Philip Bachman, MD, is an "early adopter" when it comes to new technology. The infectious disease specialist was among the first physicians, five or six years ago, to try speech recognition software to handle his dictation.
Even though the technology was far from perfect, Bachman reasoned that it would still be better than what he'd been doing: using a telephone-based transcription service for a while, and at another point dictating on tapes that a service would pick up, transcribe, and return a few days later.
Both of those methods were expensive, the Atlanta physician says, and it took too long to get dictation back. Besides, while it was true that the software he bought, Dragon Naturally Speaking, one of the well-known brands in speech recognition, made errors that Bachman would have to correct manually, the human transcriptionists were hardly perfect themselves. Even with experienced medical transcriptionists, Bachman says he'd still have to personally edit the documents — and often correct mistakes.
"If we'd kept using the service, it would have cost me personally $10,000 to $14,000 a year," he says. "And it takes anywhere from two to three days to get the transcription back. With speech recognition, I dictate it when I see the patient; it's on the chart that day, and I can fax a copy to a referring doctor that day. Also, if I'm out of the office the next day, my partner knows exactly what I was doing."
Speech recognition software is catching on among physicians. It's unclear how many U.S. doctors use the technology, but it's a fast-growing industry, and it appears that most professionals who use it are in healthcare. In recent years, the speech recognition software industry has had $300 million in annual sales, and is growing at a 12 percent annual clip, according to The Gartner Group, an industry research firm. And 60 percent of speech recognition sales are in healthcare.
That data indicate speech recognition products are a rarity among information technology tools: they are available to many industries yet are being adopted most aggressively by often tech-wary physicians.
According to a 2004 survey by the Healthcare Information and Management Systems Society, 28 percent of physicians said they were already using it, and another 52 percent planned to add it in the following two years. That survey may have overrepresented tech-friendly doctors such as Bachman, but other experts suggest that perhaps 25 percent or a third of physicians have begun using the technology.
Working better
It's easy to see why, even among doctors leery of change, speech recognition is popular, says Harry Rhodes, MBA, director of products and services for the American Health Information Management Association (AHIMA). As technology goes it's not very expensive, making it a relatively safe experiment for practices. It has obvious advantages over traditional medical transcription services, which rake in $10 billion annually in North America, according to the Medical Transcription Industry Alliance.
"I can't tell you that speech recognition will necessarily save you time over conventional transcription services, but it will definitely save you money," says Eric Fishman, an orthopedic surgeon who now runs 21st Century Eloquence, a Florida-based reseller of Dragon Naturally Speaking, who services Bachman's system.
And perhaps most important, speech recognition software has increased in accuracy and ease-of-use, and the computer processors needed to run it are getting faster as well.
"The technology is a whole lot better than it was," explains Rhodes. "There are still some issues, but ... it's much better than it was a few years ago. It's kind of like, one day you notice a guy on the street with a cell phone. The next day everyone has a cell phone."
Bachman, who works in a two-physician practice, says he's never regretted making the switch — especially now that the technology has improved to about 98 percent accuracy. Indeed, to the extent the system makes errors at all, he generally blames himself, saying most of its mistakes occur because he isn't speaking clearly enough.
"It's about as good as I am," he says. "If I'm tired or rushed, it doesn't do as well. If I do it slowly and enunciate, it does quite well."
How it works
Speech recognition software doesn't understand English, per se. Rather, it captures the speaker's utterances and vocal inflections, and places them in context with one another. Then it chooses a word that the speaker appears to be saying from a list of likely suspects. For example, if the speaker says, "Mary went to the store," but mumbles the last word, the software will fill in the blank by choosing from a list of words it knows, based on what it heard the speaker say and what makes sense in context with the rest of the sentence.
That process requires quick "thinking" and a vast vocabulary. The former is now possible thanks to more sophisticated computer processors, and the latter is due to years of software development. "It's still not everything to everyone," says Rosemarie Nelson, a senior consultant for the Medical Group Management Association (MGMA), "but it's gotten more useful for a lot of people."
