By Richard Elsberry
Republished with permission from Decisions in Imaging Economics, Spring 2000 Special Edition
Unpaid labor
The PACS imperative
An alternate view
Went cold turkey
Point of no return
The winds of technological change are blowing through hospital imaging departments, forcing radiologists to make major changes in the way they historically have done their job.
The reason for this behavioral shift is voice recognition technology (VRT)-also known as speech recognition software-that can transfer spoken dictation in real time direct to a computer screen with 95% or greater accuracy.
VRT, which has come of age in the past 12-18 months, is enabling many medical centers to shift the responsibility for typing dictated imaging reports away from pools of skilled medical transcriptionists to the radiologists themselves. Eliminating transcriptionists, who in recent years have become increasingly difficult to recruit and retain, represents a major cost savings for radiology administrators. The new technology is not inexpensive, but most hospital administrators believe they can recoup their initial investment in 1 to 3 years, and then reap ongoing cost savings. Historically, the fee for the technical portion of an examination has included the cost of transcribing and distributing the reports dictated by radiologists. Although that cost is being shifted to the professional side of the equation, no one foresees any change in the established Health Care Financing Administration reimbursement rates for imaging examinations.
The shift from transcriptionists to VRT not only is a cost savings for hospitals, it also is drawing kudos from clinicians, who now can receive a faxed report on their patient minutes after the image is read, rather than having to put up with the frustrating 2- to 4-day-and sometimes even longer-delays that have been commonplace in the past. Faster reporting also means expedited billing for both the technical and professional components of an imaging examination, improving cash flow.
While hospitals and referring physicians are enthusiastic about VRT, an informal survey of radiologists who have to use it indicates they have mixed emotions. The survey involved radiology departments at Beth Israel Medical Center, the university hospital and Manhattan campus of Albert Einstein College of Medicine, New York City; St Luke's Roosevelt Hospital Center, the university hospital of Columbia University College of Physicians and Surgeons, New York City; New York University (NYU) Medical Center and its affiliates, NYU Downtown Hospital and the Hospital for Joint Diseases, New York City; and Shands Hospital of the University of Florida College of Medicine at Gainesville.
The bottom line is that many radiologists believe they have been saddled with extra work that they are expected to perform without receiving any additional compensation. The average time required to prepare a report using VRT is estimated to be 10% to 40% longer, although a minority of users-primarily those who have developed time-saving macros-claim it is slightly faster than traditional transcription.
The additional time requirement results primarily from the need for radiologists to carefully read each report for accuracy before signing it. In the past most relied on transcriptionists fluent in medical terminology to flag possible errors, calling them to the attention of the radiologist. Many radiologists are worried that lost productivity could necessitate their practice hiring additional physicians to handle the workload in a timely manner.
In addition, some radiologists still have not become adept with the technology after using VRT for a year and regularly require assistance. And there are others, often with foreign accents, who are ready to give up, claiming the technology works for them only 70% of the time. That translates to 30 errors or omissions in every 100 words that have to be ferreted out and corrected manually using a keyboard.
"There is no incentive for radiologists to use this system," notes Michael Abiri, MD, chairman of the Radiology Department at the 1,234-bed Beth Israel Medical Center, where a VRT system interfaced to a PACS has been in place since late 1998 and is used by some 40 radiologists who interpret 200,000 examinations a year. "The radiologist who used to spend half an hour dictating 50 cases now spends an hour dictating and correcting the same cases before finalizing the report. The perception of most radiologists is that they are not gaining anything but losing productivity."
Still, Abiri believes the advantages outweigh the disadvantages. The main advantage is faster communication of results to clinicians using the hospital information system (HIS) or the radiology information system (RIS). "We have seen a reduction in the number of phone calls requesting patient reports, thus eliminating this disruption in the daily routine," he says. Also, the time formerly spent handling such calls is usually not taken into account when radiologists calculate that they are now spending up to twice as much time on VRT-generated reports, Abiri notes.
"If you consider everything, including reviewing the report, making corrections, and then signing it, I would say VRT probably takes 1.2 to 1.3 times longer than the classic transcriptionist method," believes Marc Katz, corporate director of radiology for Beth Israel. "It is a difficult system that requires persistence, but we have radiologists in our group who claim it is faster."
Still, Abiri is convinced that faster, more accurate reports are worth the extra time required by VRT. "The difficulty is convincing radiologists in general that this is not an administrative or cost savings issue but rather a quality issue. The radiologist's role is more than just looking at films or completing the dictation. It is to communicate his interpretation, in a timely fashion, to the clinicians.
"Once you install a PACS, you actually have no choice, since clinicians can view the images as fast as the radiologists. If reports are not timely, the clinicians will begin to question why there is a need for radiologists."
Although VRT has been around for at least 10 years, Abiri believes that only in the past 2 years has the technology reached a point where it is capable of producing reports with minimal errors. However, he is already exploring developmental software that may extend accuracy from 95% to more than 99%.
At Beth Israel, radiologists use Windows NT workstations interfaced to PACS workstations. When the images have been sent from the server to the PACS station, the user mouse-clicks on an icon, which automatically brings the patient's HIS and RIS data up on the Windows screen. The dictated report appears on the screen word by spoken word, and, after editing, is entered into the HIS and RIS and faxed to the clinician.
Katz and Abiri also direct the Radiology Department at the 729-bed St Luke's Roosevelt Hospital Center, where films are still read in the traditional manner but VRT is used to prepare reports. A PACS is being installed this summer.
While nine out of 10 radiologists at Beth Israel are using VRT, there are still a few using transcriptionists "because the computer does not understand them," Katz reports. "This is something that has to be accepted." To serve these strugglers, to provide backup should there be an emergency, and to work with staffers who have a cold or laryngitis, approximately one-fifth of the original transcriptionist pool has been retained.
"The current system, no matter what you do, does not recognize certain words when spoken by certain radiologists," Abiri says. "These are repetitive mistakes that can frustrate the radiologist. Sometimes the computer cannot recognize a very simple word." One way around the problem has been to create macros for the words the computer consistently mistakes. "We are hoping a new voice engine will solve some of these problems."
At Beth Israel formal training is emphasized, and it maintains a hot line to handle questions. "Sites that are really successful have somebody the radiologists can call on when they get into difficulties," Katz explains. "This is not like many other systems where you can wait for a technician to come and fix it. Radiologists are used to picking up a microphone and dictating. If they can't do that, they get frustrated. For the first 6 months you need an on-site person who knows the system very well and can hold their hand."
The 50 radiologists in the radiology department at New York University are using VRT to report on 250,000 examinations a year. Since the system was installed in the fall of 1998, the number of transcriptionists at the three NYU hospitals, and the practice's own outpatient imaging clinic, has been reduced by 90%, says Andrew W. Litt, MD, vice chairman and director of neuroradiology at NYU School of Medicine. "We tend to use transcriptionists for complicated cases, consultations that are not in the system, or reports that need to be retyped because they got messed up," he explains. "No system is perfect. There are always little errors that need to be dealt with."
The VRT system was extended in January 2000 to Bellevue Hospital, an 888-bed city-owned facility, where the NYU radiologists also practice. At all four hospitals VRT was a joint investment of the hospital and the practice doctors. "A major reason we did it was because we believed that if we did not provide information rapidly, within hours, then people were not going to care what we say," Litt notes. "So even if we have to do a little more work, we think this is the way medical care should be practiced."
While VRT is not adding that much time to report preparation-perhaps 10% on average-it is redistributing how the time is spent, Litt says. "In addition to dictation time, there also used to be time spent on reviewing the typed report, but because it wasn't sequential it tended to be overlooked. Now I know the report is correct at the time I still have the films in front of me. Further, I used to spend a big part of my time getting calls from doctors who hadn't yet received the reports because they were still being transcribed. Not having to deal with those phone calls has made a big difference."
Some NYU radiologists have been reading plain film and others PACS images. The PACS is only now being fully deployed. In both instances using VRT "takes a while before you get good at it," Litt says. "The worst thing you can do is to speak very slowly and use very segmented speech, or try to make your words very clear. You should speak at a constant, even pace without a lot of breaks in between."
While both NYU and Beth Israel elected to maintain a transcriptionist cadre when they made the move to VRT, Shands Hospital at the University of Florida bit the bullet and eliminated every transcriptionist job. "We went cold turkey," says Walter Drane, MD, professor and interim chair of the Department of Radiology. "It was sink or swim. We are always pushing the envelope and what we have found is that you don't throw too much too soon at anybody. But when you do it, you don't give them a bailout. If you do, they'll bail out."
While its six former transcriptionists are history, two new posts were created to facilitate the changeover: a technical administrator who spends a third of his time making sure that the server is running and that backups are available; and a nontechnical administrator who helps users when they are in trouble and makes certain the reports are going to the right place.
There are 22 faculty radiologists, 12 fellows, and 24 residents reading some 240,000 images produced annually at the 577-bed hospital and three hospital-owned outpatient imaging centers. All facilities are linked by a hybrid PACS developed in house. The group also reads images digitally transmitted from other facilities up to 100 miles distant.
The year-old VRT system at Shands-supplied by a different vendor than the two Manhattan systems-uses a Windows NT server with more than 30 workstations. It is not yet fully integrated with the PACS. As a result, radiologists have to deal with an array of three screens-two for film and one for text-and two keyboards and two microphones.
Starting with a paper request, a radiologist goes to the PACS and finds the patient's name on a directory to access the film. Then he goes to the archives to find any old films, and to the HIS to get old reports. Finally, he activates the transcription system and logs in the patient ID. In queuing up the elements needed to prepare the report, he is completely on his own.
Reports, which in the past often took 3 to 4 days to transcribe after being dictated, are now available to clinicians on the HIS an hour or less after the images are made. Even so, many physicians have developed unrealistic expectations for VRT. "When our turnaround time was 36 hours, no one ever called me," Drane says. "Now, when my turnaround time is an hour, I get complaints all day. People are calling for reports when the patient is not off the scanner yet."
The new reporting system requires that radiologists very carefully read every word of every report. This is important because some systems occasionally drop words that are extremely critical, such as "no," and the radiologists can no longer rely on feedback from transcriptionists to weed out errors. A survey conducted by Christopher Sistrom, MD, assistant professor of radiology, in which most Shands radiologists participated, concluded that the mean time to do a report has increased by a factor of 1.4. Only two thought VRT was faster, Sistrom reports. Yet, 80% of the respondents said that if given the chance to return to the former reporting system, they would not take it.
Drane questions whether the extra time radiologists believe they are spending on VRT reports is real or simply misperceived. "I am still trying to figure out how, if we are spending 30% to 40% more time, my guys can go home at 5 o'clock," he says. "They are reading reports they never read before, but they still seem to go home on time."
The extra time required to carefully read and correct voice-generated text may have another negative aspect, Sistrom says. He is concerned that less time is being spent looking at the film. "A radiologist in the past may have spent 30 seconds doing a chest report, spending 25 seconds actually looking at the film and talking about it. Now the whole process may take 2 minutes because you are not able to concentrate on just the images and the microphone. One time you are looking at the film, then you are trying to get the computer to work and produce the report, then you have to correct the report and send it off. By affecting the attention span, there is an increased chance of an error being made in the reading."
VRT is easy to implement, according to Drane. "You spend 20-30 minutes reading to the box, then off you go," he explains. "We've had very few problems, mostly intermittent interface problems with our HIS. This system teaches you how to talk to it, how to modify your behavior to get high recognition. If you have a bad accent but you always talk the same way, this system will spot you dead on. It won't miss a thing. The key is always doing it the same way."
Some VRT vendors are saying that in many settings perhaps only 70% of the reports should be done with speech recognition software and 30% the old-fashioned way. But financial forces being what they are, hospital transcriptionists appear to be an endangered species. Sistrom, however, believes professional practice groups might consider employing some transcriptionists to enhance the productivity of their radiologists. He considers that approach a better response than hiring additional radiologists to maintain a practice's throughput.
But where radiologists own and operate a diagnostic imaging center, collecting both the technical and the professional fee, VRT may prove the most cost-efficient way to run the enterprise.
Richard B. Elsberry is a contributing writer for Decisions in Imaging Economics.
Let AuntMinnie.com know what you think about this story.
