AI-Driven Monitoring Cuts COPD, Asthma Hospitalizations by 50%, Study Shows

Jill Bailey spent most of her life in hospitals. She had been intubated nine times. Steroids became her constant companion. The asthma that had shadowed her since childhood kept landing her back in a bed, kept her from work, kept her from the ordinary rhythms of living. Then, two years ago, she enrolled in a study at Intermountain Health that would change the trajectory of her illness—not by curing it, but by catching it before it could spiral.

The study, called iCARE, tested a simple but powerful idea: what if doctors could see the danger coming before patients ended up in crisis? Researchers at Intermountain Health and CareCentra equipped roughly 1,200 patients across five Utah hospitals with connected devices—digital spirometers to measure lung function, pulse oximeters to track blood oxygen, sensor-equipped inhalers to monitor technique and adherence, and wearable fitness trackers. These devices generated 11.5 million data points over two years, feeding an artificial intelligence system that watched for the earliest whispers of deterioration: a slight drop in lung capacity, a dip in oxygen saturation, a pattern of missed doses, changes in breathing or sleep.

The numbers from the two-year study are striking. Hospitalizations dropped by half. Emergency department visits fell by 20 percent. The total cost of care declined by 57 percent. For context, COPD and asthma cost the United States more than $50 billion annually to treat, much of that spent on preventable hospital stays. The researchers presented their findings at the American Thoracic Society's 2026 International Conference in Orlando, calling the results a fundamental shift in how chronic lung disease could be managed.

The innovation was not the devices themselves but what happened when the AI detected risk. The system did not wait for patients to call their doctors or for symptoms to become emergencies. Instead, it deployed what researchers call "nudge theory"—personalized prompts and behavioral coaching delivered directly to patients. When the AI detected missed doses, it addressed adherence barriers. When environmental triggers spiked, it sent real-time warnings. When multiple risk signals converged, the system escalated the case to a pulmonary disease navigator, a registered respiratory therapist trained to coordinate care across the patient's entire clinical team. This was not a call-center triage. It was precision intervention, triggered by data, delivered with full clinical context.

The efficiency gains were as remarkable as the clinical ones. Before iCARE, each navigator managed about 30 patients. With AI handling routine monitoring and flagging only the highest-risk cases, a single navigator now oversees nearly 220 patients—a sevenfold increase in capacity without sacrificing the human judgment that respiratory care demands. Peter Crossno, the principal investigator and senior medical director of respiratory care at Intermountain Health, described the system as a shift from reactive to proactive medicine. "We demonstrated that it is possible to gather signals of patient risk continuously between clinic visits, shape their response behaviors through personalized nudges, and intervene, when necessary, through escalations to our pulmonary disease navigators," he said.

For Bailey, the change was tangible. She no longer lives on steroids. She no longer spends her life at the doctor's office. "I've been in the hospital more than I've been out," she reflected on her life before the study. "I was at the doctor's more than I was home. It was a lot of expense, a lot of time, a lot of energy, and we just never could stay on top of it." Now, she said, "I feel so much better today."

Intermountain Health operates 34 hospitals and 400 clinics across Utah, Idaho, Colorado, Wyoming, Montana, and Nevada. The system is now exploring how to scale iCARE across that entire network, potentially transforming care for millions of patients with chronic pulmonary conditions. The question is no longer whether continuous AI-driven monitoring works. The question is how quickly it can be deployed.