Dyania Health’s Synapsis AI reviewed the records of more than 7,800 Cleveland Clinic heart-valve patients, running on the hospital’s own Nvidia servers. The data was used to build a model predicting when replacement valves fail. Doctors still make every decision.

An AI system has read up to 16 years of records for more than 7,800 Cleveland Clinic heart patients, Nvidia said on Friday. It ran entirely on the hospital’s own servers. The software, Synapsis AI, is made by New Jersey startup Dyania Health.

Much of a patient’s history sits in doctors’ notes and scan reports rather than tidy data fields. Synapsis reads those notes and answers clinical questions, linking each answer back to the record it came from. It runs on Nvidia hardware inside the hospital’s network, so records never leave the building, much like a UK NHS triage AI built with Nvidia.

For the study, it answered 88 questions about each patient who had an aortic valve replaced between 2005 and 2021. The resulting data was used to build a model that predicts the risk of a replacement valve failing three, five and ten years after surgery. Nvidia says the chart review would have taken clinicians years by hand.

Lars Svensson, head of Cleveland Clinic’s heart institute, called it “a promising step towards enhanced outcomes and more personalized care for patients.”

Doctors still review the AI’s evidence and make every decision. The study shows the tool can build a dataset and predict risk, not that it improves care, and Dyania says it still needs outside validation. Cleveland Clinic is also an investor in Dyania, it disclosed last year when it rolled the tool out for trial recruitment.

Big tech is pushing hard into hospital records. Google Cloud launched Gemini for healthcare this week, and Bunkerhill raised $55M for hospital AI agents. But proven gains for patients remain harder to find, and doctors have pushed back on bold claims.