AI Screening Platform Accelerates Trial Recruitment in Polycythemia Vera

Recruiting patients for clinical trials remains one of the most persistent challenges in oncology research. However, new data from Cleveland Clinic suggest that AI-driven screening may dramatically change that landscape.

In a study presented at the 2025 American Society of Hematology (ASH) Annual Meeting, investigators reported that the Dyania Health’s Synapsis™ AI platform, an artificial intelligence tool, identified seven times more eligible patients for a polycythemia vera trial than standard workflows, while achieving 100% positive predictive value following research-staff verification. This medically trained, large language model (LLM)-based end-to-end system was embedded within the electronic medical record (EMR).

Traditional trial recruitment relies heavily on manual chart review by clinicians and research nurses, a resource-intensive process that can take more than 30 minutes per EMR depending on the complexity of eligibility criteria and the volume of historical data. For rare diseases in particular, this approach contributes to chronically slow enrollment and extended timelines for therapeutic development. When it comes to trial recruitment for rare diseases, eligible participants are difficult to locate and often receive care outside major academic centers.

Recognizing these hurdles, a research team led by Aaron Gerds, MD, MS, Deputy Director for Clinical Research at Cleveland Clinic’s Cancer Institute, initiated the current study to evaluate the efficacy and accuracy of the platform. Their findings highlight the potential of LLM-enabled screening to overcome longstanding rare disease recruitment obstacles.

Research design and findings

The analysis focused on an ongoing, randomized, phase 3 study comparing givinostat versus hydroxyurea in polycythemia vera (NCT06093672). The researchers configured the AI platform to replicate the traditional prescreening workflow, beginning with broad population filtering and narrowing progressively to patients who met detailed protocol-defined eligibility criteria.

“The AI system leverages structured data and LLM-based interpretation of free-text clinical notes,” explain Dr. Gerds and his colleagues. “After filtering oncology-related ICD-10 codes from the prior three years, patients with a diagnosis of polycythemia vera were isolated by the LLM component. The system then assessed each patient’s eligibility against the trial’s seven eligibility and 20 exclusion criteria using both structured and unstructured data elements.”

Of 4.7 million EMRs active in the Cleveland Clinic database, 28,200 patients with an oncology diagnosis within the past three years were identified. From this group, the system then isolated those with a diagnosis of polycythemia vera, narrowing the population to 904 patients.

“The AI tool completed full eligibility assessments on these 904 patients within one week, against the trial’s criteria and identified 22 eligible patients,” Dr. Gerds and his colleagues reported. “Subsequent reviews by Cleveland Clinic’s research personnel verified the data, confirming 100% accuracy of the system’s assessments against the protocol’s criteria.”

In contrast, the traditional approach prescreened nine patients, enrolled four and treated three over a 12-month period. The average enrollment using this method was approximately one patient every four months. “The AI approach demonstrated a seven-fold increase in eligible patient identification within a fraction of the time, substantially reducing workload and accelerating timelines,” according to the study authors.

Implications and next steps

This comparison between AI-assisted and traditional prescreening highlights the transformative potential of medically trained LLMs in clinical research. AI demonstrated the ability to significantly accelerate and expand patient identification while maintaining accuracy.

By automating the most labor-intensive aspects of chart review, the system reduces the time and effort required of clinicians and research personnel. These efficiencies are especially valuable in rare diseases, where eligible patients are dispersed across diverse care settings and manual recruitment often falls short. As Dr. Gerds and his colleagues note, AI can “support faster development of life-saving therapies even in the setting of rare diseases.”

For clinicians and researchers, the implications are far-reaching. “Clinical trials are the standard of care in cancer medicine, but are often hampered by slow enrollment, especially in rare or less common diseases,” says Dr. Gerds. “Anything we can do to speed that process up will help patients. The faster we can enroll patients, the faster we get answers, and the faster we advance therapies. A technology like this, if developed and widely used, can shave years off the time it takes to develop new therapies and that equates to saving lives.”

Looking ahead, Cleveland Clinic and Dyania Health have announced a formal collaboration to integrate the Synapsis AI platform across the health system’s clinical research enterprise. This follows the success of multiple pilot programs and joint validation efforts. The collaboration will support broader adoption of LLM-driven prescreening across therapeutic areas, with the goal of enhancing trial efficiency and expanding access for patients across the system.

Dr. Gerds offered a final reflection on the future of AI in medicine: “AI is here; it’s inevitable, so let’s use it in the best way possible. If you’re standing still and not moving forward, you’re actually moving backward. It’s not a replacement for what we do,” he concludes. “It’s another tool that can help us do a better job and ultimately redirect our time and efforts toward higher-level strategy and idea shaping and toward spending more time with patients.”

Disclaimer: In addition to this collaboration, Cleveland Clinic has invested in Dyania Health and may benefit financially from the sale of this technology.