"Late binding," a just-in-time method, gives healthcare organizations more flexibility, say proponents.
7 Big Data Solutions Try To Reshape Healthcare
(click image for larger view and for slideshow)
A method of assembling data from disparate sources just in time for particular analytic "use cases," known as the "late-binding" model of data warehousing, is starting to gain traction in healthcare as many provider organizations gear up for population health management. The advantage of this approach is that it allows users to combine disparate data very quickly for targeted analyses without locking data warehouses into a predetermined data model.
A sign of late binding's potent appeal is that Health Catalyst, a data warehousing firm that uses this approach, has just raised $8 million in venture capital funding from Kaiser Permanente Ventures and CHV Capital, which is associated with Indiana University Health. A Salt Lake City firm founded by former Intermountain executives, Health Catalyst also received $33 million in December from Norwest Venture Partners, Sequoia Capital and Sorenson Capital.
In the wider world of information technology, late binding goes back to the 1980s. But in health care, most data warehousing still uses the early binding approach, in which data is mapped to standard vocabularies and is bound to business rules at the outset. One problem with this model, say proponents of late binding, is that business rules change and vary from one organization to another. Also, it takes time and effort to bind data that might never be needed for any analytic use case.
Providence Health & Services, a 32-hospital, five-state organization based in Seattle, has tried warehousing data both ways. Dick Gibson, chief healthcare intelligence officer for the system, told InformationWeek Healthcare that, when his team began building a data warehouse in 2011, it was intended only for Providence's Oregon operations. Then, early in 2012, it became a system-wide project.
That posed severe challenges to the early binding model that Providence started with. Although the majority of the system's hospitals and clinics were using Epic, there were other clinical systems involved, as well as 10 different costing systems, noted Gibson. "When you jump that quickly from a state-based to a system-based initiative, you have to adapt to those 10 different costing systems. And if you have an early-binding model, your model has been disrupted. So the advantage of a late-binding model is it's more flexible."
Taking the level of complexity up another notch, Swedish Health Services, another Seattle-area healthcare system, formed an affiliation with Providence in 2012. "Swedish has a whole other electronic health record and a different materials management system and a separate costing system," Gibson pointed out. "Now people say, 'I want you to produce reports for both Providence and Swedish.' If you did early binding, you'd have to go back and rebuild your model, and to some degree, redo the work. With the late binding model, you only have to start reconnecting data points that go across both facilities at the same time. Late binding allows you to connect only those separate sources that need to be there to answer the question. So it allows you to be faster and more flexible."
Providence had already brought in Health Catalyst in February 2012 to build certain elements of its data warehouse. By October, Gibson said, it was clear that the early binding approach wasn't going to work, so Providence decided to switch over to late binding for its entire data warehouse.
In November, the IS department delivered its first product to clinical end users: a data mart designed specifically to help clinicians figure out how to lower the organization's C-section rate. That data mart came with a dedicated dashboard that made it easy for users to get answers to their questions. The next product showed all surgical encounters in Providence facilities, allowing users to look up cases based on procedures, DRGs, surgeons or location. This year, Gibson and his team will tackle other major areas such as financial systems and materials management.
Gibson emphasized that the late binding approach requires a decentralized approach to report generation. For this to succeed, clinicians must work directly with analytics experts who are skilled in a number of different areas. Training these folks can be a challenge, he admitted, but clinicians love the end result, "because they don't face handoffs to the IT team, as is typical in an early binding model."
Health Catalyst says that about 80 hospitals, including Epic and Cerner customers, are already using its product. Besides Providence, they include Allina Health, Indiana University Health, MultiCare Health System, North Memorial Health Care, Stanford Hospital and Clinics, and Texas Children's Hospital. Intermountain is not a client, but uses the same principles in its data warehouse, said Health Catalyst CEO Dan Burton in an interview.
Burton said that Health Catalyst is in discussions with Kaiser Permanente about using its services. Although the investment by Kaiser's venture capital arm doesn't guarantee anything, he added, "We're optimistic that we'll be establishing a commercial relationship with them."
New delivery models such as accountable care organizations have created incentives for healthcare providers to invest in analytics and data warehouses, a recent IDC Health Insights survey found. Because only about 30% of providers currently have data warehouses, Health Analytics expects the market to expand rapidly, Burton said. "Every health system in the U.S. must put in a data warehouse if they're serious about population health management," he said.
As large healthcare providers test the limits, many smaller groups question the value. Also in the new, all-digital Big Data Analytics issue of InformationWeek Healthcare: Ask these six questions about natural language processing before you buy. (Free with registration.)
InformationWeek Tech Digest, Nov. 10, 2014Just 30% of respondents to our new survey say their companies are very or extremely effective at identifying critical data and analyzing it to make decisions, down from 42% in 2013. What gives?