On a Sunday evening in March 2020, Matthew Metsker, division director of Virginia Mason Franciscan Health’s mission control and virtual health services, was winding down at home when he got a call from the health system’s disaster preparedness manager. The first round of COVID-19 patients had begun to seek care at the Seattle-based organization, and they were coming in faster than expected. Some would need negative air pressure rooms, which seal off infectious individuals from the rest of the hospital.
Metsker booted up the health system’s Mission Control Command Center remotely, giving him full access to its number of sterilized negative air pressure rooms.
“I just pulled up the analytic tiles, and I said we have 42 rooms,” he said. “And I could tell if they were clean or dirty.”
The command center’s “tiles”—screens that include a panoramic view of bed availability throughout Virginia Mason Franciscan Health—allowed decision-makers to strategize about where to place the incoming COVID-19 patients without having to call each hospital individually.
The center is among the many data-driven tools health systems across the U.S. have used to address operations challenges exacerbated by the pandemic. By drawing information from electronic health records, employee schedules and other sources, analytics technology can offer real-time and predictive insight into patient numbers and staffing needs.
The logistical snarls the tools help untangle are not limited to the public health emergency. Cross-team collaboration, executive support and long-term monitoring will be vital for provider organizations seeking to harness their data effectively.
“Operations translate into clinical outcomes,” Metsker said. “If we don’t prepare well, patients are going to experience delays in care and suboptimal services.”
A team effort
Deploying analytics tools is a calculated process. If it’s not guided by a three-way relationship between data scientists who know the numbers, software engineers who know the technology and clinicians who know the medicine, “you’re going to solve the wrong problem,” said Dr. Ford Parsons, medical director of clinical analytics at Renton, Washington-based Providence. The system uses analytics software to streamline workflows and minimize overcrowding.
“You need that full spectrum. And these people need to be friends with each other,” he said. “They need to work as a team. They can’t just work in silos. They have to learn how to speak each other’s languages.”
Providence makes this possible by assembling teams of data and clinical experts who share information throughout the technology development process. In addition, the health system pooled front-line providers, data scientists and clinical leaders into a cross-departmental steering committee to raise awareness about ongoing projects among individuals who might not otherwise interact.
“We are at a very, very pivotal moment in healthcare. Data analytics was always important, but COVID brought to the forefront, for us, data-driven decision-making.”
Aashima Gupta, director of global healthcare solutions at Google Cloud
Sometimes, the collaboration can extend to third parties, such as Big Tech companies. The multibillion-dollar analytics industry has never been more in demand, according to Aashima Gupta, director of global healthcare solutions at Google Cloud.
“We are at a very, very pivotal moment in healthcare,” she said. “Data analytics was always important, but COVID brought to the forefront, for us, data-driven decision-making.”
The pandemic was what drove Providence to adopt cloud-based data warehousing through Snowflake and to partner with Microsoft, which allowed the health system to tailor analytics software to its needs.
One tool conveys information about how many patients are in hospitals’ emergency departments. Another uses patient vital signs to determine if a person with COVID-19 should be admitted, based on the likelihood they’ll require a BiPap machine or intubation. A third aims to predict whether patients will be back within 30 days of discharge.
To ensure the models are functioning at their best, Parsons said Providence runs them against those developed by competitors, in head-to-head tests.
Because bed space is always an issue, anticipating patient flow remains a top priority, Parsons said. Doing so using analytics enables Providence to bring on more caregivers ahead of potential surges.
When looking for a vendor, Parsons said healthcare clients must be specific with what they want from a technology company, as no two offer the same services. Providence’s interest in long-term innovation over “turnkey solutions” was the deciding factor for working with Microsoft, he said.
“I think there’s some vendors that want to sell a hospital on a solution and other vendors who want to sell a hospital on technologies that can be used to create a solution,” he said. “I find it more helpful to partner with technology companies where we bring our clinical expertise and they bring their knowledge of the stack and their ability to advise us on how to use their tools to solve our own problems.”
Virginia Mason Franciscan Health partnered with GE Healthcare in 2019 to launch its command center, which tracks bed space, patient transfers and supplies like ventilators and dialysis machines. During the pandemic, Metsker said the center was largely responsible for keeping the system’s COVID-19 patient intake under control.
“We are in lots of communities, and lots of patients required care. Having our eight sites and being able to manage everything at once—balancing the resources, the ventilators, the negative air pressure rooms—was a net asset to the community,” he said.
As the highly transmissible COVID-19 variant BA.5 spreads across the country, Metsker said the health system plans to develop a model to forecast labor shortages.
“We actually don’t have a ton of hospitalizations right now, but our staff are affected,” he said. “That’s what we should potentially model. With this next surge, how many staff members are we not going to have in place because they’re sick or they’re out pending a test?”
Analytics projects require early buy-in from the C-suite.
At Pittsburgh-based UPMC, the investment came more than eight years ago, according to Dr. Oscar Marroquin, the organization’s chief healthcare data and analytics officer.
By allocating capital, staff and resources, UPMC executives made it possible for the health system’s data analytics program to build a 50-person team of quantum computing physicists, epidemiologists, data scientists and nurses who code, Marroquin said. The team developed an arsenal of 12 predictive models over nearly a decade.
“Right now, our analytics are geared to be the sort of engine that enables all levels of our organization—from the CEO, to the C-suite, to the hospital presidents, to clinician leaders—to be the ones asking questions.”
Dr. Oscar Marroquin, chief healthcare data and analytics officer at UPMC
One model allows the 40-hospital system to flag patients at risk of missing their appointments. It gives caregivers added insight into patient attendance, so they can decide to follow up with individuals and help arrange transportation to prevent no-shows. Doing so has reduced wasted time, which is vital when handling more than 6 million outpatient visits a year.
“Right now, our analytics are geared to be the sort of engine that enables all levels of our organization—from the CEO, to the C-suite, to the hospital presidents, to clinician leaders—to be the ones asking questions,” Marroquin said.
The team tries to make answers to such questions readily available. In July, UPMC announced a new partnership with Microsoft to boost its analytics capabilities.
“We want to build self-service tools that people can use to get data on their own,” he said.
Executives at Sioux Falls, South Dakota-based Sanford Health sit on the health system’s artificial intelligence oversight committee and are directly engaged in the decision-making process about initiatives, said Douglas Nowak, vice president of enterprise data and analytics. Their involvement holds everyone accountable to get the job done, he said.
“You’ve got to have that support from the top,” Nowak said. “I have talked to other health systems that have tried to build it from the bottom, and they’ve usually failed.”
Sanford’s 62-person analytics team relied on its experience from the past seven years to create operations models in response to the pandemic within just a few days, Nowak said.
Using harmonized data sets sourced from the EHR, schedules and inventories, the team created ways to track vaccine distribution, along with personal protective equipment and staffing needs.
Sanford uses a shared governance framework that alerts leadership to operations exigencies or patient risk factors, allowing for early intervention.
“The quality of applicants that I get today, compared to even 10 years ago, is night and day. These are young people that want data and they want access.”
Douglas Nowak, vice president of enterprise analytics at Sanford Health
The rural health system is focused on the relationship between patient flow and staffing levels amid the nationwide nursing shortage. Nowak said the analytics department deployed a model to determine labor allocations based on bed utilization.
“The idea is getting the right person at the right place at the right time,” he said.
Nowak added that the health system monitors indicators like staff turnover and overtime, along with responses from employee surveys, to measure the success of their operations models.
“If you can keep an employee who’s much more fulfilled, enjoys coming to work, has fun at work—they’ll be happier and our patients will have better outcomes,” he said.
Avoiding the ‘black box’
The technology would be obsolete if it weren’t for constant monitoring of efficacy, reliability and bias. Predictive models can be less accurate for vulnerable populations and can exacerbate existing health disparities. That’s why it’s important for developers to understand every model in totality—and to grasp that analytics tools are just aids, with no autonomous decision-making capabilities, Nowak said.
The technology “is giving a provider one more tool in their toolbox,” he said.
“When you get into these algorithms that are being developed, you hear about the infamous ‘black box,’ ” he added. “It’ll spit out some predictions and you say, ‘Well, that’s what the computer said.’ ”
To prevent that from occurring, hospitals subject their models to a variety of stress tests.
Metsker said hospitals must also use sample sizes that reflect the populations to which they’re applying predictive models. Otherwise, the models will carry a bias.
“If you’re talking about a clinical model, those variables have to be validated with a reasonable population. A hospital in the middle of Chicago has a different population than a hospital in Gig Harbor, Washington,” he said. “On the operations side, you have your different set of variables [depending on location]—volumes, services, geography and finance.”
Once the models are deployed, analytics leaders say they are checked in real time for accuracy and compared with historical data to track longitudinal efficacy.
“I always remind people of how we go about doing this,” Marroquin said. Any decision to use analytics tools, he said, “is only born from the need that an organization has.”
The COVID-19 pandemic became a proving ground for health systems and Big Tech companies to crank out advanced technology in unprecedentedly short time frames.
According to Dr. David Rhew, Microsoft’s global chief medical officer and vice president of healthcare, the focus should be on expanding tools for further use cases.
“We’re seeing efficiencies gained, we’re seeing provider experiences gained and we’re seeing patient experiences improved. So that’s become an interesting element of this, because we’re no longer talking about just trying to solve one thing,” he said. “These technologies can help us address so many other aspects.”
As an example, Rhew said Microsoft has been working in the natural language processing space to develop tools that track conversations in the exam room and bundle them into a patient’s EHR. This can reduce the amount of time doctors spend taking notes, Rhew said, as well as the burnout they may experience from juggling multiple tasks at once.
Health systems are also sowing the seeds for future innovation by partnering with colleges, in the hopes students will settle into data-driven healthcare careers.
For example, Virginia Mason and Sanford joined forces with their local universities—University of Washington Tacoma and Dakota State University, respectively—to recruit the next generation of clinical analysts, data scientists and information technology specialists straight from the classroom.
At Virginia Mason, four students from the UW Tacoma’s master’s of science in business analytics program are working at the system’s Mission Control Command Center, Metsker said. They’re tasked with creating models in the programming languages R or Python to identify ancillary service delays and make staffing recommendations.
And at Dakota State University, Nowak said business and technology students can participate in different healthcare-cybertechnology programs through Sanford’s pipeline partnership with the school. The program can attract students to Sanford before another employer catches their eye, he said.
“The quality of applicants that I get today, compared to even 10 years ago, is night and day. These are young people that want data and they want access,” Nowak said. “It’s amazing, when given that access and a little bit of free rein, what they can do.”