Speech Recognition Technology Gives Voice to Clinical Data

Bradd envisions speech recognition so tightly integrated with other software that a physician could verbally command the retrieval of specific patient data on demand. Picking up a microphone attached to the computer, Bradd wants the ability to say, "Run an inquiry on all patients with a diagnosis of diabetes and hemoglobin A1C over 8." After the computer displays said list of patients, Bradd would further narrow his search: "Now, computer, give me a list of these patients with a systolic blood pressure over 140."

On the mark

Some speech recognition vendor executives, consultants and provider organization CIOs think Bradd's vision is right on the mark. They believe the technology soon will evolve from its primary use of speeding the physician dictation and transcription process. What's more, the technology

Bradd desires is available today. But it may take 10 years or more until speech recognition is widely accepted in health care, experts say.

"It will take five years for computer-based patient records software vendors to adequately embed this technology into their systems, and another five years to get the price down into the range of the average physician," says Barry Hieb, M.D., research director at Gartner Inc., a Stamford, Conn.-based research firm.

Speech recognition technology could play an important supporting role in health care's quest to cut administrative expenses and improve the quality of care.

Primary data entry methods today include punching buttons on a keyboard, highlighting choices with a light pen or writing onscreen with handwriting recognition technology. However, enabling physicians to verbally enter patient data could help provider organizations migrate more quickly to fully automated clinical systems, says Robert Schwager, chairman of Dictaphone Corp., a Stratford, Conn.-based vendor of dictation, transcription and speech recognition systems.

The technology behind speech recognition applications is better than ever; many vendors' products claim accuracy rates around 98%. But until accuracy rates reach 99.9%, hospitals still will need a supporting staff of transcriptionists to edit medical text, contends Alex Rodriguez, chief technical officer at six-hospital Health Alliance of Greater Cincinnati.

Vendors will meet the goal of virtually 100% accuracy, but other technical issues remain, says William Meisel, editor of the monthly newsletter Speech Recognition Update and president of TMA Associates, a Tarzana, Calif.-based consulting firm. "Speech technology is not the obstacle," he adds. "The obstacles are systems integration, and enabling consistent and enterprisewide input of all patient data into information systems."

Some integration of the spoken word and electronic medical records already is taking place, but at a basic level. Bradd, for instance, uses the Dragon speech software from ScanSoft Inc., Peabody, Mass. Dragon is integrated into his practice management and electronic records software from Physician Micro Systems Inc., Seattle.

Bradd dictates notes using a microphone plugged into his computer, views pertinent data onscreen, then pastes it into a patient's electronic medical record.

However, the barriers to widespread adoption of such use of speech technology are daunting. The biggest barrier could be getting physicians enterprisewide to use the technology to enter patient data into information systems.

The primary physician markets today for speech recognition are radiology, pathology and cardiology. Because these specialists interpret X-rays and tests, they spend much of their time dictating reports for referring physicians. And they use specialty-specific vocabularies that allow programmers to orient speech recognition systems to each unique medical field, consequently boosting the technology's accuracy rates.

Will physicians bite?

Some industry observers question if general physicians will in large numbers adopt speech recognition technology for dictation and self-editing of documents.

"Physician time is money, and the quickest way still is to pick up the phone and dictate," says Dennis Mahoney, president of Lanier Healthcare, the speech recognition technology division of MedQuist Inc., Marlton, N.J. "Speech recognition has to beat that time."

Hospitals and group practices will look for ways to get physicians to dictate and self-edit their transcribed documents to pare that labor cost, predicts Schwager of Dictaphone. He estimates the health care industry spends at least $6 billion on transcription labor each year, not counting overhead. "If the doctor is doing the editing, the transcription labor is now zero," Schwager says.

To get there, provider organizations must show physicians clear value to using speech recognition-value beyond administrative savings.

"Doctors will complain it will take them more time to do their job, and they are right," says Brad Block, CIO at Doylestown (Pa.) Hospital. Its radiology department uses speech technology from Agfa HealthCare Informatics, Waterloo, Ontario.

"But for us, value is clearly demonstrated in the quick turnaround time by the radiology department. Our radiologists aren't happy about the extra work, but they know they're pleasing every other doctor in the hospital with faster reports."

Showing value

Advances in technology, already here or in development, will show physicians the clinical value of using speech recognition, Hieb contends.

For instance, natural language processing involves the automated extraction of the meaning of text.

Using extensible markup language, or XML, the technology finds the facts in a dictated text-such as a diagnosis, medication or dosage-and automatically places that information in the patient's electronic medical record.

Such technology can help build a concise but useful medical record, offering physicians quick access to pertinent patient data.

Further, clinical decision support software embedded in a speech recognition system can immediately prompt a physician during the dictation process to add data or reconsider treatment decisions. The decision support mechanism, for instance, could inform a physician that a current diagnosis is not in a dictated note.

Or, when a physician gives a diagnosis, the software could display treatment options on the computer screen.