Remote Monitoring Just Failed a Randomized Trial. Post-Discharge Follow-Up Still Works.
Years ago I crossed the Australian desert with a circus. Actual circus, actual desert, feels like another lifetime honestly. And the thing I remember most isn't the heat or the driving. It's which acts drew the crowds.
Not the aerial silk. The silk was the hardest thing anyone did, months of training, genuinely beautiful, and people watched politely for ninety seconds and wandered off to buy chips.
Fire chains packed the ground every single night.
Fire chains are simpler. Louder. You get them instantly. The most technically sophisticated act was never the one that worked, and I've been carrying that around ever since, because healthcare technology keeps making the aerial silk mistake at enormous expense.
Which brings me to a randomized trial that should have made more noise than it did.
What did the JAMA remote monitoring trial actually find?
It found that remote patient monitoring after serious infection didn't help, and for older patients it actively hurt. The ACCOMPLISH trial randomized 1,286 post-discharge sepsis and lower respiratory tract infection patients across 19 hospitals to four different remote monitoring strategies or usual care. Primary endpoint was days at home at 90 days.
Median days at home: 90 in every arm. Including usual care.
Readmission at 30 days ran 37.8% under usual care and between 36.3% and 44.2% across the monitoring arms. No arm cleared a 55% superiority probability. And among patients 65 and older, both the standard and enhanced response arms produced fewer days at home than doing nothing, with inferiority probabilities of 99.6% and 97.9%.
That's the sophisticated act. Wearables, questionnaires, enhanced clinical response teams, four years of enrollment. And the crowd wandered off.
Why did the monitoring arms fail?
Look at one number buried in the results. Of the 887 patients assigned to remote monitoring, 529 actually enrolled in the program. That's 59.6%.
Four in ten never showed up to the thing being tested.
I'm not saying enrollment explains everything, because the trial was intention-to-treat and that's the honest way to run it. But it points at the fault line. These programs asked patients to have a smartphone, download something, answer questionnaires on a cadence, and stay motivated during the worst month of their year. The design put the burden on the person who just got out of the hospital.
Compare that to the heart failure outreach pilot published in JACC, which deliberately required no app and no patient portal. Just texts and emails, automated, with a single question that mattered: are you worried your health might send you to an emergency room? Two yeses connected the patient to clinical staff automatically. Overall readmission fell to 15.5% against a 24% national average. Highly engaged patients came in at 7.7%. Low-engagement patients at 21.6%.
Same intervention. Nearly threefold difference. Engagement was the whole ballgame.
What's the difference between reach and engagement?
Reach is whether your system touched the patient. Engagement is whether the patient did something as a result. Most health systems instrument the first and assume the second, and then they're surprised when a well-funded program produces a null result.
The distinction shows up in the deployments that are working. UPMC just expanded an AI-native transitional care model from one campus to 18 hospitals, and the design detail worth stealing isn't the AI. It's that they reach 100% of enrolled patients within 48 hours, including the ones who don't pick up the first time and the ones with no risk flag. They also moved the automated check-in up to day 15, which is earlier than most programs bother, because that's when problems are still fixable.
And two of the biggest improvements in the program came from a patient education flyer and co-branding the outreach so patients recognized it was coming from their own hospital.
Fire chains, not aerial silk.
What are the operating numbers worth watching?
UPMC's model reports a 63% cut in clinician time on non-clinical work per encounter, 82% of non-actionable alerts removed by triage, a 180% increase in patients seen after discharge, and a 45% reduction in total cost of care. Vanderbilt's Discharge Care Center, which is much lower tech, moved a hospital-wide 30-day readmission rate from 10.6% to 9.9% and held it for two years across 80,000 discharges.
Neither of those is a monitoring device story. Both are contact-with-a-human-when-it-matters stories, with automation handling everything before that point.
That's how we've built HANA's post-discharge and chronic care workflows too. The agent calls, holds a real conversation in the patient's language, documents the outcome, and escalates the second something looks clinically wrong. Across a million-plus interactions we've seen 85% weekly engagement and zero critical adverse events, and we publish the outcome data rather than gesturing at it.
What should a health system build first?
Start with reach that doesn't depend on patient effort. Voice and text, no app, no portal, no login, in the patient's own language. Then instrument engagement rather than delivery, so you know who actually responded and not just who received.
Then build the escalation path before you build anything clever, because the escalation path is where the clinical value lives and it's the piece everyone underinvests in. Everything upstream of it is logistics.
Run it inside your own perimeter if you can. HANA is fully open-source and self-hosted with no OpenAI dependency, and the integration path is documented publicly, which matters more every quarter as governance requirements tighten.
Key Takeaways
A well-run randomized trial just showed that remote monitoring after sepsis and respiratory infection doesn't increase days at home, and reduces them for patients over 65. The likely culprit isn't the concept, it's the burden the design placed on recently discharged patients, with only 60% of assigned patients enrolling at all. The programs that are working right now reach everyone rather than the top risk tier, do it within 48 hours, require nothing from the patient, and route to a human the moment something looks wrong. Measure engagement, not delivery. And pick the fire chains over the aerial silk, every time.
FAQ
Does remote patient monitoring reduce readmissions? The evidence is mixed and got weaker in 2026. Structured RPM combined with disciplined nurse escalation has shown reductions in observational syntheses, but the ACCOMPLISH randomized trial found no benefit after sepsis and lower respiratory infection, and harm in patients 65 and older. Enrollment and engagement appear to matter more than the monitoring hardware.
Why does automated outreach outperform monitoring in some studies? Because it removes the patient effort barrier. Text, email, and voice reach people where they already are, with no device to charge or app to open, so completion rates hold up in exactly the population that struggles with more demanding programs.
How quickly after discharge should outreach happen? Within 48 hours is the working benchmark across the programs showing results, with a second structured check around day 15. Earlier contact catches the problems that are still solvable at home. If you want to pressure-test your own transition workflow against these numbers, book time with me directly.