Vol. 6 •Issue 8 • Page 43
Improving Medical Transcription
Brigham and Women’s Hospital implemented a new system that has helped to improve the transcription process while lowering costs.
At Brigham and Women’s Hospital we continuously look for ways to improve our infrastructure. In particular, we focus on using new technologies to improve our patient outcomes, physician efficiency and quality of life, while lowering operating costs. Medical transcription stood out as a process within our infrastructure noted for increasing expenditures — with decreasing satisfaction. This article describes what we did to turn medical transcription into a positive process.
Currently, our physicians dictate office notes, letters to patients and other physicians, and some phone interactions. We use a computerized medical record for all our primary care patients.
However, prior to January 2001, we were using three different transcription services for our nine primary care and specialty practices. Our practices budgeted $1 million per year for transcription for just these primary care and specialty practices. We budgeted another $750,000 for discharge summaries and operative notes. Moreover, physicians and office staff were not satisfied with the outpatient services being provided. The transcription companies we used at the time returned documents in 4 to 7 days. If a document had to be edited, it would often take up to one additional week to receive the final document. We knew we needed to cut expenditures on this part of our infrastructure.
We also wanted a process that would achieve the following goals:
• produce an accurate document;
• decrease our costs;
• decrease the interval between dictation and available document;
• require no training or change for the physician; and
• upload the documents accurately into the correct electronic folder.
One of our physicians had experimented with two available voice recognition software programs. These required loading the software onto one computer, “training” the computer for at least one hour, dictating into a microphone at that computer, and then editing the document. A higher cost option involved using digital hand-held devices for physician dictation, and downloading the dictation to one computer. Both applications required upgrades to typical desktop configurations.
Further research uncovered a newly available technology that was being used at two other medical centers. As we looked closer at a new system from eScription, we began to see that we could possibly achieve our goals.
The new process
The eScription solution for medical transcription uses speech recognition and Internet technologies to automate and reduce time required to transform medical dictation into medical transcription.
Here’s how it works:
Health care providers pick up the telephone and call the IntelliScript Telephony Server to dictate a medical report. IntelliScript forwards the dictated audio to the EditScript Server, where the dictation and the associated transcripts are managed through their life cycles.
The AutoScript Speech Recognition Server receives the audio and demographic information needed to create a draft transcription of the dictation. The processing is done offline, which enables AutoScript to spend as much time as needed to accurately transcribe the speech.
EditScript Server forwards the draft transcription – along with the synchronized audio and registration information – to a qualified transcriptionist’s personal computer.
The medical transcriptionist reviews and edits the draft transcription using EditScript Client, a suite of tools specially designed for editing automated medical transcriptions.
In our experience, this process is significantly faster and less expensive than typed transcription. The result is accurate and timely medical transcriptions that do not require health care providers to change the way they work. And the transcription companies benefit by using new Internet communication and improved editing software.
The improvements in the transcription process are due to the tight integration of the components:
• IntelliScript, the voice capture unit, records high-quality digital audio designed for use with AutoScript, the speech recognition product.
• AutoScript, the speech recognizer, uses specialized information from language and acoustic models to create a draft transcription of the dictation. Other systems using voice recognition for automating medical transcription have attempted to immediately recognize physicians’ speech. The model we use does not work under this constraint. Unlike other automated medical dictation products, AutoScript does not work in real time and, therefore, has the time required to produce highly accurate transcriptions. Like human medical transcriptionists, eScription’s recognition process listens to the audio multiple times. This enables the speech recognizer to clarify what it thinks a physician said based on a larger context and to apply physician- and organization-specific rules and formatting.
Through this process, AutoScript provides medical transcriptionists with a near-final draft of the dictation. The transcriptionists can then review, edit and send this draft on to the clinician for verification, enabling dictated reports to be turned around quickly and cost-effectively.
From a technical standpoint, AutoScript uses highly accurate statistical models of medical dictation. Billions of words of medical transcription text have been used to develop these models. In addition, the solution supports unobtrusive speaker-dependent and speaker-adaptive acoustic models. The speech recognizer models specific characteristics of the voice (e.g., vocal tract length) without the physician specifically training the system. The training is a by-product of the process. Physicians do not need to “train” the speech recognizer as in traditional systems by reading poems or short stories; AutoScript adapts to the physician without the need for the physician to change his/her workflow.
• EditScript is a client/server transcription and document distribution system. EditScript manages the lifecycle of dictation as it transitions from audio recording to an item in a prioritized work queue to a draft document generated by the automatic speech recognizer, and finally to a transcribed and distributed document. Transcriptionists use EditScript client to retrieve work from the document distribution server’s dynamically adjusted workflow queue. Transcriptionists increase productivity using the power of a specialized Microsoft Word editor to transcribe reports from anywhere over the Internet with only a dial-up modem.
We’ve noted not only improved workflow, but also reduced turnaround time and increased productivity. For example, stat dictations can be made available to medical transcriptionists within 30 minutes. For a stat dictation, this leaves about 1.5 hours for a medical transcriptionist to review, rather than our current four hour availability.
Alternatively, it is possible for draft transcriptions and the synchronized audio to be made available to medical providers immediately.
As of July 2002, 200 primary care and specialty physicians at Brigham and Women’s were using eScription’s products. We are achieving our goals with this pilot project. The documents produced by the new process are extremely accurate. Our costs have been reduced by 30-50 percent per line. And, as the software learns the speech of specific physicians, the cost can be further reduced, if no editing is required. The documents can be made available for viewing and printing in 30 minutes.
In addition, no training has been necessary. Physicians who previously used a tape recorder are now using the telephone. We are experimenting with digital hand-held devices. For those who had been using a telephone dictation system, there has been no change. The system accurately uploads documents through Web-encrypted technology to the correct electronic file.
The practice administrative staff personnel are highly satisfied with the process. They now spend less time looking for and relocating lost notes. And physicians like the improved availability of their notes, the quality of the documents and the satisfaction of decreasing costs for their practices.
We’re also seeing, on average, a doubling of transcriptionist productivity. The average transcriptionist types between 150 and 200 lines of text per hour. Compare that to the average transcriptionist using our new system, who edits between 400 and 500 lines of text per hour.
We have contracted with two local typing services to edit our notes and transmit the documents via eScription software. The expenses include the purchase of a dedicated server, the software and ongoing service. We incurred a one-time startup fee for each physician. We also pay a fee per line for the use of AutoScript software. These expenses were delineated in our contract with the vendor.
We plan to use our new medical transcription solution for other physicians at Brigham and Women’s over the next year. We see significant opportunity to reduce expenses in departments where we do a large amount of dictation. These include the emergency department, radiology and our operating rooms. We expect to decrease our transcription expense by 30-50 percent, while maintaining quality and physician/staff satisfaction. In short, this new technology enables our organization to achieve many of our goals in the area of clinical documentation.
Dr. Goldszer is the associate chief medical officer at Brigham and Women’s Hospital (BWH).
Mr. Chigier is the president of eScription.
Mr. Coughlin is director of revenue operations at BWH.
Ms. Schade is the CIO at BWH.
Ms. Raymond is the director of health information at BWH.