What if hospitals could predict which heart failure patients will most likely be readmitted? Which moms are more likely to hemorrhage during labor and delivery, or which home care patients absolutely need a visit on an icy day? With the growing use of hospital data and predictive technology, such as advanced artificial intelligence, the technology is already out there.
Here’s how three healthcare systems jump-started their predictive analytics programs, how they used lessons learned from early rollouts, and how they made sure that everyone along the pipeline of patient care bought into their vision.
1. Tap Clinical Perspectives to Enhance Predictive Analytics in Healthcare
Advocate Aurora Health, which serves patients in Illinois and Wisconsin, uses predictive analytics to identify heart failure patients who are most likely to have another acute event shortly after discharge. Those patients are prescribed an aggressive readmission prevention solution. So far, the health system has engaged more than 350 patients and found 23 percent fewer acute encounters in the treatment group. The heart program has been so successful that it is now out of pilot stage and is being used in standard care. Advocate Aurora Health is also about to roll out a similar predictive analysis model for patients with chronic obstructive pulmonary disease in the next month.
As crucial as it was to get the predictive analytics right, Tina Esposito, chief health information officer at Advocate Aurora Health, says the result is what needs to be top of mind for her team.
“Data structure, at the end of the day, is a means to an end,” she says. “The end isn’t creating a data analytics group. The end is whatever strategy or place your organization sees itself in.”
Engaging clinicians from start to finish has been key to their success, adds Fran Wilk, clinical process designer for Advocate Aurora Health, who serves as a clinical resource for the team, creating the workflow and educational rollouts. She’s also been a nurse for 24 years. “We needed to have a lot of scientists on our team, but you also need the clinician’s perspective,” she said.
2. Keep Healthcare’s Core Mission at the Forefront of Predictive Analytics Programs
Texas Health Resources’ predictive analytics program has focused on areas that most affect their patients, starting with using data to create a modified early warning system, or MEWS.
MEWS identifies patients who might be in distress but don’t show an obvious sign like a sudden drop in blood pressure or spike in temperature.
“Sometimes it’s a more subtle decline and not readily detectable,” says Dr. Ferdinand Velasco, Texas Health’s chief health information officer. “When there’s a concern about a MEWS score, that’s a trigger for a nurse or physician to see the patient at bedside and make sure everything’s OK.”
MEWS has been used successfully in patient care for seven years, Velasco says. Now, Texas Health Resources is about to roll out two predictive models aimed at decreasing maternal mortality and morbidity, a critical need in the U.S., which has one of the highest maternal death rates among the world’s developed nations, according to a USA Today report. “That article was a call to action,” Velasco says.
One predictive model is a version of MEWS specific to women in labor, as their vital signs will differ from other kinds of patient during that time. The second is applied to mothers just about to deliver, and identifies those who are at risk for hemorrhage. Those with an elevated risk have one or two IVs or catheters put in place for a potential blood transfusion (since those are larger than a typical IV or catheter), and the blood bank prematches blood type in advance.
While Velasco follows developments in predictive analysis, he said that at Texas Health Resources, they are more conservative with their approach because they want to use tried and true methods that are going to save people’s lives immediately. “It’s not a space to be on the cutting edge because we’re dealing with patients’ lives. It’s not something to fool around with,” he says.
3. Give Practitioners Data Literacy Training to Promote the Best Use of Predictive Modeling
As accurate and helpful as predictive modeling can be in healthcare, it’s not going to help patients unless it’s adopted by practitioners. That’s why Betsy McVay, vice president and chief analytics officer at UnityPoint Health in Iowa, focuses on what she calls data literacy. It’s crucial to have that “doctor or operational leader who talks to their peers about how important it is to start looking at data a different way and help with change management,” she says.
“The ‘build it and they will come’ bit, it will fail in this space if you don’t engage people as a true partner and collaborator through the process,” she says. “Data literacy is a huge focus for us right now and will continue to be for a while.”
UnityPoint also found, by working with end users, that the algorithms were solving problems they didn’t anticipate. They’ve used predictive analytics to identify who was at risk for readmission within 30 days of discharge, but the model unexpectedly helped the home care branch of their practice.
“Winter weather in Iowa gets a little iffy. We used this model to prioritize who we went out and saw on days when we were short-staffed because the weather was bad,” she says.