Agentic AI Meets Data Simulation: A Framework for Self-Optimizing Clinical Trial Digital Twins

About this session

Part of Session 5, Track C: Innovation/AI (Chair: Asif Karbhari, AstraZeneca).

Agentic AI systems observe, reason, and act autonomously toward a goal. What if we applied this framework to clinical trial data quality? Instead of manually configuring simulation parameters and hoping synthetic data satisfies hundreds of edit checks, an agentic system could analyze the rules, generate test data, evaluate violations, diagnose root causes, and adjust parameters iterating until convergence.
We present a working implementation of this concept: a self-teaching simulation engine that reverse-engineers CDISC-CORE and other quality rules into field-level generation constraints, runs optimization loops with learning rate decay and oscillation detection, and produces digital twins that satisfy conformance requirements with measurable convergence. The system maintains a complete audit trail of every parameter change, enabling human-in-the-loop oversight of autonomous decisions.
How can AI assist in data quality planning? By transforming quality rules from passive validators into active generation guides closing the loop between specification, simulation, and data quality.

Speaker

• WL

  Ward Line

  Clinventive

  Speaker