Pamela Peele knows that people who subscribe to cooking magazines have a much higher risk of going to the emergency room. But how she knows that is a whole other story.
Peele's insight can all be tied back to a $1.6 billion investment in 2008 by healthcare giant UPMC in its data, analytics and information technology infrastructure. The result of that upgrade has given Peele, who serves as chief analytics officer in UPMC's Insurance Services Division, a window into the health needs of the system's populations and positioned the 125 hospitals in the UPMC system to not only survive, but thrive in the era of pay-for-quality and population health.
"I don't see how health systems stay alive if they don't do this," said Peele. "That investment in IT made us a powerhouse. We gained a $1.6 billion advantage."
The UPMC Insurance Services Division, of which UPMC Health Plan is a part, has nearly 2.9 million members. UPMC is the second largest provider/payer system in the country behind Kaiser Permanente.
But with that many beneficiaries, using analytics to reduce risk is a massive money saver.
"We use our data enabled by our investment in IT to understand the risk behind members," she said. "We've already developed models on which members we should focus our outreach: who needs care coordination, hooked to which doctor."
Peele's an analytics advocate, that's certain, but the benefits of data are bearing out in many ways for UPMC, especially since it manages its own health plan.
For starters, having access to both provider and payer data cuts down on the time between adjudication of a claim and payment.
"By making a direct pipe to providers," Peele said, "we turn those claims around in a week," she said.
And since UPMC providers do not use a clearinghouse for billing, the system saves money on the small fee for that service.
The more transformative benefit is the fast, direct access to claims data that can then be mined for better population health management.
"The backbone of data is claims data. We get that and other data from the pharmacy, Census … we bring all of that data together."
Instead of using off-the-shelf formulas based on national averages, Peele and team invented their own models using predictive mathematical formulas combined with the claims and local demographics data.
"We model on a lot of different kinds of data," Peele said. "We model on household data, we use everything we can lay our hands on."
Making sure the data is local is important, Peele said, since broader databases often include profiles for the kind of consumers that don't exist within every system's community.
All that data is then analyzed using machine learning, a mechanical process where the software looks for the kind of trends that the system can act on.
"We made no assumptions, no priors, we purely used machine learning," she said. "In EHR medical records, we took about half a million clinical notes, did pure text mining, we were looking for words or word signals that show the people who will show up in the emergency room in the future."
For example, when the clinician writes the word "mother" somewhere in the notes, those patients usually do very well, but when the clinician writes the abbreviation, "ww," which means the patient has a wheeled walker, there's cause for concern.
It's not that the patient has a wheeled walker, but that the clinician has included it in the notes, that ups the risk.
"That person is at elevated risk of going to the ER," said Peele, so UPMC will more closely monitor and follow up with that patient to avoid that.
John Showalter, MD, chief health information officer at the University of Mississippi Medical Center. (Photo courtesy UMMC.)
The vendor answer
Most health systems do not have a billion to invest in information technology. But for the smaller, three or four hospital systems operating on $2 billion a year, sometimes the answer comes from renting, and not buying a new system.
For example, the University of Mississippi Medical Center leases much of what it needs through a subscription service, according to John Showalter, MD, the medical center's chief health information officer.
It all began when UMMC sunk $500,000 to lease an Epic system that has its own data warehouse.
"We did not start from scratch," he said. "We did some predictive analytics around ICD-10 to help neurosurgeons and orthopedics who are most at risk. We showed gains on that."
Today, that investment has paid off in a return on the investment. Showalter's original two-person analytics team, including himself, has now grown to 20 people.
"I was surprised at how specific some of our problems were," he said. "We were able to move the complexity of our neurosurgery group to top ten complexity by tackling four or five major documentation issues. Not that documentation is poor, but there are specific things physicians don't know about administrative billing rules. I'm surprised how much of an impact specific changes make."
Physicians lagging on completing documentation made a big difference in cash flow. The modeling showed that If they completed documentation two days earlier, it brought in $5 million more in cash flow.
"Our spend this year, we're probably at 2.5 and 3 percent return," Showalter said.
"We work with vendor partners to work with data in bulk, they're purchasing on the block level. We're using it on the block level," he said. "It's not as robust as individual (information) but it's a lot closer than using the EHR data. We're trying to bridge that gap."
The main aim of most modern analytics programs is to move from a model in which a system reacts to problems, such as a high rate of readmissions, or hospital-acquired infections, to being able to predict issues from the data before they occur.
At UMMC, Showalter said they are now focusing on denials; the clinical documentation that's driving those denials; and the value-base to understand infections, falls and pressure ulcers.
"We've reduced our expense related to pressure ulcers," he said. "We've been very pleased to move from reporting to predictive analytics."
Showalter said health systems should start small on a very specific initiative where they know they have the volume to change behavior.
"To me it's a lot about partnerships," he said. "We're working with vendors as opposed to developing things in-house. You don't need to hire an expensive programmer."
UMMC also just completed a geographical analysis around its Medicaid population.
"We're very quickly able to see in what counties we can put nutritional intervention," said Showalter, who got his start in analytics doing EHR documentation while a student at Penn State.
UMMC is also putting together analytics around physician utilization, surgical and medication costs and workflows in labs and radiology.
"Analytics can greatly improve patient care," he said.
The underlying value
Ten years ago, Peele was the vice chair of the Department of Health Policy and Management at the University of Pittsburgh Graduate School of Public Health when she was approached by UPMC Insurance Services Division President Diane Holder to do some economic modeling.
"She wanted to be a data-driven organization," Peele said.
But now that Peele is part of UPMC, what it means to be data-driven has taken on a whole new layer of importance.
The Centers for Medicare and Medicaid Services has pledged that it will tie at least 50 percent of all payments to quality, or value-based reimbursement by 2018. Private health plans are also adopting value-based models. So it stands to reason that a health system that doesn't do all it can to improve risks its bottom line in the process.
"Data are our eyes and ears," Peele said. "It's how to be smarter about what we're doing and manage the financial risk."
But even though providers have little choice in taking on the management of risk, it's not what clinicians were trained to do. However, they do need to be trained, especially when it's the data that they generate that's most valuable to the health system.
Like Healthcare Finance on Facebook
"You cannot make the transition to value without analytics," Showalter said. "Without a structure of analytics for improvement, you won't see gains."
Peele said it's better to spend the time up front to clean data, rather than investing in analysts later to do the job.
"You're not going to get stand-up analytics if you don't have your data fit for consumption," she said. "Groom it up and you immediately get cost savings."
But just because UPMC has become a "powerhouse" when it comes to data analysis, machine learning and predictive models, that doesn't mean that everything the system learns is can be acted upon.
"Care is local, disease is local, but within a span of 100 miles of the health system there are different prevalences; they have different care resources. You can't provide that until you see disease prevalence."
She also can't figure out why consumers who subscribe to cooking magazines are more likely to land in a UPMC emergency room.
But the data keeps turning up, Peele said.