The most award winning
healthcare information source.
TRUSTED FOR FOUR DECADES.
There is increasing buzz around advanced analytics and artificial intelligence (AI) in healthcare, with data analysis becoming critical in addressing quality of care and efficiency within a hospital. A network of hospitals, clinics, and home care services in Iowa, Illinois, and Wisconsin has improved clinical effectiveness, care quality, and patient experience by looking at risk analytics and applying the findings to patient care.
UnityPoint Health, based in West Des Moines, IA, uses advanced analysis to model solutions that address issues such as length of stay, readmissions, and no-shows, says Betsy McVay, vice president and chief analytics officer at UnityPoint Health, a 30,000-employee healthcare system.
McVay leads a team of 60 people involved in data analytics, including everyone from the highly technical side to clinicians. The team has influenced quality of care in a number of ways: One UnityPoint hospital reduced readmissions by 40% using a data analytics solution that includes descriptive and predictive capabilities, and the analytics team contributed to reducing sepsis events that saved 50 lives from June to October 2017.
Data applications also have contributed to saving $60 million through operational improvements.
“A big lesson for us was the value in looking at the people we already had in place and their skill sets, and then seeing how we could retool and retrain as necessary. About four and a half years ago when we started on this journey, we were trying very hard to stay out of the descriptive reporting world and wanted to be in the high-level, advanced, predictive area as quickly as possible,” she says. “Our business was all over the place in terms of our understanding of analytics and what they needed to drive improvement, so it was important for us to bring in things like EPIC clarity reporting because until we created credibility in that space, we were going to have a problem bringing people along with us to something more advanced.”
Rather than the data analytics team determining what is important and sending their analyses downstream, they first try to identify challenges and then serve those needs.
The data analytics team always tries to partner with a business or clinical leader in the space to be analyzed, McVay says, asking them what the hurdles are and what keeps them from being able to reach their quality goals. Once her team understands that, they can start to look at how they can assist with data and analysis.
Coordinating care and controlling costs are top priorities because UnityPoint is part of a Next Generation Accountable Care Organization, notes Christopher Hill, DO, emergency medicine medical director.
“Along with that, we want to ensure that we are always increasing quality and decreasing variation in the care we provide. The challenge, of course, is how you use the plethora of clinical, demographic, and other data to get to that point of influencing patient care decisions at the point of care,” Hill says. “You really have to have best practices, actionable clinical analytics, and the adoption of that data into your care system. The key in many ways is transparency of front-line data delivery to promote healthy comparisons and competitions, using the data to identify the highest-level priorities.”
The most improvement usually comes when clinicians find a way to take high-level data and apply it to very specific processes, Hill notes. For example, the analytics team at UnityPoint health helped identify a hospital in Peoria, IL, as the top performer on patient safety measures and isolate the key factors that led to those good metrics. One of the top factors was a just culture, one in which near misses were reported freely.
The analytics team also tries to provide the level and amount of data required for success — which doesn’t always mean providing a constant stream of the most sophisticated data.
“If providing a descriptive report two times a year is the solution for this group of people or this issue, that’s OK. We don’t need to create something that is more advanced or more robust,” McVay says. “We think of ourselves as an enabling partner bringing a huge skill set, but without our business and clinical leaders helping us understand what they need, we won’t be able to drive improvement effectively.”
UnityPoint has a large team and more resources for AI and advanced analytics than most healthcare systems or hospitals, but McVay says some of the same approaches can still be useful on a smaller scale and with fewer resources.
“Using data and information to drive improvement doesn’t have to mean leveraging AI or something highly advanced and cutting-edge. Sometimes it means creating transparency with the tools and information available to you,” she says. “The first step is to determine what tools and information are available to people, if they’re using them, and if not, why not. Is it because they don’t know it’s there, or they don’t trust it?”
McVay also emphasizes that no matter the size of the organization, it is critical for the people performing the analysis to get close to the clinicians who can identify the challenges. That can be even more critical when resources are scarce because they must be spent where they can produce the most improvement in quality, she says.
“Being able to focus and not try to solve every problem for everyone is a huge priority in a small facility. You have to know where you can do the most good with the people and resources you have, and you’ll only understand that from talking to the people you’re trying to help,” she says.
Vendors can address many of the issues you may not be able to address in-house, McVay says. Their offerings may not be a perfect solution, but it often is practical to take advantage of the data analysis a vendor can provide rather than nibbling around the edges of several issues when one could focus internal efforts on the few he or she can truly affect.
“Even at UnityPoint Health, we are very purposeful about looking at what our vendors are investing in and whether a solution they have will get us 80% of the way to where we want to be. If so, is that good enough now so we can use our resources in a different space?” McVay says. “None of us have unlimited resources, organizational capacity, or willingness to take on more and more projects.”
A finite number of clinicians means predictive modeling is most useful for managing or prioritizing workflows, says Benjamin Cleveland, a data scientist at UnityPoint who builds some of the internal solutions and works with vendors. The modeling helps ensure UnityPoint deploys staff to the patients who need the most help, and enables the patients to choose the most effective care, he says.
All patients in UnityPoint hospitals are managed with analytic models that provide individualized risk assessments, Cleveland says. Those models not only improve the care of individual patients, but also produce more meaningful metrics for the hospitals, the health system, and population health initiatives, he says.
“Each patient is unique and brings a lot of variety to the table, so we use modeling techniques to control for a lot of variation within a patient’s case, and once you control for those variables whatever metrics you’re tracking tend to be much more clear,” he says.
One component is UnityPoint’s “heat map,” a data analytics model that illustrates not just a patient’s overall risk of readmission in the 30 days after discharge, but also the varying risk on each of those 30 days. The data analysis revealed that the risk of readmission varied greatly from day to day and patient to patient. Some were more likely to be readmitted early in the 30 days, and some were more at risk later in that period.
That data were used to create a map showing “heat zones” in the 30-day period, and clinicians could then use that map to more effectively provide the interventions that might avoid the readmission, Cleveland says.
An individualized heat map is created for each patient, showing the days on which readmission is most likely.
The health system also addressed costly no-shows, combining those data with the readmissions heat map to help clinicians factor in the likelihood that a discharged patient is not actually going to go to a scheduled follow-up visit with a primary care physician.
That can help the clinicians better intervene with those patients rather than simply scheduling the visit and hoping for the best, Cleveland says.
“To further inform that care coordination and the patient trajectory, on top of the heat map we overlay their follow-up appointment schedule within UnityPoint so that we can see all of their past and future appointments as well as the outcome of the appointments,” Cleveland says.
“So, if they missed the appointments those will show up in red. And for the future appointments, we have a risk model that depicts the likelihood of the patient showing up. That’s an example of integrating a lot of data from different sources and models into a dashboard that can be visually consumed quickly and effectively by our care coordinators.”
Hill says the heat map has completely changed how UnityPoint conducts follow-up care after discharge. “The traditional model we used was, with not much success, based on standardizing follow-up care for patients of the same type,” he says.
“Every COPD patient was scheduled for follow-up visits at seven days and 24 days, using these blanket rules to guesstimate when we should be doing follow-up after hospitalization,” Hill adds.
“We are seeing fantastic results using the heat map to reduce readmissions. I can’t stress enough what a game-changer that has been,” Hill says.
A length-of-stay model also improves discharge planning and resources by helping care teams predict how long a patient will be hospitalized. For example, advanced analytics also were used to develop a sepsis model that could identify when the medical record should trigger best practice alerts.
“The model allowed us to decrease our overall false positives and overutilization of tests on the wrong patients, while increasing the number of patients we’re including in our cohort appropriately. The model has a greater sensitivity and a sharply improved specificity over what we were doing before,” Hill says.
“But you also have to continue the machine learning approach and not just take a look at the past two years and then create the model you’re going to use for the next 10 years,” Hill adds. “Epidemiology changes and diseases in the community change, so you want to have ongoing learning in your systems and modify your processes appropriately.”
One of the biggest surprises for McVay has been the amount of change management required to apply data analytics to quality improvement. She and her analytics team had to gain respect by listening to stakeholders and providing the solutions they needed, and she says they now have seats at any table when it comes to discussing organizational strategies.
That was not the case in the beginning, she says.
“We started with creating transparency around organizational key performance indicators. We have a toolkit that goes out monthly that includes the metrics our senior leadership has determined are most determinative of our success,” she says. “We push that out to almost 3,000 individuals. That transparency and consistent conversation about how we want to be a data-driven organization helps us be included in strategic and operational discussions, where we make sure the data is a part of that planning.”