A Cleveland Clinic study published this month shows that AI screening found 29 out of 30 trial-eligible patients that traditional recruitment had completely missed. The system also enrolled patients faster and from more diverse backgrounds than conventional methods.
What the Study Found
Researchers at Cleveland Clinic and Dyania Health tested an AI system called Synapsis AI on a real clinical problem: finding patients eligible for DepleTTR-CM, a Phase 3 trial for transthyretin amyloid cardiomyopathy (ATTR-CM). This rare form of heart failure primarily affects older adults, making it a challenging recruitment target.
The AI screened 1,476 patient records in one week across 25 hospitals and 250 outpatient centers in Ohio, Florida, and Nevada. It identified 46 potential matches by analyzing both structured EMR data and complex clinical notes.
The numbers tell the story:
- 96.2% accuracy on 7,700 trial-specific questions across 9 eligibility domains
- 99% negative predictive value (correctly excluded 198 of 200 ineligible patients)
- 100% accuracy in justifying its eligibility decisions
- Seven patients enrolled through AI screening in 6 days vs. ten through traditional methods over 90 days
Most striking: 29 of the 30 patients ultimately identified had not been found through traditional recruitment at all.
The Diversity Factor
The AI’s reach extended beyond what human screeners typically cover. Among AI-identified patients, 36.6% were Black compared to just 7.1% found through routine screening. Additionally, 60% of AI-identified patients lacked prior connections to heart failure specialists, compared to 92.8% of traditionally-found candidates who already had specialist relationships.
This suggests AI screening could help address a persistent problem in clinical research: trials often fail to represent the populations most affected by diseases.
“This research offers real-world evidence that AI-enabled medical chart review can improve the speed, accuracy and equity of trial enrollment,” said Dr. Trejeeve Martyn, director of Heart Failure Population Health at Cleveland Clinic and lead investigator on the study.
What This Means
Clinical trial recruitment remains one of the biggest bottlenecks in drug development. Many trials fail or run years behind schedule simply because they cannot find enough eligible patients. Rare diseases compound this problem - patients exist but remain scattered across health systems, their eligibility buried in unstructured clinical notes that human reviewers lack time to analyze.
The Cleveland Clinic study, published in The Journal of Cardiac Failure, provides evidence that AI can address this at scale. The system did not replace clinical judgment - validation by the clinical team remained part of the workflow - but it surfaced candidates that would otherwise have remained invisible.
The Fine Print
This was a single-trial study at one health system. The AI system is proprietary, developed by Dyania Health, so independent replication is not possible. Performance on other disease areas or different EMR systems remains untested.
The study also does not address what happens after screening. Finding eligible patients faster is valuable, but enrollment depends on many factors AI cannot influence: patient willingness to participate, logistics of trial visits, insurance coverage, and trust in the research process.
Still, the core finding holds: an AI system found dozens of patients that extensive human screening had missed, did it in days rather than months, and surfaced a more diverse candidate pool. For rare disease research, where every eligible patient matters, that is a meaningful advance.