Kidney Week 2025 Early Program - Advances in Research Conference 2025: Generative Artificial Intelligence in Kidney Disease Research and Management-Generative AI in Clinical Trials: A Driver of Efficiency and Democratization of Care

Video Transcription

Video Summary

Dr. H.A. Blood of Brigham and Women’s Hospital describes work using generative AI to improve efficiency and access in clinical trials, focusing on the costly bottleneck of patient recruitment and screening. Trial enrollment is slow and expensive: many trials miss timelines, recruitment drives a large share of costs, and site screening effort is often wasted because most reviewed charts are ineligible.<br /><br />Blood’s team built an AI tool (“Rectifier”) that uses large language models to read unstructured EHR data (notes, reports) more reliably than older NLP approaches. Key obstacles were limited “context windows” (medical records can exceed what models can ingest) and high per-token costs. They addressed both using retrieval-augmented generation (RAG), which reduces data sent to the model while maintaining performance and dramatically lowering cost and resource use.<br /><br />In a retrospective evaluation embedded in a heart failure trial, Rectifier matched or exceeded study staff accuracy. They then ran a prospective, blinded randomized study (MAPS LLM) comparing AI-assisted vs manual screening. AI significantly increased the speed of identifying eligible patients and improved enrollment (about 84% more enrollments), without worsening equity metrics.<br /><br />In Q&A, Blood advises smaller sites to be cautious about building bespoke systems due to maintenance and generalizability challenges, noting sponsors may subsidize validated tools. He also emphasizes secure, HIPAA-compliant deployments (e.g., enterprise cloud environments) as a prerequisite.

Asset Subtitle

Alexander Blood

Meta Tag

Module ARC

Speaker Alexander Blood

Keywords

generative AI

clinical trial recruitment

patient screening

large language models

electronic health records

retrieval-augmented generation (RAG)

HIPAA-compliant deployment