Voice recognition has been used in medicine for more than 30 years, but recent advances in the technology’s accuracy have been key in cutting hospitals’ costs and saving physicians’ time, according to a study by Arun Krishnaraj, MD, and colleagues in the July 2010 issue of the American Journal of Roentgenology. These benefits are only increasing as voice recognition becomes more adaptive and integrated with hospital information systems, such as EMRs and PACS, and even to shave time off of administrative duties. But user experiences vary considerably.
To get the most from your system, keep these five points in mind:
1. Training—seek it and give it.
“Voice recognition does not come naturally—but you can significantly improve your performance with training,” says Howard M. Landa, MD, CMIO at Alameda County Medical Center in Oakland, Calif. Different products have different features, which can be both efficient and, at times, finicky; so the “extra money you put into training has a significant return on investment,” Landa says. Just as critical, because voice recognition software picks up your vocabulary and adapts to your style of speaking, correcting errors as they happen means the software will be less likely to make those mistakes in the future. “It’s almost like a puppy: If you don’t train it early, it becomes a lot harder to fix later,” quips John Kaschko, MD, medical director for informatics at Group Health Cooperative in Seattle.
2. Microphone placement—find the sweet spot.
“There’s definitely a sweet spot,” says John Umekubo, MD, medical director for clinical informatics at St. Mary’s Medical Center in San Francisco. There are two principal ways to locate the optimal microphone position: trial and error (with some training), or use a headset. Wireless headsets offer the added convenience of mobility without added interference, which is especially popular among ER physicians who are constantly on the move between their own computers, the central workstation and patients, according to Landa and Kaschko.
3. Context counts—speak naturally and consistently.
“Speak as naturally as possible,” counsels Gwendolyn O’Keefe, MD, CMIO, of Group Health Cooperative. Voice recognition is now advanced enough to handle most accents, so long as you’re consistent and let the system learn your ways, O’Keefe says. Moreover, “it’s one of the rare times where increasing complexity actually enables the system to work better,” Landa adds. The software guesses what you’re saying based on contextual probabilities, so speaking in phrases, using polysyllabic words and expressing complete concepts, all at a constant pace and without pause, give the software more information—a context—for deciphering your reports, according to Landa. In addition, don’t spell out difficult words and names: if you say the name and correct the software, Umekubo insists, it will learn.
4. Review reports—your system can’t.
“To me, the biggest issue—and this is critical for the quality and accuracy of your notes—is to proofread your reports,” O’Keefe says. Voice recognition does not misspell dictionary words, but early on it will often mistake what you intend to say with similar-sounding words. What’s more, as Kaschko points out, your brain recognizes misspellings much better than it does correctly spelled words placed in the wrong context, so careful rereading is key.
5. Maximize functionality—get a system that will speak to your EMR.
“To get the most accurate, quickest and best documentation, you have to find a voice recognition program that interfaces with your EMR,” Kaschko says. Selecting a program that integrates with your vendor’s EMR enables you to pull up and insert patient histories, charts, discharge summaries and meds with one-word commands. Furthermore, voice recognition can be integrated into multiple databases to recognize some of the more difficult patient and physician names. “These functions make physician notes legible instantaneously—I will never go back to handwriting,” Umekubo says.