Remote Monitoring Just Failed a Randomized Trial. Reach Is Why.

There's a screenshot on an old phone somewhere of a Stripe dashboard crossing a million dollars in a single day. My company. My product. I remember sitting in a hotel room looking at it and feeling almost nothing, because I already knew the product wasn't good. Scale had just arrived before the truth did.

Big numbers don't validate anything. They just make the problem more expensive to find.

I thought about that hotel room when I read the sepsis remote monitoring trial.

What did the sepsis remote monitoring trial actually find?

It found nothing, which is the point. Researchers randomized 1,286 adults across 19 hospitals after discharge for sepsis or lower respiratory tract infection into usual care or one of four remote monitoring arms, varying questionnaire intensity and clinical response team intensity, and measured days spent at home over the following 90 days. No monitoring arm beat usual care. Superiority probability came in under 55% for every comparison.

Then it got worse. Among patients 65 and older, the monitored arms had fewer days at home than the ones left alone, with inferiority probabilities of 99.6% and 97.9%. Readmission happened to 37.8% of the usual care group and somewhere between 36.3% and 44.2% of the monitored groups.

CMS reimburses this. Remote monitoring among Medicare beneficiaries went up tenfold between 2019 and 2022. And a well run randomized trial just said it didn't move the outcome it's paid to move.

Why did remote monitoring fail when the technology worked?

Because 59.6% of the patients assigned to monitoring actually enrolled in it. Four in ten never started.

Sit with that. Before you get to model quality, escalation logic, response team staffing, any of the interesting engineering, forty percent of the intervention evaporated at the door. And the ones who did enroll were being asked to self report daily, on a smartphone, while recovering from sepsis. Which, if you've ever watched someone recover from sepsis, is a thing you'd only ask if you'd never watched someone recover from sepsis.

My daughter is ten. A while back I told her, only half joking, that I work for her and not the other way around. She took this extremely well and has been billing me ever since.

Every remote monitoring program I've seen tells the patient the exact opposite. Come to us. Download this. Log in. Answer eleven questions before breakfast. We built the thing, now do your part.

Is this an argument against post-discharge follow-up?

Not remotely. Penn's Live Better program cut 30 day readmissions by 43% in its pilot population for cirrhosis and liver transplant patients, and dropped program cost per patient from $1,050 to under $50. Only 4% of cases needed escalation to a physician. The clinical hypothesis is fine. Catching deterioration early works.

The delivery is what breaks. A separate 2026 analysis of patients who bounced back to hospital inside 30 days found that fewer than half had engaged with their mHealth program at all before the return visit. The engaged ones returned later and more predictably, which is genuinely useful, but you can't act on a signal from someone who never joined.

So the honest reading of the JAMA result isn't that monitoring doesn't work. It's that a monitoring program which reaches sixty percent of its population and asks the other forty to do homework is not actually a monitoring program. It's a pilot with good intentions and a dashboard.

What does engagement-first infrastructure look like?

It inverts the ask. Nobody downloads anything. The system calls the patient, holds a real conversation in their own language, and does the structured questioning itself. If the patient says something that matters, a human hears about it inside minutes, not at the next chart review.

That's the whole design principle behind what we build, and it came out of watching a mental health app I'd built get 15% weekly engagement while phone calls to the same population got 85%. Same patients. Same clinical content. Different ask.

Across five countries and three languages we've run more than a million patient interactions with zero critical adverse events, and the deployments that hold up all share one trait: reach was solved before intelligence was.

What should a health system measure instead of enrollment?

Enrollment is a vanity metric and the trial proves it. Four numbers are more honest.

Percentage of eligible discharges actually reached, not offered. Percentage of conversations completed end to end, because a call that gets abandoned at question two is worth nothing. Escalations raised per hundred conversations, which tells you whether your questions are finding anything. And median time from patient signal to human contact, which is the number that separates a program that prevents readmissions from one that documents them.

Run those four for a quarter and you'll know more than a 1,286 patient trial can tell you about your own population. Our data on how those move is public, and the integration surface is documented in the open so you can pressure test it against your own EHR before anyone signs anything.

Key takeaways

The trial didn't disprove remote monitoring. It disproved a particular delivery model, one that assumes a discharged, exhausted, frequently elderly patient will meet the technology halfway. They won't, and it's unreasonable to have expected it. The interventions that work, like Penn's, work because the effort sits on the system's side of the line.

Before you buy another monitoring platform, ask the vendor what percentage of eligible patients they actually reach. Not enroll. Reach. If they don't have that number to hand, you're looking at my Stripe dashboard: a big impressive figure sitting on top of a problem nobody's found yet.

Frequently asked questions

Does remote patient monitoring reduce hospital readmissions? The evidence is mixed and condition dependent. It has shown benefit in heart failure, but the 2026 randomized trial in sepsis and lower respiratory infection found no improvement in days at home and worse outcomes in patients 65 and older.

Why do post-discharge programs have such low engagement? Because most of them require the patient to initiate, whether that's downloading an app, logging in, or answering daily prompts. Recovery is the worst possible moment to add friction, so uptake collapses in exactly the population that needs monitoring most.

What's the difference between outbound voice AI and automated text outreach? Text is a notification the patient can defer indefinitely. An outbound call is a conversation that either completes or doesn't, and completion is measurable, which makes the intervention auditable in a way a broadcast message never is.

If you're running transitions of care and your reach numbers don't look the way your enrollment numbers do, book a slot and we'll dig into where the gap is.