Vital Signs Blog

VA puts the 'big' in big data predictive analytics

VA puts the big in big data predictive analytics
In looking for examples of real-time predictive analytics using big data, I was forewarned by physician informaticist Dr. William Bria that I shouldn't expect too much.

“The thing I've been impressed with so far is how many people are talking about it—and how few are doing something with it,” said Bria, chairman of the Association of Medical Directors of Information Systems, a professional association for physicians in applied medical informatics.

And indeed, it took some searching.

In the outpatient world, there isn't a lot of real-time, predictive analytics using big data going on—at least not yet—said Dr. Steven Waldren, who heads the Center for Health Information Technology at the American Academy of Family Physicians.

“From an informatics perspective, we have to solve some of the problems to incentivize the data being gathered and getting the permissions right to collect it,” Waldren said. “And then we have to convince providers that it's worth their time.”

Two recent outpatient pilots, both conducted by Milwaukee's Aurora Health Care system, involved about 500 heart failure and chronic obstructive pulmonary disease patients scattered across 10 clinics. Several other inpatient projects, including one in Dallas involving more that 14,400 patients, at Texas Health Resources, also have been underway.

But leave it to the vast Veterans Health Administration at the Veterans Affairs Department, a pioneer in the adoption of electronic health-record systems, to put the big in big data predictive analytics.

The VHA's data warehouse, launched in 2006, has 6.6 billion lab tests, 3.8 billion clinical orders, nearly 2 billion outpatient encounters and 10.5 million inpatient admissions. It's refreshed nightly, and soon will be updated every four hours. Since late 2011, the VHA has been mining this trove to create risk assessments for virtually all of its 6.4 million patients.

Only if a veteran is “brand new in the system, we may not be able to generate a good probability because we won't have data on them,” said Dr. Stephan Fihn, director of the VHA's Office of Analytics and Business Intelligence.

There's a discussion of the VHA's progress in predictive analytics in the August issue of the policy journal Health Affairs. Fihn is the article's lead author.

Based on an analysis of about 120 variables, the VHA's predictive analytics system spits out six separate probabilities for each veteran, calculating their 90-day and one-year “all risk” scores for hospitalization, death, and hospitalization and/or death.

The VHA uses the risk scores to guide its Patient-Aligned Care Teams, which aim to apply the principles of patient-centered medical homes to a healthcare system that operates 152 hospitals and 990 outpatient clinics. Its predictive analytics system also ranks risk scores within each panel of patients assigned to a Patient-Aligned Care Team, with about 1,200 to 1,300 patients per team. Services are aligned accordingly.

The risk scores are referenced thousands of times a month by physicians, nurses and other care team members, Fihn said. But their target audience is each team's nurse coordinator, whose job it is to match patient needs with VA resources and programs, he said.

“If you're in the 99th percentile of risk, that's a 72% chance of being hospitalized or dying in the next years,” Fihn said. “What we're really trying to say, we've got a ton of care-management resources—home care, home monitoring, specialty care. Are these patients actually enrolled in services that would benefit them? That's the goal here.”

The VA's financial investment in the Patient-Aligned Care Team program has been in the realm of $200 million to $300 million, Fihn said. So far the program doesn't appear to have generated sufficient savings to offset that initial investment, but he expects that to change.

“Over time, when we do the projections out to 2020, it looks like the ROI becomes positive,” Fihn said. “There were a lot of sunk costs up front. It really is turning an oil tanker. I wouldn't want people to walk away and say this is a bad investment. I don't think that's a correct conclusion. This is a long-term investment like many big investments, and it requires a lot of work to get to that state where you are reducing demand for unnecessary care.”

Follow Joseph Conn on Twitter: @MHJConn


Loading Comments Loading comments...