Central lines are a fact of life across the better part of Indiana University Health University Hospital as they are in most hospitals. Out of a total of 605,000 inpatient days in 2016, a quarter of those days included a central line. That’s not a difficult concept to understand.
But what executives at IU Health, Indianapolis, couldn’t accept as a fact of life is that bloodstream infections are an inevitable side effect of those many central lines. After tackling the issue for a time in a more traditional fashion, IU Health officials decided that something new was needed to prevent such infections, known as CLABSIs, at the 15-hospital nonprofit system.
“CLABSI is a tough animal,” says Kristen Kelley, director of infection prevention. “A lot of preventing CLABSIs today is human behavior.” People tend to not keep up with regular tasks that have no direct, immediate effect. “The safety bundles in particular — you have to work at them constantly,“ she says. “Like a diet, you can’t try it for one week and hope the behavior change sticks after that."
As a result, IU Health leaders are stepping up the fight against CLABSI by embracing the fast-growing tool known as predictive analytics. The goal: Rather than treating the infection once it occurs, prevent it from occuring in the first place.
IU Health launched its predictive analytics pilot in CLABSIs at University Hospital on its main campus, which includes a 600-bed, Level I trauma center and 300-bed tertiary care center, and is one of the 10 largest transplant centers in the nation. Kelley says the hope is that its real-time predictive analytics tool can identify which patients are most likely to develop such an infection, enabling clinicians to intervene earlier.
Predictive analytics is an exploding area of data science, one that hospitals hope will help them improve patient care and community health in a broad range of clinical, administrative and financial arenas. Nearly 80 percent of hospital executives said they believe health care could be improved significantly with the use of predictive analytics, according to an August 2016 survey by Health Catalyst, the company that supplies IU Health with its analytics management software. Thirty-one percent of hospitals have used the technology for more than one year, the survey of 136 executives showed. Thirty-eight percent of respondents said they plan to adopt predictive analytics in the next three years.
Meanwhile, keeping staff on task with CLABSI prevention is daunting. IU Health University Hospital has 3,000 registered nurses working at its downtown facility alone. “In a complex academic medical center where you have staffing and acuity fluctuations and volume boluses, we need to go back to what the basic data are telling us,” Kelley says.
IU Health didn’t start with predictive analytics as a goal; it is an outgrowth of a process that began long before it launched the predictions pilot. While IU Health has had an electronic health record system for many years, the EHR has grown tremendously over the past seven years, which requires streamlining data across platforms and locations, says Tony Pastorino, director of decision support. “The biggest piece of technology is a data warehouse that has been in place for about two and a half years,“ he says.
IU Health also uses an e-surveillance program for hospital-acquired infections, which has been in place since 2008.
But the combination of EHR and e-surveillance data limited monitoring of CLABSI retrospectively, after patients are diagnosed with the infection. IU Health conducts a deep-dive review of every hospital-acquired infection. A multidisciplinary team meets with the unit providers where the infection occurred to conduct a retrospective on what went wrong, with the aim of improving performance next time.
But “next time” is too late in today’s age of stiff penalties for failure. “We felt we were constantly behind the ball,“ says Doug Webb, M.D., medical director for infection control at IU Health.
Keeping a step ahead of CLABSI
In 2016, IU Health launched a data visualization platform to allow providers on all units to see data in real time, in a usable and easy-to-comprehend format. It was a welcome transition away from Excel spreadsheets and other more labor-intensive ways of tracking data. Previously, nurse managers had to rely on staff to gather information, then check for accuracy and compile the data into spreadsheets. The hospital had five different reports in five different online locations, Kelley says.
“Now the system tells me red or green — if the trend lines are going up or down,“ she says. “I don’t have to think about it all day."
With the introduction of the dashboard, unit nurse managers and bedside nurses can, for example, see which hospital units have missed line maintenance activities and failed to complete CLABSI-prevention bundles.
The dashboard also enables care providers to more easily track how long a central line has been in a patient. The number of line days is an important predictor of infection, according to studies, and it is one factor being used in IU Health’s predictive analytics pilot.
“The majority of our infections are due to maintenance; these are long-term lines,“ Kelley says. Other identified risk factors include length of hospitalization (even without a central line), use of total parenteral nutrition, and a low white blood cell count.
An estimated 250,000 CLABSIs occur in the U.S. annually across care settings, according to the Centers for Disease Control and Prevention. Patient mortality rates associated with CLABSI range from 12 to 25 percent and the cost of the infection ranges from $3,700 to $36,000 per episode, according to the CDC.
Before the adoption of predictive analytics techniques, IU Health had implemented evidence-based, central-line practices, including safety bundles, checklists, line-insertion training and hand-hygiene rules. Those efforts succeeded in cutting its rate to 1.2 CLABSIs over central-line days in 2016 from a rate of 1.7 in 2015. Nevertheless, it was one of 769 hospitals nationwide that had their Medicare payments lowered by 1 percent for discharges because of low performance on hospital-acquired infections, according to data released by the Centers for Medicare & Medicaid Services for fiscal 2017.
The potential of predictive analytics to solve some of the most entrenched quality-, cost- and resource-intensive problems is driving adoption. But experts caution that predictions are only as good as the underlying data and that hospitals need the resources to respond to predicted outcomes.
“It’s very seductive,” says Michael Kanter, M.D., executive vice president of quality and chief quality officer of the Oakland, Calif.-based Permanente Federation, whose parent Kaiser Permanente has implemented predictive analytics in a number of areas. “It holds huge promise. But it’s one thing to predict the future and a whole other thing to change it.”
Implementing predictive analytics in an actionable way is one of the issues IU Health is working out now. The hospital has partnered with the transplant and trauma services lines on the CLABSI project and recruited physician leaders. The hospital is considering whether to launch the project with a small team, or merge the patients with the highest risk into the physician rounding.
“Maybe it will be a mix of both,“ says Kelley. “We are in a resource-limited environment. We have to get the most out of the tools that we have." — Rebecca Vesely is a freelance writer based in San Francisco. •
Executive Corner: How leadership can usher in predictive analytics
Establishing an effective predictive analytics program requires leadership from all areas, from the board to the C-suite to the clinical experts. Here’s advice for all three.
Trustees: Governance support is needed to implement robust data analytics and allocate the required resources to change predicted patient outcomes.
Hospital executives: Executives today have the ability to see trends in patient outcomes more clearly thanks to better data and tools that present the data in a more user-friendly format. Trend lines can be seen in close to real time. The challenge for executives is supporting data validation efforts, and allocating resources appropriately. This might mean adding more staff initially, including care coordinators, nurses and home health workers, to respond to predictions in patient health.
Physicians: Integrating predictive analytics into workflows can be difficult. Providers should conduct small tests of change to see what works best in terms of implementation. This could require extra training and changes to daily practices, such as rounding. If providers don't act on the prediction, patient outcomes won't change.