Radiology’s Next Move: Bigger Data

In the 1990s, it was easy to be a success. You had to work hard not to be a success. That’s not true any more,” according to Michael P. Recht, MD, Louis Marx professor of radiology at New York University School of Medicine and chair of the radiology department at NYU Langone Medical Center (New York, New York). Heightened competition, declining reimbursements, and pressure from payors to cut unnecessary imaging have forced the radiology department to watch every penny. Recht says, “We are looking at denials by site and by insurance company, asking whether there is something incorrect that we are doing. We can’t leave any money on the table.” To heighten its chances of financial success, NYU Langone Medical Center is doing what so many other radiology providers of size are doing: investing in business analytics to streamline operations and unveil opportunities perhaps otherwise missed. Analytics—a computerized way of searching for patterns—is assuming a higher profile in radiology business. The move to big data in radiology will require another leap. Characterized by the quantity and variety of data processed, as well as the speed of analysis (see article, page 12), the big-data movement seeks correlations, rather than causes, and patterns, rather than sums. It is as fundamental a management/technological revolution as the move from film to PACS was—and equally impossible to ignore. Concepts such as data governance and pattern recognition are joining key performance indicators and balanced scorecards in the radiology management lexicon. Access to data has already proven its worth. When Hurricane Sandy knocked out a CT scanner at NYU Langone Medical Center (a four-hospital complex anchored by the 705-bed Tisch Hospital), real-time workflow measurements were used to reassign patients to the three remaining scanners more or less seamlessly: 95% of cases were completed in 15 minutes and 99% were done in 30 minutes. “These data have given our people autonomy and mastery,” Recht says. Analytics programs are being run from billing systems, RIS, PACS, and voice-recognition systems, and multiple vendors offer turnkey solutions for both business and clinical analytics. Some patterns and processes are similar in all business-analytics rollouts, but beneath the surface, it’s a mixed bag. With fragmentation being endemic to health care, there is a demand for solutions that stitch everything together—a necessary precursor to a big-data solution. Usable Data Thought leader James H. Thrall, MD, FACR, is an analytics pioneer. Formerly chair of the ACR® board of chancellors, Thrall stepped down as chair of the radiology department at Massachusetts General Hospital (MGH) in Boston (a job he had held for a quarter century). Thrall, who is active on the MGH medical staff and holds a professorship at Harvard Medical School, now spends more time on research, including data analytics. “There is a construct,” Thrall begins, “that goes: data to information to knowledge to wisdom. Data are individual facts, or a collection of individual facts. Information is organization of those data, at some level. Knowledge is a higher-level organization of information. Wisdom is the ability to assimilate knowledge and also to draw inferences based on assimilated knowledge and experience.” Thrall says that business analytics for radiology has reached the information stage—and sometimes, the knowledge stage—but never the wisdom stage. “The big shortcoming I see in information systems—RIS and PACS—is that they provide no analytic capability,” Thrall says. “For example, we practice in nine different locations and we have CT scanners in seven different locations. I’d like to know how many of what kind of CT scans we did in each place. It sounds simple, but the RIS can’t give us that.” Thrall says that to get the data, MGH has to export the numbers from the RIS daily and feed them into a separate server that has the software to provide the information needed to run the radiology department. “How do you do hand-cleansing protocols or timeout protocols? There are no commercial systems available that encompass all these elements so vital to managing a department,” Thrall adds. MGH calls its dashboard The Same Page, and it gave Thrall the ability to see how many exams of what type were done at each location. “I could then trend that for a month, a year, or two years,” Thrall says. “That is very basic, tangible analytic feedback you use as the department leader. It took me 15 years to have that on my desktop every day because of how clumsy these systems are.” This is not to say that MGH has gained minimal value from its business-analytics program in radiology; Thrall is quick to note that it has been indispensable. Tools have been developed to track changes in referral patterns, radiologists’ productivity, modality turnaround times, and much more. Dashboard-accessible heat maps display changing patterns; MGH has even developed reading-room dials to keep modality workflows proceeding optimally. The workflow dials are an example of what can be done with later-generation digital devices. Thrall says, “You mine data in real time and send that back to fine-tune the work process.” Vendors, he says, need to do more to improve data analytics by making all data sources compatible and creating accessible data exchanges to link information from different hospitals. For now, each hospital works largely with its own data. “There are so many vendors that harmonizing the data formats and installing a management or analytics system between one department and another would be a big challenge,” Thrall observes. Trusting the Numbers Hospital systems, radiology departments, and radiology practices are working to meet the challenge of structuring isolated data by organizing enterprise-wide business-intelligence units that adopt a top-down approach to disseminating information to those who need it. Administrators and executive committees determine the questions that they need to have answered and then assign individuals or teams to make sure that needed data are gathered in a trustworthy way. Gathering reliable data for an enterprise data warehouse is a function of data governance, which assigns responsibility for the collection of specific datasets to specified individuals. The individuals are accountable for the data and their trustworthiness. Thrall compares this to any business-intelligence effort undertaken by a corporation. He says, “The strategy is to use key indicators in limited numbers and assign responsibility into the ranks.” A key indicator might be turnaround time; another might be hand hygiene. Thrall and his team identified about 150 key performance indicators and selected 30 when they built the Same Page system. “We have a person who is responsible for each key indicator and the time frame for reporting—daily, weekly, or monthly,” Thrall explains. “Each key performance indicator is fully defined and has a clear chain of responsibility. It could be the outpatient manager; it could be the quality manager or someone more senior.” The Data Governor Christopher J. Donovan, MBA, is executive director of fiscal services for the Cleveland Clinic in Ohio and is the architect of its enterprise business-intelligence strategy (which includes its facilities in Florida, Canada, and the Middle East). Donovan acknowledges that radiology operations are just one focal point in the business-intelligence effort. “We collect data from all across the enterprise,” he says. “We probably have close to 100 data projects. They come in as requests and are evaluated. Some are activated, and some are waiting for resources.” Anyone can request a data project, he says. A strategic multidisciplinary council reviews, scores, and ranks the requests as they are received and then prioritizes them. Once the data projects are activated, measurements for each project are developed and the results are displayed on dashboards. “We did a big project on blood use,” Donovan says. “We’re a big user of blood, with many operating rooms. There are cost and patient-safety aspects for any transfusion. We were able to reduce the use of blood by using electronic business intelligence and by partnering with our clinical teams.” The business-intelligence department was careful not to limit physicians in the ways they used transfusions. Instead, physicians were given computer displays with information on cost, safety, and trends in transfusion use. “That led to a lot of thinking, and it changed the way they used the resources,” Donovan says. “Blood use went down from 305 units to 206 units per 1,000 patient days.” Donovan says that one radiology project involves matching sites where imaging is done with sites where patients are served to see whether some imaging sites can be eliminated, based on patient convenience. “With value-based clinical care, there is a cost savings by decreasing sites,” he says. “That project uses our warehouse data. That’s ongoing right now.” A Cultural Transformation Kathryn A. Whitmore, MS, is founder and managing principal at STS Consulting Group. Whitmore consults on management analytics and overall enterprise intelligence, and she consulted with Cleveland Clinic on its business-intelligence strategy. Distribution of data across the whole enterprise is central to a business-intelligence initiative. By taking exclusive management of the data out of the IT department, you empower others to use it, she says. In this scenario, users in each relevant area (an access layer, or datamart) would have their own dashboards displaying the information that they need to make departmental decisions. Radiology would be one such datamart, accessing data through the enterprise data warehouse, with a strict data-governance policy limiting and granting access. The advantage of the enterprise approach is that it allows collaboration between departments (and from top executives down to department leaders). Whitmore says that the technology of data management is less important than the questions that are asked. “Data governance is the ability to have a high-value corporate asset that everyone can share,” she says. “It’s as much a cultural transformation as anything else.” It isn’t about data and technology, Whitmore says, but about deciding what will give you the highest value. “The concept of a datamart is knowing that you have a business case to answer,” she adds. “With that business question in mind, you want just enough data components to answer that statement.” Nearly every vendor has a dashboard product as part of its software offerings, Whitmore adds. There are applications that can link data sources and drill down from the enterprise level to answer questions. She says, “You can install the analytics product on your own server or use most over the Internet, as software as a service.” Already underway is an evolution to mathematics modeling and natural language processing that will expand the types of data at our disposal and what can be accomplished with them, Whitmore says. Analytics in Action Data governance and enterprise business intelligence sound good, but what can business analytics actually accomplish? What might it do in the future? Can analytics open a door to solvency that not only is new, but that gives radiology a special place in the health-care hierarchy? At NYU Langone Medical Center, Recht is studying scheduling—not just of patients, but of how the department’s dozen or so schedulers are rotated to meet the ebbs and flows of scheduling demand. If Fridays are traditionally slow, Wednesdays might call for longer hours for the schedulers, Recht suggests. Another project is computing turnaround time: Recht says it’s not just the time it takes for a report to be completed, but how time is spent on that exam by everyone in the department. Are patients handled efficiently? How long are billers on the phone? Kirk Lawson, MBA, is NYU Langone Medical Center’s administrative director of diagnostic radiology. Lawson says that the department always keeps an eye on its balanced scorecard—a ranking of financial incentives alongside clinical objectives, customer expectations, and internal hospital needs. To meet CMS assumptions for equipment-utilization rates, modalities must be in use a large percentage of the time, but patients don’t want to be rushed. “It’s striking that balance,” Lawson says. NYU Langone Medical Center is running analytics to see if it might overbook modalities (in the way that airlines overbook flights) to balance no-show and add-on patients, Lawson adds. “Our dashboard is presenting, in a very aggregated manner, tons of data,” he says. “We have 25 or 30 things we look at every day. It’s how we survive in a competitive marketplace, where we want to get paid quickly because we’re getting paid less.” NYU Langone Medical Center uses a highly customized vendor solution that sits on top of the RIS and PACS to create reports and dashboards. The easiest questions to answer are the strictly numerical questions. What’s hard is developing algorithms that enable the department to change its patterns, Recht says. He uses the example of intervening when decreases are noticed in a physician’s radiology referrals. It might be that the physician is on vacation; the medical center doesn’t want to dispatch marketers if factors on the referring physician’s side are causing the decrease—but if the reason is something that the medical center is doing, it does want to know. “We have to act quickly, before we lose the referrer, but when should we intervene?” Recht asks. “We haven’t come up with that trigger yet, but we’re analyzing the question.” Socioeconomic Measurements Russell Cain, DBA, CRA, CIIP, CRT, is director of imaging services for Atlanta Medical Center in Georgia. The data that Cain looks at flow from the RIS and PACS, the hospital’s electronic medical record (EMR), the modalities, billing, and corporate headquarters. He can monitor productivity, workflow, staffing, and all of the things that hospitals use data analytics to assess. He also gets daily reports from financial analysts at the hospital’s parent company, Tenet Healthcare Corp, that show reimbursements and the payor mix. He can spot surge patterns in the emergency department and can staff accordingly; he says, “I should be able to respond to imaging requests in 15 minutes.” Things also happen by surprise, and this shows up in the data, too. The hospital lost a contract for mammography that accounted for 20% of the mammography workload. “In terms of staffing and revenue, that was a tremendous impact,” Cain says. Cain can see decreases in referrals as a referring physician gets ready to retire, and to make sure that this is what is happening, he can respond by sending a marketer to the practice. “I don’t use one data point,” Cain says. “I look for trends.” Watching socioeconomic trends is important, he adds, because they will affect demand for imaging. “You can’t just look at imaging; what about outpatient surgery? The trends are out there,” he says. “Of baby boomers, a lot are still working, but they’re looking at preventive health, and they’re getting imaged more regularly. We’ve got a lot of young people in our downtown. They’re athletic, and they’re interested in preventive measures. Do we have a sports-medicine practice? How do I support that? That affects what we look at when hiring; we’ll want radiologists with musculoskeletal skills.” Analytics for the Practice The potential and uses of business analytics in the radiology practice are as vital as in any other provider setting. A private practice is likely to be hooked into several enterprise systems through the hospitals that it serves, and might be a participant in a huge data network. It is this characteristic that sets some private-practice radiology directors dreaming. Is there some way to use data analytics to create a new profit center out of the information that a radiology group can provide? At Central Illinois Radiological Associates (CIRA), Peoria, Illinois, the decision was made to partner with a billing-services vendor that was able to provide the radiology group with round-the-clock data access and analytical software—to build meaning from the practice’s CPT^® coding. Gregory Q. Hill, JD, CIRA’s CEO and an adjunct associate professor of radiology at the University of Illinois College of Medicine, says that CIRA has 77 radiologists who interpret for 16 hospitals in four separate hospital systems. CIRA also owns three interventional clinics, but not the equipment in them, Hill says. In addition, the group reads for 21 ambulatory centers operated by physicians with their own imaging equipment. CIRA interprets more than 1.3 million procedures annually. “With the data analytics we have today, the decision-making process is more credible,” Hill says. “The ability to aggregate data has been a game changer because we are able to make decisions based on data from all 16 hospitals. The economies of scale and efficiencies are significant.” CIRA uses analytics to track key performance indicators such as charges, imaging volumes, work RVUs, and receipts. The group also can track denials. “If I have a significant denial issue, then I can approach the hospital or the payor,” he says. “Think how powerful those data are. Prior to this I, would have had to obtain the information from multiple sources and aggregate the data myself. Now, they all come from one source.” Some analytics problems are easy, but others are more complicated; Hill says, “I can have productivity information for a site in five minutes. If we receive a request for proposal to provide services at a new hospital, however, obtaining the data not owned by CIRA requires additional time and effort.” Canopy Partners Greensboro Radiology in North Carolina is taking its analytics program in two directions. It is using analytics within its own practice of 55 radiologists, but it has spun off Canopy Partners, a managed-services organization that provides analytics, billing, and other services to hospitals, radiology groups, and physician practices of all kinds. Worth Saunders is CEO of Greensboro Radiology and Canopy Partners; Stephen Willis is CIO of Canopy Partners; and John A. Stahl, MD, is chair of the best-practices oversight committee at Greensboro Radiology and a shareholder in Canopy Partners. According to Stahl, Greensboro Radiology began its analytics effort with a focus on quality. The practice took radiology quality measures from CMS, the National Quality Forum, the ACR, and other sources and used them to develop its initial analytics program. Today, Greensboro Radiology is running its analytics effort using a system (developed in-house) that makes use of a single vendor’s voice-recognition transcription system and natural language processing software to pull data from radiology reports. Other measurements come from the PACS and the billing system, but the heart of the data gathering is the voice-recognition system, deployed across all client sites. A single PACS is also used to view and distribute all images. “A single voice-recognition system leads to results from a single source,” Willis says. “Text and messaging coming out of the voice-recognition system contain the exam code, RVUs, the radiologist’s name, several different time stamps, and when the exam was started and completed.” Stahl says, “With natural language processing, the system has the potential to provide a radiologist with real-time alerts: Did you include this information in your report for this diagnosis, did you put all the information in for correct coding and billing, and did you notify the referrer and document that in the report?” While this ability isn’t in hand yet, it’s on the way, Stahl says, and it will reduce delays in reimbursement. Saunders notes that Greensboro Radiology recently merged with an eight-physician practice that had a different arrangement for night coverage. Greensboro Radiology was able to use data stored in its analytics warehouse to integrate night coverage for both groups. “Whether it’s business quality or clinical quality, the difficulty is to assemble the measures that you think are useful in a fashion where the data can be obtained consistently, over time,” Stahl says. Save a Dollar and Earn a Dime Radiology Associates (RA) of North Texas in Fort Worth, the largest private radiology practice in the country, owns 13 ambulatory centers and interprets for hospitals at 26 sites. The practice’s 122 radiologists read about 2 million exams annually. Mark Kleinschmidt, RA’s CEO, says the group is now in the infant stages of building its analytics capability, still identifying key measurements and constructing a data warehouse. It is the big-data vision that Kleinschmidt offers—radiology as information broker—that is compelling. “Radiology is, and always has been, an information business, so radiology analytics is a natural step,” Kleinschmidt says. “Radiology has always been about using pattern analysis.” Currently, RA is collecting data from its own systems, the EMRs of the hospitals that it serves, and even public-health sources, Kleinschmidt says. These data will be standardized and put in a single data warehouse to create a meaningful data pool that can be used to assist hospitals in studying radiology use, outcomes, and other topics. “At what point in a patient’s episode of care is an imaging study best performed?” Kleinschmidt asks. “Is it best on the front end, or is it better to wait and watch for six weeks?” There are instances in which RA radiologists have spotted orders from one hospital for an exam that the same patient just had at another hospital. “We want to try to use data to support whether those health-care costs are necessary,” Kleinschmidt says. RA is well positioned as an information broker, Kleinschmidt says, but the challenge is getting the competing hospitals that it serves to see the advantages of combining data. Kleinschmidt says, “There is a move away from fee-for-service reimbursement to paying for population health. We think we can provide an overall analysis based on the entire population of Fort Worth. We’re just at the stage of communicating and getting people to buy into that.” Kleinschmidt acknowledges it won’t be easy. “Health care has been held back by competition, privacy, and security issues,” Kleinschmidt says. “Retailers are collecting data right and left. We think radiology can lead the way in health care.” After all, he says, radiology is the only specialty that deals routinely with almost every other specialty. The information gatherable through radiology is potentially huge. Recently, Kleinschmidt asked his IT department to give him the number of patients at any site who’d had more than one CT exam of the head in the past 30 days. “We found 38 patients from one hospital system who’d had more than one such exam—and one patient who’d had 14,” Kleinschmidt says. “What we found was that one hospital had a significant stroke program. It has a protocol that all stroke patients get a head CT every morning to see whether there have been changes. The question is this: What is magic about one day? We have raised the issue. We don’t tell the hospital how to do this, but as people on the radiology business side, we think that’s useful information to have.” He adds, “It’s worth looking for patterns. In someone with a bad back, what’s the likelihood of knee trouble? In the long run, if we can save money for the system, we’ll get paid for it. If we save them a dollar, they’ll pay us ten cents.” That’s an argument in favor of radiology business analytics all by itself.

George Wiley is a contributing writer for Radiology Business Journal.