When AI Saves Lives

When CHI Franciscan launched its Mission Control Center, it became the first health system in Washington state and only the fifth globally to use artificial intelligence and advanced analytics to coordinate patient care across its entire system.

Four years later, the NASA-inspired effort — created in collaboration with GE Healthcare — is credited with shortening hospital stays, reducing wait times in the ER, and coordinating care so efficiently that the system now has the capacity to serve an extra 1,000 patients per year.

CHI merged with Virginia Mason in early 2021 and is now called Virginia Mason Franciscan Health, a 10-hospital system with about 300 care sites across the Puget Sound region.

“When someone’s in a bed longer than they might need to be, it’s keeping somebody else out of the hospital,” says James Terwilliger, chief operating officer of Virginia Mason Franciscan Health (VMFH). “We also know that the longer you spend time in the hospital, the more potential there is for complications.”

Having a bed available to a patient when needed is a complex job affected by schedules, staffing shortages, medicine orders, the availability of diagnostic tests, and a multitude of other factors. It all has to line up. Until recently, it was often handled by decidedly low-tech tools such as whiteboards, dry erase markers, and lots of phone calls, texts, and pages from one hospital to another, one floor to the next.

Using predictive analytics and artificial intelligence, GE Health- Care’s Command Center software pulls together information from disparate electronic medical records as well as data often siloed in different systems and records, such as radiology, bed management, and nurse and physician scheduling. It crunches that data and delivers it in 30-second intervals.

Picture a huge screen at an offsite location, numbers flashing and constantly updating, telling clinicians where in the system a bed might be available, or where diagnostic tests are creating a backlog. A bed-matching algorithm finds beds to match incoming demand to available beds, and flags delays in the discharge process that can lead to long waits in the emergency room.

It uses machine learning to constantly improve and makes predictions in census and staffing as far as 48 hours out.

Real-time data and predictive analytics track each patient’s hospital journey, including identifying delays or bottlenecks, reducing wait time, and prioritizing tests and procedures. For example, if a patient needs a CT scan before being discharged, Mission Control will flag the patient to expedite the scan.

Officials couldn’t have known it back in 2019, but the system also proved indispensable when the pandemic hit. Negative air pressure rooms, which suck air out of a hospital room to prevent the spread of infection, were one of the first big-demand resources. Problem was, until the Mission Control center was established, few people knew where they were or how many were available at any given moment.

“It was around 7 or 8 o’clock on March 20, 2020,” recalls Matthew Metsker, division director of the Clinical Command Center, “and everyone started getting on their phones to call their people asking, ‘Do you know how many of these rooms we have?’ All I had to do was pull up one of these tiles,” he adds, pointing to the screen on the wall. “We have exactly 48.”

Many hospitals were overrun with Covid patients, and many were forced to refer patients to the Washington Medical Coordination Center, which was set up to take patients who could not be accommodated at their local hospital. But partly because of the information that was available to clinicians in almost real time, CHI Franciscan (which deployed the system before its merger with Virginia Mason) was able to meet the demand for beds throughout the Covid crisis. System officials never had to call into the coordination center.

“In fact, we were able to be a resource to others,” Terwilliger says. “It wasn’t just a matter of having the beds, but also of having the information, of knowing what is available at any given moment.”

Another unseen benefit was the system’s ability to coordinate care after Covid dramatically increased an already troubling staffing shortage. Recent reports show that many Seattle-area hospitals are today operating at 120% to 130% capacity.

The crunch was especially acute last fall when long waits at Seattle Children’s emergency room made headlines, while emergency room clinicians at St. Michael Medical Center in Silverdale resorted to calling firefighters to the ER to help.

While big data and AI cannot solve staffing shortages or prevent a spike in emergency room visits, it has moved the needle in other ways.

Terwilliger points to a full-day reduction in the average length of stay to 4.7, a 10% system-wide improvement in the time it takes to move a patient from the ER to a room; and a 20% reduction in lost cases, meaning patients transferred from other hospitals that the system had lost due to an apparent lack of beds or other resources.

Real time, actionable data have allowed VMFH to, in effect, create 4,380 additional patient days annually, which in hospital terminology means the ability to take in and care for 973 additional patients a year. More patients in the door mean more surgeries and more testing, which in turn lead to still more admissions.

“It saves money on the cost side, and it creates the opportunity on the revenue side,” Terwilliger notes. The extra revenue is crucial at a time when Washington state hospitals are grappling with financial losses that collectively amounted to $2.7 billion last year, according to the Washington State Hospital Association.

Equally important, he adds, is “that’s 900-some patients that wouldn’t have gotten the same level of great care. “We are at the point where we can accept 90% of the patients who present to us. It’s another example of being more accessible to the community.”

Such efficiencies are a godsend for hospitals that can afford sophisticated data analytics. Smaller, rural hospitals, however, still find themselves at a disadvantage. Across the country, 631 rural hospitals, about 30% of all rural hospitals, were at risk of closing due to financial losses, according to the Center for Healthcare Quality and Payment Reform. In 2022, the group estimated that Washington was at risk of losing 13 rural hospitals, or 33% of the total.

Virginia Mason is no longer the only hospital in the region to harness big data to run operations this way. Nor has AI solved other challenges facing VMFH, such as the recent bacteria outbreak that has infected 31 patients and killed seven.

As recently as last fall, the hospital system faced crisis-level bottlenecks and delays at St Michael’s, leading Kitsap County residents to start a petition of no confidence in St. Michael leaders. Longtime patients were reconsidering St. Michael as their go-to place for care, according to The Seattle Times. VMFH officials say they have worked hard to reverse the issue, and thanks to Mission Control and other measures, the weekly median arrival-to-triage time at St. Michael has decreased from 15 minutes in December to four minutes or less in April.

GE HealthCare, however, says the efficiency created by data analytics and machine learning is worth the investment for hospitals, payers, and patients. Across 25 health systems and 300 hospitals nationwide, where the command center software is now used, hospitals have reported big gains.

Tampa General Hospital cited a half- to full-day reduction in lengths of stay to 5.5 days; Johns Hopkins Medicine in Baltimore reported a 40% to 70% reduction in patient wait times in the emergency department and operating rooms. The system can spot and correct imbalances in staffing, send patients to sister hospitals where there may be more availability, and importantly, allow clinicians to focus on their patients rather than hunting down information.

“A huge benefit is the time it saves caregivers,” says Jeff Terry, who founded and leads the Clinical Command Center group at GE HealthCare. “We are giving caregivers back their time because instead of working till midnight to get a report out or finishing a shift report. The information is already there.”