The Top Five Medical Imaging IT Projects of 2010
When Radiology Business Journal was founded four years ago, it was with the understanding that IT represented not just the platform for image interpretation, exchange, and archiving, but also a broad foundation for practice operations, communications, and financial analysis. Earlier this year (and with that in mind), we approached the Society for Imaging Informatics in Medicine (SIIM) to collaborate on a competition to recognize the remarkable innovation we were witnessing in medical imaging, across practice settings and practice domains. We believed that the combined resources of our two organizations would generate more interest and offer greater results legitimacy for a competition to identify the Top Five Medical Imaging IT Projects of 2010. Our sincere thanks go to Anna Marie Mason, SIIM’s executive director; to the SIIM board, for approving the idea and agreeing to judge and publicize the contest to SIIM members; and (last, but not least), our panel of six judges. We received 28 entries in the categories of clinical, interoperability, communications, business-intelligence, and security projects: All of them were very interesting, many were excellent, and five of them received the highest marks from the judges. The criteria were innovation/ingenuity; meeting a critical, urgent, or unmet need; improving quality; validating/evaluating a tool; and having the potential to be generalized to other institutions. Here are the winning entries (edited for length and style), along with some insight into the work of the innovators who submitted them. We thank all of you who took the time to enter and invite all imaging informaticists to look for our second annual contest early in 2012.
Peer-review System With Brains
Yun (Rob) Sheu, MD, is a radiology resident at the University of Pittsburgh Medical Center in Pennsylvania.Peer review is often looked upon as inefficient and without objective results by many radiologists, despite its educational value and improvement of patient care, Sheu notes in his winning entry. He was convinced, however, that there was a better way. “Applying a mathematical cost model to guide the selection of radiology exams for peer review is feasible,” he says. Barton Branstetter, MD, was the principal investigator, and three mathematicians collaborated with the radiologists: Elie Feder, PhD; Igor Balsim, PhD; and Victor Levin, PhD. While peer review is an essential component of radiologists’ practice, the increasing constraints on a radiologist’s time require this process to be as efficient and effective as possible. “Our eventual goal is to streamline the process more and build the model into one of the new electronic peer-review systems, ACR® RADPEER™ being an example,” Sheu continues. “To our knowledge, this has not been done. Currently, the informatics department at our institution is working on an in-house peer-review system that is incorporated in the PACS, and we hope, eventually, to use our model in the system.” Winning Entry 1 Problem: Although advances have been made in incorporating peer review into the daily workflow, cases to be reviewed are still selected at random, without consideration of prior errors or the consequences of those errors. Solution: Starting in 2009, and using data collected over a period of several years, we created a computer model that calculated the cost for 12 categories that can be used to target areas of weakness; cost is defined as the liability addressed per unit of peer-review time. Given a unit of peer-review time, the cost function represents the expected cost (both financial and medical) to the hospital and patient, if the error is not fixed. Four attributes of past errors were used to calculate cost: morbidity, financial expenditures, probability of occurrence (based on past data), and the time needed for peer review of the study in question. Our model determined the modality and body part for each radiologist who had, based on past errors, the greatest potential for future liability. This information would then allow a peer-review committee to pick review cases selectively for a given radiologist to achieve a more efficient review, maximizing the statistical likelihood of discovering a true area of weakness. A large sample of more than 64,000 significant discordances—based on overnight preliminary reports—over a five-year period was compiled. Discordances were adjudicated by specialty-trained radiologists. The preliminary and final diagnoses were categorized into approximately 20 broad categories per body part. Each error type consisted of an ordered pair of diagnoses for which the preliminary diagnosis and final diagnosis were different. Each of these error types was then assigned a numerical financial-cost value and a morbidity-cost value based on standardized scoring criteria. The computer model calculated the total cost as morbidity cost times financial cost times probability of error, with that product then divided by time needed for peer review of the study. The total cost was compiled for 246 on-call attending radiologists and residents in each of 12 categories spanning three modalities (CR, CT, and MRI) and four body systems (neurological, abdominal, thoracic, and musculoskeletal). The category with the highest cost was then selected as the one that should be preferentially examined in future peer reviews. The total number of cases read for each category was also determined for each attending radiologist and resident. The universal probability of an error for each category was compiled using data from all radiologists, and the total cost was calculated. Last, the average cost per category and the range of costs per category were tabulated. Technology: The model supplies the morbidity and mortality costs of errors committed during overnight calls. It then tabulates the total error cost for a particular radiologist and for the radiology department as a whole. The program is easily modifiable, constantly updates the cost functions as new data are received, and suggests the best test to review (whenever an opportunity for review arises). Results: “Applying a mathematical cost model to guide the selection of radiology exams for peer review is feasible,” Sheu says. In addition, he notes, the study creates a baseline evaluation, for each of the participating attending radiologists, against which future errors can be compared. The advantages of the model are that it is flexible, with easy adjustment of scoring values, and that additional gradations can be added as new data become available. Clinically, such information is useful in many ways. It is useful to radiologists and residents because it directs the focus of future study, and it is useful to radiology departments because it allows the administration to monitor the performance of all staff members, relative to their peers. Global mistakes can be detected and communicated to staff members to allow more care to be taken with high-risk studies. These data also provide a quantifiable way to ensure improvement in weak areas. For a residency-program director, the model can detect global deficiencies among the residents that can be remedied using targeted teaching. Cases for daily case conferences can be selected from the category with the greatest average cost. From residents’ standpoint, RADPEER provides information on what mistakes they are making, what impact the mistakes had, how many other people in their peer group made the same mistakes, and what to work on in the future. With targeted case selection, we hope to reduce the number of cases needed for a successful peer review, saving radiologists’ time and providing concrete evidence that peer review is having a positive effect. Conclusion: By identifying trends in errors, as well as their costs for patients and hospitals, this proposed model of peer-review case selection will allow a more relevant selection of future cases for review, in addition to providing statistically accurate ways of monitoring physician improvement and resolving areas of weakness. Interoperability: PACS Experanza
Li Lillian Hou, MS, CIIP, is a specialist programmer, radiology information services, at the University of Chicago Medical Center in illinois.Postacquisition workflow is a major bottleneck in busy hospital CT suites, and the University of Chicago Medical Center (UCMC) was no exception. “While it takes only seconds to scan a patient, it requires many minutes to generate the reconstruction and advanced visualization series on the scanner,” according to Hou, who collaborated with her PACS and CT modality vendor to develop an automated workflow that offloaded postprocessing—freeing the scanner to scan—and eliminated paper. Phase one of this project, called closed-loop imaging (CLI), began in February 2008. Phase two was launched in April 2009 and went live in April 2010. Deeming the project complete, given that it underwent three months of clinical trials, Hou currently is working on the hospital’s EMR system business-intelligence team to support meaningful use and other analytics projects. Winning Entry 3 Problem: A major bottleneck in UCMC’s CT imaging workflow is CT postacquisition workflow. In addition, existing workflow requires technologists to push the study to its destinations, fill in paper forms and scan them into the PACS, and manually set up the presentation state in the PACS. The paper-based information that is scanned into the PACS as a secondary image cannot be easily leveraged for subsequent business intelligence or analytics. Solution: The UCMC imaging-informatics team and the CT and PACS vendor jointly developed an optimized postacquisition workflow model that offloads from the scanner all advanced visualization series to an image-processing and -routing service, thus allowing the CT scanner to perform only scanning, imaging quality assurance, thin-slice reconstruction, and autosending of the thin-slice series to the image-processing and -routing service. Once that service receives the study, it asks an intelligent imaging protocol service to get the rules that apply to the study, including protocols for advanced visualization series, routing destination, and notification requirements. This completely frees the CT scanner for the next patient, once all actual scanning is complete, and it significantly improves the efficiency of CT-scanner utilization. The solution makes PACS autohanging-protocol rules easier to define, and automated presentation states can now be created by metadata supplied by the image-processing and -routing service, thus significantly minimizing manual intervention by technologists. To address the manual scanning of paperwork, a document-scanning service was developed: Documents are scanned into the system at the point of the reception by clerical staff, rather than by a technologist. A technologist portal service was developed to preload patient/exam information, prevent repeated manual data entries, reduce errors, and save time for technologists. Technology: The radiology order service, document-scanning service, intelligent imaging protocol service, image-processing and -routing service, and technologist portal service were built for optimizing the postacquisition workflow, based on service-oriented architecture. The radiology order service provides radiology orders from the hospital information system (HIS) or RIS to the other services, based on such query criteria as order status, appointment time, modality, specialty, and anatomic region. The document-scanning service is a tool for scanning paper forms, such as those for external laboratory results and contrast screenings. The intelligent imaging protocol service is a rules engine that provides clinical decisions on imaging protocols, as well as workflow rules based on clinical needs. The image-processing and -routing service processes imaging protocols and workflow rules based on the query results from the intelligent imaging protocol service; the technologist portal service provides a Web tool for technologists, replacing paperwork. An application service provider, SQL server, DICOM library, and other technologies were used to create the solution. Results: A clinical trial using abdomen protocols was conducted from April 20 through July 20, 2010. Six attending radiologists, eight residents, 13 technologists, and front-desk coordinators participated the trial. In the control (non-CLI) cases, the radiologist, technologist, and front-desk coordinator used normal workflow, in which paper forms—including requisitions, contrast-screening forms, protocol forms, charge forms, and technologists’ logs—were carried and operated from during each step of the imaging process. Technologists waited for the CT scanner to finish the advanced visualization series, pushed the series to the PACS, waited for all images to arrive in the PACS, and manually completed presentation setup and scanning of all paper forms into the PACS. In the experimental (CLI) cases, appointment staff used the document-scanning service to scan the paper forms. Technologists used the online technologist portal, and the manual steps after scanning were automated. During the trial, all abdominal outpatient appointments scheduled between 8:30 am and 4 pm on Tuesdays and Thursdays were handled using CLI workflow, while other appointments were handled using non-CLI workflow to generate a random population of measurement data. An image-quality tool was used by the radiologist for both CLI and non-CLI cases to evaluate accuracy and quality. One CT scanner was used for CLI cases; the other six, for non-CLI cases. The trial demonstrated multiple improvements. Paper was eliminated, and the workflow was streamlined by having paper forms online and access-ible, instead of having technologists manually scan the paper forms into the PACS. This also eliminated unnecessary waste of PACS storage, since the scanned paper was stored in the PACS as a secondary captured-image series. In addition, autoextraction of laboratory results and preloading of patient/exam information provided the information needed by radiologists for protocols, using a single portal to improve accuracy and efficiency. The online protocol worklist is available at any time and anywhere, eliminating manual processes and physical/time restraints, and auto-ontology mapping from radiologist protocols to CT machinery protocols eliminates manual translation, avoids errors, and saves both CT-scanner and technologist time. The trial also demonstrated that offloading advanced visualization from the CT scanner saves additional CT-scanner and technologist time. The presentation state was set automatically in PACS, also saving technologist time, and image quality was improved, with respect to missing series in PACS, because the protocol is automated. Additional data, collected using both stopwatch time/motion and system timestamps, corroborate these improvements. All abdominal protocols were divided into two categories: a routine protocol, consisting of one scout and one or two scanning phases (for example, with or without contrast), and a complex protocol, consisting of one scout and more than two scanning phases (such as angiography protocols, which typically comprise one phase without contrast, one arterial and one venous phase with contrast, and one delay phase). For the routine abdominal protocols, CT-scanner efficiency mean time (from the start scan scout command to the availability of the scanner for the next patient) improved by 78.06%, from 14 minutes 17 seconds to 3 minutes 8 seconds. For the complex abdomen protocols, mean time improved by 61.17%, from 27 minutes 54 seconds to 10 minutes 50 seconds. Conclusion: Using integration based on service-oriented architecture, including the image modality, can effectively integrate heterogeneous medical devices and information systems to improve workflow. The use of open-platform software architecture should be encouraged for advanced modalities to empower users to maximize workflow efficiency. Pumping Up Productivity
Michael P. Recht, MD, PhD, is chair of the department of radiology at NYU Langone Medical Center in New York, New York.In May 2009, Recht and his colleagues decided that departmental workflow, communications, and productivity needed to be kicked up a notch. The key to attaining this objective, they concluded, was to build a customizable, flexible, easily updatable application that would reside atop the PACS, RIS, and other information systems and integrate a variety of functions—from electronic protocols to data mining and much more. The project, which kicked off in May 2009 with a request for proposals and went live in October and November 2010, was a highly collaborative effort by members of NYU Langone Medical Center’s department of radiology (radiologists, technologists, administrators, and radiology IT personnel) and its IT department. “The initial stages have been completed and have had a major impact,” Recht says. “Our current work is centered on using our new systems to mine our data to manage our department more efficiently. For example, over the past year, we have developed a strategy map and balanced scorecard for our department. We also are working with our new system to provide data related to our key performance indicators, as well as to understand deviations from our benchmarks.” Other related initiatives underway include using the system to mine data for research—such as radiation-dose, utilization-management, and outcome-analysis studies. “The changing economic climate has made active management of a radiology department imperative,” Recht says. “Such active management requires the availability of real-time data. The analytic module integrated into our system gives our leadership team the ability to access such information in a flexible, user-friendly way.” Winning Entry 4 Problem: Previous PACS and RIS implementations lacked the workflow, communications, and analytics applications needed to execute exam protocols, rapid and accurate image interpretation, data mining for research and operational efficiency, access to online journals, and other critical daily tasks effectively. Solution: We first identified the categories of systems needed, including PACS, integrated viewers for nuclear medicine and ultrasound, a vendor-neutral archive, voice-recognition systems, advanced visualization, and document management. We used our existing RIS. We then identified an applications specialist vendor to create a customized, tightly integrated (but loosely coupled) software application called PRISM, layered on top of these systems. This layer offers extensive functionality, including customizable reading worklists, electronic protocols, radiology and technologist synchronous and asynchronous communications, quality-control systems, residents’ workflow, search and research tools, teaching files, document management, real-time metrics, and (soon) radiation-dose tracking. This layer is now the front end interface used by all members of our department. Technology: PRISM was built atop industry-standard Web-server databases, computer languages, and communication protocols. A variety of application programming interfaces and messaging systems were used to communicate with, control, and respond to underlying components in the PACS, RIS, and other systems. A thin-client configuration with an automatic updating infrastructure was used to allow growth and change over time. Reliability was ensured though the use of high-availability systems with redundant servers. Results: Electronic protocols and reliable synchronous and asynchronous electronic communication (instant messaging, integrated email, and technology notes that open automatically with the launching of each exam) between technologists and radiologists allow for more accurate exam protocols and monitoring. Technologists have become a more integral component of the clinical care team, increasing their morale. Integrating the application portal with the EMR allows the EMR to be opened in context when each exam’s protocol is set and/or the exam is launched. This gives us more informed and optimal protocols, the convenient (and rapid) ability to check laboratory values, and better access to clinical concerns during image interpretation. An embedded document-management system allows for easy and reliable integration of outside reports and other relevant documents. Electronic protocols will soon incorporate radiation-dose–management features. Worklists with advanced filtering and real-time faceting/subfiltering have led to a major restructuring of daily workflow, with turnaround times decreasing by more than 40%. Radiologist collaboration has increased, and instant access to search engines embedded within the reading environment—which instantly search the entire archive of historical reports—yields more rapid, reliable data mining for research projects. Searches that formerly required several hours and the creation of dedicated reports now are accomplished by radiologists and house staff in seconds. The integration of a teaching-file tool with the ability to capture key images and text easily from reports and to index each case with either free-text keywords or RadLex™ greatly facilitates the creation of teaching files. Conclusion: A customized, integrated layer on top of best-of-breed underlying systems has provided NYU Langone Medical Center’s department of radiology with optimized functionality in the areas of workflow, communications, and analytics. In addition, the architecture of this solution allows for customization, rapid change, and growth. A Balanced Workload
Jon Copeland is CEO of Inland Imaging Business Associates in Spokane, Washington.Like many large radiology practices, Inland Imaging was struggling to balance workloads across groups and varied locations, to meet increased regulatory requirements for proof of quality, to respond to shifting payment models, and to increase the productivity of the partners. Beginning in 2006, Copeland began work on a streamlined, data-rich, Web-based workflow system that integrates with any RIS or PACS. “We started work in 2006 and did the implementation in 2008 and 2009,” he says of the practice’s enhanced communications system. It was time well invested. In addition to other benefits, radiologist workloads are now distributed far more equitably, and workflow is smoother. Moreover, while Copeland considers the project to be complete, enhancements and the addition of features are ongoing. “We sold the system to a vendor, and it is doing great things with it that were beyond our abilities—like DICOM enhancements and other decision-support and office-based patient systems that we want or need, but did not have the capacity to build ourselves,” Copeland notes. Winning Entry 5 Problem: There was a need for an improved IT workflow infrastructure with intelligence beyond that of any single PACS, RIS, HIS, or other system, along with a need to provide a Web-based form that technologists and others could use to submit updates. Solution: A streamlined, data-rich, Web-based workflow system was developed to function in multispecialty clinics, rural hospitals, major tertiary-care hospitals, and the practice’s imaging centers. The same workflow system and common subspecialized worklists are used for and by all technologists, radiologist assistants, dispatchers, and radiologists. A three-tiered worklist allows a specific work assignment to be made to an individual radiologist, a shared subspecialty worklist, or a catch-all worklist for general-radiography exams. The system can track work RVUs, and we developed our own algorithm for daily equivalency of work performed. Technology: The system uses Microsoft® .NET components and resides on a single independent server, integrated via our own interface engine. It is built on and supported for the Microsoft platform. It is designed to integrate with any RIS/PACS that supports integration. Results: Since implementation, we have seen a 14% improvement in radiologist productivity. The ability to balance workloads quickly across the system has dramatically reduced variation in the work performed by radiologists. Before implementation, we had more than a 50% daily variation in work in some subspecialties. Variation is now less than 10%. The system includes a scheduling and credentialing component that knows everything about our radiologists, including their subspecialties and what shifts they can work, in which cities. The system improves productivity, and there are quality-tracking functions. Our report-turnaround times have improved significantly due to our ability to balance workloads and identify exams by urgency category (routine, urgent, emergency, or stroke). There is also a peer review, based on ACR standards, within the workflow system. We perform both retrospective and prospective peer review. The system helps to optimize technology investment and drive clinical and operation results that make a difference. Our radiologists have benefited from a balanced workday and from increased productivity, and they no longer waste time on administrative issues. Conclusion: The rules of radiology are changing and will continue to change. It is critical, as a radiology group, to provide added value—including workflow systems—in addition to having accurate data to measure productivity and quality. The Judges These SIIM members scored the entries in the Top Five Medical Imaging IT Projects competition. Raymond J. Geis, MD, is a radiologist with Advanced Medical Imaging Consultants, PC, Fort Collins, Colorado, and is chair-elect of the SIIM board of directors. David S. Hirschorn, MD, is director of radiology informatics in the radiology department at Staten Island University Hospital in New York and is a member of the SIIM education committee. Woojin Kim, MD, is a radiologist in the department of radiology at the University of Pennsylvania School of Medicine in Philadelphia and is a member of SIIM annual-meeting program committee, 2011–2014. Elizabeth A. Krupinski, PhD, FSIIM, is a research professor in the department of radiology physics at the University of Arizona in Tucson and is chair of the SIIM board of directors. Christopher D. Meenan, CIIP, is director of clinical information services in the department of radiology at the University of Maryland Medical System in Baltimore and is treasurer of the SIIM board of directors. James T. Whitfill, MD, is CIO, information services, at Scottsdale Medical Imaging in Arizona and is a member of the Journal of Digital Imaging editorial board.
Peer-review System With Brains
Yun (Rob) Sheu, MD, is a radiology resident at the University of Pittsburgh Medical Center in Pennsylvania.
Beginning in 2008, Goldszal’s goal was to create one enriched reading environment for a practice that reads a million studies annually from five distinct hospital systems, 10 outpatient imaging office locations, and three teleradiology clients, aggregating studies from all sites for the purpose of accessing prior studies. University Radiology Group went live with the solution in 2008, starting with one site. Other sites are being gradually phased in at intervals of approximately three months. “It’s definitely doing what it was designed to do,” Goldszal says. “It’s an organic project, however, and it keeps on growing and changing.” Currently, the practice is fully engaged in developing a regional imaging exchange in Central New Jersey. There, imaging studies originating from multiple unaffiliated organizations will be logically grouped by patient and presented at the point of care whenever required. Winning Entry 2 Problem: University Radiology Group wanted to provide fast, efficient, accurate subspecialty interpretations based on the most complete clinical information available. Solution: We developed a cross-institutional, patient-centered, longitudinal imaging database that delivers patients’ entire imaging history (regardless of image-acquisition site) and uses a standard viewing platform. The PACS aggregates clinical results and longitudinal imaging studies across health-care organizations, using networking technology, as well as aggregated DICOM-based datasets; aggregated HL7-based RIS datasets; and a combination of DICOM, RIS, and nonstandard data, such as scanned documents. The system combines patient datasets stored in multiple unaffiliated sites under different medical-record numbers. Technology: At the heart of the cross-institutional PACS is a dynamic electronic master patient index—an algorithm that performs probabilistic matching of patient data scattered across health-care institutions, in the absence of a common and unique health-care identifier. The solution is implemented using off-the-shelf components, in addition to a commercial PACS, reporting system, and interface engine. No customization was necessary. Results: We have developed and implemented a longitudinal, patient-centered imaging database of radiological studies acquired at multiple, unaffiliated health-care organizations. A live system aggregates and provides the final interpretation for more than a million studies that originate from 18 locations and from nine distinct, unaffiliated PACS/RIS implementations. Cases are presented to geographically distributed interpreting radiologists, who access all studies, via global worklists, through a standard diagnostic workstation. Aggregated patient data presented to each radiologist include historical exams with final reports, ancillary clinical data (such as technologist and nursing notes), and patient demographic data originating from multiple data sources. Upon interpretation, final results are automatically transmitted to the native RIS and/or electronic medical record (EMR) system for the permanent record, and to clinical departments (such as the emergency department) for immediate patient-care decisions. The consolidation of patient records is possible due to the development and implementation of HL7-based orders/results interfaces or HL7-based registration and admission, discharge, and transfer (ADT) interfaces between University Radiology Group’s system and the RIS (or equivalent order-entry system) of the image-acquisition site. Based on the patient demographic information (including name, date of birth, gender, Social Security number, and address) available in the order or registration stream drawn from the RIS at each facility, we are able automatically to perform a DICOM Query/Retrieve or DICOM C-Move to fetch the corresponding new and old imaging studies stored in the corresponding hospital’s PACS. In case an HL7-based ADT or electronic order is not available, our system is capable of abstracting enough demographic information from the DICOM header (available in the imaging study) to trigger the fetching of prior images and historical medical records. The system—using the probabilistic matching algorithm—continually links the patient’s identity across all health-care facilities with which we connect, allowing all-encompassing access to relevant prior images and clinical information stored across regional health-care organizations. Conclusion: Our solution represents a major departure from traditional methods that rely on a single PACS and an institutional medical-record number to store and manage patient data. These schemata only provide access to a limited set of imaging and clinical history. We believe the lack of complete patient history can lead to duplicative (and often, unnecessary) exams, lower-quality diagnoses, and/or delayed results. Our solution accounts for these shortcomings and promotes better multisite data integration. Liberating CT Workflow
Alberto Goldszal, PhD, is CIO of University Radiology Group, PC, in New Brunswick, New Jersey.
Alberto Goldszal, PhD, is CIO of University Radiology Group, PC, in New Brunswick, New Jersey.
Li Lillian Hou, MS, CIIP, is a specialist programmer, radiology information services, at the University of Chicago Medical Center in illinois.
Michael P. Recht, MD, PhD, is chair of the department of radiology at NYU Langone Medical Center in New York, New York.
Jon Copeland is CEO of Inland Imaging Business Associates in Spokane, Washington.