This article is adapted from a webinar hosted on August 5, 2026, featuring Dr. Jacob Strand, Chair of Palliative Care at Mayo Clinic in Rochester, in conversation with Bayesian Health CEO Suchi Saria. Watch the full conversation here.

Quotes have been lightly edited for length and clarity.

Palliative care has a math problem.

The patient population is getting older, sicker, and more complex. Palliative care programs are now near-universal in large hospitals and the norm even in smaller ones. Demand keeps climbing, but fellowship training has not kept pace, and the specialty is heading into a sustained workforce shortage.

At Mayo Clinic in Rochester, that tension is concrete. The palliative care team handles roughly 14,000 to 15,000 inpatient encounters a year with three physicians and five nurse practitioners and PAs, plus interdisciplinary staff on service on a day to day basis. At some hospitals in the Mayo Clinic Health System, the entire palliative care presence is two clinicians.

For most of its history, the program worked the way nearly every palliative care program in the country works: it waited for referrals.

"We were reliant on referrals from clinicians who knew about palliative care, or perhaps recognized the need for palliative care, to identify patients for us to see," said Dr. Jacob Strand, who leads the program in Rochester and previously served as enterprise chair across Mayo's sites.

The problem with a referral-based model is that it depends on the right clinician noticing the right patient at the right time. Dr. Strand recalls a budget meeting in 2017 where an administrative leader asked a simple question: how many patients in our area should you be seeing in the hospital every year?

"I just sat there, completely not sure what to say. I had to be honest that we don't have a way to know that."

Why the usual fixes didn't work

Before partnering with Bayesian, Mayo tried the approaches most health systems try.

Manual scoring collapsed when the grant funding did. The team piloted a validated heart failure risk score, but it required a person to review charts, calculate scores, and route them to someone who could act. That person was a research assistant funded by a grant. "When that research grant ended, that person went away and our pilot ended," Dr. Strand said. "It's not just about finding the person-power. It's how do you make it sustainable and scalable beyond one specific condition."

Disease-specific triggers missed the patient. Published triggers exist for individual conditions, but patients are not a disease. "Most of them are focused on a disease, not a patient," Dr. Strand said. A trigger built for one malignancy or one stage of heart failure says nothing about a patient's function, psychosocial context, or caregiver needs.

Mortality prediction answered the wrong question. Mayo had access to commercially available end-of-life prediction models within its EHR. The team pushed back hard on using them. "Whether we're predicting six-month mortality or 30-day mortality, we have missed a significant window to provide high-impact interventions for patients and caregivers," Dr. Strand said.

There was an ethical dimension too. "One of the concerns that came up around mortality-based triggers was, what does it look like to disclose that to patients? Do I tell the patient that an algorithm said you have a high risk of death in the next six months? That's a profoundly different conversation. What I worry is that those conversations don't happen."

And the models were opaque. "A lot of these scales are often black boxes. As a clinician, when you see a patient surfaced up, it's like, well, what about this patient made this happen? There's a significant risk of breeding mistrust."

Why Mayo partnered instead of building

Mayo has one of the strongest internal data science organizations in healthcare. The team built and validated its own models. Then the clinical trial ended, and the data science and IT resources moved to the next study.

"Systems are often good at pilots. They're good at studies. But translating that into sustainable practice over time is a challenge," Dr. Strand said. "I was writing extension grants every six to twelve months just to maintain, not improve, not customize, but maintain what we had."

Every workflow change, like moving from hospital location to clinical service, meant a new IT request and a slow-moving queue. What Mayo needed was not another model. It was a real-time clinical intelligence platform that could sustain the program, adapt as clinicians asked for changes, and scale across conditions and sites.

The integration itself took roughly 100 hours of Mayo IT time, with the platform running against Mayo's real-time Epic infrastructure within six to ten weeks.

What the program looks like now

Bayesian Health’s platform continuously reads the full patient record, drawing on nearly 300 indicators across the comprehensive chart, to identify patients with unmet palliative care needs early in the hospital course. Not patients likely to die. Patients likely to benefit.

"I can open up my electronic health record, look at the inpatient census here in Rochester, and see all of the patients who are flagging with high scores for unmet palliative care needs," Dr. Strand said. "It gives me a window within the hospital. It's like a heat map."

The workflow runs through the EHR, not a separate system. A palliative care trained nurse reviews flagged patients each morning. For every patient, the platform surfaces the why: contributing conditions, medications, advance care planning documentation, and the context the team chose to prioritize. "It prevents that clinician from having to go deep into the chart, scrolling through note after note."

If the nurse agrees the patient should be seen, one click sends a secure chat to the primary service with a pended consult order and the same transparent context. The primary team can agree with one click, ask for clarification, or decline. Two clinicians stay in the loop on every referral, and both can see exactly why the patient was flagged. That structure was deliberate. Interruptive alerts that fire without explanation get dismissed on reflex, and a referral a clinician cannot explain to a patient never gets made.

That design decision paid off in adoption. The conversion rate from signal sent to consult placed rose from roughly 10% to north of 40%. "That is because you're able to provide that clinician with the autonomy to engage with their patient, to decrease their cognitive load, and to feel like they have agency and understanding of where this is coming from," Dr. Strand said.

The results

These results land where clinical impact and system economics meet. In a hospital where readmitted patients once consumed 200 beds and crowded out surgical volumes, the program has delivered:

• A five-day reduction in patient length of stay for patients reached by the program. "We're seeing a five-day reduction in patient length of stay, which is really incredible," Dr. Strand said.

• A 25 to 30% reduction in readmission rates, on top of the length-of-stay gains.

• A 50% efficiency gain in the nurse review workflow, part of what Dr. Strand described as incredible efficiency gains across the staffing team.

The benefits extend beyond the measurable. Dr. Strand pointed to qualitative gains as well: reduced cognitive burden for individual clinicians, a team that feels supported and excited about the work, and a level of visibility within the institution the program never had before.

The driver behind these numbers is timing. "The ROI from a cost avoidance perspective for palliative care teams is optimized within that first, often three days, of a patient's hospital journey," Dr. Strand said. Everything about the program is built to act inside that window: surfacing need at admission, putting context in front of the right clinician, and reducing the referral to a single click. It hardwires the habits, so that "making the right thing became easy."

Scaling beyond the specialty team

The workforce shortage means specialty palliative care cannot simply hire its way to coverage. So Mayo is extending the same identification and workflow infrastructure to primary services that want to run their own first-line palliative interventions, with specialty palliative care as the escalation path.

"We can build up different pathways for the bedside clinicians," Dr. Strand said. Projects that once required a department to build its own score and identification process can now stand up in a short period of time.

It is also changing how the program plans growth. "I don't think we are going to be able to grow at Mayo Clinic without using this type of construct. Look at our hospital and say, here's where the patients are, here's what those needs are, and here's the hiring or the incremental growth we need to meet that need."

Mayo’s playbook: making the case inside your health system

Most health system leaders already believe in the patient value of palliative care. The hard part is turning that belief into funded headcount and technology investment, when the program competes for resources against specialties with more visibility and more direct revenue. Dr. Strand has built or expanded palliative care programs at multiple sites over the past decade. His advice for leaders advocating for this investment:

1. Speak the C-suite's language, consistently. Pair the patient quality story with utilization and cost avoidance. Length of stay, readmissions, throughput, mortality attribution. Different executives carry different pressures. The message has to hold across all of them.

2. Bring the numbers that show the gap. At one point, 200 beds in Mayo's hospital were consumed by readmitted patients, crowding out surgical volumes. Only about one in three readmitted patients had advanced serious illness, and only about half of eligible patients were receiving a palliative care consult. Those numbers made the case tangible.

3. Get the technology right before you hire. The instinct is to hire the team first, then build workflows, then add technology. Dr. Strand argues that sequencing is program-focused, not patient-focused. Visibility into true patient need should drive who you hire and how you deploy them.

4. Answer the "we already have alerts" objection with precision. More alerts are not the goal. The goal is making sure a scarce clinical resource reaches the patients with the greatest need, which disease-based triggers and mortality models cannot identify.

5. Plan for sustainability, not a pilot. Internal builds tend to end when the grant or the trial ends. Ask what the program looks like in year three, and who maintains and improves it when priorities shift.

To learn how health systems are using real-time clinical intelligence to identify unmet palliative care needs, contact us or watch the full conversation with Dr. Strand here.