How to Do Big Data in Healthcare

Healthcare is replete with big data analytics use cases that offer measurable results, including reduced hospital readmissions, better medication management, improved strategic planning and heightened fraud detection.

That’s all well and good, but for one key factor: How do you start? Most healthcare data remains unstructured, proprietary and siloed — and creating a clinical data warehouse is a complex task that time-crunched healthcare CIOs can’t always justify.

Fortunately, there are lessons to be learned from healthcare’s big data implementations. Here’s some advice from providers who have been there, done that and lived to tell the tale.

Speaking at the recent Oracle Industry Connect, the Mayo Clinic‘s James Buntrock recalled a time when organizations simply added applications, databases and point-to-point interfaces between them. The result? A plethora of disconnected data warehouses.

Rather than focus on apps, Mayo takes a data-centric approach, Buntrock says, treating data as a critical asset for research, clinical and other needs. Getting this right means aligning IT and business objectives, he says, as well as supporting informatics and data mining in addition to traditional business intelligence.

[ Analysis: Can Healthcare Big Data Live Up to Its Promise? ] [ Also: Healthcare Needs Analytics for the ACO Model to Succeed ]

Eliminating redundant data warehouses represents one short-term goal of healthcare analytics initiatives. Dr. Gwen O’Keefe of the Group Health Cooperative says organizations also need to implement a cost-efficient analysis architecture and provide front-line operational managers with business dashboards to provide better visibility into necessary metrics. Once that happens, organizations can focus on adding external data (from providers as well as patients) to internal clinical and financial data, implementing predictive analytics and using the subsequent data analysis to drive clinical and operational change.

[ More: Healthcare Sees Big Data As More Than a Bandage ] [ Also: Big Data From Federal Agencies Will Drive Health IT ]

Of course, healthcare CIOs must sell IT to executives. When it comes to analytics, O’Keefe recommends first finding business leaders who know what problems they need to solve. (More on that later.) Take a rapid, agile approach to solving the problem, she says, and build a forward-compatible solution that will work in this instance and many others to come.

Those line-of-business leaders must also work with both IT and executives to define data governance and quality standards, O’Keefe adds. This collaboration will help everyone avoid “analysis paralysis” by determining at the outset when decision makers will need ballpark figures versus more concrete figures.

[ More: Being a Successful Healthcare CIO Requires Vision ]

The conversation can’t be limited to the top of the org chart, though. Successful analytics requires “relentless communication” about use cases as well as the importance of training, O’Keefe says.

As you do this, Buntrock says, you’ll “add a lens” that lets you tap the expertise of your workforce. Doing so will “raise maturity” about analytics themes and projects, he adds, which will help support data discovery as well the operations that benefit from such discovery.

But be warned, says Lisa Khorey of the University of Pittsburgh Medical Center (UPMC): Don’t measure something if your organization will be too afraid to act on it.

[ Feature: Early Adopters Poised for Big Data Advances in Clinical Medicine ]

What to measure, then? Khorey offers several suggestions. Many, admittedly, are regulatory, so they’re less about disruptive innovation and more about mere survival. Such initiatives could include measuring physician performance and financial modeling, which are covered under federal regulations for quality reporting and shared saving.

That said, analytics can help healthcare organizations make a business case for genomic research — or, as UPMC has done in its personalized medicine program, bring together clinical and genomic data from disparate sources (and vendors) to research causes of and treatments for breast cancer.

[ Related: Can Watson, IBM’s Supercomputer, Cure Cancer? ]

This kind of powerful analytics helped researchers find the cause of French Canadian Leigh Syndrome, which remains largely fatal if untreated but presents symptoms similar to other diseases, making it hard to diagnose. As the Broad Institute’s Michael Reich described at the MIT Information and Communication Technologies Conference, looking at a gene expression data set and a database of known mitochondrial proteins (available in the public domain) helped researchers discover an overlap in genes that, when re-sequenced, showed mutations on the LRPPRC gene that cause the disease. Subsequent advances — sequencing can be done in 36 hours, for $4,000 — have yielded an “explosion” in genomic data, Reich says; soon such discoveries will be the norm.

On a smaller scale, Khorey offers 10 analytics “accelerators” for healthcare organizations. Some have already been covered. Here are the rest:

* Buy a data model. Don’t build one yourself; it’s complicated and expensive. * Exploit existing technology, including that data warehouse. When you buy, though, do so for the future, not the present. * Be addictive. Build interfaces, reports and results that turn skeptics into believers and get everyone interested in what analytics can do. * Prepare your organization for change. Analysis and the insight derived from it represents change management — and threatens traditional power and influence hierarchies. * Lead with medicine. But go deeper than “Diabetics take medicine.” What do they take? How often? Does it work?

Four years ago, Boston Medical Center fell to the bottom 10 percent of the University HealthSystem Consortium (UHC)’s mortality rankings. Now the safety net hospital ranks in the top 25 percent. Analytics played a big part in this turnaround, officials said at Medical Informatics World. BMC reviewed patient safety reports, as well as deaths that had otherwise presented a low probability of mortality, and used that to provide a risk adjustment methodology, says Roshan Hussain, the hospital’s director of analytics and public reporting. Also key: Separating observed (clinical) and expected (analytical) mortality, reframing causality (who owned the patient, not who caused the death) and emphasizing institutional improvement over individual punishment.

[ Feature: How Big Data Will Save Your Life ]