CHICAGO - If radiologists have mixed feelings about switching from tape transcription to voice recognition-based reporting, there's reason for the reticence.
In a series of studies presented Sunday in the health services policy and research scientific sessions at the RSNA conference, doctors concluded that using voice recognition systems complicated and lengthened the reporting process. Still, these problems were seen as being outweighed by benefits such as same-day report availability and other benefits to be realized in the future.
Dr. B.D. Gale of the SUNY Health Sciences Center in Brooklyn reported on a study determining the time required to produce reports on an IBM MedSpeak system compared to tape transcription. The study found that three radiologists took significantly longer to create MedSpeak reports - an average of 139 seconds for the system-generated reports versus 27 seconds for reports dictated on tape. The study comprised 70 reports including 27 plain films, 25 mammograms, and 17 GI/GU exams.
The three radiologists were then informally polled to determine the reasons for the excess time needed with MedSpeak. They attributed the time difference to several factors, the most significant of which was time spent correcting errors in transcription, followed by the need to retrain the application vocabulary, and recovering from system crashes. He said they found it was often faster to type the reports than to correct errors in the system.
Dr. Gale said that the time difference has narrowed slightly since the study was completed in January 1999.
MGH findings
Dr. Amit Mehta presented the findings of a broader study conducted at Massachusetts General Hospital in Boston, a large public facility that has relied on voice recognition reporting for more than four years. The study sought to measure the impact of voice recognition on radiologists' efficiency by comparing the time required to produce 100 randomly selected reports on MedSpeak compared with 100 reports on Sudbury Systems' RTAS dictation system.
It took 27.7% longer to produce reports on the voice recognition system than with the conventional dictation system, a far smaller difference than was noted in Dr. Gale's study.
Despite the small productivity difference, however, Dr. Mehta said the researchers uncovered several major problems with voice recognition, including the need for radiologists to "spend less time looking at films and more time correcting errors." He said doctors with high voices or foreign accents have severe problems using the system as well, and that simply correcting dictation errors took so long that he could no longer tell the staff how much time they should spend reading films, "which is a scary issue," he said.
Nevertheless, he saw promise in the system down the road, and noted that voice recognition did offer important ways to improve productivity, from the use of macros and templates in dictating reports, to the time radiologists saved by not having to write preliminary notes on the film bag. The most important difference in using MedSpeak, he said, was that written reports that once took an average of 4.3 days to become available were now available in an hour, and as quickly as five minutes for urgent reports.
Dr. Mehta said he remained completely committed to voice recognition technology. "We've supported it for four years and we're going to continue to support it because of these added benefits," he said.
Dr. Mehta then presented a second study that was undertaken to determine whether voice recognition technology shortens radiologists' reports. The study used 100 studies, mostly normal, and measured both the impression and the body of the report.
The unexpected result was that, while the average MedSpeak report length was close to those produced with tape dictation, 36 and 24 words each, respectively, the composition of the reports was very different. The impressions produced with MedSpeak were 40% longer (15 words) than those produced with conventional dictation (7 words) while the body of MedSpeak-produced reports was 22% shorter than those produced with conventional dictation.
Future potential
Keynote speaker Dr. Curtis P. Langlotz outlined the problems associated with the traditional tape transcription process, including the high cost of employing transcriptionists, errors in dictation and transcription, and reports that were unclear, unstructured or ambiguous.
Current voice recognition systems offer several advantages over traditional tape systems, he said, including lower personnel costs, and faster, better structured reports that are linked directly to the images being examined.
Even more benefits could come in the near future, he said, with the addition of natural language processing and artificial intelligence to tag or extract the text, "so you can in effect extract the various concepts out of the report, and then structure data so that you can provide it back to the other information systems."
If radiologists were to extract diagnostic data in this manner and feed it back into the information systems, Dr. Langlotz said, "the information could be used to provide diagnostic advice to doctors at the time of dictation."
By Eric BarnesAuntMinnie.com staff writer
November 30, 1999
