A.I. Shows Promise of Reducing Length of Hospital Stay—a Growing Heath Care Concern

At several U.S. hospitals, A.I.-powered electronic health record (EHR) systems helped reduce the average patient's length of stay by days.

While going to the hospital is, in most cases, a matter of necessity, staying there doesn’t have to be. Reducing the length of a patient’s stay in a hospital is a growing concern in the health care space, both for the benefit of the patient and the health care provider’s financial wellbeing. Artificial intelligence-powered systems in hospitals, it seems, may be helping drive reductions in the length of stay.

Concord Hospital, a regional health care system based in New Hampshire, uses an A.I.-powered electronic health record (EHR) system called Wellsheet. The hospital found the EHR’s A.I.-powered user interface has contributed to a 1.25-day decrease in the average length of stay for patients. New Jersey’s RWJBarnabas Health, using the same A.I. system, also reported a reduction in the length of stay. HonorHealth, a non-profit health system in Arizona, found similar results with an A.I.-powered EHR system, called Qventus. Over a three-year period ended in the first quarter of 2024, HonorHealth found that 86 percent of its patients received early discharge.

“Your length of stay will define your capacity, which will basically define your ability to provide access to the community,” Christopher Fore, MD, emergency medicine and chief quality officer at Concord Hospital, told Observer.

During the Covid-19 pandemic, Concord Hospital saw its average patient’s length of stay balloon from about 4.8 days to six or seven days. While Fore said that there were a lot of uncontrollable factors contributing to that (including the pandemic itself, which increased the average length of stay at hospitals around the country), he admitted, “We knew that a lot of it was actually just us. It was people, process and technology.” Using an A.I. system that streamlines all three of those things by optimizing communication, he said, was the secret sauce.

Wellsheet CEO Craig Limoli, who entered the health tech space around 2015 after working at IBM’s Watson Health division, said the fact that Wellsheet presents full access to patient record in a highly contextualized way—specific to each end user—plays a crucial role in its success. But decreasing length of stay is just one of A.I.’s aspirations for health care systems.

A.I.’s possible side effects

Dr. Josh Fessel, director of the office of translational medicine at the National Institutes of Health, is enthusiastic about A.I. but remains skeptical about the technology solving problems in an environment as multifaceted as health care. A.I. may be optimized for certain areas but come at the expense of unanticipated side effects, he warned. “On the surface, it sounds really good,” Fessel told Observer. “[But] what if it turns out that hospital length of stay is reduced […] but readmission rates increased by some unacceptable factor? A.I. doesn’t necessarily know to look for that unless you trained it to look for that.”

At Concord Hospital, Fore said it’s not the technology working in a vacuum, but rather the clinicians at the hospital using the A.I.-powered system in a way that elevates their performance. “It seems like the secret sauce for us has been teams communicating, identifying: What are the barriers? What are the goals? What information is needed? How do we talk about that? How do we plan for the patient’s departure and how do we create a successful treatment plan?” He said. By giving people more efficient workflows and reduced software burden, teams find themselves in a place to think more critically and make more impactful decisions.

In a 2024 survey of health care decision makers by Sermo, a medical social media and crowdsourcing platform, 23 percent reported that they currently have integrated A.I. at their organizations, and 71 percent forecast A.I. integration in the next five years. Long-term implications of A.I. remain unknown, but there will at least need to be built workarounds for things like degradation of performance over time, for example.

Fortunately, this is something health care professionals are already familiar with. “We already have cognitive models in place for this. That’s why we monitor antibiotic resistance,” said Fessel. “Things do change. It’s just that the timescale may be different for A.I.-enabled tools. We’re still figuring all that out.”

Academic studies on A.I. systems’ impact on the length of hospital stay are not yet prevalent, but other use cases are filtering into the conversation. For example, research suggests implementing generative A.I. text assistance for patient replies in messaging systems decreases provider burnout, but doesn’t actually save providers time in the long run. In this case, it’s up to humans to decide if reduced burnout is valuable enough in and of itself (burnout is notoriously prevalent among health care professionals and, in some cases, can lead to the end of a career—which can ultimately reduce a community’s access to health care).

Additionally, in a study looking at real-time alerts generated by machine learning mechanisms, these alerts led to potential reduction in patient mortality, but had no impact on the incidence of escalating action regarding patients’ health concerns.

Zeroing in on the length of hospital stay feels virtually impossible because of all of the factors simultaneously at play, like changing economic, social and wellness factors across communities, and potential unintended consequences. However, A.I.’s impact on something that could measurably improve a community’s quality of life is worth a conversation.

“When you think about the universe of possible positive impact, it’s actually pretty large,” said Fessel. But, he added, it’s crucial to ask: “Are we really improving the things that we would hope to be improving? […] The people you serve getting better are always your north star, in my opinion.”