R Submission Pilot 6

Pilot 6 - AI-Assisted ADaM and TLF Generation

Goal: Leverage AI/LLM tools to generate a more comprehensive set of ADaM datasets and Tables, Listings, and Figures (TLFs) based on the original CDISC pilot study, expanding the scope of analysis artifacts available for future regulatory submissions.

Key evaluation aspects:

  • For the Development Team
    • Use AI coding assistants (e.g., GitHub Copilot, CodeZen, Keiji AI TrialMind) to generate R programs for ADaM dataset construction and TLF production
    • Evaluate the effectiveness and efficiency of AI-assisted programming within the clinical submission context
    • Ensure all generated programs and outputs follow CDISC standards and open-source principles
  • For the Broader Community
    • Provide a publicly available, expanded submission package to support future Pilot submissions
    • Demonstrate open and transparent use of AI tools in clinical data programming

Data and analysis scope:

  • Builds upon the original CDISC Pilot 1 dataset (5 datasets, 3 tables, 1 figure) to produce a broader collection of ADaM datasets and TLFs
  • Uses CDISC Pilot 1 SDTM data as the reference source
  • Not intended as a formal submission to the FDA; serves as an exploratory and preparatory effort for future pilots

Links:

R Submission Pilot 6 Development Repository

Key team members:

Developer team:

  • Ben Straub (GSK)
  • Brandon Theodorou (Keiji AI)
  • Camilla Calmasini
  • Dmitry Kolosov (Parexel)
  • Eli Miller (Atorus)
  • Jeff Dickinson (Navitas)
  • Phani Tata (Syneos)
  • Robert Devine
  • Sam Parmar (Pfizer)
  • Steven Haesendonckx (Johnson & Johnson Companies)