CHICAGO - While less accurate than human transcriptionists, voice recognition technology results in speedy transcription-to-approval time and offers tangible financial incentives, according to research presented at this week’s RSNA meeting.
To compare the performance of a voice recognition system with a conventional transcription service, researchers at the Duke University Medical Center in Durham, NC, prospectively and randomly reviewed VR dictations in each diagnostic session during a six-week study period. Faculty radiologists, fellows, and senior radiologists performed the dictations, with all radiologists having formal VR training and a minimum of one year of experience with the system, according to Dr. John Pappas, who presented the study.
The radiologist dictating the examination was blinded to the date and time of review, preventing alteration of dictation style and format. All dictations were recorded on the computer hard drive to allow review of every word, Pappas said.
One day, Duke’s VR system crashed, and all dictations were performed on tape recorders and transcribed by a conventional transcription service. This day served as the researcher’s control comparison for the manually transcribed reports, Pappas said.
The researchers then retrospectively reviewed transcription-to-approval times of VR dictations over a two-month period, and compared the results with those of the transcription service during a prior seven-month period. The complete start-up VR cost (hardware, software, technical support, and licensing and installation fees), was retrospectively determined by billing records, and was then compared with the annual cost of the transcription service.
Eighteen radiologists dictated 125 reports. Of the 19,901 words dictated, there were 895 transcription errors (770 computer, 125 human) for an accuracy rate of 95.5%, according to the study team. Of the 125 reports, 118 (94.4%) had at least one error.
The transcription service performed better in transcribing the 153 reports dictated by eight radiologists. Of the 15,194 words dictated, there were only 33 errors, giving them an accuracy rate of 99.8%. Only 21 of the 153 reports (13.7%) had at least one error.
The VR system did offer performance advantages, however. The mean transcription-to-approval time for VR was 26 hours (median 8.5 hours, 79.5% within 24 hours), while the average time for the transcription service was 35.7 hours.
Voice recognition also conferred financial benefits, paying for itself within one year and eight months, according to the researchers. The total startup cost for the VR dictation system was $789,599, compared with one-year costs of $475,059 for the transcriptionist service.
“There’s a large initial capital investment (for voice recognition), but once this is paid, we estimate annual financial savings of over $475,000,” he said.
By Erik L. Ridley
AuntMinnie.com staff writer
November 28, 2000
To compare the performance of a voice recognition system with a conventional transcription service, researchers at the Duke University Medical Center in Durham, NC, prospectively and randomly reviewed VR dictations in each diagnostic session during a six-week study period. Faculty radiologists, fellows, and senior radiologists performed the dictations, with all radiologists having formal VR training and a minimum of one year of experience with the system, according to Dr. John Pappas, who presented the study.
The radiologist dictating the examination was blinded to the date and time of review, preventing alteration of dictation style and format. All dictations were recorded on the computer hard drive to allow review of every word, Pappas said.
One day, Duke’s VR system crashed, and all dictations were performed on tape recorders and transcribed by a conventional transcription service. This day served as the researcher’s control comparison for the manually transcribed reports, Pappas said.
The researchers then retrospectively reviewed transcription-to-approval times of VR dictations over a two-month period, and compared the results with those of the transcription service during a prior seven-month period. The complete start-up VR cost (hardware, software, technical support, and licensing and installation fees), was retrospectively determined by billing records, and was then compared with the annual cost of the transcription service.
Eighteen radiologists dictated 125 reports. Of the 19,901 words dictated, there were 895 transcription errors (770 computer, 125 human) for an accuracy rate of 95.5%, according to the study team. Of the 125 reports, 118 (94.4%) had at least one error.
The transcription service performed better in transcribing the 153 reports dictated by eight radiologists. Of the 15,194 words dictated, there were only 33 errors, giving them an accuracy rate of 99.8%. Only 21 of the 153 reports (13.7%) had at least one error.
The VR system did offer performance advantages, however. The mean transcription-to-approval time for VR was 26 hours (median 8.5 hours, 79.5% within 24 hours), while the average time for the transcription service was 35.7 hours.
Voice recognition also conferred financial benefits, paying for itself within one year and eight months, according to the researchers. The total startup cost for the VR dictation system was $789,599, compared with one-year costs of $475,059 for the transcriptionist service.
“There’s a large initial capital investment (for voice recognition), but once this is paid, we estimate annual financial savings of over $475,000,” he said.
By Erik L. Ridley
AuntMinnie.com staff writer
November 28, 2000
Copyright © 2000 AuntMinnie.com
Click here to view the rest of AuntMinnie’s coverage of the 2000 RSNA conference.
Click here to post your comments about this story.
