Accurate speech recognition: mHealth's Holy Grail for docs

mHealth penetration of healthcare isn't a done deal. iPads and smartphones may seem ubiquitous--and there are plenty of them in circulation in hospitals today--but the truth is most hospital electronic medical records are still evolving mobile capability, Wi-Fi access is still sketchy at many facilities, and mobile security/enterprise management systems are still in their infancy.

Those are all factors CIOs are rightly focusing on, to meet physicians' growing appetite for mobility, and drive toward meaningful use and other regulatory IT goals. But there is another factor that's getting short shrift in all this, and it's a central reason healthcare providers aren't using mobile technologies to their fullest potential--speech recognition.

Doctors constantly tell me how much they love their iPhones and Android tablets, but they also complain about the difficulty of data input. Few find the touchscreen keyboard handy for inputting notes or updating their comments about patient progress. Most use the devices as data viewers.

To make the devices--tablets and smartphones alike--truly two-way tools, CIOs must uncover a speech recognition module that works, or at least push vendors to create one.

What brought the topic to mind for me was a recent announcement by mobile vendor MedMaster Mobile, and speech recognition developer Nuance Communications, Inc. Together they've added Nuance's speech recognition function to MedMaster's iPad-compatible EMR. I spoke with MedMaster CEO Dennis Carson, as well as Ed Matthews, CFO/CIO of Victor Valley Community Hospital, Victorville, Calif., one of the product's superusers, about the broad challenges of speech recognition.

A few factors you can ask vendors about to make speech recognition work in your mobile devices include:

  • Accuracy: Speech recognition has been on a long, slow move on this front for years, and still has a ways to go, Denis Baker, CIO for Sarasota (Fla.) Memorial Hospital says. Vendors should have clear metrics for accuracy--85 percent, 90 percent, etc.
    Be warned, though: Vendors may claim 99 percent accuracy, but the Gartner "2011 Hype Cycle Special Report" projects that won't arrive for three to five years.

    "Accuracy has increased, but is still not 100 percent despite decades of improvement. So somebody has to edit the document and I suspect the doc won't be that somebody," Baker says. "Until everything a doc or nurse needs to do can be done [documentation and navigation] through speech recognition, it is a work in progress. We're not Capt. Picard on the Enterprise yet."

    What can improve any software's accuracy is its built-in learning function. Most speech recognition systems will "learn" from the user, and improve accuracy as they get used to the physicians' speech patterns, accent and word choice. The learning curve can vary widely for different users, with accuracy improving in as few as two to three uses, but also taking 10 or more to get the hang of a particular doctor's speech. Push your vendors to be specific about how long their system takes to get up to speed--and then test their claims.

    Another important factor to check for accuracy are the dictionaries a system uses to identify and recognize certain terms. Medical terminology can be difficult to recognize, so most speech recognition systems have built-in dictionaries of key terms, common phrasing, and the like, to bridge the gap somewhat. Dictionary quality can vary by vendor, and by specialty, however. Matthews points out that for some departments, like radiology, speech recognition dictionaries are well-established, but in other specialties, still need some work.
  • Connectivity: Losing signal around a hospital is common, but can frustrate doctors trying to dictate notes. Carson recommends a combination of off-the-shelf routers and repeaters to amplify signal and crosswalk devices from one area of the hospital to another. A $199 "airline router" is generally hefty enough to support up to 50 repeaters on a floor, with the cheaper $99 repeaters extending its reach and keeping mobile devices hooked into your network.
  • Voice quality: Microphones are king here. Any speech recognition module is only as good as the audio device recording the physician's voice, Matthews notes.
  • Minimal editing: This is the bane of most physicians' dictation efforts. Because no speech recognition product has 100 percent accuracy, it needs to have robust, but easy-to-use, tools for physicians to fix any mistakes made in translation.

    On-screen editing is the gold standard here, allowing physicians to see their text as they're speaking, or shortly thereafter, and simple touchscreen buttons for needed revisions.

One reality check for this emerging technology, from Matthews, is that physicians who have trouble with traditional dictation methods probably will have trouble with dictating into speech recognition modules on their mobile devices. One cardiologist at his facility, soft-spoken and with a strong accent, has never mastered traditional dictation, and continues to struggle. - Sara (@FierceHealthIT)