Synapsis AI, a medically trained, large language model (LLM)–based end-to-end system, reduced the time and effort needed to screen for eligible patients to participate in a randomized, interventional phase III clinical trial in patients with polycythemia vera (PV). Use of Synapsis AI also led to more patients being screened and enrolled in the study than standard methods.
“Once we took the [clinical trial] protocol, fed it into the LLM, and typed in the study, within seconds we received a list of about 30 patients who were eligible for the study that are already in our medical system,” said Aaron T. Gerds, MD, MS, a physician in the Department of Hematology/Oncology at Cleveland Clinic.
The ASCO Post spoke to Dr. Gerds about the LLM and the poster he and co-investigators presented during the 2025 American Society of Hematology (ASH) Annual Meeting and Exposition.1
About Clinical Trial Screenings
Identifying patients who are appropriate candidates for a clinical trial can be a resource- and time-intensive process. With studies of rare disease populations, the burden to find enough patients with that disease to adequately power the study is challenging.
Dr. Gerds, who is also Assistant Professor of Medicine at Case Western Reserve University, Deputy Associate Director for Clinical Research, and a member of the Developmental Therapeutics Program at Case Comprehensive Cancer Center, explained that the study authors calculated that standard methods of chart review can take more than 30 minutes. This is depending on the inclusion/exclusion criteria for the trial and the length of the patient history to be reviewed. Participants are usually only found and recruited during active or routine clinical care, further limiting potential candidates for an upfront treatment trial.
Aaron T. Gerds, MD, MS
Investigators at Cleveland Clinic explored the use of Synapsis AI to automate the identification of patients who were eligible for enrollment in a clinical trial. The system’s ability to accurately identify patients was assessed and a positive predictive value for accuracy was calculated. The volumes of patients identified by Synapsis AI and traditional processes were then compared.
“The stark contrast in the number of patients identified highlights the significant efficiency and time-saving advantage of Synapsis AI’s automated prescreening over conventional screening methods,” Dr. Gerds said. “By streamlining the recruitment process, tools like Synapsis AI can enhance trial efficiency and support faster development of life-saving therapies even in the setting of rare diseases.”
The ongoing, randomized phase III clinical trial is comparing treatment with the class I and class II histone deacetylase inhibitor givinostat vs hydroxyurea in patients with JAK2 V617F-positive high-risk polycythemia vera (GIV-IN-PV; NCT06093672).
Dr. Gerds reported that Synapsis AI conducted a series of progressive rejections across the Cleveland Clinic Health System’s records to find eligible candidates for the trial. First, the system filtered through cancer-related ICD-10 codes to find patients with polycythemia vera who were treated within the past 3 years, and then assessed the patient’s electronic health records against the trial’s inclusion and exclusion criteria using both structured and unstructured data.
Results
Cleveland Clinic’s health system contains 4.7 million active electronic health records, with 28,200 of these being from patients who were diagnosed with cancer within the past 3 years; 904 of these patients were identified as being diagnosed with polycythemia vera.
Full assessment of all of these patients for eligibility criteria was completed by Synapsis AI within 1 week. The system identified 22 patients who fit the inclusion criteria. In total, the system identified 50 patients before the trial was closed.
Research personnel then verified the eligibility of the system’s choices and confirmed that all patients were accurately eligible candidates, for a positive predictive value of 100%.
Comparatively, by traditional methods, clinical staff members prescreened nine patients, enrolled four patients, and treated three patients within 12 months. The team achieved an average enrollment rate of about one patient enrolled every 4 months.
Synapsis AI achieved a sevenfold increase in candidate identification in just a fraction of the time spent by the medical team. This resulted in a substantially reduced workload for the clinical trial team and an ability to accelerate the timeline for the trial.
“If we can enroll more patients more quickly on studies for rare disease, we can really push the ball forward. Instead of taking 5, 10, 15 years to develop a drug, it may take half that amount of time,” Dr. Gerds said. “Once these systems are built out it’s going to make clinical research much more efficient.”
DISCLOSURE: Dr. Gerds reported receiving research funding from Kratos Pharma, Ionis Pharmaceuticals, Imago BioSciences, Constellation Pharmaceuticals, Accurate Pharmaceuticals, and CTI Biopharma. He also reported consultancy, including expert testimony, for AbbVie, Novartis, Disc Medicine, Rain Oncology, Karyopharm, Bristol Myers Squibb, Constellation Pharmaceuticals, Agios, Sierra Oncology, GSK, Kartos Therapeutics, Inc., MorphoSys AG, Pharma Essentia, and Keros.
REFERENCE
1. Gerds A, et al: 2025 ASH Annual Meeting. Abstract 4340. Presented December 7, 2025.
EXPERT POINT OF VIEW: Kenneth L. Kehl, MD, MPH
Artificial intelligence technologies applied to clinical data will likely play an increasingly important role in identifying potential clinical trials for patients with cancer in coming years. Many such AI tools are now in
Kenneth L. Kehl, MD, MPH
development. It will be critical to evaluate approaches to integrating them into clinical research workflows, especially across cancer types and institutions. A frequent challenge in this space is the general lack of a central source of truth for which trials have current slots for which types of patients, given the frequency with which trial cohorts open and close. Often, developing an AI infrastructure that can surface reasonable trial options is the easy part; the hard part is implementing them, connecting them to the electronic health record and to clinical trial databases, and demonstrating measurable impacts on trial accrual.
DISCLOSURE: Dr. Kehl reported no conflicts of interest.
Kenneth L. Kehl, MD, MPH, is on faculty at the Lowe Center for Thoracic Oncology and the Division of Population Sciences at Dana-Farber Cancer Institute, Boston.
