Predictive Analytics Help ID Patients Headed for Problems

Feb. 01--Jessica Taylor's computer knew her patient was in trouble before she did. Even the patient, a Bangor-area man with chronic illness, was in the dark.

In all likelihood, he was about to wind up in the emergency room.

Taylor, a nurse care manager at St. Joseph Internal Medicine, spotted the warning signs with help from new technology believed to be the first of its kind in the country.

By combing through the electronic medical records of nearly all of Maine's 1.3 million residents and applying sophisticated modeling software, the computerized tool predicts which patients are most likely to land in the ER. Not only that, but also who is headed for a hospital admission or even well on their way toward a stroke, heart attack or type 2 diabetes.

The "predictive analytics" tool red flags patients headed for serious illness, allowing doctors and nurses to intervene and hopefully prove their computers wrong. While still unproven, it holds the promise of averting needless pain and suffering, as well as millions of dollars in unnecessary health care costs.

Taylor called her patient on the phone, learning he'd just lost his health insurance and was working two jobs to keep his family afloat. She helped him sign up for a new health plan under the Affordable Care Act and, while he waited for the policy to kick in, connected him with programs that provided three months of his asthma medication for free and his other drugs at much lower prices, she said. She also gave him a hand in finding transportation to his doctor's appointments.

"He didn't realize these resources were there," Taylor said. "He was just trying to keep his head above water."

Soon, the man dropped out of the tool's "high risk" category. It no longer saw an ER visit in his future.

"He hasn't been in the hospital, he hasn't been in the ED, he hasn't been here," she said. "He's been home with his family."

Pilot Project

Taylor's health system has been piloting the predictive analytics tool for the last nine months. St. Joseph Healthcare was among the first in the state to sign on, along with the Bangor Beacon Health Accountable Care Organization, a coalition led by Eastern Maine Healthcare Systems that's working to keep patients healthier while reducing costs. Three other health organizations also jumped on board.

"The tool alerts me, but that human element is still so necessary," Taylor said. "It allows me to reach out to the patient and get the story."

The new tool was spearheaded in part by HealthInfoNet, a Portland-based nonprofit organization that oversees the secure computer exchange housing the electronic health records of 1 million Maine residents. HealthInfoNet sought to go beyond delivering health record information to hospitals and other providers, to put the data to good use helping them give better care, said Devore Culver, executive director and CEO.

HealthInfoNet teamed up with a Silicon Valley firm that employs researchers at Stanford University, who are working to predict variations in the human genome, he said. The firm, HBI Solutions, suspected its team could apply the same pattern-hunting to HealthInfoNet's massive repository of medical information, he said.

"We started with a fairly simple question," Culver said. "Can you predict who's going to show up in the emergency room?"

The software identified Maine patients who had already visited ERs and dug backwards. After some number crunching, HBI provided HealthInfoNet with a list of all the patients it expected to land in the ER again during the six months that followed its last data check. They were right 74 percent of the time, Culver said.

"We said, 'Whoa, that's pretty interesting,'" he said.

Now, they're testing the analytics tool in the field with St. Joseph and others. As of recently, the tool also considers who's at risk of a heart attack, stroke and diabetes, though the jury's still out on its crystal ball abilities for those measures.

Fresher Data

Predictive analytics in health care is nothing new, but so far has largely relied on data from health insurance claims, which can be anywhere from 60 to 90 days old, Culver said. The new tool is the only one in the country that pulls from real-time clinical data, which provides a fresher and fuller picture of each patient than insurance claims for tests and procedures, he said. The information is at most 24 hours old.

HealthInfoNet is at the leading edge of the technology, said Joe Wivoda, chief information officer at the National Rural Health Resource Center, a Minnesota nonprofit that helps states adopt electronic health records. No other state's health information exchange has reached that level, but they want to, he said.

"Wal-Mart uses data to predict how quickly something is selling to make sure the shelves aren't bare," Wivoda said. "We have no way in health care of doing this predictive analysis at all, so this is a really exciting project."

Culver credits Maine's progress to the forethought of organizers who standardized the state's health record data from the beginning. That allows the modeling software to sift through the entire population's information and deduce from patient histories which hospital visits and illnesses are likely to occur, he said.

The new tool pulls data from HealthInfoNet's entire electronic health records network, which includes 32 of the state's 36 acute-care hospitals, more than 300 outpatient facilities, as well as long-term care and behavioral health providers, among others. That means it knows when a patient bounces around within the same health system or jumps to another. So, for example, a patient could be discharged from a nursing facility in Hampden, visit his primary care doctor at an Eastern Maine Medical Center practice, then get admitted to St. Joseph Hospital for surgery and never fall off the system's radar.

That makes the tool's predictions more accurate, Culver said.

"Seventy-five percent of the people walking into St. Joseph's Hospital have medical care at Eastern Maine or somewhere else in the community," he said.

Reducing Readmissions

Not just anyone can peer into a patient's health records through the exchange or the new predictive tool, Culver said. Only doctors, nurses and other medical providers treating that patient can view his or her personal data, he said.

Already, the tool has led to a reduction in hospital readmissions -- when patients wind up back in the hospital within 30 days -- at St. Joseph, according to Taylor.

Hospitals have begun focusing on readmission rates due to concerns about patient care and cost. Medicare now financially penalizes hospitals that readmit too many patients with heart failure, heart attack or pneumonia.

The government's move to tie hospitals' payments to the quality of care they provide got a shot in the arm earlier this week. The Obama administration on Monday unveiled an ambitious plan to rein in health costs by paying for value rather than the volume of tests and procedures, beginning with Medicare, the health insurance program for the elderly and disabled.

Without better ways to analyze their performance, like the new HealthInfoNet tool, making that shift will be hard for health organizations to accomplish, Wivoda said.

"Right now, we provide sick care," he said. "When someone's sick, they get care. This moves us closer to that ability to provide wellness care."

Maine hospitals pay a subscription to use the new tool, starting at $14,000 a year for the smallest facilities and rising based on additional beds, Culver said. HealthInfoNet must prove to hospitals -- many already squeezed by declining government payments and rising free care for patients who can't pay their medical bills -- that the service is worth the money. It remains to be seen whether the tool's predictions can avert enough of the costliest care, such as ER visits, to make up for the upfront expense.

HealthInfoNet and HBI also are marketing the tool to health information exchanges in other states.

So far, along with St. Joseph and the Beacon organization, St. Mary's Health System in Lewiston, Mount Desert Island Hospital in Bar Harbor, and Franklin Memorial Hospital in Farmington have signed on, Culver said. He predicts -- without the benefit of impartial computer software -- that at least half of Maine's hospitals and a significant number of large physician practices will be using it by the end of the year.

For Taylor, the tool's proven most helpful in identifying patients who aren't yet in dire straits but could soon take a turn for the worse. Simple steps, such as double checking a patient's medications after they leave the hospital or teaching them how to properly test their blood sugar, can prevent a crisis down the road.

"The folks who need a lot of support, we know them by name," she explained. "It's the folks who were in that middle band that we were missing before the tool."

As a nurse care manager, it's Taylor's job to see the full picture of each patient's treatment, sniff out trouble spots, and try to keep them healthy. She has other ways to identify patients within her practice headed for trouble, "but it's a small net," she said. "This tool is a wide net, and it allows me to catch many people and help them."

Taylor has worked with patients weakened by chronic illnesses for 15 years, first as a home care and then hospice nurse. She's seen the toll diabetes, asthma, heart disease and other ailments can take. There's nothing worse than seeing a patient sick, struggling to afford their care, and depressed when an ounce of prevention could have forged a new path, she said. She hopes a better way to harness their health data holds some answers.

"I'm looking at them thinking, 'If I could have just connected with this person a month ago, I could have prevented this,'" she said.

Copyright 2015 - Bangor Daily News, Maine