• Making Good on the Promise of Big Data in Health Care

    Laura Ramos Hegwer Nov 02, 2015

    Healthcare organizations are harnessing the power of big data as they prepare to take on greater accountability for clinical and financial outcomes. From platforms that monitor the real-time status of heart failure patients to tools that score the level of risk in an ICU, the latest data-driven innovations emphasize the importance of data quality and integrity across the enterprise.

    A patient with heart failure steps into a fast-food restaurant and, on his GPS-enabled phone, receives a reminder text from his healthcare provider to choose a low-sodium selection from the menu. Some may view this type of intervention as invasive, yet it is not unlike the data-driven recommendations offered by online retailers that have gained consumer acceptance.

    Although health care has lagged behind other sectors in efforts to harness "big data," the advent of new payment approaches may be pushing the industry to catch up.

    Preparing for Greater Risk Sharing

    "If you are in a fee-for-service model, big data is nice to have, but it isn't necessarily going to affect the bottom line," says Douglas B. Fridsma, MD, PhD, president and CEO, American Medical Informatics Association (AMIA), Bethesda, Md. "In alternative payment models where there is shared risk, you will absolutely make money or lose money depending on the power of your analytics." For instance, big data applications could give providers greater insights on the most cost-effective treatments for specific patient populations.

    Douglas B. Fridsma, president and CEO, American Medical Informatics Association

    Douglas B. Fridsma, MD, PhD, president and CEO, American Medical Informatics Association, Bethesda, Md., believes alternative payment models will spur the adoption of more analytics tools. (Photo: American Medical Informatics Association)

    Fridsma, formerly the chief science officer for the Office of the National Coordinator for Health Information Technology, credits the move toward value-based care for spurring providers' interest in adopting more predictive analytics. For example, the Centers for Medicare & Medicaid Services has set a goal of making more Medicare payments through alternative payment models (30 percent by the end of 2016, 50 percent by the end of 2018). At the same time, commercial payers continue to engage providers in risk-based population health management strategies.

    Seeing opportunities. David W. Bates, MD, MSc, senior vice president and chief innovation officer for Brigham and Women's Hospital (BWH) in Boston, says most U.S. healthcare organizations can do a much better job of realizing the potential of big data. "We are really not sophisticated compared with other industries in terms of using big data, and there are opportunities everywhere we look," Bates says.

    David W. Bates, senior vice president and chief innovation officer, Brigham and Women’s Hospital.

    David W. Bates, MD, MSc, senior vice president and chief innovation officer for Brigham and Women’s Hospital, Boston, says organizations should strive for a “self-serve” approach to data. (Photo: Brigham and Women’s Hospital)

    Bates believes many healthcare organizations in risk-sharing arrangements are using data only to identify which patients are high-cost. "Unfortunately, the algorithms are not that sophisticated, and the way these patients are actually managed is not driven by data," he says. "For example, we look at which pharmaceuticals we are spending the most on, and we put programs in place to manage those medications, but we don't have tools in place to figure out which patients are most likely to benefit from those drugs."

    Bates coauthored an article last year that identified six ways in which big data could help manage patients (Bates, D.W., et al., "Big Data in Health Care: Using Analytics to Identify and Manage High-Risk and High-Cost Patients," Health Affairs, July 2014). In the article, Bates and his colleagues identify how predictive analytics can reduce healthcare costs in the following categories:

    • High-cost patients
    • Readmissions
    • Triage
    • Decompensation
    • Adverse events
    • Treatments for diseases affecting multiple organ systems

    Overcoming challenges. Bates does not blame interoperability issues for the healthcare industry's slow adoption of predictive analytics. "You can do a great deal with just your own data," he says. Rather, the problem has to do with personnel. "Healthcare organizations don't have groups with the right training to understand how to use data to reduce costs and improve care," he says. "If they do, the groups are relatively small and completely consumed with meeting external requirements, such as reporting quality data. They just don't have the bandwidth."

    Another problem is that up-to-date analytics software and tool kits—especially those that take a more "self-serve" approach to data—have not been available until recently. "Ideally, you'd like someone who is a mid-level manager, not just an analyst, to be able to consult the data and ask questions," Bates says. "Making that possible requires that you put in place the governance so data are protected."

    Knowing where data come from. Understanding the origin of a piece of art can dramatically affect its value, and the same is true for data. Data can be blurred if organizations that are trying to share data lack the ability to put the data in the proper context.

    For example, the diagnosis of high blood pressure may have a different threshold in patients with renal failure than in healthy patients or those who have diabetes, creating different definitions in different data sets. "If you understand the original source of the data sets, you have a better chance of being able to integrate them effectively," AMIA's Fridsma says. "If you don't have the ability to merge data, then the data are less useful for predictive analytics."

    Healthcare organizations are finding they have a lot of data that may be of questionable value. "The data sets are probably not high-quality enough for the kinds of clinical questions providers want to ask," Fridsma says. However, that is changing thanks to data governance efforts at healthcare organizations across the country.

    Laying the Foundation for Data Governance

    When Chris Harper came to The University of Kansas (KU) Hospital, Kansas City, Kan., to lead a new business architecture and analytics team in 2013, his first step was establishing good governance practices throughout the academic medical center and its 80 care sites. One obstacle was that the organization used 60,000 separate databases to store data. "Before we spent a dime on analytics, we needed to come to an agreement on how we were going to manage our data," Harper says. Before investing in a new data warehouse and in analytics technology that KU Hospital needed to prepare for value-based care, Harper and his colleagues spent 18 months creating an enterprise-wide framework for data governance.

    The University of Kansas Hospital’s Data Governance Committee
    The University of Kansas Hospital’s Data Governance Committee

    Creating a data governance charter and committee. Harper, with the help of key operational partners, established a data governance committee that is responsible for ensuring data quality, usability, and accessibility across the organization. The committee includes an executive group, composed of C-suite executives and key physician leaders from the university and health system, that has decision-making authority on all matters related to data. Having support from senior leaders has been critical. "A lot of organizations hold their data close to the vest because knowing and understanding how to get information is associated with power," Harper says. "As you start to move toward enterprise-wide data governance, you need some of that top-down authority to let groups know that it is OK to share, consume, and standardize data."

    Chris Harper, director of business architecture and analytics, University of Kansas Hospital.

    Chris Harper, MBAi, MPM, director of business architecture and analytics at the University of Kansas Hospital, Kansas City, Kan., recently implemented a data governance committee. (Photo: University of Kansas Hospital)

    The committee also includes a data advisory group composed of data producers and operational, clinical, and technology leaders who provide recommendations. Data work groups also are convened to address specific issues, such as standardizing a particular data definition across the organization.

    One of the committee's first initiatives was writing a data governance charter that outlines the organization's goals for information governance as well as data quality, usability, and availability. As Harper says, "Technology is really the least important part of this. It's really about people's ability to collaborate and make decisions together."

    Standardizing data definitions. For a large organization such as KU Hospital, Harper says, creating enterprise-wide data definitions is time consuming. Even defining length of stay was more difficult than expected. "We needed a work group to determine how we were going to measure the date and time of admission," he says. "Is it the time the patient walks in? Is it the first time a clinician interfaces with the patient? Is it when the encounter is captured in the electronic health record?"

    Data Definitions at the University of Kansas Hospital
    Data Definitions at the University of Kansas Hospital

    Since establishing the data governance committee, KU Hospital has formalized 70 enterprise-wide data definitions. "If anyone is reporting length of stay or a readmission, for example, we expect them to use the enterprise definition," Harper says. "Our goal is not to create roadblocks but rather to make reporting consistent at the enterprise level." Leaders at KU Hospital also have reduced the number of databases outside the data warehouse to 40,000, with the goal of reducing that number to zero. Beyond improving transparency and usability, having fewer databases also helps reduce risk. "For data that are considered high-risk, we want to make sure we have those in a secure, enterprise-wide tool like the data warehouse, rather than in an individual database that is more likely to be compromised," Harper says.

    Finding the right people. For Harper, a primary concern is the difficulty of finding experienced data workers and managers in Kansas City. "Although Kansas City is a wonderful place, it is hard to find healthcare technology workers, and it took 18 months and 30 interviews to hire one data manager," Harper says.

    To help identify whether KU Hospital had the talent in-house, Harper created a resource skills matrix that mapped out the competencies the organization needed to move forward with analytics. "We actually had 80 percent of the right skills in-house; they just needed new tools and training," Harper says. Today, the organization has 15 data workers and managers in three groups: data and technology (which Harper oversees), quality and process improvement, and strategic planning and finance. These individuals are segmented into two main roles: data management and information delivery. KU Hospital also partners with a key strategic vendor on its analytics and data warehouse needs.

    Testing Smartphones and Sensors to Improve Heart Failure Outcomes

    Each year, more than 1 million heart failure patients are hospitalized, and the total cost of caring for heart failure is estimated to be $30.7 billion (Mozaffarian D., et al., "Heart Disease and Stroke Statistics—2015 Update. A Report from the American Heart Association," Circulation, Jan. 27, 2015). Researchers at The Ohio State University (OSU) believe mobile sensors, cell phones, and analytics software could help prevent unnecessary heart failure hospitalizations and, ultimately, help contain the costs of caring for these patients.

    Making the most of mobile apps. OSU is one of 12 universities participating in the Center of Excellence for Mobile Sensor Data-to-Knowledge (MD2K). MD2K has received $10.8 million from the National Institutes of Health as part of the Big Data to Knowledge (BD2K) initiative to study the use of sensors and mobile technology to gather, analyze, and interpret data.

    "A goal of the MD2K project is to use modern sensor technology, deep computing and analytics, and smartphone platforms to get an early warning of worsening heart failure so clinicians can adjust therapies remotely and keep patients out of the hospital," says William T. Abraham, MD, FACP, FACC, FAHA, FESC, director of cardiovascular medicine at OSU's Wexner Medical Center. Abraham is studying a unique biosensor designed by OSU bioengineer Emre Ertin, PhD. The biosensor is placed on the chest for a few minutes and uses electrical waves to detect fluid in the lungs and chest. This measurement can help clinicians determine whether a heart failure patient is stable or getting worse.


    William T. Abraham (left), MD, FACP, FACC, FAHA, FESC, director of cardiovascular medicine at The Ohio State University Wexner Medical Center, is testing a predictive analytics tool to prevent heart failure hospitalizations. (Photo courtesy of The Ohio State University Wexner Medical Center)

    "Part of the challenge of managing heart failure patients is determining which data are truly actionable," Abraham says. "We need to create data-driven algorithms for patient care, and we cannot rely on our intuitions, which are often wrong," he says. Daily weight monitoring is a prime example. "For some time, we believed that having patients monitor their weight every day would be a great way of knowing when they were getting worse," Abraham says. "But it turns out that the sensitivity of using daily weight to predict heart failure hospitalizations is only 10 percent to 20 percent. We hope that a single measurement, such as a lung fluid measurement, or combination of measurements—such as lung fluid, heart rate, and breathing rate—will be most predictive of how a patient is doing and guide therapy so we can adjust medications and keep patients out of the hospital."

    Empowering clinicians and patients. OSU's biosensor technology is being tested among inpatients, and the study will be expanded to include outpatients next year. Ideally, the final product will issue smartphone alerts to clinicians only when intervention is needed. "The idea is to let the computer platform analyze the data and then present what really counts to the clinicians on their smartphone," Abraham says. "If the algorithm determines the patient is at risk, the clinician might alert the patient to take an extra dose of a water pill." Such a tool also could be used by heart failure patients for self-monitoring, similar to how patients with diabetes use glucometers, he adds.

    Using Data to Create a Safer ICU

    Many big data initiatives are focused on using analytics to determine how an individual patient's risk can change. But researchers at Boston's Beth Israel Deaconess Medical Center (BIDMC) are harnessing analytics to predict how a complex environment such as an intensive care unit (ICU) can become riskier.

    Specifically, Kenneth Sands, MD, MPH, chief quality officer, and his colleagues are looking at how a combination of clinical parameters derived from ICU patients—as well as environmental factors such as staffing and the "churn" of patients coming in and out of a unit—affects a patient's risk of harm. These harms include cardiac arrests, falls, medication errors, acute bleeding episodes, and similar events that can happen immediately. Problems that tend to manifest over time, such as catheter-associated urinary tract infections, are more difficult to attribute to the short-term ICU environment and are not being measured as part of the study, Sands says.

    Helping staff make decisions to reduce risks. Now in the "proof of concept" stage, BIDMC's "Risky States" application could allow managers to make decisions that would lower the risk of immediate harms in their ICU. How it might work: A web-based status dashboard in a nurses' station would be updated every 15 minutes to show the current risk of harms across several ICUs, using a red, yellow, and green color-coded intensity index. If the intensity index is yellow in the cardiac care unit (CCU) and green in the medical ICU, a nurse manager might decide to float a nurse from the MICU to the CCU. Or the nurse manager may choose not to take a patient transfer to the CCU until another patient is discharged from the unit the following morning.

    Sands expects the tool to go live in the ICUs at BIDMC this fall.

    Cleaning the data. Sands says the most challenging part of the project has been "cleaning the data" from multiple sources for use in the project. The team analyzed two years of retrospective patient data from seven BIDMC ICUs to determine which scenarios were associated with higher patient risks. "We underestimated the time it would take to get clean, analyzable data," Sands says. "Ensuring the data were legitimate for analysis required multiple steps of making sure the fields were valid and that the information was complete." To help with the analysis, Sands partnered with systems engineers at the Massachusetts Institute of Technology.

    Getting data ready for research queries. Ensuring data integrity—essentially, its accuracy and validity across the life cycle—is a critical step in research projects that aim to harness big data, Sands says. "Even before you generate the research question, make sure that your information systems are set up so they can be queried in the aggregate so you can get that information into a data mart," Sands says. "You also need a way to merge data sets and an emphasis on making sure the data has integrity at the times it is collected."

    Lessons Learned

    Using analytics to transform care and reduce costs requires an enterprise-wide approach to ensuring data quality and integrity. Healthcare leaders who are making strides in their big data efforts offer the following advice.

    Establish a data warehouse. "Make sure that people who want to use data to make care better can actually get at the data," BWH's Bates says. The architecture of the data warehouse should include several "data marts" so different data are available to different stakeholders. He also advises adding robust "self-serve" tools so non-analysts can query the data in the warehouse.

    A Look at an Enterprise Data Warehouse
    A Look at an Enterprise Data Warehouse

    Create a culture that supports interdisciplinary teams. "Medicine cannot be practiced anymore in isolation or by subspecialty," OSU's Abraham says. "We need to bring together information technologists, bioengineers, physicians, and clinical scientists to make this work."

    Bates recommends hosting team-building training for leaders in IT, research, quality, finance, performance improvement, and other areas.

    Include senior leaders on the data governance committee. KU Hospital's Harper credits the support of the C-suite for creating the political capital needed to make decisions across the enterprise.

    If using a third party for analytics, pick a vendor that understands your goals around data. KU Hospital chose a vendor to automate data collecting and reporting via its new data warehouse. As part of the collaboration, leaders at KU Hospital have shared their "big data road map" with the vendor. The road map spells out specific goals for the organization over the next two years.

    Conduct market comparisons to make sure your organization compensates data analysts fairly. Bates says healthcare organizations often do not realize that data analysts can make more in other industries, and that organizations need to adjust their salaries to attract and maintain the best talent.

    Healthcare Analytics Adoption Model
    Healthcare Analytics Adoption Model

    Maintaining Trust with Patients

    AMIA's Fridsma believes that as big data initiatives gain more traction, leaders need to keep the focus on the patient and be transparent about how they are using data.

    "The most powerful innovation that we have in health care is the patient," Fridsma says. "The worst thing that could happen for big data and analytics is for patients to lose confidence that the systems are there to improve their care experience. The real power of big data and analytics is in using them to engage the patient in a conversation in which they are first-order participants in their care."

    Laura Ramos Hegwer is a freelance writer and editor based in Lake Bluff, Ill.

    Interviewed for this article: Douglas B. Fridsma, MD, PhD, president and CEO, American Medical Informatics Association, Bethesda, Md.

    David W. Bates, MD, MSc, is senior vice president and chief innovation officer for Brigham and Women’s Hospital, Boston.

    Chris Harper, MBAi, MPM, is director of business architecture and analytics at The University of Kansas Hospital, Kansas City, Kan.

    William T. Abraham, MD, FACP, FACC, FAHA, FESC, is director of cardiovascular medicine at The Ohio State University Wexner Medical Center, Columbus, Ohio.

    Kenneth Sands, MD, MPH, is chief quality officer, Beth Israel Deaconess Medical Center, Boston.

    CommentsPlease login to add your comments

    Add Comment

    Text Only 2000 character limit