Kidney Week 2025 Early Program - Advances in Research Conference 2025: Generative Artificial Intelligence in Kidney Disease Research and Management-Applications of AI in Event Adjudication

Video Transcription

Video Summary

The talk discusses how AI, especially large language models (LLMs), could streamline clinical trial event adjudication—one of the slowest, most expensive, and least transparent trial tasks. Central clinical event committees (CECs) review lengthy site-submitted “dossiers” to decide whether events meet endpoint criteria, but this process is slow, costly, variable across reviewers, and often lacks clear reasoning documentation.  <br /><br />Bhadrushi’s team developed AI adjudication systems trained on thousands of CEC-labeled cases. In validation on the DELIVER heart failure trial, AI agreed with CEC decisions 83% overall, and 96% when the model was confident; using a “human-in-the-loop” approach could reduce manual review by ~84% without changing treatment effect estimates. Similar results were shown for major adverse cardiovascular events, with strongest performance in confident subsets, though numeric lab extraction (e.g., troponin types/values) remained challenging.  <br /><br />AI also extracted 51 structured variables from unstructured dossiers with ~96% accuracy, enabling secondary analyses and prognostic insights. Risks include generalizability, bias (e.g., handwritten/scanned/translated records), privacy, and reduced human oversight; mitigation requires diverse validation, bias checks, privacy safeguards, and regulatory pathways.

Asset Subtitle

Samarra Badrouchi

Meta Tag

Module ARC

Speaker Samarra Badrouchi

Keywords

clinical trial event adjudication

large language models (LLMs)

central clinical event committees (CEC)

human-in-the-loop adjudication

major adverse cardiovascular events (MACE)

structured data extraction from clinical dossiers