Technology can be a valuable assistant, but the collection of data necessary for better outcomes often takes a back seat to bureaucracy.
Like most physicians, pediatric cardiologist Ricardo Muñoz, MD, remembers the patients he couldn’t save. Actually, it’s not that he couldn’t save them, but rather, he sometimes couldn’t see their cardiac events coming.
“I have been a physician practicing this specialty for almost 20 years, and I know from experience that most events could be prevented,” says Muñoz, who works at Children’s Hospital of Pittsburgh. “When you look in retrospect, you can say, ‘Well, I could have done this and this if I had known that; I could have asked the nurse to pay attention to this detail.’ ”
Now Muñoz is working to correct it. A computer program Muñoz is developing with University of Pittsburgh bioinformatics experts aims to take past events and use the data to prevent them from happening again. Part artificial intelligence, part computer learning — call it predictive analytics at the bedside. When fully functioning, it’s meant to download real-time data from the machines monitoring heart rate, blood pressure and other vital signs, and literally flash a red light when certain combinations in the normal range trend toward trouble.
Then, physicians and their teams can use clinical judgment to prevent crisis events.
Across the country, medical systems and academics are looking at the terabytes of data generated by vitals machines, lab tests and electronic health records and asking if the data can be used for more than connecting care teams or satisfying federal regulations. As the age of predictive analytics dawns, the questions remain whether the biggest institutions in the land will stay behind them.
The Data-to-Value Algorithm
If you ask Farzad Mostashari, MD, former national coordinator for health information technology at the U.S. Department of Health and Human Services, you can’t have predictive data without payment reform.
“In a fee-for-service world, predictive analytics don’t matter very much,” says Mostashari, now CEO and founder of Maryland-based Aledade, which works with physicians to build and lead accountable care organizations. “In fee-for-service, you don’t care who has a higher risk of being readmitted — economically, at least. The doctor may care, but economically, the incentive is not there” to prevent repeat events.
Accordingly, it shouldn’t be a surprise that major health care technology upgrades became mandatory around the same time the Centers for Medicare & Medicaid Services began experimenting with payment models.
In 2009, the American Reinvestment and Recovery Act, an economic stimulus package, included the Health Information Technology for Economic and Clinical Health Act, better known as HITECH, which changed reimbursements to encourage health care organizations and individual practices to take their data digital. This, Mostashari says, laid the groundwork for meaningful use of predictive analytics.
“You simply can’t do analytics, predictive or otherwise, if medical records are on paper,” he said. Before HITECH, medical systems might have analyzed claims data for trends. But, he said, “there was no way to get that information into the clinical workflow.”
The meaningful use requirements in HITECH also were about payment, Mostashari says. In 2010, the Affordable Care Act passed and included a number of experiments to change payment models from fee-for-service to value- and quality-based care. That included creating the CMS Innovation Center, shared savings payments and the risk-sharing programs such as accountable care organizations.
In these environments, Mostashari says, predictive analytics become key to getting paid.
“When you have a system where you’re being paid on outcomes, then it becomes super important, with limited resources, to use data to decide where to concentrate those resources,” he says.
Finding Meaning in Use
If physicians have time to call only one patient at the end of the day, they need to call the right patient — the one most in need, most likely to be readmitted, or most likely to have an adverse event. And they need data to decide that.
Unfortunately, right now, all those boxes physicians have to click after eight hours in clinic can feel “meaningless,” says Srinivasan Suresh, MD, MBA, chief medical information officer and a pediatric emergency room physician at Children’s Hospital of Pittsburgh.
“They probably meant well,” Suresh says. “CMS convened a lot of panels to put it together, and they did it the right way. But for some reason, the outcomes are not the best-quality metrics.”
Mostashari, who was in HHS during the implementation of these programs, insists the measures are meaningful — just not for fee-for-service.
“Implementation of meaningful use lurched ahead, when payments are still overwhelmingly fee-for-service,” he says, adding that he understands physicians’ frustration with so many currently meaningless measures. “A lot of doctors will say, ‘Look, this is slowing me down.’ In a fee-for-service world, ‘slowing me down’ is synonymous with, ‘I’m going to make less money.’”
But, he added, “the payment systems are catching up now.” And that means that the value of predictive analytics is about to grow.
Predictive Measures, Improved Outcomes
Indeed, because of the CMS readmissions penalty program — which also is part of the ACA — Children’s Hospital of Pittsburgh began developing a program in 2014 to use EHR and claims-denial data to identify people who might return to the emergency department. Some 12,000 patients did. But 88,000 didn’t.
“So what’s different in the 88,000 who didn’t come back?” Suresh asks. They’re trying to figure that out. The idea, he says, is not that predictive analytics will replace clinical judgment, but that it will allow all physicians, no matter how long their daily shift or their experience in a specialty, to make the best decisions for patients.
It’s also a solution, he says, to the staffing-to-acuity issue in hospitals. If a physician can work with just one acutely ill patient, they might have the time and focus to identify the things that might lead to an episode. But technology can help physicians with up to six patients do the same — ideally without affecting quality.
“It’s not man vs. machine,” Suresh says. “It’s man vs. man-plus-machine.”
The Limits of Analytics
The challenge with predictive analytics, of course, is capturing the right information. Edwin Zhao, who spent a decade helping medical systems implement their EHRs and now works for a geriatric primary care company, says he’s seen firsthand how EHRs can hinder rather than support predictive analytics and value-based care.
“The thing with EHRs is that they’re like a box of Legos,” he says. “The usability of an EHR doesn’t have as much to do with the software as implementation and who gets to be involved.”
Optimization, he says, is often an afterthought. For in-stance, he’s seen medical systems implement EHRs while maintaining the same ticket-based structure for their IT departments. Much like fee-for-service, he said, IT specialists are evaluated based on the number of tickets they close, not on solving endemic system problems.
For example, if your organization keeps receiving errors saying that the bar code on a medication bottle is one digit short, do you fix that one drug? Or do you step back and say, “How many drugs do we have this problem with?” Often, he says, it’s the former.
Likewise, he says, EHRs are optimized for hospital use but by the time they get to Zhao’s company, a value-based organization that proves its worth by demonstrating how many emergency department admissions it avoids each year, the configuration leaves out key data that his company needs to identify the frail elderly who don’t leave home.
“The measure that’s standardized for that is ‘activities of daily living’ — ADLs,” he says. “But that’s not something that’s captured [in the EHR] right now.”
If you don’t collect the right piece of information, he says, “how do you make that up? Maybe a history of staying in an acute rehab facility, or visits from a home nurse. Those may be indicators of difficulty getting out of the home. It’s more of the squishy social work side of things.”
What the Future Holds
The potential good news, Suresh and Mostashari agree, is that current policy on meaningful use and value-based payments don’t appear to be on the chopping block, even though Congress has tried to repeal the ACA.
CMS has moved to delay implementation of some features of the meaningful use requirements, and it has expanded the amount of data that practices must provide to be reimbursed under its Quality Payment Program. But the programs appear to have survived so far. Recently, CMS Administrator Seema Verma emphasized the organization’s commitment to “the value of care rather than the volume of tests.”
Mostashari is watching what happens.
“If you look at what Seema Verma had said in writings and speeches, and if you see what leadership in Congress said, and if you see what the White House has said, the idea of getting more value from our health care dollar is not under question,” he says. “And once you get into serious value-based payments, you can’t not start talking about predictive analytics.”
Heather Boerner is a freelance health care writer based in Pennsylvania. She covers health law and policy for the Physician Leadership Journal.