >10 different arguments then chances are it is going to be difficult to understand if this function could be applied for your specific need. The intention is that arguments/parameters can influence how the output of a function is calculated, but not change the purpose of the function.

  • We try to combine similar tasks and algorithms into one function where applicable to reduce the amount of repetitive functions with similar algorithms and to group together similar functionality to increase usability (e.g. one study day calculation rather than a function per variable).

  • We strive to design functions that are not too general and trying to fulfill multiple, complex purposes.

  • Functions should not allow expressions as arguments that are used as code snippets in function calls.

  • We recommend to avoid copy and paste of complex computational algorithms or repetitive code like checks and advise to wrap them into a function. However we would also like to avoid multi-layered functional nesting, so this needs to be considered carefully to keep the nesting of 3-4 functions an exception rather than the rule.

Findability

All admiral functions are easily findable.

  • In a growing code base, across a family of packages, we make every effort to make our functions easily findable.
  • We use consistent naming conventions across all our functions, and provide vignettes and ADaM templates that help users to get started and build familiarity. Each admiral family package website is searchable.
  • We avoid repetitive functions that will do similar tasks (as explained above with study day example).
  • Each package extension is kept focused on the specific scope, e.g. features that are relevant across multiple extension packages will be moved to the core admiral package.

Readability

All admiral functions follow the Programming Strategy that all our developers and contributors must follow, so that all our code has a high degree of consistency and readability.

  • We encourage use of tidyverse (e.g. dplyr) over similar functionality existing in base R.
  • For sections of code that perform the actual derivations (e.g. besides assertions or basic utilities), we try to limit nesting of too many dependencies or functions.
  • Modularity is a focus—we don’t try to achieve too many steps in one.
  • All code has to be well commented.
  • We recognize that a user or a Health Authority reviewer may have the wish to delve into the code base (especially given this open source setting), or users may need to extend/adapt the code for their study specific needs. We therefore want any module to be understandable to all, not only the admiral developers.

References and Documentation

Pharmaverse Blog

If you are interested in R and Clinical Reporting, then visit the pharmaverse blog. This contains regular, bite-sized posts showcasing how admiral and other packages in the pharmaverse can be used to realize the vision of full end-to-end Clinical Reporting in R.

We are also always looking for keen admiral users to publish their own blog posts about how they use the package. If this could be you, feel free make an issue in the GitHub repo and get started!

Recent Conference Presentations

For a full collection of admiral conference presentations over the years, please travel to our Presentation Archive.

Contact

We use the following for support and communications between user and developer community:

  • Slack—for informal discussions, Q&A and building our user community. If you don’t have access, use this link to join the pharmaverse Slack workspace.
  • GitHub Issues—for direct feedback, enhancement requests or raising bugs.

Acknowledgments

Along with the authors and contributors, thanks to the following people for their work on the package:

Jaxon Abercrombie, Mahdi About, Teckla Akinyi, James Black, Claudia Carlucci, Asha Chakma, Bill Denney, Kamila Duniec, Alice Ehmann, Romain Francois, Ania Golab, Alana Harris, Declan Hodges, Anthony Howard, Shimeng Huang, Samia Kabi, James Kim, John Kirkpatrick, Leena Khatri, Robin Koeger, Konstantina Koukourikou, Pavan Kumar, Pooja Kumari, Shan Lee, Wenyi Liu, Iain McCay, Jack McGavigan, Jordanna Morrish, Syed Mubasheer, Thomas Neitmann, Yohann Omnes, Barbara O’Reilly, Hamza Rahal, Nick Ramirez, Tom Ratford, Sukalpo Saha, Tamara Senior, Sophie Shapcott, Ondrej Slama, Andrew Smith, Daniil Stefonishin, Vignesh Thanikachalam, Michael Thorpe, Annie Yang, Ojesh Upadhyay and Franciszek Walkowiak.