• Using Business Intelligence Intelligently

    May 25, 2012

    By Lola Butcherpg12_-art-for-feature-business-intelligence

    Data-savvy providers are using business intelligence to improve care at the bedside, coach physicians to use resources more wisely, manage populations of patients, and improve financial performance.

    Key Takeaways

    Providers at the forefront of business intelligence share a number of lessons learned:

    • Identify the best data storage approach for your organization: Some providers are using physical data warehouses; others are employing virtual approaches
    • Enlist the help of clinicians and other frontline staff in identifying the right data to capture, the right performance targets, etc.
    • Employ straightforward performance dashboards that use visual cues to help leaders and staff quickly identify when they need to take action
    • Institute a process improvement process for addressing issues identified through business intelligence

    Across the country, health systems are turning data into useful information in surprising new ways. By combining data from a myriad of sources, analyzing it to inform specific processes, and presenting it in easy-to-understand formats, they are improving efficiency, lowering costs, and saving lives.

    For example, Denver Health is using business intelligence to identify patients who are likely to have a heart attack or other serious cardiac event. A programmed "workflow" continually scans the hospital's data warehouse in search of patients who are deteriorating. If certain factors-for example, oxygen levels or systolic pressure-hit predetermined thresholds, clinicians are notified so they can intervene.

    "We can now predict hours ahead of a cardiac event that a patient is likely to code," says Gregory Veltri, the system's CIO. "As a result, in our non-ICU areas, we have experienced zero cardiac resuscitation events for some quarters."

    While quick to praise the noteworthy improvements attainable with business intelligence, those in the trenches warn that these efforts are time-consuming and often frustrating. As the industry adopts value-based performance approaches, many healthcare leaders-from physicians and nurses to administrators and IT-will likely see their responsibilities for data collection, analysis, and monitoring increase in future years.

    "The most important thing to know is that business intelligence is a work in progress," says Michael C. Lindberg, MD, chair of the medicine department, Hartford Hospital. "You can't just say, 'This is it,' and leave it alone. You have to be looking at the data all the time and making adjustments."

    Collecting and Mining Data

    One key upfront decision that healthcare organizations have to make about their business intelligence strategy is where to house all the data. Denver Health and Northeast Georgia chose different approaches.

    Creating a single data warehouse. Denver Health built a data warehouse for financial data in the mid-1990s and began adding clinical data in 2005. Since then, the data warehouse has grown to support registries that help manage chronically ill patients, inform quality reporting and benchmarking, and develop patient management reports-not to mention dashboards, or the easy-to-read interfaces that allow clinicians and executives to quickly assess progress on key measures.

    "We have produced CFO dashboards, CEO dashboards, core measure dashboards, and ambulatory indicator dashboards," says Veltri. "And we're now actually producing a hospital dashboard that tracks all of our meaningful use indicators, plus all of our core measure and regulatory indicators."

    Veltri calls the warehouse "the single source of truth" for Denver Health, and access to that truth is disseminated throughout the organization.

    "We wanted analytical tools that were easy to use with the idea that departmental super users or physicians would eventually mine the data," he says. "Today we have approximately 50 physicians and departmental users plus 12 IT staff who leverage the data warehouse, enabling improved quality, safety, and patient outcomes."

    Building a data warehouse is a huge IT project that can cost millions of dollars, depending on the size and scope of such a project. Costs include not only the tools for development, but hardware, storage, training, and resource costs. For an organization the size of Denver Health that wants to centralize organizational reporting as much as possible, the cost of a data warehouse is an ongoing expense, says Veltri.

    "A lot of organizations are not willing to invest millions in a project without an upfront definable ROI, but there's no return in building a data warehouse until the foundation is built and data integration in some form is completed," says Veltri. "And the return received is based on how the platform is leveraged by organizational leadership. If you're not willing to take some risks, then just don't start."

    Pulling data from multiple sources. At Northeast Georgia Health System in Gainesville, Ga., the business intelligence journey began two years ago. Instead of a traditional, physical data warehouse, Northeast Georgia employs a so-called "virtual" data warehouse strategy, using technology that allows data to be pulled from multiple sources without first being aggregated inside a warehouse.


    "Knowing how fast health care is continuing to change, we believe we can accomplish our virtual enterprise intelligence strategy with this newer technology rather than having everything in a traditional data warehouse," says CIO Allana Cummings.

    Northeast Georgia's business intelligence tool uses "in memory" computing, an approach that is less expensive than building a data warehouse because there are no costs for the warehouse hardware, software, implementation, or support. Cummings says it allows for data to be analyzed more quickly. However, because data is pulled from disparate sources, there is no central point of control that ensures that all data is "normalized" for analytical purposes.

    In Cummings' view, that does not present a problem as data can be normalized on an as-needed basis, which is required in a small percentage of analyses. Northeast Georgia's business intelligence work is overseen by a data quality council that is chaired by the system's chief data architect and includes representatives from clinical, ancillary, quality improvement, revenue cycle, and other areas of the health system. "The council's goal is to help us ensure the integrity of data being reported in the organization, and that starts at the point of actual data collection," says Cummings. "We're looking at standardizing definitions, how the information is entered, how it is pulled, and how it is used."pg16_photo-doctors-miller-baily

    From left, Northeast Georgia's Zan Miller, chief data architect, and James M. Bailey, MD, chief medical information officer and chief quality officer, serve on the health system's Data Quality Council that oversees business intelligence work.

    Transforming Data into Useful Information

    Connecticut's Hartford Hospital, an 867-bed regional referral center, is gaining market share as its ability to move patients through the hospital improves. As a result of monitoring pertinent data and taking needed actions, the hospital is now discharging patients in a more efficient manner. Every Sunday night, inpatient data from several systems-patient billing, bed management, and transport-is compiled and loaded into the hospital's patient throughput dashboard. "Come Monday morning, we can pull up the latest data from the last week," says Michael C. Lindberg, MD, chair of the medicine department.

    The dashboard includes three subsections:

    • Discharge metrics, including the following:
      • The percentage of patients discharged by 11 a.m.
      • The percentage of orders for discharge that are in by 10 a.m.
      • Average length of stay
      • The number of opportunity days (the days that patients stay beyond their predicted length of stay)
    • Bed management measures, including the length of time between when a bed is requested and is assigned
    • Capacity metrics, including average daily census, the number of beds that are closed, and the readmission rate for the previous 30 days

    Since the dashboard came online in 2009, the percentage of patients meeting the hospital's early discharge target of 11 a.m. tripled-increasing from 9.5 percent to 30 percent.

    Lindberg keeps his eye on the weekly updates to ensure no one is backsliding. "If I see a physician who is starting to drift-for instance, if three data points show him missing our patient throughput goals-I will pull him into my office," he says. "I pull up the graph and show him, 'This is you. Here's how you compare with everyone else. Let's take a look at this right on the screen. Why are you having trouble?'"

    A multidisciplinary group including pharmacists, hospitalists, nurses, midlevel practitioners, and representatives from various clerical areas are tasked with combing the patient throughput dashboard for information that can be immediately put to use. "All of us go over this stuff on a weekly basis to see what the trends are," says Lindberg. "We go to individual patient floors to give them feedback. We reward the floors that are doing great, and we use the lessons learned from the great floors to help the floors that are having trouble."

    For example, the dashboard revealed that a medical floor consistently had an extremely long length of stay for its patients and a lot of opportunity days. Both results contributed to a backlog in the emergency department (ED). "We slightly restructured our hospital medicine service to put two dedicated hospitalists on that floor," says Lindberg. "That floor was then able to reduce its length of stay to below the hospitalwide mean by simply redistributing resources."

    In another case, dashboard data showed that the length of stay was unnecessarily high for medical patients who were placed on a floor that is usually reserved for surgery patients. "By getting surgical midlevels-physician assistants and nurse practitioners who were working on that floor-to not only take care of their surgical patients, but to help with medical patients, they improved throughput and helped get the patients out earlier in the day with a shorter length of stay," he says.

    Having met its early discharge target, Hartford is currently focusing on readmissions: Its overall readmission rate is currently about 10 percent-which is below the hospital's 2012 target of 12.3 percent. Its heart failure readmission rate has fallen from 24.7 percent in early 2011 to 21.4 percent this year. "We think we can get our heart failure readmissions down to 15 percent or lower," says Lindberg.

    Using Data to Save Lives

    Mercy Hospital St. Louis, the flagship hospital of the Sisters of Mercy health system, is reducing sepsis mortality with a business intelligence approach. Data pulled from electronic health records (EHRs) is being used to monitor-and improve-the hospital's handling of severe sepsis and septic shock.

    Shortly after the 31-hospital Sisters of Mercy Health System installed its EHR system, hospital leaders engaged a vendor to start mining data that could be used to improve clinical performance. The upshot: new care protocols for patients who develop one of the deadliest hospital-acquired conditions.

    Severe sepsis strikes about 750,000 Americans each year, and it is estimated that between 28 percent and 50 percent of these people die-more than the number of U.S. deaths from prostate cancer, breast cancer, and AIDS combined (National Institute of General Medical Sciences, 2009). Researchers estimate that about half of sepsis cases may be hospital-acquired (Eber, M.R., et al, "Clinical and Economic Outcomes Attributable to Health Care-Associated Sepsis and Pneumonia," Archives of Internal Medicine, Feb. 22, 2010, vol. 170, no. 4).

    Timely diagnosis and treatment are essential for survival, says Robert Taylor, MD, a critical care physician, Mercy St. Louis. Evidence-based guidelines call for a "resuscitation bundle" of treatments to be delivered within six hours of diagnosis and a "management bundle" to be administered within 24 hours.

    Like most hospitals, Mercy Hospital St. Louis did not know how well it was complying with those guidelines until EHR data was available.

    An analytics vendor extracted data from 27,000 de-identified patient records at four hospitals in the Mercy system and created a database that could be analyzed to reveal the facts: Overall, the four hospitals were less than 3 percent compliant with the entire six-hour bundle and less than 20 percent compliant with any of the individual elements.

    That was just the information Taylor needed to start an improvement initiative. The "change elements" of that initiative include the following:

    • Expediting admissions from the ED to the ICU so treatment can begin
    • Integrating a sepsis response team with the hospital's rapid-response team
    • Developing sepsis care pathways
    • Embedding sepsis order sets in the EHR
      • Arranging a stat order with the pharmacy so that antibiotics will be delivered in less than an hour after they are ordered

    Mercy St. Louis began tracking several measures on a daily report card: time from sepsis diagnosis to ICU admission, time to placement of a central-venous catheter, time to achievement of the volume resuscitation goals of the sepsis care bundle, and time to administration of antibiotics.

    The results are astounding: In the first months of 2012, the mortality rate for severe sepsis at Mercy St. Louis was 5.5 percent-down from 22 percent in early 2011. At the same time, the ICU length of stay for sepsis patients has fallen from eight days to three, and the overall length of stay for the sepsis diagnosis has decreased from 10.6 days to 8.7.

    "We are talking about a disease process that will kill one in four patients, and our survival rate has improved remarkably in just six months," says Taylor. "Our ICU and total hospital lengths of stay have also decreased. Those are important, fundamental markers that suggest that we are already making a difference."

    The sepsis treatment protocols that Taylor is developing will be disseminated throughout the Mercy system. Meanwhile, Mercy leaders are returning to the database to find other opportunities for improving performance.

    Addressing the Challenges

    "We're really excited about what we are already able to do, and what we are going to be able to do," says Tim Smith, MD, vice president of research at Mercy's Center for Innovative Care. That said, using EHR data in this way presents many challenges.

    Capturing the right data. The first is ensuring the EHR captures the right data in the right format so it can be analyzed to produce actionable information. "It takes teams of people to be able to do that, and even with the help of an outside analytics firm, it takes a lot of resources," says Smith.

    Contracting with an outside vendor that will work with patient data is time-consuming because the health system must ensure that it is complying with the Health Insurance Portability and Accountability Act. "There's a lot of work that goes into that before you can even push the first button," he says.

    Overcoming those data challenges-not to mention the process improvement challenges that follow-is worth the work when lives are being saved, says Smith: "When this whole story unfolds in the next year or two, we are going to be able to point to some pretty phenomenal success rates in sepsis management."

    Identifying the right targets. While the technical aspects of business intelligence have been worked out in other industries, the conceptual questions specific to health care are just now emerging. For example, when administrators and clinicians have the opportunity to get information they have never had before, it is not always clear what information they need or how it should be used.

    Veltri discovered this when Denver Health created a registry for hypertension patients. Clinicians could-for the first time-see that about 30 percent of patients had blood pressure under control. Veltri assumed the visual target to be used on the registry dashboard was 100 percent of patients, but clinicians pushed back: "They said, 'Is it possible to get to 100 percent?'"

    After a debate on whether targets should be achievable or aspirational, physicians agreed to set the target at 70 percent. In the three years since the registry was established, blood pressure is under control for 71 percent of Denver Health's hypertensive patients, compared to approximately 50 percent nationally.


    The achievable-versus-aspirational target is just one of many issues that must be considered when presenting data that is used for decision making. Veltri recommends that all business intelligence requests be evaluated on six questions:

    • Is there agreement on what to measure?
    • Can anything be done to affect the measure?
    • Are there specific goals or targets?
    • Are there defined critical thresholds?
    • Is an upward or downward trend "better?"
    • Is there such a thing as "too good?"

    "Learning to create the right metrics is a journey, and these guidelines can help an organization measure metrics that matter," says Veltri. "Understanding what is being measured, and why, is as important as the measurement itself."

    Helping users interpret the data. Electronic dashboards must be built with the audience in mind, an understanding of how the information will be used, and a clear indication of how frequently the data is refreshed. "Everybody assumes that the data that they are looking at is current as of the time they are looking at it, and that's not necessarily true. Some data is one-day old, some data is real time," says Veltri. "But the users have to know that."

    Most business intelligence dashboards use visual cues-such as red flags or green arrows-to help users quickly interpret the information. Denver Health embeds another layer of information. "When you mouse over an indicator, it tells you what value is good and what the target is," says Veltri. "If national benchmarks or research is available, that information is provided to help the user interpret what is being displayed."

    Getting Started

    The business intelligence pioneers profiled in this article share other lessons they have discovered in their journey:

    Do not try to engage physicians in the start-up phase. "During that time, there's really nothing for anybody to decide. You're just building data maps and dictionaries and the basic foundational elements of the warehouse," says Veltri. "If you put organizational leaders or physicians into this upfront process, they get bored because there is little progress toward the things they want, such as dashboards or reports."

    Appoint the right group of people to set priorities, depending on your organization's needs. At Northeast Georgia, an Enterprise Priority Council is made up of the system's vice presidents and medical staff members. "Based on what we are trying to accomplish strategically, they help prioritize the next opportunity, whether that is developing a new dashboard tool or taking on an improvement project with some technology need," says Cummings.

    In Denver, clinicians and revenue cycle leaders serve on the Warehouse Advisory Council that sets priorities. "They prioritized colorectal cancer screening as the first registry that we would build-even over diabetes," says Veltri. "The primary reason was that a key stakeholder had received a grant to support a colorectal cancer screening project, which paid for some of the development. Without such a council, IT never would have known that this direction was acceptable to the physicians."

    Share dashboard data broadly. At Hartford Hospital, the chief physicians of every department and every division have access to all the data-as do the nurse managers. "Any nurse who is managing a floor can go in and look at my physicians and I can go in and look at her floor," says Lindberg. "That allows us to have a conversation on how the different things are impacting one another. We can brainstorm together and come up with good solutions, working as a collaborative management team."

    Document how EHR data can be used to drive clinical improvement. Because this capability is so new, everyone is a trailblazer, making mistakes as they gain experience. "I would encourage others that want to embrace a program like this to document the processes and the failures and successes and how we can improve this," says Smith. "We can all learn from one another as we go forward."

    Veltri believes business intelligence answers the question: Can hospitals improve quality and charge less for that quality? "The answer is: 'Yes, Denver Health has proven this can be accomplished,'" says Veltri. "But the organization must manage the use of quality data, and use this data to make informed decisions based on data."

    Lola Butcher is a freelance writer and editor based in Missouri.

    Interviewed for this article (in order of appearance): Gregory Veltri is CIO, Denver Health, Denver (gregg.veltri@dhha.org). Michael Lindberg, MD, is chairman, department of medicine, Hartford Hospital, Hartford, Conn. (mlindbe@harthosp.org). Allana Cummings is CIO, Northeast Georgia Health System, Gainesville, Ga. (allana.cummings@nghs.com). Robert Taylor, MD, is a critical care and internal medicine physician, Mercy Hospital St. Louis, St. Louis. Tim Smith, MD, is vice president of research, Mercy Center for Innovative Care, St. Louis. (timothy.smith@mercy.net).