Why Virtua Health's AI Committee Gives IT The Review, Not The Vote
By John Oncea, Chief Editor, Clinical Tech Leader
At most health systems, the department that reviews the security and validity of a new AI tool is also the department that decides whether to fund it. At Virtua Health, those are deliberately two different jobs. “When we talk about our project governance, IT has no vote,” said Laura Patrone, who manages imaging informatics for cardiology and neurology at the South Jersey health system. “It’s actually a voting committee, and it’s made up of senior operational leaders.”
That separation – technical vetting in one lane, resourcing decisions in another – is the backbone of how Virtua has scaled from roughly four major technology projects a year to four a quarter, largely driven by the volume of AI tools now moving through its imaging and EHR systems.
The Approval Sequence
Every new AI tool or technology project at Virtua moves through the same gates regardless of which department requests it. First, an IT security review checks the data handling and the vendor itself: Virtua works with very few vendors based outside the US, and when it does, Patrone said, “we put rigors in place so that our data stays within the US.” From there, it moves to the privacy and safety committee, then, for new products or research tools, a products committee. Finally, it reaches the governance committee, where the project has to clear a bar unrelated to security at all: measurable outcomes.
“Nothing gets approved without a key performance indicator,” Patrone said, “unless it’s a technology being enabled for a patient safety risk or something like an upgrade of a software we already have.” Projects typically need two to three KPIs to move forward. On a vascular measurement mapping project her team is building in Epic, one KPI was concrete: an additional 10 studies a week once manual documentation was removed from technologists’ workload, phased as five more appointments a week by six months and 10 more by a year.
Only after a project clears the IT, privacy, and KPI gates does it reach the voting committee, senior operational leaders who rank competing projects against each other by need, scope, and, Patrone acknowledged bluntly, revenue: “Is it going to be an enterprise-wide solution or is it narrowed to a specific department? And also, I hate to say it, but is it revenue generating?” IT doesn’t sit in that vote. It builds the technical case; operations decides what gets resourced first.
What’s Actually Live
The governance structure isn’t theoretical; it’s running against a real portfolio of imaging AI at various stages of maturity. ICAD, which flags breast cancer and breast abnormalities, has been in place for more than a decade and is well studied. Newer tools are still being phased in: a partnership with Rapid flags suspected pulmonary embolism on chest CT scans, alerting radiologists, pulmonologists, and ED physicians to a condition where speed of treatment matters. A parallel AI process already runs for stroke, reading head CTs to flag whether a stroke is present and whether it’s hemorrhagic or clot-based, a distinction that determines treatment. Every clinical AI tool that touches patient data, Patrone noted, has to carry FDA approval before it’s deployed.
On the EHR side, Virtua has piloted ambient listening during physician appointments, generating draft notes from conversation that physicians review and sign off on, and an Epic tool that drafts procedure notes for invasive cardiologists directly from the OR’s procedure log – the record of every event, medication, and supply used during a case – with source references attached to each data point Epic pulls in. Virtua’s chief nursing informatics officer, Marianne Everett, piloted DAX with Microsoft to bring dictation tools to nursing documentation specifically.
The Pushback Was Real, And The Response Was Structural
Physician resistance to the earliest AI tools wasn’t minor. “They did not think that AI could sound like a doctor, that it would know how to look at the data, that it would be unreliable,” Patrone said. Virtua’s answer wasn’t a mandate; it was pilot groups, run specifically to surface workflow problems before a tool reached the full organization, backed by physician informaticists who kept one foot in clinical practice: a primary care doctor, an ED physician, a pediatric emergency room physician, all serving in informatics leadership roles.
That combination of a slow, staged rollout and physician-led governance appears to be what shifted the room. “Now they’re asking for the AI to make them more efficient at their jobs,” Patrone said, “and to also allow them to really focus on the patient rather than being stuck in documentation.”
A Working Relationship With The EHR Vendor
Governance at Virtua also extends into how the health system pushes back on its own EHR vendor. On a vascular reporting project involving carotid artery stenosis measurements, Virtua’s lead vascular sonographer questioned where Epic’s AI was sourcing its data points, since Virtua uses a different, separately validated methodology. Epic’s response wasn’t to override the site; it asked to meet with Virtua’s team to understand the discrepancy. Virtua also participates in Epic’s development immersion trips, weighing in on where EHR functionality should go next before it ships broadly.
The Misconception That Still Comes Up
Asked what misconception she corrects most often, Patrone didn’t point to a technical misunderstanding; she pointed to a fear. “It doesn’t replace people,” she said. “I think people are scared they’re going to lose their jobs. That’s been a big fear. I think it just enhances people’s ability to do their jobs. A bot can’t do what I do. AI helps me because it can do some of these mundane tasks for you. So it can make you more efficient at your job.”
For health systems earlier in their own AI rollout, Virtua’s structure suggests the governance model matters as much as the tools themselves: separating technical vetting from funding authority, requiring measurable outcomes before a project moves forward, and building physician trust through staged pilots rather than mandates.