January 28, 2008

Sound of Progress
By Mary Anne Gates
Radiology Today
Vol. 9 No. 2 P. 24

How speech recognition technology is making a difference at these healthcare organizations, including back-end transcription in radiology.

Speech recognition system implementation in hospitals and other healthcare facilities is becoming more widespread. As more facilities turn to the technology, medical transcriptionists are being asked to learn and use editing skills to complement their transcription expertise.

Editing a document includes monitoring for correct grammar and language usage, as well as adding punctuation. Various headings or subheadings may also need to be inserted.

As older dictation systems fail or are unable to meet a user’s needs, adoption of newer technology is driven by the promise of achieving faster turnaround times and reducing transcription costs. Some departments have been so successful in achieving these twin goals that they are able to eliminate the costs associated with their department and have even begun generating revenue by offering transcription services to other physicians and healthcare facilities.

For example, Mercy Health Center in Oklahoma City went from a transcription department with excessive expenditures associated with many hours of overtime, as well as monthly outsourcing costs anywhere between $20,000 and $60,000, to its current state of actually generating revenue after implementing a new dictation system in 2004, which included speech recognition.

“We went from a department with actual costs of estimated 18 cents per line to our current state of between 11 to 12 cents per line, which has made a significant impact on the transcription cost center’s contribution margin to the health center,” says Health Information Management Director Cheryl Doudican, RHIA. “It took two years to reach that point, but once you get there, you see the great benefits. We have had very limited to no overtime for our MTs [medical transcriptionists], as well as no outsourcing costs at all since September of 2006.”

Similarly, Poudre Valley Health Systems in Fort Collins, Colo., implemented eScription speech recognition technology in February 2006 and is now considering ways to generate a positive cash flow.

“We are exploring the possibility of doing transcription for various doctors and other entities, changing from an expense to the hospital to becoming a revenue-generating department,” says Bonnie Barlow, an assistant supervisor in the transcription department.

Editing and the Human Touch
Current speech recognition software often captures much of what a dictator says. Word capture estimates range from 20% to nearly 100%, says Doudican.

Despite the ability of some speech recognition programs to accurately capture nearly all spoken dictation, producing an error-free document continues to require human intervention for a final product. “Transcriptionists still listen to the whole dictation, putting in every word the system did not recognize. They follow along and insert missing or incorrect words,” says Doudican.

Besides learning a different process with additional language and punctuation aspects, a higher degree of concentration is needed during the editing process, according to Amy Bowman, RHIA, director of health information services at Jordan Hospital in Plymouth, Mass.

“We have noted that the role of the editor at times requires more concentration than the role of the transcriptionist. The MTs feel that they tend to be on autopilot when transcribing, but as an editor, [they] really need to concentrate on correct punctuation, language use, grammar, and overall quality of the final report,” she says.

Choosing Speech Recognition Features
The management at Poudre Valley determined several important elements were necessary in selecting the right system to meet their needs—digital voice capture; conversion of speech to text, including user-definable formatting templates; and a two-cursor system for speedy editing of textual transcription documents. Additionally, interfacing to the hospital’s electronic health record for record storage and e-signature was also necessary, as were client and Web-based tools for editing draft transcription documents, Barlow says.

Switching to Speech Recognition
The decision to implement speech recognition often has to do with the overall cost of transcription services, high turnaround times for completed reports, and the inherent limits of an existing system using more traditional transcription methods.

For example, a voicewriter server in use for nearly 10 years at Jordan Hospital was reaching the end of its useful life, says Bowman. “We were experiencing more and more unplanned downtime and service calls,” she says.

Additionally, other constraints further prompted the need to make changes. “We were also faced with a limited number of remote access ports to the system. This limitation did not allow us to complement our in-house staff with enough contract help to maintain our desired report turnaround time,” Bowman says.

Consequently, she headed a team of transcriptionists and IT staff to find a system that would better serve Jordan’s needs. In May, they chose MedQuist DEP, a Web-based product. “From the start, we knew speech recognition needed to be a part of our solution in order to enable current MTs to become more productive, functioning as an editor rather than transcribing all reports,” she says.

The system Bowman and her team chose required transcriptionists to learn a new transcription platform and editing skills. The education process began with employees mastering the new platform after a pair of two-hour training sessions, says Bowman.

Approximately three months later, transcriptionists began using the system’s speech recognition component. This time, a two-hour Web-based training session brought them up to speed. “We opted to begin with the medical oncology reports as we had not been able to reach our turnaround time goal of three days for these reports,” Bowman says.

Overall, Bowman likes what she sees thus far from the new system. “We are currently exceeding our turnaround time goals, which are equal to or exceed industry standard benchmarks. After two months of going live with speech recognition, we are able to produce an additional 14,000 lines per month and have completely eliminated any backlog,” she says.

Besides increasing productivity, Bowman has also reduced costs in a variety of ways. “We have been able to eliminate all contracted services, per diem hours, and overtime,” she says.

Conquering Fear of Change
Apprehension about switching to speech recognition was largely due to a fear of the unknown, says Doudican. She says transcriptionists feared their jobs would be eliminated and were cautious about veering away from traditional transcribing.

Similarly, Poudre Valley transcriptionists were resistant to change. Embracing the new system was easier as transcriptionists learned they could build their own vocabulary with an autocorrect feature. Additionally, less typing made the new system easier on their arms, hands, and fingers, according to Barlow.

Learning to use speech recognition depends greatly on the type of system and its native environment. For example, back-end, or passive, speech recognition, such as eScription or MedQuist DEP, is often seen as a seamless change in technology because doctors continue to dictate in much the same manner as before. The speech recognition software captures the dictation with an MT transcribing anything the software did not “hear” correctly and editing the final document version.

“The beauty of the system [back-end speech editing] is they [the doctors] don’t even know it is happening. They don’t necessarily have to learn anything new because it is all happening in the background,” says Doudican.

Conversely, front-end, or active, speech recognition, such as Dragon Naturally Speaking, is more noticeable to the dictator, who will also edit the document. “There is a slightly different speaking pattern to use one of these,” says Sheila Sund, MD, administrative medical director at Willamette Valley Hospice in Salem, Ore. “You have to speak slower, speak clearly, and enunciate properly. It’s not horrible, just a little slower than regular speech.” For example, a paragraph that normally took 30 to 40 seconds to dictate now takes approximately one minute, she says.

Front-end speech recognition is increasingly used in radiology. It can greatly speed radiology report turnaround because the doctors do the editing and fits well with imaging’s transcription needs. Radiology’s relatively limited vocabulary and high rate of normal reports allows rapid turnaround without burdening the physician with an excessive amount of editing. The radiologist does more editing, but often sees the benefit of report turnaround and improved service as worthwhile.

Sund, who uses the medical version of Dragon Naturally Speaking to dictate and edit her hospice care reports, didn’t take long to feel comfortable with the technology. “The learning curve took two to four weeks. The biggest part of the learning curve is you have to add punctuation. With a human, you don’t have to do that,” she says.

For example, commas, periods, and quotation marks have to be added to the document, a chore which is usually accomplished by speaking the commands. A misspeak or a wrong word in the document requires the use of the keyboard to fix, explains Sund.

Decreasing Turnaround Times
By making the move to speech recognition, Poudre Valley realized a 93% reduction in turnaround times, says Barlow. “Discharge summaries in 2005 were at 30 days turnaround time. By January 2006, they were at approximately 23 days. They decreased to less than two days by September 2006,” she says.

Turnaround times for history and physicals, emergency department reports, and other items requiring transcription went from three to eight hours to less than two hours by August 2006, says Barlow.

Turnaround times have also decreased substantially at Willamette Valley Hospice. Previously, it took roughly one week for a report to go from dictation to transcription, edit, and review. Additionally, a copy was sent to the physician for a signature before finally becoming part of a patient’s medical record.

Now, a report is dictated, edited, and signed the same day and immediately becomes part of the electronic medical record, says Sund.

Cost Savings
Implementing speech recognition can reduce the cost per line of transcription, thus reducing the overall costs associated with transcription services.

Doudican calculates transcription costs per line by applying the following formula:

Cost per line generated = Total lines transcribed/Transcription expenses

An example from a recent pay period shows that the total lines transcribed were 289,929.4. Divide that number by the transcription center costs ($36,699.03), and it reveals the cost per line of transcription to be 12.6 cents.

Besides reducing overall transcription costs, overtime, expenses associated with outsourcing, and per diem expenditures, Poudre Valley has increased income by improving turnaround times. More interest is earned on the income because the quicker turnaround times have translated into faster billing by physicians and the hospital, resulting in receipts coming in sooner, says Barlow.

Perhaps an even bigger savings came from decreasing overtime. “There was an 85.7% reduction in overtime pay, saving $300,000 per year,” Barlow says.

Besides reducing overtime, outsourcing has also been cut back. “We’re able to take back a significant percentage of outsourced transcription from the Family Medicine Center and Mountain Crest [Behavioral Healthcare Center], saving further costs,” she says.

Including Radiology
Finally, more savings were realized when Poudre Valley took the radiology transcription from a separate department and integrated it into the transcription department’s workflow. This decision better utilized the transcriptionists’ skills to cover all work types, says Barlow.

“We wanted to get critical medical information into the hands of the providers in real time, reduce expenses, improve physician and transcriptionist satisfaction, and achieve greater transcriptionist retention, specifically telecommuting,” Barlow says of Poudre Valley’s goals for implementing speech recognition.

At Mercy Health Center, reducing costs was a central component in the decision to make changes. “We were outsourcing thousands of dollars per month, as well as [working] many hours of overtime, and we still couldn’t keep up with the volume,” says Doudican. “My goal was to eliminate overtime and eliminate outsourcing by changing our transcription platform completely to include the use of speech recognition.”

Enhancing Productivity
Besides enabling transcriptionists to more quickly produce an accurate and complete document, speech recognition systems can keep track of the shortcuts that help increase productivity.

“Through our reporting system, we are able to monitor what tools in EditScript gave transcriptionists the greatest benefit to transcriptionist productivity,” Barlow says. “We have discovered transcriptionists who use the shortcut tools can greatly increase the average number of lines edited per hour as well as their editing accuracy. Editing through these shortcuts is seamless and allows you to do more work in less time.”

— Mary Anne Gates is a medical writer based in the Chicago area.