What Does AI Follow-Up Actually Do to Your Readmission Rate? A New Study Has the Answer

We threw out a mental health app once. It had beautiful UX, good science behind it. Patients loved it in the clinic.

Fifteen percent did. Weekly.

That number haunted me. Because the other 85 percent weren't failing. They were just disappearing into their lives and no app on their phone was going to cut through that noise. So we tried something different. We called them. With AI. Weekly check-ins, voice, personal. Eighty-five percent weekly engagement.

Same patients. Completely different number.

I think about that gap a lot when I read new clinical data on AI follow-up. Because we're now past the point of asking if AI can engage patients. The question is: what does it actually do to outcomes?

A study published last week in npj Digital Medicine has one answer. And the Intermountain Health / CareCentra data has another. Together, they build a picture that's harder to argue with than most of what's been published in this space.

What Does the Research Actually Show?

The Intermountain study tracked patients with chronic pulmonary conditions over two years. AI-driven continuous monitoring. The results: 50% reduction in hospitalizations. 20% fewer emergency department visits. 57% reduction in total cost of care: from $36,837 to $15,899 per patient per year.

One navigator went from managing 30 patients to 220. That's not a productivity improvement. That's a structural change.

The npj Digital Medicine study looked at virtual nursing-assisted discharge across nine hospitals. VN-assisted discharges had a 30-day ED readmission rate of 3.7%. Traditional discharge: 13.3%. That's a risk ratio of 0.28.

These aren't pilot studies. These are multi-site, propensity-matched, peer-reviewed results published in June 2026.

Why Does Continuous Outreach Change the Math?

The failure mode in transitional care is always the same. Patient leaves. Feels okay. Doesn't call. Problem compounds quietly. By the time they're back in the ED, the window for cheap intervention is gone.

Traditional follow-up breaks here. Nurses have 30 patients. The call gets deprioritized. The portal message sits unread. The reminder text is ignored because it looks like every other text.

AI follow-up changes the economics of contact. It doesn't get tired. It doesn't deprioritize. It reaches out at the right moment, post-discharge, post-procedure, at the gap in care that matters, and it asks the right question in a way the patient actually answers.

In the CareCentra data, when a pulmonary patient reported worsening symptoms, the system escalated. One navigator, 220 patients, and the humans only showed up when clinical judgment was actually required.

That's the model. Not AI replacing care. AI making sure care happens at all.

What This Means for Your Practice

If you're running a specialty practice, orthopedics, cardiology, GI, dermatology, you have patients leaving your clinic every day who won't follow through. Referrals that sit open. Medication adjustments that don't stick. Post-op instructions that get forgotten by Tuesday.

The gap isn't patient motivation. It's contact. Nobody's calling them consistently. Nobody's checking in on Wednesday, asking how the new med is going, flagging that they haven't scheduled their 2-week follow-up.

AI follow-up fills that gap at scale.

At HANA, we see 85% weekly engagement versus the 15-20% baseline you'd get from a patient portal or app. Over 1 million patient interactions with zero critical adverse events. In five countries, three languages. A 31:1 ROI for the practices we work with.

The readmission data from Intermountain isn't surprising to us. It's confirmation. When you reach patients consistently, in a way that feels human rather than automated, outcomes move.

The Infrastructure Question Nobody's Asking

Every readmission costs a payer $15,000-$20,000. Every care gap that doesn't close is a quality score that doesn't improve. Every coordinator hour spent on manual follow-up is an hour not spent on complex cases.

The question isn't whether AI follow-up is worth the investment. The Intermountain data puts that to rest. The question is: what's your current infrastructure actually doing for the 80% of patients who need follow-up and aren't getting it?

Most practices have an answer. It's uncomfortable.

Is AI Follow-Up Ready for Specialty Care?

The concern I hear most: "My patients are complex. AI can't handle that."

It doesn't have to. That's the point of the model.

In the CareCentra data, AI handles the 90% that's routine: check-ins, symptom monitoring, medication adherence prompts. It escalates the 10% that needs a human. The navigator's judgment isn't diluted. It's concentrated on the cases that actually need it.

HANA is built the same way. Our AI handles the follow-up volume. Clinicians see the exceptions. The system is open-source, self-hosted if you need it that way, designed to fit inside your existing workflows rather than replace them.

The patients who disappear after discharge, the ones you don't hear from until they're back in the ED, aren't a staffing problem. They're an infrastructure problem. The research is now clear on what the right infrastructure looks like.

Key Takeaways

The latest clinical data shows AI-driven follow-up reducing readmissions by 60-70% and cutting total cost of care by more than half. The mechanism is consistent contact at scale: reaching the patients who won't call you, before they need the ED. For specialty practices, the model is AI handling routine volume and escalating complexity to clinicians. HANA delivers 85% weekly engagement versus 15-20% baseline, with a 31:1 ROI and more than 1 million interactions across five countries. The technology is no longer experimental. The question is whether your current follow-up infrastructure is doing what these results suggest is possible.

FAQ

Does AI follow-up actually engage patients, or do they just ignore it like they ignore everything else?

The engagement gap is real, but it's not inevitable. The difference between 15% and 85% weekly engagement comes down to channel and feel. Patients ignore portals and apps because they're passive. You have to remember to open them. AI that reaches out by voice, at the right moment, feels more like a person checking in than a notification to dismiss. When AI follow-up is done right, engagement is higher than almost any digital alternative.

What happens when a patient flags a serious symptom?

Escalation is the whole point. In the CareCentra / Intermountain model, when a patient's responses suggest clinical deterioration, a human navigator is looped in immediately with full context. In HANA, our protocols flag urgent responses in real time so your clinical team can act. The AI doesn't make clinical decisions. It makes sure the right person has the right information at the right moment.

Is the ROI real, or is it just modeled?

The Intermountain numbers are from a two-year peer-reviewed study across a real health system. The cost drop from $36,837 to $15,899 per patient per year is measured, not projected. HANA's 31:1 ROI figure is derived from avoided readmissions, recaptured appointments, and reduced coordinator burden in live deployments. Neither number is a model. Both are outcomes from production.

Want to see what this looks like inside your practice? Book a discovery call, 30 minutes, no deck, just your numbers.

Related reading: HANA use cases · Clinical research · Case studies · Pricing

Sources: Intermountain Health / CareCentra study · npj Digital Medicine virtual nursing study