There's a study out in JAMA Network Open that should stop a lot of people mid-sentence. Researchers looked at roughly 16,000 Medicare fee-for-service patients, about 4,200 of them treated through hospital-at-home and the rest admitted the normal way. The hospital-at-home group did better on the things you'd hope. In-hospital mortality was 0.4% versus 3.6%. ED visits ran 8.8% versus 10%. ICU escalations were less than half. Real wins.
Then you get to readmissions. Hospital-at-home patients came back at 11.7%. The inpatient group came back at 11%. No meaningful difference. We took the entire acute hospital, wheeled it into someone's living room, and the number everyone actually gets judged on didn't budge.
Here's the thing nobody wants to say out loud. The location of care was never the problem. You can deliver world-class acute care in a bedroom instead of a ward and the patient still ends up back in the ED three weeks later, because the thing that drives readmission isn't where the IV pole stands. It's what happens in the twenty-something days after the acute episode ends, when the monitors are gone and the visiting team has moved on to the next patient. That gap is where readmissions live. Moving the walls doesn't close it.
I've watched a version of this play out for years, and I'll be blunt about it. Health systems keep buying the setting and the hardware, then acting surprised when outcomes flatline. Hospital-at-home is the acute version of a mistake the industry made first with remote monitoring.
Look at the monitoring evidence and it tells the same story. A 2025 systematic review in JMIR of 40 randomized trials found remote patient monitoring produced a hospitalization risk ratio around 0.86, and the authors graded the certainty as low. A 2026 heart failure meta-analysis pooling 65 trials and roughly 23,000 patients found a mortality benefit, but the hospitalization effect had a prediction interval that crossed 1.0, which in plain English means in plenty of real-world settings the effect vanishes. And the most telling detail: invasive, implantable monitoring beat app-and-cuff monitoring, something like a 0.72 risk ratio versus 0.83. (I'm pulling those numbers from the study summaries, so verify the exact confidence intervals before you quote them anywhere.)
Why would an implant beat a Bluetooth blood pressure cuff? It isn't the sensor. It's that implantable programs are built around a clinician who is contractually on the hook to act on the reading. The cuff programs too often just pile data into a dashboard nobody's paid to watch. The device was never the intervention. The human response to the device is the intervention. Take the human out and you've bought a very expensive step counter.
This is exactly why I keep telling people we're not a hospital-at-home company and we're not an RPM vendor. Those are settings and tools. We sell the thing that actually moves the number, which is a clinical relationship that doesn't clock out when the acute episode ends. When we see systolic blood pressure come down and hold under 130 inside a few months, or A1c drop 20 to 30 percent in a population, it isn't because we shipped a shinier gadget. It's because a nurse called on day three, caught the medication the patient stopped taking, adjusted, and called again on day ten. Boring. Repeatable. It works.
The RPM bubble already taught this lesson and the industry mostly refused to learn it. A whole generation of vendors sold devices, billed the codes, and generated data lakes with no one downstream to drain them. Hospital-at-home is now walking into the same trap wearing a nicer suit. Same conviction that if you just relocate the care or upgrade the equipment, the outcomes follow. They don't. The JAMA readmission line proves it with the strongest version of the model we've got.
So what do you actually do Monday morning if you run a system or a physician enterprise?
Stop shopping for platforms and start shopping for follow-through. When a vendor pitches you, don't let them open the laptop and show you the dashboard. Ask them one question and watch them squirm: when a patient's reading goes red, who acts on it, in how many hours, and how do you prove it happened? If the answer is "the care team is alerted," that's not an answer. Alerted isn't managed. Push until you get a name, a time, and a documented action, or walk.
Then instrument the gap, not the episode. Most systems obsessively measure what happens inside the four days of an admission and go blind the moment the patient goes home. Flip it. Track time-to-first-contact after discharge. Track how many flagged patients got a clinical touch inside 48 hours and what changed because of it. Track thirty-day readmission by whether a human actually intervened in between, not by whether a device was in the house. You'll find the readmissions clustering exactly where the follow-through broke down, and you'll stop paying for setups that make you feel modern while your numbers sit still.
And be honest with your board about what you're buying. If you're standing up hospital-at-home purely on a readmission thesis, this study just knocked the legs out from under that pitch. Buy it for the mortality and ED and ICU wins, which are real. But if readmissions are the target, you need a care model that owns the patient through the whole vulnerable window, not a location change and a monitoring kit.
The house call feels like progress. The wearable feels like progress. Both are easier to buy than the thing that actually works, which is why they sell so well. The patients who don't bounce back aren't the ones with the best equipment at home. They're the ones somebody was still paying attention to after everyone else stopped looking.
