One Nurse Watching 220 Patients: What Continuous AI Monitoring Actually Changes for a Clinic

A respiratory therapist I met last year told me she used to keep a paper list of her sickest patients taped to the side of her monitor. Thirty names. She'd call them when she could. Between charting, between rooming, between the phone ringing off the hook at the front desk. Some weeks she got to all thirty. Most weeks she didn't.

That list haunted her.

Not because she was lazy. Because she was one human being, and there were only so many hours, and the patients who slipped off the bottom of the list were exactly the ones who'd show up in the ER three weeks later, gasping.

I keep thinking about that list, because a study just came out that puts a number on what changes when you stop relying on it.

What did the Intermountain study actually find?

Researchers at Intermountain Health and CareCentra ran a two-year study on continuous AI monitoring for two of the most common chronic pulmonary conditions in the world. The results are hard to wave away. Hospitalizations dropped 50%. Emergency department visits fell 20%. Total cost of care per patient per year went from $36,837 to $15,899. That's a 57% reduction in cost, backed by real outcomes, not a vendor slide deck.

But the number that stopped me cold was this one. Navigator capacity went from 30 patients per navigator to nearly 220. Sevenfold.

That's the paper list, gone. That's the therapist getting to everyone.

Why does continuous monitoring beat a periodic check-in?

Because illness doesn't wait for your next appointment. A heart failure patient doesn't decompensate on a schedule that matches your clinic calendar. The danger lives in the gaps. The two weeks between visits where nobody's watching and the patient is quietly retaining fluid, skipping a med, telling themselves it's fine.

The Intermountain team called their system a "smart check-engine light." I love that. You don't wait for the engine to seize. You catch the warning before the breakdown. The AI senses rising risk between visits, shapes behavior with a nudge, and escalates to a human respiratory therapist only when the signals converge toward a real exacerbation.

That last part matters more than people think. The human isn't removed. The human is aimed. You can read more about how that escalation logic works in our clinical use cases, because the routing is the whole game.

Isn't this just another patient app nobody opens?

Fair question. I've been burned by exactly that.

Years ago I built a mental health app for people with bipolar disorder. Beautiful thing. Clean design. I was proud of it. Engagement was 15%. Fifteen. People downloaded it, opened it twice, forgot it existed. I'd built a gym membership, not a treatment.

So I threw it out and tried something stupid and obvious instead. I called the patients. With AI. A voice on the phone, not an icon on a screen they'd have to remember to tap. Engagement jumped to 85%. That gap, between 15 and 85, is the entire reason HANA exists today.

The Intermountain data echoes it. The win wasn't more devices or more dashboards. It was reaching people where they already are, before they got worse. Monitoring only works if the patient actually engages with it. Otherwise you've built another app nobody opens.

What does this mean for a clinic that can't hire its way out?

It means the math finally moves in your favor. You're not being asked to add ten coordinators you can't afford. You're being asked to let one coordinator cover the work of seven.

Look at the staffing reality. Every clinic I talk to is short-handed. The front desk is drowning, the nurses are burning out, and the answer everyone reaches for is "we need more people." But more people is slow, expensive, and the talent isn't there to hire. The Intermountain model didn't add bodies. It multiplied the ones they had. That's the difference between a cost center and a 31:1 return, which is the kind of number that survives a conversation with your CFO.

HANA has run more than a million patient interactions with zero critical adverse events. Across five countries, three languages. The point isn't the scale for its own sake. The point is that the safety holds when you stretch one human across hundreds of patients, as long as the system knows when to tap that human on the shoulder.

What should you do before you buy anything?

Set your baseline first. This is the part everyone skips and then regrets.

Before you turn anything on, write down your current numbers. Thirty-day readmission rate. ED utilization. How fast your team responds when a patient flags they're feeling worse. Without those, you'll never prove the program worked, and you'll lose the budget fight next year. A separate analysis of remote monitoring programs found the same thing: programs with disciplined enrollment and clear response protocols beat technology-only pilots every time. The tech is a third of the job. The workflow is the rest. We wrote up how teams instrument that in our research notes.

Key Takeaways

The Intermountain study is one more data point in a story that's getting hard to argue with. Continuous, AI-driven monitoring of chronic patients cut hospitalizations in half and let a single navigator cover seven times the patients. The mechanism isn't magic. It's catching the slow slide before it becomes a crisis, and pointing scarce human attention at the patients who actually need it right now.

The thing that breaks these programs isn't the AI. It's engagement and workflow. If patients don't engage, monitoring is theater. If your team doesn't have a clear escalation path, alerts pile up and get ignored. Get those two right and the outcomes follow. Get them wrong and you've bought a very expensive dashboard.

If you want to see whether your patient panel and your staffing reality fit this model, grab a discovery call with me. I'll show you the math on your actual numbers, not a generic case study.

FAQ

Does AI monitoring replace nurses and care coordinators?

No. It changes what they spend time on. The Intermountain study kept the human respiratory therapist fully in the loop and used AI to handle routine monitoring, so the navigator's attention went to the highest-risk patients instead of working through a list at random. Capacity went up sevenfold without removing clinical judgment.

How fast can a clinic see results from continuous monitoring?

The strongest programs show movement within the first few months, but only if the baseline is measured before launch and patients are enrolled with a clear first-contact window, usually within 48 hours of discharge. Programs that skip baseline measurement struggle to prove value even when outcomes improve.

What conditions are the best starting point?

Start where avoidable admissions are frequent and the response is well understood. Heart failure, chronic pulmonary conditions, and high-risk diabetes are practical first waves because the intervention pathways already exist. Expand once the workflow is proven on one population.