HMN 2026: How AI-driven chart review accurately identifies potential rare disease trial participants

New research by Cleveland Clinic and Dyania Health demonstrates how a medically trained large language model system can accurately and efficiently screen electronic medical records (EMRs) to identify patients who are eligible for a rare disease clinical trial. Published in The Journal of Cardiac Failure, the study offers real-world evidence that artificial intelligence (AI)-enabled medical chart review can improve the speed, accuracy, and equity of trial enrollment.

The study assessed the performance of an AI system—developed by Dyania Health and deployed at Cleveland Clinic—tasked with pre-screening participants for DepleTTR-CM, a Phase III trial for transthyretin amyloid cardiomyopathy (ATTR-CM), a type of heart failure mostly seen in older adults.

In one week, the system reviewed 1,476 patients and identified 46 as potential matches. Among the findings:

• 29 of the 30 patients identified as trial matches through the AI-system and clinician-in-the-loop review had not been identified through traditional recruitment.
• The AI-system achieved 96.2% accuracy when answering 7,700 trial-specific questions across nine domains.
• Seven patients ultimately enrolled through AI-assisted screening prior to ending enrollment due to meeting site enrollment goal, compared to 10 via traditional screening over 90 days.
• The AI’s justifications for criterion conclusions were found to be 100% accurate and interpretable by physician reviewers.
• The system correctly excluded 198 out of 200 non-eligible patients, achieving a 99% negative predictive value (NPV).

Importantly, the AI-driven process resulted in a more diverse patient population. Of the 30 AI-identified patients, 36.6% were Black, compared to just 7.1% identified through routine screening. Additionally, only 60% of AI-identified patients were previously connected to a heart failure specialist, compared to 92.8% of those found by traditional methods, suggesting that AI can expand access to trials among traditionally underenrolled populations.

“This study shows how medically trained AI can support chart review at scale, transforming what has traditionally been a labor-intensive process,” said Trejeeve Martyn, M.D., lead study investigator and director of Heart Failure Population Health at Cleveland Clinic.

“By rapidly identifying high-quality trial candidates across a large health system, we can increase enrollment efficiency and increase enrollment of patients from different backgrounds and from a broader geographical area.

“We are optimistic that this technology can be used across our health system and are looking at how the platform can help accelerate observational research, disease registries, and evidence-based implementations of approved therapies that are underutilized.”

The AI system used a combination of structured EMR data and natural language processing to analyze complex clinical notes and lab reports. It also provided detailed, auditable justifications for each inclusion or exclusion decision, enabling research coordinators to verify eligibility with confidence.

“Clinical research is often limited by how efficiently and equitably we can match patients to trials,” said Eirini Schlosser, CEO and Co-founder of Dyania Health. “This study provides compelling evidence that AI can help solve that bottleneck—not just by improving workflow efficiency, but by helping surface eligible patients who may otherwise be missed, especially those from historically underrepresented groups.”

The AI model, Synapsis AI, was embedded within Cleveland Clinic’s EMR system and screened data across 25 hospitals and 250 outpatient centers in Ohio, Florida, and Nevada.

Validation by the clinical team remained an essential component of the workflow to ensure safety and accuracy. The real-world implementation of AI in a live clinical trial setting and the performance metrics and diversity findings suggest an opportunity to expand AI-enabled tools more broadly for clinical trial matching, population health registries, and real-time quality reporting.

The content is provided for information purposes only.