Skip to content Skip to navigation

At iHT2-Boston, HealthInfoNet’s Dev Culver Offers a Success Story Around Predictive Analytics for Population Health

May 21, 2015
by Mark Hagland
| Reprints
Innovation is proceeding apace in Maine, where HealthInfoNet, Maine’s statewide HIE, is now leveraging predictive analytics to support care management for its stakeholder provider organizations

Innovation is proceeding apace in Maine, where the Portland-based HealthInfoNet, Maine’s statewide health information exchange (HIE), is forging ahead to provide value to HIE stakeholders along a number of dimensions. On Wednesday, May 20, the closing day of the Health IT Summit in Boston, being held at the Hyatt Regency Cambridge (Massachusetts), and sponsored by the Institute for Health Technology Transformation (iHT2—a sister organization to Healthcare Informatics under the same Vendome Group LLC corporate umbrella), Dev Culver, HealthInfoNet’s executive director and CEO, offered a case-study presentation entitled “Using Real Time Clinical Data to Support Patient Risk Stratification in the Clinical Care Setting.”

For background, Culver, who is very well-known in the HIE world, began by sharing some current statistics about HealthInfoNet. Originally created in 2006 and having gone live with data exchange in 2009, the HIE now has contracts with all 37 hospitals in the state, with 35 of those hospitals exchanging data live (and the remaining two set to go live within the next couple of months). In addition, 38 federally qualified health centers and 400-plus other ambulatory care sites are also exchanging data live.

Dev Culver speaks to the iHT2 audience on May 20

What’s more, HealthInfoNet’s database contains 1.5 million lives, and runs analytics tools on 1.2 million lives, for a variety of purposes. As Culver noted, the state of Maine’s total population is 1.3 million, so that total includes some non-permanent visitors as well.  One innovation that has proven to be very popular with the stakeholder groups involved—hospitals, physician groups, health insurers, and consumers—has been its alerts service, which sends participating providers alerts when individual patients are hospitalized or visit an emergency department (ED).

The core of Culver’s presentation on Wednesday, though, focused on a fascinating case study around predictive analytics. As Culver explained it, he and his colleagues at HealthInfoNet have been working with the leaders of a 112-bed community hospital in Maine who wanted to be able to predict events such as hospital admissions and readmissions, ED visits, and other important patient events. That hospital’s leaders recognized that ED visits and readmissions were leading to actual and de facto penalties coming from federal, state and private payers. The hospital’s leaders hoped to be able to avert such readmissions and ED visits through predicting such events in advance.

“The impact at this community hospital has been fascinating,” Culver told his audience. “They’re not in a lot of risk-based contracts; they’re doing it to focus on self-pay and Medicaid populations, because every time any one of those individuals visits them, it costs them money.”

What became clear was that, while “care managers tend to know who the high-risk people are, what they’re really interested in is those patients who are 40-percent or 50-percent probable, but are ‘moving’” from lower levels to higher levels, of risk for readmissions and ED visits. “Can you tell me who my top ten patients are who are in motion? We had to figure this out.” Culver and his colleagues began by analyzing which patients were at risk for 40 percent or greater risk for readmission, within the next three months, based on a six-month window of analysis.

Culver and his colleagues began by focusing on predicting which patients were at highest risk for readmission and ED visits. Over time, they’ve introduced disease-based predictions: which patients were at risk for developing type 2 diabetes, for having a heart a attack, for having a stroke, or for dying. As Culver noted, the community hospital client involved provides a great deal of palliative care, but was finding that they were initiating the provision of palliative care too late—thus the need to predict the probability of death.

The process has become more sophisticated over time, with Culver and his colleagues helping the hospital’s leaders to identify higher-than-expected inpatient lengths of stay, and to identify utilization trends and disease prevalence, as well as the psychosocial and socioeconomic factors that might be influencing outcomes for patients. The ability to help that hospital’s leaders combine clinical and claims data, Culver added, has been extremely helpful.

In the end, Culver told his audience, the ability to leverage predictive analytics to support care management and population health, will be one of the most important things that HIE organizations like his can do, and he pointed to the fact of continually new frontiers that provider leaders will discover over time.