For people living with chronic lung disease, the riskiest stretch of any given year isn't the hospital stay itself. It's everything that happens before it — the weeks and months between scheduled appointments, when symptoms drift, inhalers get used incorrectly, oxygen quietly dips, and no one is watching. By the time a patient ends up in the emergency department, the problem has usually been building for a while.
That gap is what iCARE — the Intermountain COPD and Adult Asthma Remote Evaluation — was built to close. Over the past two years, CareCentra and Intermountain Health have run one of the longest real-world studies of AI-enabled connected care in respiratory medicine, enrolling 1,200 patients across five Intermountain hospitals in Utah. The findings, presented at the American Thoracic Society 2026 International Conference and recognized at the Respiratory Innovation Summit, suggest that continuous monitoring paired with predictive intelligence and a clinical escalation pathway can reshape both the economics and the experience of chronic care.
The headline numbers
Across the cohort, total cost of care per patient per year fell by 57 percent — from $36,837 to $15,899 — across all payer types. Hospital admissions were cut in half. Emergency department visits dropped by 20 percent. Observation-stay costs collapsed by 73 percent. Outpatient costs declined nearly 29 percent even as outpatient visit volume only fell about 12 percent, which suggests patients weren't simply being shifted between settings — they were stabilizing.
Retention told a similar story. Ninety-two percent of patients were still enrolled at twelve months, and nearly 90 percent at eighteen. Daily active use of the CareCentra app averaged 54 percent, and — counterintuitively — engagement climbed with age. Patients in their eighties were the most consistently active users in the entire study, with a 62.5 percent daily active rate. Every additional decade of age correlated with a 6–7 percentage point bump in daily activity, a relationship that held with very high statistical confidence across the cohort.
How the system works
Each enrolled patient carries a digital spirometer that tracks lung function, with subsets of patients also using a connected pulse oximeter, a sensor-equipped inhaler that logs technique and adherence with every dose, and a consumer-grade activity tracker. Symptom check-ins from the patient round out the picture. Together those inputs generated more than 11.5 million data points over the course of the study — roughly 24 readings per patient per day.
CareCentra's platform evaluates that stream against each patient's individualized baseline and against GOLD and GINA guideline-based protocols. The point isn't to log the data; it's to recognize the early shape of trouble. A drop in FEV1. A dip in oxygen saturation. Two or three skipped doses in a row. A shift in sleep or breathing pattern. When signals start to converge, the system responds with the lightest intervention that the situation will accept — a behavioral nudge, a coaching message about inhaler technique, a reminder, or a real-time warning about an environmental trigger.
When the data suggests something more serious is taking shape, the platform hands the patient off to a Pulmonary Disease Navigator — a registered respiratory therapist who has full clinical context and can coordinate across the patient's care team. The handoff is data-triggered, not call-center triage. It happens because the algorithm has reason to believe a real-world exacerbation is forming.
What changes when AI handles the routine work
The clinical model also reorganized the workforce. Before iCARE, each Pulmonary Disease Navigator could realistically manage about 30 patients. With AI doing the continuous watching and surfacing only the highest-risk situations, that same navigator now supports nearly 220 — roughly a sevenfold increase in capacity without losing the human judgment that high-acuity respiratory care requires.
The data also points to where the AI is most useful. About two-thirds of the platform's high-severity alerts came from spirometry, making FEV1 the most sensitive early-warning signal in the dataset. The sensor-equipped inhaler generated more than 900 technique-correction alerts over the study, catching cases where the medication was being prescribed correctly but delivered ineffectively.
A quieter finding about disease severity
One of the more useful side effects of running continuous monitoring is that it generates longitudinal evidence that administrative claims data alone can't. Over the course of the study, 43 patients who had been coded as lower-severity COPD — 28 of them as GOLD B — were reclassified to GOLD E, the highest-risk category, once their actual clinical trajectory could be matched against their records. The program didn't make these patients sicker. It revealed how sick they had been all along.
That kind of accurate severity classification matters well beyond a single study. It changes how health systems plan capacity, how payers price risk, and how clinicians prioritize their attention.
Why this matters now
The United States spends close to $50 billion a year managing COPD, and most of that spending lands in the hospital — on admissions and readmissions that, in many cases, were preventable. The structural problem is that traditional care is episodic. Patients are visible to their clinicians during an appointment and effectively invisible the rest of the time. For a condition where deterioration unfolds gradually, that's a dangerous design.
iCARE points toward a different design. Continuous monitoring backed by predictive intelligence acts as a kind of clinical check-engine light — surfacing problems early enough that small interventions can replace large ones. The combination of connected devices, behavioral nudging, and precision escalation appears to produce both better outcomes and lower cost, at a scale and over a duration long enough to take seriously.
What's next
Intermountain Health is now exploring how to scale iCARE across its broader network of 33 hospitals and 385 clinics, which serve patients across Utah, Idaho, Colorado, Wyoming, Montana, and Nevada. Final 24-month outcomes will be reported when data collection closes in March 2026. The numbers cited above reflect patients enrolled for at least 90 days, with a median follow-up of 358 days.
For other health systems, payers, and digital health stakeholders, the iCARE program offers a model that has now been validated in a real-world cohort over two years — the longest digital-health AI connected-care study in respiratory medicine to date. The takeaway is straightforward: closing the gap in chronic disease management, using a combination of smart devices, precision nudging, and a digitally aware clinical team, is both clinically achievable and economically compelling.