Remote Monitoring Just Failed A 1,286 Patient Trial. Now What?

Years ago I watched a Stripe dashboard cross a million dollars in a single day. My company. My product. I remember the specific feeling, which was not joy exactly, more like vertigo, because I already knew the product wasn't good.

The number went up anyway. Numbers do that.

Scale doesn't fix a broken thing, it just makes the breakage load bearing. It took me another eighteen months and a fair amount of personal wreckage to fully absorb that lesson, and I think about it every time a category in healthcare gets adopted faster than it gets evidence.

Which brings us to remote patient monitoring.

Did remote patient monitoring just fail?

Not exactly, but a big trial just told us it doesn't work the way most health systems are deploying it. The ACCOMPLISH trial, published in JAMA Network Open in June, randomised 1,286 adults discharged after sepsis or lower respiratory tract infection across 19 hospitals, and none of the four remote monitoring arms beat usual care on days alive at home at 90 days.

Readmission rates landed between 36.3% and 44.2% across the monitoring arms. Usual care was 37.8%. The differences didn't reach significance. You can read the trial writeup here and the full paper in JAMA Network Open.

CMS reimburses this. Adoption has been climbing for years. The dashboard has been going up.

Why did older patients do worse?

This is the finding that should genuinely unsettle people. Among patients 65 and older, both the standard and enhanced response arms had fewer home days than usual care, with inferiority probabilities of 99.6% and 97.9%.

Worse. Not neutral. Worse.

The trial authors are appropriately careful about causation, and so am I, but the plain reading is uncomfortable: if you hand a complex 78 year old a smartphone questionnaire and an internet connected scale, you may be generating alerts that pull them back into the system rather than keeping them out of it. Detection isn't the same as benefit. Somebody has to be able to act on what you detect, and acting sometimes means admitting.

So what's the difference between monitoring and engagement?

Monitoring collects data from a patient. Engagement means the patient participated. Those are not the same event, and almost every RPM business case quietly assumes they are.

Look at the enrolment number in that trial. Of 887 patients assigned to remote monitoring, 529 enrolled. That's 59.6%. Four in ten patients randomised to the intervention never really got the intervention, and that happened inside a well funded academic trial at UPMC with staff whose entire job was making it happen. Now imagine that ratio in a community hospital with one transitions of care nurse.

This is the same wall I hit with a mental health app years ago, at 15% engagement, before we started calling patients instead and watched it go to 85%. We publish those splits on our research page because the gap between "deployed" and "used" is where almost all of healthcare's digital spend goes to die.

Why did virtual nursing work when remote monitoring didn't?

Because a human conversation happened. A multi site study in npj Digital Medicine covering nine hospitals found virtual nursing assisted discharges had 30 day ED readmission rates of 3.7% versus 13.3% for traditional in person discharge, after propensity matching on 4,662 encounters each.

Different study design, different endpoint, so don't stack them directly. But the shape of the difference is instructive. One intervention asks the patient to send you data. The other has someone talk them through what to do. Structured conversation beats passive telemetry, and it beats it by a lot.

That's the entire premise of what we build. Not a device that watches, a call that asks. The clinical use cases where it works best are the ones with a scripted clinical question set and a real escalation path behind it.

What should a health system change in the 2027 budget?

Three things, and none of them are "buy more sensors."

Fund the response layer before the sensing layer. An alert with nobody behind it is a liability, and the ACCOMPLISH enhanced response arm shows that even staffing it well doesn't rescue a modality patients don't complete. Second, stratify by age and complexity before you enrol, because the subgroup that did worst is exactly the subgroup most programmes target hardest. Third, measure completion rate as your primary operational metric, not enrolment. Enrolment is a signature. Completion is care.

We've run over a million patient interactions with zero critical adverse events across five countries and three languages, and completion rate is the number I'd stake the whole thing on. The routing and escalation logic that makes it defensible is documented openly in our technical docs, self hosted, open source, running inside the provider's own infrastructure. When completion holds, the economics come out around 31 to 1. When it doesn't, no ROI model in the world saves you, which is precisely the story ACCOMPLISH is telling.

Key Takeaways

A large randomised trial just found remote monitoring didn't increase days at home after sepsis and pneumonia hospitalisations, and actively reduced them in patients over 65. The likeliest explanation isn't that post discharge follow up doesn't matter, it's that passive data collection isn't follow up. Only 59.6% of assigned patients even enrolled. Meanwhile virtual nursing, which puts a human conversation into the discharge, cut 30 day ED readmissions dramatically in a nine hospital cohort. The lesson for health systems is to stop buying sensing and start buying conversation, then measure completion rather than enrolment, because a growing adoption curve tells you nothing about whether the thing works. I learned that from a dashboard once. It was an expensive lesson.

FAQ

Does this mean we should shut down our RPM programme? No, it means you should audit it. Pull your completion rate by age band and by condition. If completion is under 60% or your over 65 cohort isn't showing benefit, you have a design problem, not a technology problem.

Is voice AI follow up different from remote monitoring? Structurally, yes. Monitoring waits for the patient to send data. An outbound call reaches the patient on their own phone with no app, no login, and no device, asks a bounded clinical question set, and escalates to a human when something's off. Completion rates aren't comparable.

What about older patients and AI calls? They're often the strongest cohort, which surprises people. No digital literacy barrier exists on a telephone. The thing that changes outcomes is whether somebody picks up and answers honestly, and a ringing phone gets that from a 78 year old far more reliably than a questionnaire app does.

If you're rebuilding a transitions of care programme for next fiscal year and want to argue about any of this with the actual data in front of us, book time with me here. Bring your completion rates.