248: UT Health San Antonio's Dr. Edward Sankary: Why Pilots Can't Be an Excuse to Avoid Strategy, How a "Mundane" AI Agent Hit Near-100% Compliance in Days, and When to Pause or Kill an AI Project

Subscribe on: RSS | SPOTIFY | APPLE PODCAST | GOOGLE | BREAKER | ANCHOR

On this episode of The Digital Patient, Dr. Joshua Liu, Co-founder & CEO of SeamlessMD, and colleague, Alan Sardana, chat with Edward Sankary, MD, Vice President, Chief Health Informatics Officer and Chief Value Officer, UT Health San Antonio, and President, Regional Physician Network ACO, about "Why Pilots Can't Be an Excuse to Avoid Strategy, How a "Mundane" AI Agent Hit Near-100% Compliance in Days, When to Pause or Kill an AI Project, and more..."

Audio:

Guest(s):

• Edward Sankary, MD, Vice President, Chief Health Informatics Officer and Chief Value Officer at UT Health San Antonio
• Joshua Liu, MD (@joshuapliu), Co-founder & CEO at SeamlessMD

Episode 248 - Show Notes:

[00:06:41] Why Ed is graded on value, not just whether the technology works — earlier CHIOs were judged on whether the technology was working. Ed's role is measured on whether it improves healthcare, the revenue cycle, and the health system's MSSP accountable care organization. Technology and AI can help decide which patients need chronic care management and care managers, an expensive resource for an ACO, so the right patients reach them.

[00:08:33] How every technology project starts with a problem and a metric — whether it's inpatient virtual care, AI agents, or EHR applications, each project begins with the problem being solved and the one or more metrics it should improve. Ed says Epic is a "terrific electronic health record, but it ain't cheap," and many healthcare leaders don't realize the value the EHR brings to the revenue cycle, operations, supply chain, and patient engagement. "At the end of the day, you don't have a mission without a margin."

[00:10:40] How AI matches each patient to the right outreach channel — the goal is using AI to determine whether email, text, a MyChart message, a person, or an autonomous agent is the best way to reach each patient, since each carries a different cost. The team tied a customer relationship management platform (Epic's Cheers) to care management and multimodal outreach. Patients like it, no-show rates are falling, and patients can now schedule when they want a care manager to call, so care manager templates are full.

[00:13:07] What today's conversational AI platforms can do — Ed's team is evaluating options, including Epic's own and platforms like Hippocratic, for verbal outreach to patients. He says the technology "is really there": latency has been reduced, conversation quality is excellent, and complex questions can be escalated to another agent or a person. Hyperscalers such as OpenAI and Anthropic are also developing patient-facing agents integrated with the EHR.

[00:16:48] How to get clinicians confident in voice agents for clinical conversations — using colonoscopy prep as an example, agents can confirm patients completed their prep correctly and know when to arrive, addressing a significant no-show rate even for endoscopies. The key is detecting red-flag responses and escalating them to a supervisory agent and ultimately a human, which requires pressure testing the escalation pathway. Ed notes some top platforms have performed as well as, and sometimes better than, human nurses. Humans still need to review conversations at the start, and consistently reaching 99% successful responses is what will make clinicians accept the technology.

[00:18:36] Why relational AI matters for patient conversations — Ed's eight-year-old twins have developed a relationship with Alexa, which talks to them at a level that's right for them. Conversational agents now pick up on who they're speaking to and modify their responses as they listen. "Those agents never get fatigued, they don't get frustrated, they don't get short."

[00:22:01] How inpatient virtual care pays for itself regardless of reimbursement — at UT Health San Antonio's new hospital, virtual nurses handle intake, medication reconciliation, discharge education, and transitions of care. Because the platform is integrated with the EHR, they can pick up earlier signs of sepsis, and AI flags fall risk and line pulling. Patients can press a button on their pillow speaker to bring a virtual nurse into the room, and floor nurses gain a partner. Ed expects preventing nurse turnover, reducing readmissions, and improving quality to pay for the system.

[00:23:59] Why remote monitoring is ripe for expansion in ambulatory and rural care — as technology costs fall and broadband access grows, remote tools can help patients take medications correctly and track weight, diet, and blood pressure at home, where they spend most of their time. Sending the important data back into the EHR, and to clinicians at the appropriate time, can reduce cost of care and utilization. The inpatient platform uses an NVIDIA GPU for edge AI and video language modeling, and falling costs could bring it into the ambulatory space.

[00:26:07] The tension between procedure-based revenue and value-based care — tertiary and quaternary health systems make money doing procedures, while value-based care makes money by not doing them and reducing cost. Ed says "somewhere in there in that upper right quadrant is the right place to be, but you can do both," and technology deployed well delivers value either way by making patients' lives better.

[00:27:32] Why UT Health San Antonio skipped the pilot for virtual nursing — "There's absolutely a place for pilots, but those pilots can't be a way to avoid making these strategic decisions." Virtual care had proven itself elsewhere, and a few rooms would have been just "a small test bed," so the hospital equipped every room. Requirements included deep Epic integration so virtual nurses work from one set of screens. Virtual nurses developed workflows applied across med-surg, cancer, and intensive care units, with floor and virtual nurses rotating between roles in a hybrid model.

[00:30:04] How Epic bedside TV replaced the grease board — working with Epic, the team placed a large TV at the foot wall where patients see the virtual nurse come in and view their care team, upcoming events, and vitals. All the information that used to be on a grease board now lives on that single screen, with a planned overlay of critical information even when patients switch to entertainment.

[00:31:47] Why an enterprise rollout created more ownership than a pilot would — "This wasn't a hobby." Every patient got the same technology regardless of room, so staff took pride in building workflows they knew would affect the whole hospital. Borrowing from Apollo 13, Ed says "failure wasn't an option." He sees pilots as useful for early, evolving technologies or head-to-head vendor comparisons, and says talking with health systems that have already pressure-tested a technology "saves you many times from doing pilots."

[00:35:02] How a "mundane" AI compliance agent was built in days — converting ambulatory clinics to hospital outpatient departments and pursuing Joint Commission certification meant hundreds of standing delegated orders, far too many to build into order sets in time. The team added a drop-down question to every order identifying its source standing delegated order, then used AI to review 100% of orders every day and maintain a dashboard. Staff who didn't follow the process received non-punitive coaching, and compliance reached basically 100%. The project worked because the problem was clearly defined, suited to AI, and fit the existing workflow with just one additional click.

[00:38:53] Why AI meant as temporary scaffolding sometimes becomes permanent — the compliance agent was expected to be switched off once order sets were built. Instead, it became a way to test standing delegated orders before the lengthy Epic order set build and approval process, and some orders remain in the easier-to-change online form. Where AI finds and fixes a workflow error, such as a growing work queue, Ed wants to retire it: "Remove the scaffolding, and you got a strong building."

[00:42:06] How Ed decides whether to scale, pause, or kill an AI project — early AI-drafted responses to patient medical advice requests were long and cost physicians more time, so after watching peers, the team chose not to start; now that the technology saves time, piloting and scaling makes sense. A virtual care platform that doesn't integrate with the EHR, leaving staff in two environments and increasing error risk, may need to be killed. The key is setting a clear metric, such as reducing controllable write-offs, and stopping if an agent isn't hitting the mark. Monitoring and being willing to say something isn't working also helps clinicians accept new technology.

[00:47:32] How UT Health San Antonio approaches AI governance and monitoring — Ed says both the technology and processes are not yet mature. The University of Texas system's eight health-related institutions are evaluating an AI monitoring platform to tier risk, assign owners, and detect performance decay. Risk depends on the lens: a level of service calculation seems low risk to a clinician but high risk to revenue cycle if CMS finds bias. The AI committee includes clinical, operational, legal, risk, and governance members plus experts from the UT College of AI, Cyber and Computing, supported by an intake questionnaire and an inventory of AI tools in use.

[00:52:05] Why clinicians will remain critical as AI evolves — "These technologies aren't gonna take our jobs away. They're gonna only make them better." Ed encourages listeners to stay connected, share how their organizations approach the technology, and talk openly about what doesn't work: "I like talking about our failures 'cause it only makes us better."

Fast 5 Lightning Round:

1. What is your favorite book, or the book you've gifted the most?

The Years of Lyndon Johnson by Robert Caro

2. If you could instantly master any skill, what would it be?

"Throwing a football and playing pool"

3. Would you rather have super strength, super speed, or the ability to read people's minds?

"Ability to read people's minds"

4. What is something in healthcare you believe that others might find insane?

"Medical errors equal two jumbo jets crashing every day, and administrators have grown 4,000% versus 150% for physicians"

5. What is the last movie or TV show you saw, and what did you think of it?

"Lakers Celtics. A terrific series."

The Digital Patient has been recognized as Feedspot's #1 Patient Engagement Podcast of 2025. Thank you to our listeners for making this happen!