The tough part of predictive analytics is making the information actionable, says Michael Kanter, M.D., executive vice president of quality and chief quality officer of the Oakland, Calif.-based Permanente Federation, an arm of integrated health system giant Kaiser Permanente.
“There are a lot of people running around with predictive algorithms, but the question is: 'What are you going to do with that?'“ Kanter says.
For instance, a hot area for predictive analytics is hospital readmissions. If hospitals know which patients are most likely to be readmitted, they can place resources such as care coordinators and extensive outpatient services with that minority of patients. This is often measured using the LACE index scoring tool, which broadly stands for length of stay, acuity, comorbidities and emergency department visits [See below].
It's a new type of predictive gauge, one that shows which patients need intensive services so they don’t wind up in the ED in the next 60 to 90 days.
Being able to target patients more efficiently is very appealing, says Jose Azar, M.D., medical director for quality improvement and patient safety for Indiana University Health, Indianapolis. “We have 750 patients every day,“ he says. “Instead of looking at everybody, if we can look at 20 patients, that would be a great advantage."
Some caution that the quality of predictions are only as good as the underlying data, which can be extremely limited when it comes to social determinants of health and other underlying reasons for complex cases.
“You have to have reliable data to make predictions about patients,“ says Mark Jacobs, chief information officer at the Delaware Heath Information Network, the oldest statewide health information exchange in the country.
Delaware hospitals, for instance, exchange patient data securely through the DHIN. “So when I look at predictive analytics, I think, 'What are the data you are going to start with to make predictions about the patients?'" Jacobs says. “Most organizations have clinical data and most have claims and administrative data. Most don’t have information about social determinants and lifestyle and most don’t have the ability to combine all that data."
Christiana Care Health System, a nonprofit in Wilmington, Del., has been doing predictive analytics for five years. Called Christiana Care Link, the system is homegrown and started with a $10 million grant from the Centers for Medicare and Medicaid Innovation Center in 2012. Officials at Christiana Care were reluctant to share their “secret sauce“ on how their predictive analytics system works and what results it has garnered, except to say that they have been able to move the needle on a number of utilization and financial metrics.
Terri Steinberg, M.D., chief health information officer at Christiana Care, cautioned other health systems that they should have the “human to human“ resources in place to respond to predictions. These resources include care managers, bedside nurses, care coordinators, social workers and even appointment clerical staff to respond to whatever is predicted to happen to an individual patient.
“You need to be able to respond to and receive information in real time,“ Steinberg says. “That’s the cost of entry. Without a robust care management program, there’s no point.”
Calculating risk is science, while “making it actionable is magic,“ Steinberg adds. “It's finding Waldo and then finding what Waldo needs."
Small hospitals, too, are testing predictive analytics tools to improve patient outcomes.
Northwestern Medical Center, a 70-bed facility in Saint Albans, Vt., is launching a 30-day hospital readmissions predictive analytics program in 2017.
Leading the project is Chris Giroux, manager of data management and integration services, who is completing a master’s degree program in predictive analytics at Northwestern University in Evanston, Ill.
Small hospitals can participate in predictive analytics if they have the foundational pieces, Giroux says. These include an electronic health record system, an integrated data system, a data warehouse and active data governance to clean up and maintain good data.
With all those components in place, Northwestern Medical Center will launch the use of the LACE index scoring tool to identify patients at risk for readmission or death within 30 days of discharge. The LACE index comprises four parameters: length of stay; acuity; comorbidities; and number of emergency department visits. A patient’s LACE score (risk for readmission) ranges from 1 to 19; the higher the score, the most at risk a patient is.
Some studies have shown that implementing the LACE index can result in a moderate to high reduction of 30-day readmissions.
Kaiser Permanente of Oakland, Calif., implemented the LACE index and has seen positive results, says Michael Kanter, M.D., executive vice president of quality and chief quality officer at the Permanente Federation. The system pulls data on patients from the EHR to make the LACE calculation and, ultimately, drive clinical decision support for each hospitalized patient, he says.
About a year ago, Northwestern Medical Center’s community board of directors approved the creation of a division within the information technology department to develop data initiatives including predictive analytics and a user-friendly data dashboard.
“We have a really supportive board and leadership team,“ Giroux says. “That's how we've gotten this far. You give them data and they want more."
Giroux says the LACE tool is just the beginning of how her organization will apply predictive analytics. “I believe in the next five to 10 years, predictive analytics will be a necessary component in health care,“ she says. “It is going to help contain costs and produce better patient outcomes."