The Largest Remote Monitoring Trial Ever Run Just Failed. I Think I Know Why.

I built a mental health app once. Beautiful thing. Mood tracking, sleep logs, medication reminders, a little chart that turned green when you were having a good week. The users were patients with bipolar disorder. I was a clinical psychologist, I'd sat with hundreds of them, I knew exactly what they needed and I was extremely confident about it.

Fifteen percent were still using it after a month.

Fifteen.

So when the results of the ACCOMPLISH trial landed in June, the largest randomised trial of remote patient monitoring anyone has ever run, I felt something I'm not proud of. Relief. Somebody finally measured, properly, with 1,286 patients and four years and a real control arm, the exact mistake I made six years ago with a hundred people and a hunch.

What did the ACCOMPLISH trial actually find?

It found nothing. That's the finding. Across four different remote monitoring strategies tested against usual care for patients discharged after sepsis or serious respiratory infection, none of them increased the number of days patients spent alive and at home. Readmission rates ran from 36.3% to 44.2% across the monitoring arms. Usual care sat at 37.8%. Statistically, a wash.

And then the part that should stop everyone cold. Among patients 65 and older, the monitored groups did worse. Fewer days at home. Higher readmissions. In the exact population CMS reimburses you to monitor.

Why didn't remote monitoring work?

Because the trial didn't test whether monitoring is a good idea. It tested whether sending someone a questionnaire twice a week is the same as caring for them. Look at the engagement numbers buried in the results. Of 887 patients assigned to remote monitoring, only 529 actually enrolled. Of 10,561 questionnaires sent, 5,922 came back. Just over half.

Meanwhile the control arm, the boring one, got a phone call from a nurse. Sixty four percent got the call. Seventy eight percent of those answered it.

A human being rang, and four out of five people picked up.

What's the difference between monitoring a patient and engaging one?

Monitoring is something you do to a patient. Engagement is something that happens between two parties. One requires the patient to show up for your system. The other requires your system to show up for the patient.

Everything I've learned in healthcare since that failed app comes down to that sentence. When we stopped asking bipolar patients to open something and started calling them instead, engagement went from 15% to 85%. Same patients. Same clinical content. Different direction of effort. We've since run over a million patient interactions across five countries and three languages and the pattern hasn't moved. Outbound beats inbound. Every single time.

The uncomfortable implication for anyone with an RPM budget: your dashboard fills up with data from the patients who were least likely to be readmitted anyway. The ones you're worried about are the ones not answering the questionnaire.

Does anything actually reduce readmissions?

Some things do, and the shape of them is instructive. A multi-site study across nine hospitals found that virtual nursing at discharge dropped 30-day ED readmissions from 13.3% to 3.7%. Not a questionnaire. A nurse, remote, talking to the patient at the moment of discharge, making sure the plan landed.

Notice what that has in common with the boring control arm that beat four sophisticated monitoring models. Somebody initiated contact. Somebody asked the questions out loud. Somebody heard how the answer sounded, which is a different signal entirely from a tapped radio button.

That's the thing a questionnaire structurally cannot do. It can't hear that the patient is out of breath while answering. It can't catch the pause before "yeah, I'm taking them all."

So what should health systems build instead?

Not more surfaces for patients to log into. Fewer. The strategy that survives contact with reality is proactive, conversational, and initiated by you.

That's the entire architecture behind what we've built. Voice, outbound, structured clinical protocol, escalation to a human when something's off. Across a million-plus interactions we've had zero critical adverse events, which matters more than any engagement number because safety is the only thing that lets you scale. Clinics running it see roughly 31 dollars back for every dollar spent, and honestly the ROI is almost beside the point. The point is that the patient at highest risk, the one who would have ignored the app, answers the phone.

If you're building on top of this rather than buying it, everything is open source and self-hosted, and the integration path is documented. No OpenAI dependency, no patient data leaving your infrastructure.

Key takeaways

The biggest RPM trial ever conducted found no readmission benefit for sepsis patients, and a possible harm signal in adults over 65. The likely reason isn't that remote care doesn't work. It's that asking patients to come to your system filters out precisely the patients who need you most. The interventions that do move readmissions, nurse calls and virtual nursing at discharge, share one property: the clinic initiates. If your engagement strategy depends on the patient's motivation, you're measuring motivation, not care. Build outbound. See the clinical use cases where this holds up and where it doesn't.

FAQ

Does remote patient monitoring reduce hospital readmissions?
The evidence is far weaker than the reimbursement policy implies. The ACCOMPLISH trial, the largest randomised trial to date, found no reduction in 90-day readmissions for sepsis and lower respiratory infection patients across four monitoring designs. Some chronic conditions like heart failure show better results, but broad deployment isn't supported by the data.

Why did older patients do worse with remote monitoring?
Researchers don't know yet, and they've flagged it for further investigation. One plausible mechanism is that alerts generated by symptom questionnaires trigger reflex readmissions rather than effective outpatient intervention. The signal was strong enough that the trial authors advise caution before enrolling older or high-complexity patients.

Is voice AI a better fit for post-discharge follow-up than app-based monitoring?
For engagement, yes, and the gap isn't subtle. Voice is outbound and requires nothing of the patient except answering a phone they already own. HANA sees 85% weekly engagement against a 15-20% industry baseline for app and portal-based programs. Whether that converts to readmission reduction depends entirely on what happens after the call, which is the part most programs underbuild.

If you're rethinking a post-discharge program that isn't moving the numbers, book a call with me and I'll tell you honestly whether voice is the missing piece or whether your problem is downstream.