Targeting Chronic Illness Together | Healthcare Informatics Magazine | Health IT | Information Technology Skip to content Skip to navigation

Targeting Chronic Illness Together

April 18, 2012
by David Raths
| Reprints
Health plans support providers through predictive analytics

Executive Summary:
Health insurers that used to focus on individual claims are now looking across their entire membership to better define the appropriate level of financial risk to assign and to drive the right kinds of interventions. Many are making a determined effort to share their predictive analytics findings with clinicians.

Can predictive analytics software help health insurers and clinicians identify which patients could most benefit from more attention and better-coordinated care? As the patient-centered medical home and accountable care organization movements get rolling, provider organizations are starting to work more closely with insurers to manage population health for improved clinical outcomes, and the predictive tools may play a key role.

Payers have some experience using predictive analytics (also sometimes called clinical analytics) to stratify group populations by risk, identify high-cost conditions, and approach members for participation in disease management programs. Of course, integrated health systems, especially those with affiliated health plan entities, such as Kaiser Permanente (Oakland, Calif.) and Geisinger Health System (Danville, Pa.), have an advantage when it comes to sharing data. Most payer and provider organizations have to overcome cultural, financial, and technical differences.

On their own, the data systems of providers and insurers each have their strengths and weaknesses, explains Dan Coate, a principal with Pittsburgh-based consulting firm Aspen Advisors. Payers must largely rely on claims data. “The good news is that it is structured data. They can search it and parse it out,” Coate says. “The bad news is that it is often delayed. By the time they do analyses and run a report, it may be five or six months since an identified member was supposed to have that hemoglobin A1c test,” he says.

Providers are starting to use the more real-time data from their clinical systems to search for gaps in quality of care, but the data in those systems is currently less codified, notes Coate. “There is no way to quickly run analytics on it to see gaps in care. You can write a wonderful clinical note, and I could read that about the one patient, but I can’t yet search across all your patients with a large-scale analytics engine.”

To cross the divide, medical executives of some payer organizations are making a determined effort to identify trends and find ways to share their findings with clinicians as those provider organizations take on more risk.

One such pioneer is Brian Wolf, M.D., senior medical director for medical affairs for the Providence-based Blue Cross & Blue Shield of Rhode Island. “There are aspects of healthcare where if we are more efficient in identifying patients and have a free flow of information about best practices, we can eliminate money wasted on missteps such as repeat procedures,” he explains.

Brian Wolf, M.D.

Starting with models developed at Johns Hopkins University, Wolf’s analytics team came up with risk-adjusted scores to identify “impactable” patients who have complex conditions such as cancer or severe behavioral health issues. Starting in 2008, Blue Cross & Blue Shield of Rhode Island began sharing this data and helping groups of local providers set up patient-centered medical homes following the concepts delineated by the National Committee for Quality Assurance (NCQA).

“We now have 200 practitioners in five large physician groups in the patient-centered medical homes,” Wolf says. The insurer also pays to fund on-site case managers and in some cases co-located behavioral health specialists. Although they haven’t published any findings yet, Wolf says the early research suggests that the patient-centered medical homes have been able to lower rates of hospital readmission, especially among the complex patients the predictive analytics identified. “Readmission is a good measure,” Wolf says, “because it tells you that probably there wasn’t good resolution of the complaint initially.”

Tufts Health Plan
For their Medicare Advantage offering, executives at Tufts Health Plan in Watertown, Mass., realize that disease management efforts are labor-intensive and expensive, because they involve more direct interaction with members. “So we want to make sure the interventions work and that we are targeting the right population in order to get the return on investment,” explains Jonathan Harding, M.D., senior medical director of senior products.

“We needed to have a predictive model to identify the members to target,” he recalls. “We went to the vendor of analytic software used on the commercial side but weren’t impressed with what they had because it was not designed for the Medicare population, so we developed our own product.”

Harding and his team had to work through many variables before identifying what was truly predictive. They rejected some ideas that seemed intuitive, such as the number of doctors seen in a given period, which might indicate discontinuity of care. “It didn’t turn out to predict anything,” he says. But recent hospital admission or re-admission, age above 80, or a recent fall all were good predictors. “We came up with this complex report that lists for physicians all patients at high risk of being admitted or re-admitted. And it gives the member a score.”


Get the latest information on Health IT and attend other valuable sessions at this two-day Summit providing healthcare leaders with educational content, insightful debate and dialogue on the future of healthcare and technology.

Learn More