Arthur Talansky, M.D., a self-described "techno guy," talks to his computer.
Like most doctors, the Long Island gastroenterologist used to dictate patient records and medical correspondence into a tape recorder and send out the tapes for transcription. Now he saves time by dictating documents directly to his computer during or immediately after an examination and editing them on-screen.
"If I need a letter, I can have it dictated, typed and enveloped before a patient leaves my office," he says. What's more, he adds, when doctors complete reports while patient information is fresh in their memories, better patient care results.
The technology Talansky uses is known as continuous voice recognition. His particular product is made by Dragon Systems of Newton, Mass., a leader in the field. Several similar products are being marketed specifically to healthcare, most introduced within the past year. The same firms also sell products specific to other professions, such as law, and for general use.
In lieu of voice recognition, physicians send tape-recorded patient reports to a medical transcription service, which types the document and returns it for review before completing a final report. That can take 48 hours or more. Some physicians handwrite patient reports, which can result in errors if they forget information before they have time to prepare documents or if billing coders can't read their writing.
Voice recognition could change all that. Its widespread use would eliminate problems associated with illegible handwriting. It also could replace transcription services altogether or turn a medical transcriber into an editor who proofreads computers' work.
Companies believe they have a hot product -- especially in healthcare, where pressures to reduce costs are extreme. As a result, many companies made healthcare their first priority when designing programs for specific uses, and they claim it is the field with the fastest-growing sales.
Voice recognition programs have advantages other than reducing the need for medical transcription services. For example, programs can prompt doctors to provide specific information needed to obtain reimbursement. And in the long run, getting physicians to speak to computers might be the easiest way to get medical data into information systems, particularly because some doctors are uncomfortable with computers or don't know how to type.
Voice recognition software with a medical dictionary costs about $1,000. Computers sophisticated enough for the program now sell for about $1,600 apiece, making the total price about $2,600. Such programs typically would be used in the office by an individual physician.
More complicated systems are available to connect entire hospital departments, such as radiology and pathology, and can be customized to work with a healthcare organization's existing information systems. The higher-end networkable products of Lernout & Hauspie, another leader in the field, sell for $5,000 to $8,000.
That's a far cry from the $200,000 spent annually on transcription services by an individual department, such as radiology, in an average-sized hospital, according to Todd Kuoni, a regional manager at Dragon.
In all, the potential annual value of the U.S. medical transcription market is thought to be $6 billion. Currently only about $1 billion worth of transcription occurs yearly; the rest of the work is handwritten or typed. Of the $1 billion spent on transcription services, about 5% is committed to voice recognition systems, estimates Richard Oldach, director of product management for healthcare solutions at Lernout & Hauspie.
The Burlington, Mass.-based company is one of three major players in the voice recognition market. Dragon and IBM are the others, says Hans Von Braun, an analyst in the San Francisco office of Creative Strategies, which tracks small industries. Several firms incorporate those companies' technologies into their own products.
Right now, voice recognition primarily is seen as a way to save time and reduce errors in writing reports. But the future payoff likely will be the technology's ability to facilitate the creation of electronic medical records. As a result, current products are designed in accordance with industry standards so they can interface with a healthcare organization's information systems.
In all voice recognition systems, a physician speaks into a microphone attached to his computer or to a digital recorder. His speech is broken down into sound bites, called phonemes, which are compared with a dictionary of phonetic spellings. All interpretations of the phonemes are rated according to probability, based on acoustics, vocabulary and grammar. The most likely appears on-screen almost as the physician dictates. If necessary, he or she can tell the program to correct a word or phrase.
Companies market individual workstations, which generate immediate reports but must be used in a specific location; portable systems, which later upload digital recordings for transcription; and higher-end networks for hospital departments.
Voice recognition programs were introduced in the early 1980s. Early programs required discrete dictation, meaning users had to pause after each word.
Needless to say, most users found them awkward. Today faster computers allow continuous dictation. Since the introduction of the first continuous dictation products for healthcare in late 1997, sales have grown rapidly.
For example, Oldach says Lernout & Hauspie has sold roughly 50 copies of its Voice Xpress for Medicine program in six months on the market. The company offers applications for several specialties, such as pathology, radiology and emergency medicine, as well as general medicine.
Fonix Corp., which uses the Dragon voice recognition engine, has signed up 42 hospitals for the radiology product it introduced last winter, says John Oberteuffer, vice president for technology. The Draper, Utah-based company is testing a product for emergency departments at three hospitals, Oberteuffer says.
Von Braun recommends that doctors use PCs with at least 450 MHz capacity for voice recognition systems. While faster is better, companies say slower computers with 200 MHz capacity also can accommodate their products.
Von Braun, a former computer teacher who never learned to type, says he uses voice recognition programs from each of the three major players almost daily for reports and correspondence. That's partly because he follows the field but also because he likes the systems, which have reduced his errors.
Typically it requires one hour or more to train the program to recognize voice intonations and speech patterns, Von Braun says. After about two weeks, the program should be 98% to 99% accurate, because it learns as the user corrects errors, he says.
For example, Talansky says when he started using the Dragon system, it transcribed the phrase "more formed stools" into "more formal stools" in a letter to a colleague. Thankfully, he caught the error before mailing the letter. He says such mistakes are increasingly rare, and he completes his paperwork in much less time.
The biggest issue in adopting a voice recognition system no longer is the technology, but physicians' comfort with it. Talansky says his three partners responded with varying enthusiasm when he offered to teach them to use the product. One physician isn't interested, another is leaning toward it, and the third is using the technology daily.
"This is not a package where you can just buy it and use it," Talansky says. "You need someone to hold your hand, to teach you, to train you."
Training provided by companies varies. Lernout & Hauspie, for example, says most of its resellers offer one to two hours of training for about $75 total. For its higher-end networkable products, the company sends out teams to provide training to end users and hospital information system experts. It charges $1,500 a day for the service.
The systems have limitations. For example, dictation must be fairly precise, and the system microphone must be placed in much the same position (close to the speaker's mouth) each time it's used to minimize interference.
Still, Talansky says the more he uses voice recognition, the more he believes it is for "doctors who don't love technology," as well as those who do, because it can significantly speed up their practices.
Despite being a techno guy, Talansky doesn't take advantage of what may be voice recognition's biggest selling point: easy data entry for an electronic patient record. Right now, he dictates patient records, prints out a hard copy for his files and eliminates the electronic version. "I believe this has got no way to go but up," he says. "It's not only sort of a fun thing, but it's helpful to patient care."
Lisa Scott is a Chicago-based freelance writer who specializes in healthcare business topics.