November/December 2018

Harnessing Data
By Sarah Elkins
For The Record
Vol. 30 No. 10 P. 14

Organizations that can corral data stand to benefit the most from their intrinsic value.

The world has become a vast collection of data. Every perceivable thing, from finance to desire, is now broken into its smallest identifiable components and assigned a value. Seemingly every changeable item can be tracked, right down to the activity of every single gene in every single cell in the human body.

These data are at our fingertips—and growing. Indeed, "by 2016, 90% of the world's data had been created in the last 12 months," according to a report generated the same year from IBM Marketing Cloud.

In short order, data scientists have scrambled to come up with new ways to take data from identifiable to actionable. The big question driving big data has been: What can we do with this information?

Criminals are using big data to commit fraud. Financial institutions are preventing it with the same tools. The data are the weapons on both sides of every battle, and they've snuck into our most intimate relationships. For example, Netflix is better than your spouse at predicting what movie you want to watch tonight.

Big data has delivered the ease—and the paranoia—of our modern virtual lives, but it has the potential to do much more.

The human benefit of harnessing data has no greater potential than in the health care field. Data analytics is revolutionizing what is possible, from better appointment scheduling to more effective cancer treatments.

Here's how three organizations and their respective experts are driving health care forward through data.

Clinical Data Disorder
Before data can be used to solve any problem, they must be made usable. At the most basic level, that's what Diameter Health is doing. Its team of engineers and scientists have developed a platform that aggregates and normalizes EHR data for use in a myriad of clinical and analytical applications.

"Industry experts estimated that data scientists spend up to 80% of their time doing manual and semimanual 'data wrangling,' or simply organizing unruly data in preparation for analysis. At Diameter Health, we call this condition 'clinical data disorder,' says Eric Rosow, CEO of Diameter Health.

That's where Rosow's team comes in. "By normalizing and enriching clinical data, health care data analysts spend less time data wrangling and more time providing analytic deliverables and insight," he says.

But what does it mean to normalize data? Rosow explains, "There are 91 different ways to code heart failure within multiple code systems. If an analyst were asked to study heart failure using raw clinical data, the analyst would need to hunt and peck for all the variants of heart failure."

Therein lies the problem with big data—it's just so huge. Organizations such as Diameter Health harness technology to do the "hunting and pecking" so the analysts, researchers, and providers can reach their ultimate goal: improve outcomes for patients suffering from heart failure.

From Diameter Health's position serving health care organizations, health information exchanges, accountable care organizations, and other technology vendors, they enjoy a broad view of the many ways data are being used. According to Rosow, clients are tackling problems such as "identifying opioid prescription patterns, predicting chronic kidney disease, ambulatory quality reporting using clinical data from across the continuum of care, and incomplete risk adjustment."

Through technology, Rosow is able to watch "chaotic clinical data" evolve into "actionable analytics in support of better care at lower costs."

Some of the most noteworthy actionable analytics projects involving Diameter Health have resulted in collaboratively published peer-reviewed research. For example, in partnership with the VA, Diameter Health published "Interoperability Progress and Remaining Data Quality Barriers of Certified Health Information Technologies." The partnership's goal is to "improve the quality and consistency of clinical data received from [the VA's] trading partners representing 9 million veterans," Rosow says.

If producing and parsing data are the first two Ps of data analytics, the third is prediction. Once large data sets can be reliably delivered and organized in a meaningful way for analysis, the next logical step is to find connections and follow trends to make predictions.

That's exactly what Brigham and Women's Hospital in Boston had in mind when the organization sought to predict the risk of kidney failure by extracting EHR data using the consolidated clinical document architecture interoperability standard.

In pursuit of that goal, Diameter Health assisted in the development of an application used by primary care providers. The resulting study, "Implementation of a Scalable, Web-Based, Automated Clinical Decision Support Risk-Prediction Tool for Chronic Kidney Disease Using C-CDA and Application Programming Interfaces," was published in the November 2017 issue of the Journal of the American Medical Informatics Association.

Data analytics isn't a perfect science because data—at least when they emanate from an EHR—are filled with gaps and errors. "Useful conclusions may be drawn from imperfect data, but enhancing the data improves the value derived from all analytic tools," Rosow says. "While you cannot say that true conclusions cannot be drawn from imperfect data, we can say that false and potentially even harmful conclusions can result from imperfect data."

To combat imperfect data, Diameter Health developed an application that measures the completeness and accuracy of clinical data. As a result, it can assess whether a data set holds value.

In the long term, Rosow believe big data will give health care a better chance of meeting its lofty goals. "[Big data] will drive us closer to the industry-stated goals of improving patient outcomes at a lower cost," he says. "We will see better allocation of resources, physicians will gain more value from clinical decision support, and we will continue to remove waste from the system."

Provider Attribution
Officials at Dignity Health are zeroing in on ways to improve patient outcomes through the application of data analytics. Hospital executive leadership discovered issues with several hospital metrics that referenced physician attribution using the discrete EHR data field "Attending Physician." The information was throwing a wrench in the organization's attempts to evaluate important metrics such as patient satisfaction, mortality, and surgical complication rates.

For internal benchmarking and quality analytics, leadership turned to the Dignity Health Insights (DHI) division to help create an algorithm—dubbed Substitutable Medical Approaches and Reusable Technology (SMART) Provider Attribution—that would help determine which provider was most responsible for a patient's care. By doing so, key metrics could be attributed to a specific provider.

"Because being able to accurately identify the provider responsible for a patient's care helps paint a clearer picture of hospital metrics, hospitals needed to have a better way of attributing care. This was the catalyst for the algorithm," says Angelia Chanco-Larios, MD, a clinical analyst at Dignity Health. Furthermore, "a patient may see a lot of providers—physicians, nurse practitioners, physician assistants—during their stay. A patient may have a complex hospitalization stay with multiple involved specialties or may be admitted in a teaching hospital with multiple residents, fellows, and consultants."

In short, there are numerous reasons why it's challenging to identify a patient's attending physician, but the DHI department resolved to come up with an algorithmic approach that would resolve the issue. SMART Provider Attribution proved extremely effective in predicting the provider attributed to a patient's care.

As data analysts are wont to be, Chanco-Larios and her team were curious how they could further utilize SMART Provider Attribution. "We wondered if we could leverage the thinking behind it to create [an algorithm] that predicts the attributable provider for a patient's length of stay. Happily, the algorithm performed admirably as well," she says.

Developing the algorithm was only half the battle. The data still needed to be compiled in a digestible format to allow the areas in need of improvement to be identified. The next step for the DHI team was to build an in-house app that would pull data and identify outlier encounters, such as patients whose stay was longer than expected.

"Through the dashboard, we were able to bucket these long-staying patients into diagnosis-related groups, view them by payer types, evaluate if these patients are outliers due to physician practice or placement issues, and drill down to a more granular level to identify providers with exceptional metrics," Chanco-Larios explains.

By doing so, Dignity Health was able to gauge the performance of regions, hospitals, and individual physicians, and learn the keys to both success and failure.

Chanco-Larios doesn't lose sight of the humanity behind the mathematics. "This data work must be done with the human element in mind because the answers that data analytics provides must be used in very human situations," she says. "In the end, data should be interpreted correctly, implemented strategically and ethically. Doing so will make an immeasurable difference for both the patient and their care team, making their lives easier and safer."

Cancer Prediction
The human element of health care is perhaps no more striking than in the struggle to abate global cancer rates. According to figures released by the World Health Organization in September 2018, nearly 10 million people will die of cancer this year. Elana Fertig, PhD, an associate professor of oncology, biostatistics, and bioinformatics and associate professor of applied mathematics and statistics at Johns Hopkins University, knows better than most how challenging the cancer conundrum is.

"Cancers change over time. It's not a constant, static system," she says. This is a simple analysis of a complex problem, but nothing about Fertig's work is simple. As the assistant director of the research program in quantitative sciences at Johns Hopkins' Sidney Kimmel Comprehensive Cancer Center, she helps the lab develop and apply new quantitative methods for cancer biology. The big data at their fingertips is the human genome, all 3,088,286,401 base pairs in the human DNA sequence.

Fertig seeks to understand, for example, which T cells will respond to immunotherapy and how that response will change over time. The lab monitors how cancer cells change, then combines the recorded changes with new computational tools. The team examines gene activity within a single cell. "The ultimate goal," she says, "is to be able to predict how a cancer will change."

Interestingly, Fertig brings a background in weather prediction to her work in immunology. From that perspective, she has a greater appreciation for cancer research. "Biology came from a field of naming and describing and reducing things to their smallest parts. Meteorology came from a field of more abstraction. [Weather] is limited in terms of four or five variables," she says.

Fertig believes biology will evolve in the next 10 to 20 years, eventually taking a big-picture view of how connected systems work together, an approach borrowed from weather prediction.

"We really need to combine these systems. In both fields we're getting this influx of data at ever, ever finer scales. The trick is merging that microscopic view of what the data are giving us with the bigger picture fundamental rules about how systems work in order to get a clear picture of what's going on. In either field, it can't be one or the other," she says.

For now, Fertig has plenty to fill her time. She works with researchers and practitioners who bring her their most vexing problems, what she calls "the questions that can't be answered with current tools."

"That's where my work comes in," Fertig notes.

Sometimes she has a quick answer for her colleagues' problems. They deliver data and she runs them through a standard analytics pipeline. Other times, Fertig is pressed to develop the mathematics to analyze the data. In those cases, one question can turn into a long-term project.

Right now, Fertig is most interested in how different systems and scales of biological processes work together in therapeutic responses.

"How do the data at all these different scales fit together to give us a picture of what is going on in the cancer. It poses both biological challenges and … really interesting mathematical challenges—what's the right data?" she says.

As health care comes up with more interesting ways of using data to answer complex questions, the greatest challenge may be in zeroing in on what's most important while simultaneously expanding our view of what's possible.

"Everyone in big data wants one right answer. This is what the data are telling me," Fertig says. "The reality is, if we have these big data sets, the data have the power to tell us multiple things. It's a question of where to focus your investigation. Which is most critical? How do you filter that?"

— Sarah Elkins is a freelance writer based in West Virginia.