Last year I wrote about how Cleveland Clinic is working with (and investing in) an AI company called Dyania Health to help speed up clinical trial recruitment. Now Dyania has announced a partnership with Nvidia and an expansion of the use cases for its Synapsis AI platform.nded to a few of Healthcare Innovation’s questions about the announcement
Jersey City, N.J.-based Dyania said its Synapsis AI reads a patient’s longitudinal record and answers clinical questions with evidence from the chart. The company added that Nvidia’s accelerated computing makes it possible to do that across large patient populations behind the hospital firewall, without transferring patient records outside the health system.
HCI: Last year I interviewed you and Cleveland Clinic’s Lara Jehi, M.D., about using Dyania to improve the clinical trial recruitment process. The results Dr. Jehi was describing sounded impressive. Any further updates on your work with her group? Any uptake by other large health system research arms?
Schlosser: Yes. Since that conversation, the work has progressed from successful pilots to enterprise deployment. Cleveland Clinic deployed Synapsis AI across its enterprise in November 2025 and recently reported that study teams using the platform enrolled, on average, 40% more patients, 47% faster. The collaboration also received a 2026 Gartner Eye on Innovation Award. Those results are meaningful because they reflect progress in getting patients into studies, beyond the speed of reviewing charts.
We have also continued building the published evidence. Research published in the Journal of Cardiac Failureexamined Synapsis AI’s use in screening patients for a transthyretin amyloid cardiomyopathy trial across Cleveland Clinic’s network. The system provided supporting evidence for its eligibility assessments so research coordinators could review the reasoning behind each potential match.
What we are most ecstatic about is the impact and success we have had in clinical care. Now days, in addition to clinical trials and registries, Synapsis AI is primarily scouring medical records to find exact moments when patients need care and should get contacted, routed, referred, or treated per guideline concordance. We realized how incredibly large the gap in U.S. healthcare is and have spent the last year focusing not only on the precise finding of patients but also on how those insights can serve to triage which patients need care more urgently and how that can fit in a limited capacity system.
We have also established relationships with other academic medical centers, though we’re not able to publicly disclose their names quite yet. That said, the Nvidia collaboration is intended to support further expansion across health systems in the U.S. and internationally.
HCI: What are some benefits for Dyania in partnering with Nvidia?
Schlosser:For us, the biggest benefit is being able to bring specialized clinical reasoning to more patients while keeping the work inside the health system. Reading one patient’s history can involve connecting information across hundreds of documents. Applying that process across a large patient population — and updating the analysis as new records arrive — creates a substantial computing challenge and the opportunity for optimization of compute usage based on medical logic.
Dyania brings the proprietary clinical models, physician-led development and implementation expertise. Nvidia provides the accelerated computing and software needed to run those models efficiently at scale. We use NVIDIA NeMo in model development and training, and TensorRT and TensorRT-LLM to support inference, which is the process of running the models against records to generate answers.
The collaboration also creates a path to make deployments more repeatable across institutions, publish joint case studies and evaluate infrastructure for more continuous analysis. The clinical goal is to shorten the distance between information being recorded and that information becoming useful to a researcher or care team.
This is an important part of the AI opportunity in healthcare. A capable general-purpose model is one component; interpreting a complete medical history requires clinical context, evidence tracing and a medically specialized system built for the task. Nvidia allows us to scale that specialized work within the environment where hospitals already govern their patient data.
HCI: How much effort would it involve for new health system customers of Dyania to connect their longitudinal patient data to Nvidia’s computing infrastructure and software?
Schlosser:The first distinction is that the hospital is not sending its records to Nvidia. We deploy Synapsis AI on Nvidia-powered infrastructure within the health system’s own environment. The patient data stays within that environment.
The implementation effort depends on how accessible and consistently organized the longitudinal records are. A health system may have information spread across different EHRs, acquired hospitals and historical databases. The work includes connecting the relevant sources, preserving dates and relationships between records, and ensuring that equivalent clinical information is interpreted consistently.
We handle software deployment and ongoing maintenance, working with the health system’s IT, security and clinical teams. The institution’s role includes approving access, provisioning the environment and helping define and review the clinical use case. Hardware requirements depend on the patient population, the volume of records and how frequently the organization wants to run analyses.
There is real implementation work involved, so I would avoid giving every hospital the same timeline. An organization with accessible longitudinal data and suitable computing capacity has a different starting point from one that needs to consolidate data sources or procure infrastructure. The goal is to establish a foundation that can support additional applications over time, rather than build a separate data integration for every clinical question.
HCI: A press release describes the first work from the collaboration — a Cleveland Clinic study on aortic valve replacement. It quotes Lars G. Svensson, M.D., Ph.D., chief of the Heart, Vascular & Thoracic Institute at Cleveland Clinic, as saying, “Using technology to bring together and analyze information that can help us predict the duration of implanted heart valves is a promising step towards enhanced outcomes and more personalized care for patients.”
That seems like a great example of the potential uses of Synapsis AI. Could you describe a few other potential applications of the technology that could help a health system improve outcomes?
Schlosser: A particularly important application is identifying patients whose records suggest they could benefit from a treatment reassessment or specialist evaluation. The relevant information may be spread across several encounters, with no single note capturing the full picture. Synapsis can bring those details together and surface patients for clinical review, with the supporting evidence.
Another is identifying gaps between a patient’s clinical profile and the care documented in their record. That could help a health system find people whose circumstances warrant closer attention under an evidence-based care pathway. The clinician determines what is appropriate; the technology helps make those patients and the reasons for reviewing their care visible. These are applications we describe in our evidence-guided care pathway work.
There is also tremendous potential in understanding long-term treatment outcomes. The valve study illustrates how information collected during ordinary care can support a more individualized assessment of risk. Similar approaches could help researchers examine which patient characteristics are associated with complications, how treatment outcomes vary and which changes over time warrant closer follow-up.
The common thread is that hospitals already possess a great deal of the information needed to investigate these questions. Making it usable at scale creates the opportunity to learn from more of their experience and bring that learning back to patient care.
HCI: The press release says that the companies are evaluating working on a few things together, include Nvidia Confidential Computing for training and inference, Nvidia Dynamo for distributed inference serving, Multi-Instance GPU for partitioning GPUs across concurrent clinical workloads, and scaling to many Nvidia Blackwell Ultra GPUs as deployments move from batch processing toward near-real-time clinical insight at the point of care. Are all of these things examples of the way Synapsis AI would take advantage of Nvidia’s infrastructure to improve its performance?
Schlosser:Yes, although they address different needs. Some are about processing capacity and responsiveness, some are about using infrastructure efficiently, and one is specifically about adding protection for data and models while they are being processed.
Nvidia Confidential Computing would add hardware-based protection for sensitive data and model information during computation. Keeping records inside the hospital establishes where the work happens; confidential computing adds protection while that work is running.
Nvidia Dynamo could help coordinate inference across multiple GPUs and servers, improving how work is distributed as demand grows. For us, that could support more concurrent analyses and more efficient use of computing resources.
Multi-Instance GPU allows a supported GPU to be partitioned into isolated computing resources. That could help a health system run several appropriately sized workloads concurrently, rather than dedicate an entire GPU to each task.
Scaling across Nvidia Blackwell Ultra GPUs is an avenue we are evaluating for larger workloads and more frequent analysis as deployments move toward near-real-time clinical insight.
These are areas of evaluation and collaborative work, rather than a claim that every capability is already deployed. More computing power does not automatically make a clinical answer more accurate. It can allow us to process more histories, support more simultaneous work and return results sooner. The clinical reasoning, evidence behind an answer and review of its usefulness remain essential.
Ultimately, the goal is to make analysis timely enough to matter. A retrospective study can produce valuable knowledge; being able to revisit a patient’s history as new information arrives could bring that knowledge closer to the moment a care decision is being made.
