Turning raw data into better health outcomes is no easy task

Developing sound data models is challenging enough, but operationalizing them can be even trickier.
By Mike Miliard
10:12 AM
Share

Jonah Comstock moderates a panel with Mark Poler from Geisinger Health System, Ken McCardle from Mount Sinai Health System and Joe Bluechel from Sirius at the HIMSS Big Data and Healthcare Analytics Forum on Tuesday.

BOSTON – Though analytics technology is well-developed and widely deployed, putting those tools to work for better patient outcomes is still a big project as providers grapple with complex and sometimes competing priorities.

At the HIMSS Big Data and Healthcare Analytics Forum on Tuesday, clinicians from two major health systems and an expert in data systems compared notes about turning troves of numbers into financial and operational improvements and – most importantly – better healthcare for their patients.

One key factor in determining an organization's chances for analytics success has to do with their approach to the myriad problems that can be solved with data, said Joe Bluechel, vice president, data and analytics solutions at Sirius, a consultancy that helps with technology and data systems integration.

"A lot of it boils down to how you are organizationally aligned," he said. 

Is your process an ad hoc one, reactive and marked by one-off requests to disparate challenges, for instance? Or are you able to take a more strategic view, deploying analytics as "competitive advantage to improve quality and costs and the patient experience?"

How you answer that probably depends, in part, on how attuned you are to some of the finer points of how data models are put into practice, said Ken McCardle, senior director of clinical operations at Mount Sinai Health System.

There are teams that create models and teams that operationalize models, said McCardle, and oftentimes those people have very different skill sets.

"At my organization, for the data science and biostatistical analytics professional that are developing all these models, that operational world is sort of scary to them – it means you have to deal with IT and change management and applications and vendors and interfaces and all kinds of other different things," he said.

On the other hand, those on the operational side have their own set of challenges.

"It's another aspect of these model developments we have going on today: How do operationalize it?" said McCardle. "Who's keeping an eye on these models when you do the next (IT) upgrade? Did it break your new model? When you install a new device at the lab, does it send data to your system like you were expecting it to? There's a lot for us to think about, about how we operationalize these models."

Another big variable has to do with where the demand for various analytics projects comes from:  Is it demands from the C-suite to find innovative ways improve the bottom line? Or is it more targeted project initiated by those on the clinical side? Each demands a different approach.

"Sometimes the charge will come from an executive level, whether it's readmissions, or some other problem on a large scale, like supply chain," said Mark Poler, physician informaticist for enterprise data strategy at Geisinger Health System. "Other times it's clinician champions, who are focused on a certain area like heart failure or chronic lung disease or diabetes who provide the stimulus to create something new.

"Then it becomes an interactive dance between those who are motivated to create and to use things and make them useful," he added, "and governance over what all the processes are that are in play and discovering that perhaps people in different parts of the organization that are doing similar things could do those things together and create something that's more reproducible and usable and modular, instead of a lot of little things that are hard to operationalize and support."

That sort of "dance" is a critical thing to keep in mind when trying to chart lasting real-world changes from analytics projects, no matter what they might be, said Bluechel.

"There's always a trade-off between organizational agility and quick wins, versus long-term charters that really have impact," he said. "Our organization focuses a lot of time and energy with our customers on where to start: What are the different dependencies on these different initiatives and what's the business value and impact."

As they strive to show incremental wins from their data projects, more "self-aware" customers are able to identify their challenges, said Bluechel.

"Everything we do has to be focused on foundational improvement," he said. "A lot of that is not a technology problem. Policy and process and compliance issues are also road-blockers."

McCardle agreed. Although the excuses can be easy to grasp for – "my patients are so much sicker, my data is so much dirtier" – it's often not hard to realize that people and process are at the root of the problem.

"Data is imperfect," said Poler. "There's always huge defects in the data. But it's what we got, we have to work with it."

Add to that a fast-changing data and technology landscape, and the challenges become more acute.

"The world has changed dramatically over the past 25 years," said Bluechel. "If you think about traditional data warehousing, and Kimball and Inmon philosophies about a centralized data repository, some of the best practices and universal approaches still hold true," said Bluechel.

But with the explosion of new data in recent years – Internet of Things, social determ – the world has "morphed where you have to look at your data warehousing and analytics platforms truly as a distributed system," he said.

"We talked a lot about logical data warehousing, and using that as a reference architecture and building common semantic layers across a wide variety of different sources and systems to organize and integrate that data in a logical fashion using things like data virtualization and some of the data discovery tools that can help with that. The days of being able to suck it all into the central repository, those days are dwindling."

However the data is warehoused, an absolute must-have for effective analytics is good data visualization and UX, said Poler.

"You can show people numbers all the time, and people fall asleep," he said. "But then you show them a picture of the gaps in the operating room and the senior administrators go, 'Gasp! We can't have that.' The picture is the thing that changed the impact of the data. It's the same data.
”

"Eighty percent of the cost of an analytics solution is in data preparation: the ingestion, data quality and data organization," said Bluechel. "But organizations don't appreciate that value, because it's just the plumbing: 80 percent of the final impact is in the visualization. There's a reason they call it the final mile. You have to take that data and visualize it in a way that can be actionable."


 Read our coverage of HIMSS Big Data & Healthcare Analytics Forum in Boston.
⇒ Change management is IT's greatest challenge in delivering needed innovation
⇒ Healthcare is swimming in data, but what to do with it?
⇒ It takes a lot more than analytics to make population health work
⇒ Go figure: The key to big data is actually small data
⇒ Geisinger, Mount Sinai execs offer best piece of advice on succeeding with analytics