Skip to content Skip to navigation

Patients in Motion: Maine HIE Rolls Out Real-Time Predictive Analytics

January 7, 2015
by Gabriel Perna
| Reprints
Dev Culver, CEO of HealthInfoNet

One of the most successful health information exchanges (HIEs) in the United States just got a little better.

In an era where many HIEs have struggled to remain sustainable, HealthInfoNet, the statewide health information exchange (HIE) in Maine, is doing just fine. It’s got 32 out of the state’s 36 hospitals and 326 ambulatory locations on board, sending data to it. It’s gotten long-term care, specialty, and behavioral health facilities and providers, along with a host of others, to view or submit data to the HIE. Over the years, it has introduced a number of value-added services for participants, including sending preventative care data to public health officials, creating a statewide image archive, and a clinical data-facilitated patient event notification system.

In conclusion, it’s done well for itself.

The HIE’s latest initiative might be its most impressive value-add yet. This week, HealthInfoNet announced it is rolling out a predictive analytics service option for members. Five have already signed on. The analytics service, created in partnership with the Palo Alto, Calif.-based HBI Solutions, uses real-time clinical data to determine and identify the potential costly patients that are in the middle range. They’re not yet the high-cost, frequent users of the system, but they very well could be.

“The analytics platform can tell you within the year, who will be admitted to inpatient, who will be in the emergency room, who will be among the top 10 percent of most expensive patients, who is going to be returning for admission within 30 days, and who will return to the ER within 30 days,” Devore Culver, the CEO of HealthInfoNet, says in an exclusive interview with Healthcare Informatics.

The predictive analytics came about when HBI, which came out of an academic setting in Stanford University, approached HealthInfoNet about using its standardized data set for predictive analysis. It took them a year to clean up the data, says Culver. Once they did, the HBI team tested out a dataset with 74 percent accuracy, a number that is pretty high, he says.

To test it out, HealthInfoNet turned to St. Joseph’s Hospital, a general medical and surgical facility in Bangor.  The hospital runs risk scores on patients every morning using real-time data. For those who are being discharged and have a risk score of higher than 30 percent of returning within 30 days, the reasons why are communicated to their primary care physician.

Care managers at the hospital are already seeing an impact from the predictive analytics tool, says Culver. They’re finding more potential at-risk patients than before and lowering readmissions rates in the process, he notes.

“It picks up patients we may not have thought about,” stated Darcy Bond, R.N. care coordinator at St. Joseph’s. Amber Sloat, R.N., another care coordinator at the hospital, noted that if a patient comes in with a diagnosis of anemia, they won’t typically be followed. “But the tool flags them and then we’re able to use the health information exchange to get a fuller picture of their health and the clinical background for why they are here,” she said.

Not only is this kind of alignment unusual, it’s extremely rare to use it in conjunction with real-time clinical data for predictive analysis. Culver notes that because claims data is at minimum, 30 days old, it won’t be able to predict outcomes at the same level as real-time clinical data. And also, he adds that it can’t change your risk score, for instance, on a bad lab score.

“Both types of models have value. For the care manager or doctor, the value proposition of knowing a risk score based on what was going on through yesterday is huge,” Culver says. “In the field, they know the really sick ones who is high risk, those at 80-to-90 percent and above. What they’re seeing here are the people who are in motion towards that.”

In the world of value-based reimbursement and risk-based models, having this kind of predictive analysis is vital, says Culver. HealthInfoNet has worked with Eastern Maine Health System, also in Bangor, which leads one of the Centers for Medicare and Medicaid Services’ (CMS) Pioneer accountable care organizations (ACO), on tailoring the predictive analytics to support specific patient populations. The analytics platform can filter the population by provider and by contractor. “There are so many ways to filter and shape the data,” Culver says.

It will go deeper. The predictive analytics tool is currently being filtered for the payer community, Culver says. He adds they’re working on creating predictive analysis that will determine who will have a stroke, who will have a heart attack, and who will develop Type 2 diabetes. While they’ve made some progress on that front, he is waiting for more clinician engagement to determine its accuracy.  

Like most successful HIEs, HealthInfoNet is run like a business. Culver says he believes in the concept of the lazy asset, taking something you need to run your business and repurpose it as value-add. In the world of HIE, he points to the master-patient index (MPI) as an example. “An HIE has to run an MPI in order to survive, you’ve got to match the patients up. Providers, especially large hospitals, have multiple MRNs on the same patient. Since they’re sending me all those medical record numbers anyway, I can send them back a single identifier for that patient they can use to manage across all points of service,” he says.