From Quality to Outcomes: Deploying Clinical Analytics

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon

Although radiology has employed clinical analytics for more than a decade, the field is in its infancy. Nonetheless, the possibilities are tantalizing—if technological, economic, political, and interoperability hurdles can be cleared. David Ecanow, MD, is the radiology department’s vice chair for quality at NorthShore University HealthSystem (Highland Park, Illinois). He reports that analytics methods are used across the spectrum of his responsibilities: to measure and improve access, utilization, exam quality, safety, interpretive accuracy, outcomes, communication, basic patient care, and regulatory compliance. “By measuring and analyzing the reasons for exams, we attempt to ensure that the correct test is ordered and performed,” he explains. “For example, we tracked the reasons for CT exams ordered with and without contrast, and we were able to decrease unnecessary exams.” To ensure exam quality and safety, Ecanow says, his team has been tracking CT-exam radiation dose by protocol and has used that information to reduce dose across multiple types of exams—particularly in pediatrics, obstetrics, and urology. By tracking peer review of exam interpretations, as well as clinical and pathology outcomes, Ecanow’s team is better able to pinpoint difficult diagnoses and generate inservice education to sharpen accuracy.  “We audit specific critical-exam or critical-result communications to help ensure timely and appropriate communication of results,” he adds. “We audit our mammography services extensively for both exam quality and clinical outcomes.” Improving the Product Woojin Kim, MD, is interim chief of the division of musculoskeletal imaging at the Hospital of the University of Pennsylvania in Philadelphia. He says that clinical analytics helped him increase his section’s study RVUs by 16%, and he also has used analytics to drive quality improvement for reports. “It’s important to remember that reports are a radiologist’s main product,” he emphasizes. “For example, by using analytics tools that can leverage the power of natural language processing, I’ve been able to monitor the laterality errors in radiology reports.” Kim says that his team—after discovering the presence of errors that were being overlooked by coders—has implemented a process for continuous monitoring that allows rapid error correction (using a tool that was developed for the department and subsequently commercialized). “The tool keeps reminding us, if a given report with an error does not get revised,” he explains. “It also keeps track of the time it takes for each report to be corrected. These functions contribute to enforcing and improving compliance.” Within a month of implementing that monitoring process, Kim notes, the error rate dropped by 48% at one of the sites within the system—purely, he insists, because people knew that the process was in place. Analytics can help an organization go beyond the typical measurements to make a meaningful impact on patient care, Kim adds. “The more advanced analytics tools, combined with natural language processing, can mine the radiology reports to detect various quality elements like laterality errors, degree of uncertainty, gender errors, coding errors, and follow-up recommendations,” he says, increasing the referrer’s confidence in the radiologist’s report. “Because many referrers and patients don’t know their radiologists, small errors in reports can have a significant impact on their confidence in their radiologists,” he notes. The Data Path Dan Wassilchalk, executive administrator of the department of radiology at the University of Pittsburgh Medical Center in Pennsylvania, views clinical analytics within the broad sweep of the entire radiology workflow and how it connects with each patient’s data path—from registration to exam acquisition, interactions with other departments, results delivery, and follow-up care. Wassilchalk notes, “Our radiologists will evaluate exams on a peer-review basis and will measure concordance or discordance between interpretations. From a patient-safety perspective, dictations contain key clinical findings that are critical in nature—that need to be reported to the ordering physician in a very timely manner to effectuate clinical intervention. Images, as well as clinical findings, are captured and stored as part of the patient’s electronic record.” Down the line, Wassilchalk says, his department’s goal is to determine whether certain exams for particular diagnoses are useful in the sense of having an impact on treatment. Going forward, he adds, clinical analytics could be used to improve treatment across the board. “We hope to reduce the number of repetitive exams and reduce or eliminate exams that are not useful,” he says. “We want to pool radiology data with clinical-outcomes data to improve the health-care system further. Pay for performance will be dependent on reducing admission rates and improving the quality of care.” Wassilchalk adds that one of the biggest challenges is breaking down data silos to promote interoperability among information systems. “We’re working hard to pull it together to give the analytics more power,” he notes. A colleague of Wassilchalk, Christopher Deible, MD, remarks that these data could be fed back to radiologists to show them how they are doing. The best example of this sort of information use, Deible says, is time stamping. “By doing that, we improve patient satisfaction and throughput, from exam completion to finalization of the report,” he says. “Timing is everything, and the sooner we can provide an effective diagnosis, the faster the patient can move through the process to discharge. We want to avoid duplication of effort. We also want to use analytics to improve the quality of documentation and reporting of critical results—to make sure that if I make a critical finding, I communicate it.” Supporting Decisions Clinical analytics can be used in the interpretation process to support radiologists’ decisions. Bradley J. Erickson, MD, PhD, director of radiology informatics at the Mayo Clinic (Rochester, Minnesota), reports that his team uses common mammography and breast MRI computer-aided detection tools, as well as more novel internally developed tools (such as an algorithm for detecting aneurysms from MR angiography data). He says, “Computer-aided detection is a subset of decision support, usually focusing on detecting lesions. Examples we use today include breast-lesion detection and polyp detection on CT colonography, and there are several 3D laboratory applications (such as liver- and kidney-volume measurements) that are important for surgical decision making.” He adds, “We’ve developed additional analytics tools that are helpful for detecting significant—but subtle—changes in brain tumors, intracranial aneurysms, and interstitial lung disease. Those hits are then sent to the 3D system and are rendered much as potential polyps are highlighted in the advanced visualization product,” he explains. Erickson’s department also developed and uses a brain-tumor change detector that highlights very subtle changes that are not easily perceived by humans, as well as an tool that characterizes and measures the amount of interstitial lung disease seen on CT exams. “All of these are visual tools that highlight things that we want to make sure the radiologist pays close attention to; however, in the end, the radiologist decides whether there’s a disease, and what to do about it,” Erickson says. “The tools are based on imaging data—from CT or MRI—and at present, we don’t include other information. We’re working on improving brain-tumor decision support that will include nonimage data such as the tumor type and the nature of the therapy that a patient has had.” The field still has plenty of room for improvement, Erickson says. The number and quality of images being produced continue to increase, and better tools are needed to analyze them. The Compliance Juggernaut Richard Grzybowski, MD, a radiologist with Diversified Radiology of Colorado (DRC) in Lakewood, is chair of the practice’s quality committee. Using data to improve clinical performance probably began with the introduction of the ACR BI-RADS® system, Grzybowski believes. With widespread adoption of enterprise PACs and RIS, radiology departments and practices became the custodians of enormous caches of clinical data. Ecanow agrees that the BI-RADS system represents radiology’s earliest foray into clinical analytics, but he says that the system could be improved through integration with enterprise-level information systems. “Computerized tracking of multiple quality elements has been in place for at least 15 years,” he notes. “It’s progressed to a state of maturity where it’s a routine method for benchmarking our clinical quality, managing resources, and providing a high volume of the best quality of subspecialized care. The future lies in more automated data entry, with some of the elements that are now manually entered being more reliably shared from the overall electronic medical record (EMR).” “Clinical analytics is still in its early phases of utility,” Ecanow remarks. “Only recently have the EMR, RIS, and national databases begun to coalesce into a useful matrix of data. The tools to mine those data robustly are only on the cusp of utility. As these tools become more widespread and routine, we hope that they will allow analytics to have an impact on health-care providers closer to the point of contact with patients.” Acting on Information The ways in which the information gathered is acted upon, Erickson says, are legion. One example is volume measurement for surgical planning. In cases of liver tumors that must be resected, or if using a living donor for liver tissue is being contemplated, it’s critical to know whether the amount of tissue that will be left will suffice to keep the patient (and donor, if applicable) alive. “One is further restricted by having to resect along segmental boundaries,” he adds. For liver donors, Erickson says, state-of-the-art clinical analytics can take life-threatening guesswork out of the equation. “Another example is that some computable properties of a brain tumor correlate with molecular markers that are important for predicting responsiveness to certain classes of therapies and for predicting overall expected survival,” he says. Erickson predicts that the next big steps in clinical analytics will be integration and prediction of molecular properties. “Molecular imaging is a hot area of interest, and while some forms allow direct visualization of some molecular marker, many require additional processing or integration of other information to be useful,” he says. Ecanow points to CT radiation-dose auditing as the clearest example of how data mining can produce actionable clinical results. “A calculated dose from each CT exam is entered into the RIS,” he explains. “We then routinely audit multiple data points—patient age, exam type, dose, and so forth—letting us pinpoint exam protocols that can be improved (or practice sites or staff that need more focused improvement). By addressing those exam protocols, we’ve reduced radiation dose, particularly in pediatric and urology patients, while retaining excellent diagnostic quality.”  New Frontiers and Tools Most of the data currently used in radiology to perform clinical analytics are sourced at the department level, but richer meaning and benefits will be derived through access to information beyond the department—or practice—walls. Barriers to access are technological, political, and economic in nature. “By leveraging other systems (such as in pathology), one can perform tasks like pathology–radiology correlation analytics,” Kim says. “In fact, one’s ability to obtain truly meaningful analytics can be vastly improved by tapping into data from other specialties in medicine, such as pathology and cardiology.” Erickson says, “Analytics will improve by getting a richer set of information to work on: more information about the patient and about the health-care environment. Information about the patient that we need today includes the results of simple laboratory tests, such as those for creatinine level or pregnancy. In the future, we’ll want genetic and epigenetic information, and in turn, we’ll start providing some of that type of information.” Ecanow acknowledges that the origin of most of the information that he is analyzing is the RIS. “As we gain more experience in working with the overall EMR and hospital quality infrastructure, that pool of information will widen,” he says. Interoperability among information systems remains an obstacle to data sharing. “More robust data-mining tools are slowly becoming available, but haven’t yet hit their stride,” Ecanow says, adding that simpler, more effective tools for communicating data (and then auditing that communication) are needed. The Golden Ring Despite the challenges of integrating clinical analytics with existing technologies, Grzybowski believes that the benefits to be derived outweigh the time and development efforts required to overcome the barriers. “Access to clinical data isn’t the limitation anymore,” he says. “The industry will see more transparency in quality analysis and decision support, with the growing access to clinical data. The use of clinical analytics is incredibly beneficial to patient care. Health care hasn’t pursued the benefits of data integration because it’s largely uncompensated time and expense, given current payment models. This is changing, though, as systems become more uniform and data sharing becomes more standardized.” Another factor fueling the need for clinical analytics is health-care reform: Health care is transitioning from fee-for-service to value-based payment models, for which using clinical analytics is likely to be an imperative. “As people focus more on population-health management,” Kim notes, “data mining and analytics will take us to new frontiers, allowing us to mine data we could not before and to discover things we could not before—at a much faster pace.”

Joseph Dobrian is a contributing writer for Radiology Business Journal.