Quality

Hospitalist Groups Extract New Solutions Via Data Mining


 

One hospital wanted to reduce readmissions among patients with congestive heart failure. Another hoped to improve upon its sepsis mortality rates. A third sought to determine whether its doctors were providing cost-effective care for pneumonia patients. All of them adopted the same type of technology to help identify a solution.

As the healthcare industry tilts toward accountable care, pay for performance and an increasingly

cost-conscious mindset, hospitalists and other providers are tapping into a fast-growing analytical tool collectively known as data mining to help make sense of the growing mounds of information. Although no single technology can be considered a cure-all, HM leaders are so optimistic about data mining’s potential to address cost, outcome, and performance issues that some have labeled it a “game changer” for hospitalists.

Karim Godamunne, MD, MBA, SFHM, chief medical officer at North Fulton Hospital in Roswell, Ga., and a member of SHM’s Practice Management Committee, says he can’t overstate the importance of hospitalists’ involvement in physician data mining. “From my perspective, we’re looking to hospitalists to help drive this quality-utilization bandwagon, to be the real leaders in it,” he says. With the tremendous value that can be generated through understanding and using the information, “it’s good for your group and can be good to your hospital as a whole.”

So what is data mining? The technology fully emerged in the mid-1990s as a way to help scientists analyze large and often disparate bodies of data, present relevant information in new ways, and illuminate previously unknown relationships.1 In the healthcare industry, early adopters realized that the insights gleaned from data mining could help inform their clinical decision-making; organizations used the new tools to help predict health insurance fraud and identify at-risk patients, for example.

Cynthia Burghard, research director of Accountable Care IT Strategies at IDC Health Insights in Framingham, Mass., says researchers in academic medical centers initially conducted most of the clinical analytical work. Within the past few years, however, the increasing availability of data has allowed more hospitals to begin analyzing chronic disease, readmissions, and other areas of concern. In addition, Burghard says, new tools based on natural language processing are giving hospitals better access to unstructured clinical data, such as notes written by doctors and nurses.

“What I’m seeing both in my surveys as well as in conversations with hospitals is that analytics is the top of the investment priority for both hospitals and health plans,” Burghard says. According to IDC estimates, total spending for clinical analytics in the U.S. reached $3.7 billion in 2012 and is expected to grow to $5.14 billion by 2016. Much of the growth, she notes, is being driven by healthcare reform. “If your mandate is to manage populations of patients, it behooves you to know who those patients are and what their illnesses are, and to monitor what you’re doing for them,” she says.

Practice Improvement

Accordingly, a major goal of all this data-mining technology is to change practice behavior in a way that achieves the triple aim of improving quality of care, controlling costs, and bettering patient outcomes.

A growing number of companies are releasing tools that can compile and analyze the separate bits of information captured from claims and billing systems, Medicare reporting requirements, internal benchmarks, and other sources. Unlike passive data sources, such as Medicare’s Hospital Compare website, more active analytics can help their users zoom down to the level of an individual doctor or patient, pan out to the level of a hospitalist group, or expand out even more for a broader comparison among peer institutions.

Some newer data-mining tools with names like CRIMSON, Truven, Iodine, and Imagine are billing themselves as hospitalist-friendly performance-improvement aids and giving individual providers the ability to access and analyze the data themselves. A few of these applications can even provide real-time data via mobile devices (see “Physician Performance Aids,”).

Thomas Frederickson, MD, MBA, SFHM, medical director of the HM service at Alegent Creighton Health in Omaha, Neb., and a member of SHM’s Practice Management Committee, sees the biggest potential of this data-mining technology in its ability to help drive practice consistency. “You can use the database to analyze practice patterns of large groups, or even individuals, and see where variability exists,” he says. “And then, based on that, you can analyze why the variability exists and begin to address whether it’s variability that’s clinically indicated or not.”

When Alegent Creighton Health was scrutinizing the care of its pneumonia patients, for example, officials could compare the number of chest X-rays per pneumonia patient by hospital or across the entire CRIMSON database. At a deeper level, the officials could see how often individual providers ordered the tests compared to their peers. For outliers, they could follow up to determine whether the variability was warranted.

As champions of process improvement, Dr. Frederickson says, hospitalists can make particularly good use of database analytics. “It’s part of the process of making hospitalists invaluable to their hospitals and their systems,” he says. “Part of that is building up expertise on process improvement and safety, and familiarity with these kinds of tools is one thing that will help us do that.”

North Fulton Hospital used CRIMSON to analyze how its doctors care for patients with sepsis and to establish new benchmarks. Dr. Godamunne says the tools allowed the hospital to track its doctors’ progress over time and identify potential problems. “If a patient with sepsis is staying too long, you can see who admitted the patient and see if, a few months ago, the same physician was having similar problems,” he says. Similarly, the hospital was able to track the top DRGs resulting in excess length of stay among patients, to identify potential bottlenecks in the care and discharge processes.

Some tools require only two-day training sessions for basic proficiency, though more advanced manipulations often require a bigger commitment, like the 12-week training session that Dr. Godamunne completed. That training included one hour of online learning and one hour of homework every week, and most of the cases highlighted during his coursework, he says, focused on hospitalists—another sign of the major role he believes HM will play in harnessing data to improve performance quality.

You can use the database to analyze practice patterns of large groups, or even individuals, and see where variability exists. And then, based on that, you can analyze why the variability exists and begin to address whether it’s variability that’s clinically indicated or not.

—Thomas Frederickson, MD, MBA, SFHM, medical director, hospital medicine service, Alegent Creighton Health, Omaha, Neb., SHM Practice Management Committee member

Slow—Construction Ahead

The best information is meaningful, individualized, and timely, says Steven Deitelzweig, MD, SFHM, system chairman for hospital medicine and medical director of regional business development at Ochsner Health System in New Orleans. “If you get something back six months after you’ve delivered the care, you’ll have a limited opportunity to improve, versus if you get it back in a week or two, or ideally, in real time,” says Dr. Deitelzweig, chair of SHM’s Practice Management Committee.

In examining length of stay, Dr. Deitelzweig says doctors could use data mining to look at time-stamped elements of patient flow and the timeliness of provider response: how patients go through the ED, and when they receive written orders or lab results. “It could be really powerful, and right now it’s a little bit of a black hole,” he says.

Based on her conversations with hospital executives and leaders, however, Burghard cautions that some real-time mobile applications, although technologically impressive, may be less useful or necessary in practice. “If it’s performance measurement, why do you need that in real time? It’s not going to change your behavior in the moment,” she says. “What you may want to get is an alert that your patient, who is in the hospital, has had some sort of negative event.”

Data mining has other potential limitations. “There’s always going to be questions of attribution, and you need to have clinical knowledge of your location,” Dr. Godamunne says. And data mining is only as good as the data that have been documented, underscoring the importance of securing provider cooperation.

Dr. Frederickson says physician acceptance, in fact, might be one of the biggest obstacles—a major reason why he recommends introducing the technology slowly and explaining why and how it will be used. If introduced too quickly and without adequate explanation about what a hospital or health system hopes to accomplish, he says, “there certainly is the potential for suspicion.” The key, he says, is to emphasize that the tools provide a valuable mechanism for gleaning new insights into doctors’ practice patterns, “not something that’s going to be used against them.”

Paul Roscoe, CEO of the Washington, D.C.-based Advisory Board Company's Crimson division, agrees that personally engaging physicians is essential for a good return on investment in analytical tools like his company’s suite of CRIMSON products. “If you can’t work with the physicians to get them to understand the data and actively use the data in their practice patterns, it becomes a bit meaningless,” he says.

We’re looking to hospitalists to help drive this quality utilization bandwagon, to be the real leaders in it.…It’s good for your group and can be good to your hospital as a whole.

—Karim Godamunne, MD, MBA, SFHM, chief medical officer, North Fulton Hospital, Roswell, Ga., SHM Practice Management Committee member

Roscoe sees big opportunities in prospectively examining information while a patient is still in the hospital and when a change of course by providers could avert a bad outcome. “Suggesting a set of interventions that they could do differently is really the value-add,” he says. But he cautions that those suggestions must be worded carefully to avoid alienating physicians.

“If doctors don’t feel like they’re being judged, they’ll engage with you,” Roscoe says.

Similar nuances can affect how users perceive the tools themselves. After hearing feedback from members that the words “data mining” didn’t conjure trust and confidence, the Advisory Board Company dropped the phrase altogether in favor of “data analytics,” “physician engagement,” and similar descriptors. “It’s simple things like that that can very quickly either turn a physician on or off,” Roscoe says.

Once users take the time to understand data-mining tools and how they can be properly harnessed, advocates say, the technology can lead to a host of unanticipated benefits. When a hospital bills the federal government for a Medicare patient, for example, it must submit an HCC code that describes the patient’s condition. By doing a better job of mining the data, Burghard says, a hospital can more accurately reflect that patient’s contdition. For example, if a hospital is treating a diabetic who comes in with a broken leg, the hospital could receive a lower payment rate if it does not properly identify and record both conditions.

And by using the tools prospectively, Burghard says, “I think there’s the opportunity to make a quantum leap from what we’re doing today. We usually just report on facts, and usually retrospectively. With some of the new technology that’s available, the healthcare industry can begin to do discovery analytics—you’re identifying insights, patterns, and relationships.”

Better integration of computerized physician order entry with data-mining ports, Dr. Godamunne predicts, will allow for much better attribution and finer parsing of the data. As the transparency increases, though, hospitalists will have to adapt to a new reality in which stronger analytical tools may point out individual outliers. And that level of detail, in turn, will require some hospitalists to justify why they’re different than their peers.

Even so, Roscoe says, he’s found that hospitalists are very open to using data to improve performance and that they make up a high percentage of CRIMSON users. “There isn’t a physician group that is in a better position to help drive this quality- and data-driven culture,” he says.


Bryn Nelson is a freelance medical writer in Seattle.

Physician Performance Aids

Physician Performance Aids

Company: The Advisory Board Company

Sample product: CRIMSON Continuum of Care

Claim: “Places credible, severity-adjusted performance profiles directly in the hands of physicians, enabling the hospital-physician collaboration needed to advance quality goals and secure cost savings.”

Number of users: More than 850 hospitals

Real-time data? No, but available as add-on service

To learn more: www.advisory.com/Technology/Crimson-Continuum-of-Care

Company: Panacea Developments

Sample product: Iodine

Claim: “An easy-to-use mobile system that analyzes a hospital’s sea of data for knowledge that physicians, case managers, and documentation specialists can act upon to improve length-of-stay, readmissions, and documentation.”

Number of users: More than 50 hospitals

Real-time data? Yes

To learn more: www.panaceadevelopments.com/iodine

Company: Truven Health

Sample product: CareDiscovery

Claim: “Provides powerful insights into a hospital’s or health system’s quality measures and resource utilization to support improvement with objective, fact-based information.”

Real-time data? No

Number of users: More than 3,000 users in 400 hospitals

To learn more: http://truvenhealth.com/products/carediscovery

Company: Ingenious Med

Sample product: Imagine

Claim: “A comprehensive analytics dashboard platform that provides actionable decision making data to optimize physician performance and revenue.”

Real-time data? Yes

Number of users: 25,000 total users (not separated by individual products)

To learn more: http://ingeniousmed.com/what-we-do/imagine

Reference

  1. Yoo I, Alafaireet P, Marinov M, et al. Data mining in healthcare and biomedicine: a survey of the literature. J Med Syst. 2012;36:2431-2448.

Next Article:

   Comments ()