Loud and Clear
Radiologists articulate the advantages and challenges of speech-recognition software.
Radiologists have adjusted to many new technologies, but few have posed a bigger challenge than speech recognition (SR).
Using computers to aid in translating speech into text, SR strikes at the heart of radiology practice and work routine. SR performs its primary function precisely at the moment radiologists perform theirs.
Many radiologists dislike SR and prefer the reassuring backup of a medical transcriptionist. In fact, SR forces radiologists to correct mistakes and edit their work, which sometimes makes interpretation more time-consuming.
They are also wary because of SR's tendency to make rare, but dangerous, mistakes. Nonetheless, the service has evolved, and the most recently released software's error rates have fallen to about 2.5 percent
Although radiologists still debate SR's merits, many hospital administrators are convinced it saves money — often hundreds of thousands of dollars annually in reduced medical-transcription expenses. Vendors typically promise a positive return-on-investment in about a year.
SR also dramatically cuts imaging-report turnaround times, creating goodwill among referring physicians whose loyalty and patients are key to a hospital's financial success. Recent product developments are designed to capitalize on SR as a workflow tool to automatically load information, boost productivity, and draw data from the system for billing, patient management, and quality control.
Despite the misgivings, SR is on its way to becoming as ubiquitous as digital workstations in reading rooms. Indeed, it's hard to find a U.S. academic radiology department that continues to rely mainly on traditional transcription, notes Arun Krishnaraj, M.D., M.P.H., clinical fellow at Massachusetts General Hospital and Harvard Medical School, who studied SR practices at the University of North Carolina.
Medium-sized acute care hospitals are adopting SR swiftly, and small hospitals are freestanding imaging services have much to gain. "I would suspect the improvement in report turnaround that you get with speech recognition would be even greater with small practice groups and hospitals," Krishnaraj says.
One group with a positive SR experience ins Radiology Consultants of Iowa (RCI) in Cedar Rapids, Iowa. RCI first applied the concept in 2005 as part of a wide-area network PACS/RIS implementation for 12 small, rural hospitals. The goal was to view images and dictate reports any time, any place, says RCI partner John L. Floyd, M.D., FACR. In fact, senior partner W. Jay Friesen, M.D., uses the system to dictate reports during frequent visits to his second home in Arizona.
The new configuration delivered on RCI's promise to its rural hospitals for report turnaround times as short as those at its large facilities in Cedar Rapids. With the exception of emergency CT transmitted via rudimentary teleradiology, the rural sites were accustomed to turnaround times as long as five days. After installation, however, most studies were signed and transmitted back in less than two hours.
RCI then took steps to collaborate on proposed SR installations with its two largest clients, Mercy Medical Center and St. Luke's Hospital, both in Cedar Rapids. Software was installed at Mercy after the group contracted to manage the facility's in-house transcription service and promised to charge the hospital less per report than it paid when operating the service itself. The hospital continued to manage its own PACS and RIS.
St. Luke's came on board in January 2007 when Floyd showed its management how SR was dramatically improving report turnaround times for its nearby competitor. The change at St. Luke's was impressive. Previously, only 20 percent of reports had been available to referring physicians in less than two hours after imaging. After implementation, 90 percent arrived in less than two hours, says Dennis E. Winders Jr., St. Luke's director of imaging services. Since mid-2008, mean turnaround time for final reports for ER imaging has been less than 15 minutes.
Improvements were especially evident in the ICU. "Physicians who arrived in the mornings to make rounds among ICU patients found that their radiology reports were waiting for them," Floyd says. "When they came in to make afternoon rounds, imaging reports were sometimes showing up in patients' records before the patients returned to their rooms."
Referring physicians were especially pleased with the improved performance, notes Kathy Epley, RCI's chief administrative officer. A Press-Ganey survey conducted a year after implementation found that St. Luke's and Mercy scored with the top 1 percent of U.S. hospitals for referring-physician satisfaction with report turnaround times.
Radiologists also benefitted professionally and financially from the new SR system. Faster turnaround led to fewer interruptions for wet reads, Floyd reports. Attending radiologists were more likely to be available to field questions when referring physicians read the findings. And RCI's transcription services at both hospitals made a profit.
Like many SR implementations, Floyd notes that his group's effort started acrimoniously. Deep divisions emerged from initial discussions about RCI's strategy. "Now, you won't find any radiologist in our group who would prefer to return to traditional transcription, mainly because of the backend benefits of immediate filing," he says.
Michael N. Brant-Zawadzki, M.D., FACR, tells a similar story about his group's initial skepticism and eventual acceptance of SR at Hoag Hospital in Newport Beach, Calif., in 2005. To overcome resistance, Brant-Zawadzki argued that self-editing improved on traditional transcription methods. "Before SR, reports at Hoag Hospital were dictated in a stream-of-consciousness fashion without a conscientious attempt by the radiologist to structure them properly," he says.
Most reports were accurately transcribed, but there was no guarantee that the transcriptionist understood crucial details, so errors could find their way to the referring physician. Confirming clinical results was nearly impossible because of the way work was organized, Brant-Zawadzki adds.
Now, about 80 percent of the reports are self-edited after SR implementation. This extra step forces radiologists to invest 10 to 20 percent more time into each report, but it also delegates responsibility for correcting errors to the person most qualified to find them, according to William J. Van Dalsem, M.D., Hoag Hospital's medical director of radiology.
While Van Dalsem doubts that every radiology group will embrace SR, he is committed to the self-editing concept. "The style and quality of our group's reports are generally much improved," he shares.
New Features Save Time
Despite some lingering skepticism, labor-saving upgrades are encouraging broader use of SR. Radiology reports are now frequently prepopulated with demographic and clinical history, as well as imaging protocol data from RIS and PACS. Other time-savers include a growing library of report templates providing shortcuts to reporting normal findings, and dictation that can be limited to exceptions only, such as the presence, size, and location of abnormalities. And, the source of all material in a report can be color-coded before self-editing so radiologists can focus on critical sections where significant errors may reside.
Newer software also helps radiologists more accurately structure reports, add punctuation, and correct bad grammar, which means less editing time. Another key advantage is that SR users can be guided by master work lists that alert them of the status of studies awaiting interpretations at multiple hospitals and imaging centers served by the practice.
New, computer-assisted coding engines that draw data from the final report for automatic coding and insurance billing can provide additional efficiencies. These tools can also mine PACS, RIS, and electronic medical records for data to assist administrative management-by-exception reporting. And, future enhancements may even support structured reporting. Although new features continue to improve SR, it may take time for radiologists to fully embrace it; in the meantime, case studies such as those in Iowa and California can help illuminate its value and future ways to refine its implementation.
By James Brice