Voice recognition technology has made huge strides in the past decade, enabling overnight shippers to track packages and reporters to write their stories simply by speaking into a microphone.
In doctors' offices, however, speech technology has made only modest inroads in the estimated $10.5 billion a year medical transcription business.
One manufacturer of the new wave of all-in-one, computerized physician office work flow devices says the time is right for voice recognition to fulfill its potential, allowing a physician to dictate clinical notes into one end of a machine and a printer to produce perfectly transcribed medical records at the other end.
Yet concerns about the inaccuracy of speech technology remain a barrier for general medical applications.
Physicians in some specialties, particularly radiology, have been using voice recognition software for years and have dramatically improved their work flow.
Radiologist Joseph Collins, M.D., of the Mayo Clinic in Scottsdale, Ariz., says he is one of the beneficiaries of voice recognition technology. Collins has worked with an IBM-based system for more than six years. The system cut turnaround time on transcription of physician reads of X-rays and scans from an already fast 3 hours to 10 minutes, he says.
"I start dictating and I don't even look at the text coming up on the screen," Collins says. "My eyes are focused on the X-ray or the CT. When I'm done interpreting the images, I turn and look at the text."
Collins says it takes just a few seconds to edit the report for an X-ray and a minute or two to go over the computer transcription of a more complicated CT scan, but neither take much longer than the time he used to spend reading the returned work of his transcriptionist.
"I haven't seen anybody claim it's saving us any money," Collins says, adding it would cost between $100,000 and $200,000 to install a similar system today in a radiology department with 10 to 15 doctors. "What it does is improve services."
Collins says he spends more time improving the clarity of his own words than he does correcting the occasional mistake made by the transcription software.
For radiology and pathology, "the complexity of the vocabulary is such that it fits very well with speech recognition," he says.
Bill Hawkins, senior vice president of corporate development for MD Productivity of Austin, Texas, says his privately held company spent 2 1/2 years tweaking a speech translator from Dragon Systems, one of the pioneers in voice recognition. The MD Dictate system is now ready for use by general practitioners.
MD Dictate will translate up to 180 words a minute with 97% to 99% accuracy after a physician has trained the device to recognize his or her speech patterns. The calibration period takes an hour a day for two weeks.
If there are doubters, Hawkins says, "it's because they've not seen our product."
Count Charlie Koo among the doubters. Koo is CEO of iMedica, a Mountain View, Calif.-based company that makes an all-in-one system, which combines practice management systems and electronic medical records.
He is one of several manufacturers who say they plan to leapfrog the need for dictation entirely by designing systems that create electronic medical records through portable computers.
Koo says he looked hard at voice recognition because his physician customers said they wanted it as an alternative to dictation. He can add voice recognition to his device on request but rejected making it standard equipment.
"It didn't work," Koo says. "If you have a quiet room, it's OK to about 95% accuracy." But if a physician tried to make a recording in a noisy hospital or clinic, the accuracy plummets. "As it is now, it's not good enough."
To work around the problem, iMedica and other manufacturers say they have designed portable, touch-screen computer charting devices that guide physicians in creating electronic patient records so complete they all but eliminate the need for dictation and transcription.
These systems have several pricing models, ranging from monthly fees to fixed prices that range up to $7,500 per unit.
For the patient whose circumstances don't fit the preprogrammed charts--Koo says that's less than 1%--he and other makers have added on-board voice recorders. With them, the attending physician can either attach a digital voice clip directly to the patient's electronic medical record or send the clip off via the Internet to a traditional transcription service.
Carolyn Kopenhaver is vice president of marketing for Marietta, Ga.-based JMJ Technologies, maker of EncounterPRO, a notebook-based work flow system. She says her company's device will interface with voice recognition software, but there is little need to use either that or a transcription service.
"We've found we can eliminate them with structured data entry," Kopenhaver says. "The pediatricians I know have all given up on transcriptionists."
Jeffrey Cooper, M.D., a pediatrician in Norcross, Ga., says he's used the system for five years and has abandoned transcriptions entirely. "It's revolutionized my practice," Cooper says.
Steven Schramm, M.D., of Atlanta says he has been beta testing an EncounterPRO version for family practitioners since December. Schramm says he hasn't bothered to use the voice recognition software available with the system and uses dictation for about 1% of his patients. Schramm says when the other physician he works with stops using dictation and switches to EncounterPRO, they expect to save about $30,000 a year in transcription costs.
"That pretty much pays for it," Schramm says.
Frank Rhie, M.D., co-founder and chief medical officer of Irvine, Calif.-based Alteer, a manufacturer of EMR devices, says voice recognition is "not ready for prime time" because it still is susceptible to producing "weird errors in weird places."
Rhie recalls witnessing a demonstration in which a user dictated "a skin abrasion to the forehead" into the machine only to find the wound changing in translation to an abrasion of the foreskin.
Still, Rhie says, his Pocket Alteer work flow management system offers dictation and speech recognition because some patient cases are too complicated to be handled by a preprogrammed chart.
Positioned between traditional transcription services and the new wave of all-in-one devices are companies like SpeechMachines of Alpharetta, Ga. Its DictationNet system uses both transcription and speech recognition technologies.
President and CEO Michael von Gray says so far the barrier to entry for speech technology has been the ability of an experienced transcriptionist to beat any computer-based voice recognition system in terms of speed, accuracy, convenience and cost.
"It can't get a lot more efficient for that doctor," von Gray says. "He picks up the phone and rattles away for five minutes at 90 miles per hour and (he's) done."
With a typical voice recognition program "he would have to stop every eight words and go back."
So even though it takes a good transcriptionist four to five minutes to type out a single minute of a physician's recording, the difference in pay grades between the doctor and the transcriptionist makes the service economically viable, von Gray says.
The DictationNet system is being marketed not to doctors but to transcription companies that network with upwards of 300,000 transcriptionists, von Gray says.
The DictationNet model provides transcriptionists with a browser-based Internet connection, the software on which they will work and the stream of doctors' recorded voices they will transcribe.
As they stream in, the physicians' voices are run through DictationNet speech recognition software configured with a medical dictionary.
Transcriptionists have two choices: They can use the computer-generated word flow as it streams in, adding or editing only those words the program misses, or they can ignore the computerized stream and transcribe in the traditional manner.
The choice is up to the transcriptionist and will depend on how well the physician's speech patterns can be translated by the voice recognition program.
"We project that on average we'll be able to get 50% of doctors to generate a level of accuracy to get some form of economic value," he says. That economic threshold is about 85% accuracy.
Jay Vance, director of the American Medical Transcription Guild in Yuma, Ariz., says he is aware of the inroads that have been made in radiology, but for now, for most medical applications, the technology is not able to replace a good transcriptionist.
"It's definitely not as fast and under the best circumstances it's not at the state where you could set it up and have a perfect product come out of the printer," Vance says. "It's not even close."