A 1,286-Patient Trial Just Found Remote Monitoring Doesn't Reduce Readmissions
I built a mental health app once. For people with bipolar disorder. I was a clinical psychologist, I knew the literature, I knew the patients, and I was certain I knew what they needed.
Engagement: 15%.
Fifteen. Out of a hundred people who genuinely wanted to get better, eighty five never opened the thing twice. I remember sitting with that number for about a week, refusing to accept it, blaming onboarding, blaming push notifications, blaming everything except the fundamental premise, which was that I had built a homework assignment and called it care.
So when I read the ACCOMPLISH trial results in JAMA Network Open this summer, I felt something uncomfortably close to recognition.
Why didn't remote monitoring reduce readmissions?
Because the patients didn't use it, and the ones who did were answering questions instead of talking to anyone. That's the short version. The trial randomised 1,286 adults discharged after sepsis or lower respiratory tract infection across 19 hospitals, testing four remote monitoring strategies against usual care. None of them increased days alive at home at 90 days. Readmission rates ran 36.3% to 44.2% across the monitoring arms against 37.8% for usual care.
Of 887 patients assigned to remote monitoring, 529 actually enrolled. Of 10,561 questionnaires sent, 5,922 came back.
Look at those two numbers next to each other for a second.
What's the difference between monitoring a patient and engaging one?
Monitoring extracts data. Engagement creates a relationship the patient wants to maintain. One of those survives contact with a sick, tired, overwhelmed human being at home on day four post-discharge, and it isn't the twice-weekly symptom questionnaire.
The trial critique at Critical Care Reviews makes a point that I keep rereading: nurses responded to over 94% of alerts. The system worked. The clinicians worked. The limiting factor wasn't neglect on the provider side at all. It was that the patient interface was a form, and forms don't ask how you're sleeping and then actually listen to the answer.
We measured this ourselves before we had any right to call it a company. Our numbers are on HANA's research page, and the gap between a form and a conversation is not subtle.
Why did older patients do worse with remote monitoring?
This is the part that should make everyone uncomfortable. Among patients 65 and older, the monitoring arms had fewer home days than usual care. Inferiority probability of 99.6% for the standard response arms. Same signal for patients discharged via skilled nursing facilities.
The population Medicare reimburses you most enthusiastically to monitor is the population where monitoring appeared to make things worse.
My guess, and it's a guess, is that a smartphone questionnaire is a cognitive tax, and taxing an 80 year old recovering from sepsis produces either silence or a panicked yellow alert that ends in a reflex readmission. Nobody in that loop is being careless. The loop is just badly shaped.
What should health systems do with their remote monitoring budget?
Stop buying data collection and start buying contact. The University of Pittsburgh team said it plainly: health systems should reevaluate broad deployment, and technology companies should build systems better tailored to complex care needs.
That's an invitation, honestly. It's also a warning, because the same trial found UPMC's usual care was strong. A nurse phone call and timely primary care follow-up beat four flavours of remote monitoring. The phone call is the control arm that won.
So the question stopped being whether to call patients. It became whether you can afford to call all of them, every week, in their language, at 7pm when they're actually home. That's the arithmetic we run on HANA's pricing page, and it's the only reason a clinic gets to 31:1 instead of a pilot that dies quietly in month four.
Does voice AI actually solve this?
Not automatically. It solves the coverage problem, which is a different thing from solving the care problem, and anyone selling you otherwise is selling you a 15% engagement app with better graphics.
What it does solve: we called patients instead of asking them to open something. Engagement went to 85% weekly against the 15 to 20% industry baseline. Over 1M+ patient interactions across 5 countries and 3 languages, zero critical adverse events. The deployments and how they were structured are written up in our case studies, including the unglamorous parts.
The Australian circus taught me this, weirdly. I crossed the desert with one. The performers pulling the biggest crowds weren't doing aerial silk. They were doing fire chains, which is easier, louder, and legible from forty metres away in bad light. Accessible beats sophisticated. A ringing phone beats a dashboard.
Key Takeaways
Remote monitoring is not a readmissions intervention, it's an infrastructure decision, and ACCOMPLISH is the largest trial to say so out loud. The failure wasn't the clinical response teams and it wasn't alert fatigue. It was 59.6% enrolment and a form standing between a sick person and a clinician who was ready to help.
If you're a health system with a monitoring budget and a readmissions penalty, the trial hands you a strange gift: your nurse phone call already works. The problem is you can't staff it at population scale, and older patients, the ones most at risk, are the ones least served by asking them to type. Fix the channel before you buy more sensors. Voice is not a feature here, it's the whole argument, and the clinical use cases that hold up are the ones where somebody actually gets talked to.
FAQ
Did the ACCOMPLISH trial show remote monitoring is useless?
No. It showed that symptom-questionnaire monitoring added on top of an already strong usual-care pathway didn't increase home days after sepsis or lower respiratory tract infection. Physiological monitoring, wearables and hospital-at-home weren't tested, and remote monitoring in heart failure still has solid evidence behind it.
Why would remote monitoring make outcomes worse in older adults?
The trial couldn't answer that, and the authors flagged it for further investigation. The plausible mechanisms are low engagement with the questionnaire format and alerts escalating to readmission rather than to a resolvable intervention. Either way it argues for careful targeting instead of blanket deployment.
Is voice AI follow-up different enough to matter?
It changes who has to do the work. A patient answers a phone without learning an interface, which is why weekly engagement lands near 85% rather than near 15%. If you want to pressure-test that against your own patient panel, grab a slot on my calendar and bring your readmission numbers.