Can you substantiate the claim of many speech recognition (SR) companies that SR with back-end transcription yields cost savings?

Dagan: Today's speech recognition technology should be deployed as one of the tools in the reporting process rather than as the centerpiece. That's not to say that it won't become the focal point eventually. But in a specialty that requires accuracy, timeliness and most of all volume, we've found that speech recognition is best used as a back-end tool in conjunction with other elements, including transcription.

To measure speech recognition's effectiveness, we sampled 200 radiologists using the technology. Almost 24 percent didn't score high enough to qualify for consistent use of speech recognition because of their speech patterns, accents or some other variable. This number rises to 63 percent when accounting for people who are average or borderline in terms of dictation. That's almost two out of every three radiologists who had difficulties with speech recognition. Overall, we found that only 13 percent of our radiologists are excellent candidates for this form of reporting.

On the reporting side, the highest-scoring radiologists processed 38,000 dictations through our back-end speech recognition engine. While almost 95 percent of their reports were translated successfully, 77 percent of draft mode reports required editing-not to mention 1,991 reports that couldn't be processed at all. Speech recognition can actually reduce productivity by as much as 30 percent in cases of long reports or specialty studies.

If front-end speech recognition is the focal point of your reporting, as many speech recognition companies try to sell you, then your radiologists-who represent the highest cost on your ledger-will have to spend significantly more time fixing the problems in the reports they just dictated. Report turnaround has always been paramount for radiology, and the rise of teleradiology has meant that report processing is now measured in terms of time and volume. We've found that back-end speech recognition works best as a tool to produce draft mode reports that are corrected by editors.

Durlach: Background speech recognition can reduce medical transcription (MT) costs and improve clinical documentation turnaround time. Whether the software is managed on-premise or offered on-demand, results can be significant. Findings from a sampling of hospital and medical center customers show:

• The average monthly transcription rate with background speech transcription is 347 lines per hour-98 percent above the industry average of 175 lines per hour, with the top MTs producing at a rate of 493 lines per hour.

• Eighty-five percent of the work produced at customer sites is processed through background speech recognition, eliminating traditional MT processes and costs.

• Eight surveyed facilities reported a combined annual savings of more than $1.9 million in the first year.

• Documentation turnaround time improved, with eight surveyed facilities decreasing the average completion time for

"nonstat" medical reports by 53 percent (from 42 to 20 hours).

With background speech recognition, health care organizations are lowering clinical documentation costs and enhancing the productivity of transcription staff. Customer data show that financial savings begin within 60 days of deployment.

Productivity gains from the new speech system at St. Elizabeth Medical Center in Edgewood, Ky., have allowed St. Elizabeth's overall department to produce at an average rate of 258 lines per hour, and the system has shortened turnaround time for finalized medical reports, says Paula Burchett, operations manager for transcription. St. Elizabeth processes about 3.9 million lines per year, with almost 85 percent going through the speech system.

Just prior to the Dictaphone Enterprise Speech System upgrade, Mercy Memorial Hospital System, Monroe, Mich., experienced an unrelated decrease to its transcription team, dropping from 13 to eight members. With its new system, Mercy increased productivity to 354 lines per hour with speech and is steadily increasing that rate, says Nancy Tanguay, transcription supervisor, health information management at Mercy. The facility has accelerated its report creation process without changing physicians' dictation behavior, and is now at about 95 percent speech utilization, says Tanguay.

Spring: Speech recognition has matured to the point where everyday use is possible in front- and back-end recognition workflows. Both have demonstrated tremendous return on investment (ROI) tied directly to user adoption. Simply stated: The more you use it, the more you save. In some instances, users have front-end systems with editing support; in other arrangements, they have back-end solutions in which the voice file goes directly to an editor. Generally, the radiology market has adopted front-end workflow with transcription backup, although that doesn't preclude use of a handset to dictate a voice file.

Front-end savings come from the elimination of transcription costs when a physician opts not to have normal reports and templates reviewed because the speech recognition is so good. They also come in the form of time and workflow efficiency from having to only edit a document versus transcribe it from scratch-reducing the clerical burden, the need for transcriptionists, re-reads, etc.

The typical user with whom I speak has gone from about $4 a report to around $1 a report after deploying speech recognition. A 2007 study showed marked savings and productivity improvements with the technology. In six months, participating radiologists went from completing 7 percent of their reports in an hour to 82 percent, and their report errors dropped from traditional transcription. The icing on the cake was a 28 percent improvement in productivity (measured in relative value units [RVUs] and hours worked, as a baseline).¹The bottom line: Speech recognition is having a strong and lasting effect on report cost and patient care.

Reference

1. Floyd, J. (2007). Radiology reporting with speech recognition: Error rates, accuracy and turn around time. (Radiology Consultants of Iowa). Radiological Society of North America (RSNA), Chicago.

Jill Hoffman is senior associate editor of ADVANCE.

Participants

Yaniv Dagan is CEO of Integrated Document Solutions Inc. (IDS), Fort Lauderdale, Fla.

Peter Durlach is senior vice president of marketing and product strategy for the health care division at Nuance Communication Inc., Burlington, Mass.

Christopher Springis director of product management for speech recognition solutions at MedQuist Inc., Buckeye, Ariz.