Does Remote Patient Monitoring Actually Reduce Hospital Readmissions?
I built a mental health app once. Bipolar patients, daily check-ins, mood tracking, the whole thing. We had clinicians who believed in it. We had a decent design. We had 15% weekly engagement.
Fifteen percent. Meaning eighty five percent of the people who needed the app most just... didn't open it.
I remember staring at that number in a dashboard at 2am, the way you stare at a number that's telling you something you don't want to hear. The problem wasn't the content. The problem wasn't even really the app. The problem was that a screen full of sliders and mood emojis is not what a person recovering from something serious wants to interact with. So we threw it out. We started calling patients instead, with AI that actually talked to them like a person would. Engagement went from 15% to 85%. That gap is basically the whole reason HANA exists.
I think about that a lot when I read studies like the one that came out of Pitt and UPMC in June. Because it's the same lesson, at a much bigger scale, with much higher stakes.
Does remote patient monitoring actually reduce readmissions?
According to the largest trial ever run on the question, no. Not the way it's currently built. The ACCOMPLISH trial, published in JAMA Network Open, randomized 1,286 adults recovering from sepsis or lower respiratory infection across 19 hospitals. Four different remote monitoring setups, all reimbursed by CMS, all using symptom questionnaires plus a nurse response team of some kind. None of them beat usual care. Readmission sat around 37 to 44 percent no matter which arm you were in.
That's not a small study with a weird result. That's the biggest test of this exact model ever conducted, and it came back basically flat.
Why did symptom questionnaires fail older patients?
Because a form doesn't know when to worry. The trial found something genuinely alarming buried in the subgroup data: patients 65 and older who got remote monitoring had fewer home days and higher readmissions than patients who just got usual care. The monitoring made things worse for the group that needed the most help.
My read on why: a questionnaire twice a week is a checkbox exercise. It rewards patients who are already engaged and articulate enough to fill it out honestly. It does nothing for the patient who's confused, scared, in pain, or just doesn't feel like typing symptoms into a phone at 7pm. Which, if we're honest, describes a huge share of people three days out of a sepsis hospitalization.
What's the difference between monitoring and engagement?
Monitoring watches. Engagement talks back. That's the whole difference, and it's not subtle once you see it.
A questionnaire is a one-way pipe. Data goes in, sits in a queue, and a nurse eventually looks at it if the alert threshold trips. A conversation is bidirectional in real time. It asks a follow-up question. It notices when an answer sounds off, not just wrong. It can escalate on tone, not just on a number typed into a box.
I used to think this was a nice-to-have. I don't anymore. Our own data across more than a million patient interactions shows weekly engagement sitting around 85%, against an industry baseline closer to 15 to 20%. That's not a testament to better UX. It's the same lesson from my failed bipolar app, at scale, in a different clinical population.
Why does a phone call still beat an app?
Because a phone call meets the patient where they already are, which is usually on a couch, tired, not looking at a screen. Nobody explains their symptoms better through a dropdown menu than they do out loud to a voice that sounds like it's actually listening.
I spent years around the circus in Australia, and the acts that pulled the biggest crowds were never the most technically impressive ones. Fire chains beat aerial silk, every single time, because fire chains are legible from fifty feet away. A phone conversation is the fire chain of patient follow-up. It's accessible. Everyone already knows how to do it. You don't need a tutorial to answer a call.
What should clinics do instead of buying another RPM device?
Stop assuming the device is the intervention and start treating the conversation as the intervention. The ACCOMPLISH researchers said it themselves: their finding "supports reevaluating and tailoring remote monitoring... to support further alignment with patients' needs." Translation, from someone who's read a lot of these papers: build something that adapts to the patient instead of asking the patient to adapt to a form.
Concretely, that means voice-first outreach that calls at the right cadence, asks open questions, listens for the answer that doesn't sound right, and escalates a real human the moment something looks off. It means use cases built around post-discharge recovery specifically, not a generic wellness check bolted onto a billing code. And it means proving the ROI, which for the clinics we work with runs around 31 to 1, because avoided readmissions are expensive to not avoid.
Where does voice AI fit into post-discharge care?
Right where the nurse call used to sit, except it can happen every day instead of once, at 8pm instead of only during business hours, in three languages instead of one. UPMC's own usual care, the arm that beat every monitoring intervention in the trial, already included a structured nurse phone call. That's the tell. The intervention that worked best in this whole trial was a human voice checking in. We just think that voice should be available more than once, to more patients, without burning out the nurses who have to make the calls.
We've run this across five countries now, more than a million interactions, zero critical adverse events. Not because the AI is smarter than a nurse. Because it can show up more often than a nurse's shift allows, and it never forgets to ask the follow-up question.
Key Takeaways
The biggest remote monitoring trial ever run found that passive symptom questionnaires don't reduce readmissions, and actually hurt older patients. The lesson isn't that remote monitoring is a dead end. It's that monitoring and engagement got confused with each other somewhere along the way. A form that sits in an inbox until a nurse checks it isn't engagement, it's paperwork with a health tech label on it. What actually moves the needle, in this trial and in our own data, is a real-time conversation that adapts to the patient in front of it. That's a harder thing to build than a questionnaire. It's also the only thing that's actually shown to work.
FAQ
Does remote patient monitoring reduce hospital readmissions after sepsis?
Based on the largest trial to date, no. The ACCOMPLISH trial found no monitoring arm outperformed usual post-discharge care, and monitoring was associated with worse outcomes in patients 65 and older.
Why do symptom questionnaires perform worse for older patients?
Questionnaires depend on the patient being willing and able to self-report accurately on a schedule. Older, sicker, or more confused patients are less likely to engage that way, so risk signals get missed until it's too late.
What's a better alternative to app-based remote monitoring?
Voice-based conversational check-ins that happen more frequently than a nurse call, adapt in real time to what the patient says, and escalate a human the moment something sounds wrong. That combination is what HANA is built around, and it's worth a conversation if your post-discharge numbers look anything like the ones in this trial. Book a discovery call if you want to see what that looks like for your clinic.