MD Anderson turns to analytics, big data to fight cancer, boost efficiencies

Register now

MD Anderson Cancer Center is sitting on 23 petabytes of data, including more than 2 billion diagnostic radiology images, generated by its massive IT infrastructure. But Chris Belmont, vice president and CIO, isn’t intimidated by the amount of data—he’s just scared of staring at it too long.

“Our biggest fear when we decided to move into Big Data was that, like many healthcare organizations, we’d have a two-year data ‘ingestion’ process where we’d keep thinking about that massive set of data, and connect all our systems big and small together, go get even more data from external sources, and then eventually offer our users an add-on tool and tell them to go at it,” Belmont says. “By the time we’d be done ingesting all that data, the time to change the game in terms of costs or population health would have already passed.”

MD Anderson, the Houston-based health system devoted to cancer care, isn’t the type of organization to let time slip by.

The center has embarked on a large-scale analytics effort to better understand the myriad forms of cancers it treats and develop therapies and medication regimens to combat them. In the past six months, it’s also put together a Big Data infrastructure focused on pulling nuggets from every nook and cranny from the enterprise to make it more efficient, be it procurement data, cost data, enterprise resource planning or another data set hanging out there. Central to that effort is speed: The center wants to be able to take ideas for data tools, assess and quantify their value, and then build them within a matter of weeks.

To do so, Belmont says, the IT team has kept executives and other decision makers from chasing the “bright shiny objects” and kept resources focused on building competencies with the data that feed front-end data visualization apps. “Too often I’ve seen execs get wowed by a dashboard, buy it, and then when it doesn’t do everything they want, go out and buy the next dashboard and put it on top of the old one,” he says. “We want to understand the underlying data and learn how to utilize it correctly.”

MD Anderson is no stranger to large-scaled analytics initiatives. It’s worked with IBM Watson on an oncology platform for the past few years that applies predictive analytics to cancer tumors.

The center treats more than 100,000 cancer patients annually, plus thousands more through its regional and national networks. Its “Moon Shots” program, built on its Oncology Expert Advisor clinical decision support platform, uses the data generated on patients and standardizes the longitudinal history of images, therapies and medications in a centralized warehouse to advance cancer care. The program analyzes certain types of cancers that are linked at the molecular level, and based on the clinical knowledge and technology available to treat them, provides an opportunity to rapidly develop treatments that could significantly increase the survival rates for those diseases.

Part of that effort is advanced molecular profiling, which analyzes tumors to identify genetic characteristics and flag unique biological markers. That effort yields very specific, personalized treatment pathways for patient, not only for established therapies but for various clinical trials underway.

The platform also automatically stacks radiology images—a process that could take a radiologist as long as five hours—and provides a diagnostic read of a cancer tumor’s shape, size and texture. Medication and test data is embedded in the software to enable clinicians to analyze how therapies and procedures might have affected the growth of a tumor.

While the oncology platform is an example of how analytics tools and intelligence can be applied to massive data sets and images, MD Anderson’s current Big Data drive is focused on leveraging data for efficiencies across the enterprise. Belmont sees a point, for example, at which care teams can be assembled based on affinity with the patient—perhaps a nurse grew up near a patient, or they went to the same school, or Press Ganey patient satisfaction scores show that certain staff members connected with a patient during a previous encounter.

Another example of its recent efforts is a just-launched pilot of a “cognitive” help desk that’s fueled by a machine-learning platform, says Mike Antonoff, executive director of the Office of the CIO at MD Anderson.

Users can call the help desk and have a natural language conversation with the tool, which sits on a machine-learning platform from CognitiveScale, a cloud-based analytics vendor based in Austin with which MD Anderson has formed a tight partnership, Antonoff says.

The cancer center went live on a new electronic health records system from Epic less than three months ago, so the help desk has to be able to face a deluge of calls from users asking about how to place orders or document in the EHR. Belmont and the IT department have created more than 2,000 tip sheets for the Epic system, and the help desk provides information as well as punch-out links that enable users to go directly into the system to perform tasks. It also can provide information from MD Anderson and outside sources, such as Reuters and trade magazines related to the question asked.

What’s significant about the effort and other projects in the pipeline is that the Big Data platform is sifting through massive data sets and continuously adding to a knowledge base of its understanding of the type of information users need and how they need it, Antonoff says. Plus, the original plan was to develop a cognitive platform in 12 weeks from start to finish, and MD Anderson did it in less than seven weeks. “This was a test of our efforts to build a framework to go from an idea to a tool in a very short timeframe.”

The cancer center also recently launched a patient concierge app that enables patients, most of whom are coming from other parts of the country, to tap into the local networks of restaurants, housing and other services. The mobile software can analyze a patient’s list of scheduled appointments, determine they might need a six-month home lease in the area, and put them in touch with real estate agents who specialize in those services. It also can look at the dietary guidelines entered by a physician in the EHR and suggest local restaurants that can accommodate those restrictions. The application also, with the permission of the patient, can tap into their social media data.

So, for example, if the patient is coming from New York and they’re a baseball fan, the application would send them information on getting tickets from a Yankees game though MD Anderson while they’re in town, Antonoff says.

MD Anderson has built a Big Data framework with some “traditional” components, Antonoff says: It uses the Hadoop open source framework for data storage and processing, and uses tools from Informatica for extract, transfer and load functions from its myriad databases.

“It’s not so much about the tools but about the competency around the data.”

From there, Antonoff says, “things get kind of crazy.” The cancer center uses a number of open-source applications for natural language, image and video processing. Under that are machine-learning tools from CognitiveScale as well as from IBM that are used for different initiatives. Data discovery tools and visualization dashboards from Tableau are used on the front end. Wedged in that framework is “pretty much every business intelligence tool on the market today,” he adds.

However, “it’s not so much about the tools but about the competency around the data,” Belmont says. “Part of that effort is to continuously work the data to find those golden nuggets about our staff as well as the diseases we are fighting. We keep finding insights that are roughly right and not precisely wrong. But we keep putting that in front of clinicians so they know we have this capability to find nuggets in our data and rapidly build a tool around it, if that insight is significant for them.

“One of our biggest challenges right now it to educate our staff on the potential of what we can do—many of them think that data is buried way too deep to get at, or we can’t possibly connect all the dots that they’d like us to. But we can—the technology, especially the machine learning, is now available. We just have to use it the right way.”

For reprint and licensing requests for this article, click here.