Brown also noted that some physicians don't want to document in EHRs by pointing and clicking or typing, so they use speech recognition or some combination of methods. Brown made this observation during a discussion of his report on how healthcare providers perceive the leading speech recognition solutions. In the past, these solutions have cut time and costs for transcription and imaging documentation. Now, the report noted, "the hottest steam" in the market is around EMRs/EHRs.
[ Does your organization have a strategy to implement NLP? Here's what you need to know. 6 Questions To Guide Natural Language Processing Strategy. ]
KLAS' survey measured the popularity of applications in three categories: speech EMR, front-end speech imaging and back-end speech. Front-end refers to applications that require clinicians to edit the transcribed text after they dictate, while back-end applications are those in which transcriptions or editors fix the text generated by speech recognition.
In the speech EMR category, Nuance Dragon Medical, the only reviewed product, received a relatively high rating. Nuance has the most live clients at this point, with Dolbey a distant second and M*Modal third. "The ability to improve documentation efficiency across thousands of hospitals appears real," the report said.
Nuance has the broadest technology portfolio of any of the speech recognition vendors, but Dolbey's front/back-end solution is gaining market share. M*Modal, formerly known as Medquist, is also gaining momentum on the back of the speech engine Medquist garnered when it acquired M*Modal and took its name.
In the area of front-end speech imaging, used for imaging documentation, Afga and Dolbey are improving their performance while Nuance and M*Modal struggle, the report said. It attributes Nuance's problem to "delayed implementations and poor training." M*Modal lags because it is using an older SpeechQ engine created by Philips and later acquired by Nuance.
However, M*Modal is picking up momentum in back-end speech with its Fluency for Transcription product, which uses the M*Modal speech engine. The leader in that category is Nuance, followed by Dolbey, M*Modal and 3M.
Overall, Brown said, speech recognition has improved incrementally in recent years. But the real determinant of accuracy is the proper training of applications to recognize users' voices. "Most people who have the right training and applications tend to see better results," he noted.
While the survey didn't ask about natural language processing or computer-assisted coding, the speech recognition vendors are among the leaders in those categories. Nuance recently announced that at the IHE North America Connectathon 2013 event in Chicago, it successfully completed testing for extracting discrete data from paper records and automatically populating appropriate fields in medical documents, using its clinical language understanding solution. And M*Modal has just reached an agreement with 3M to turn transcribed text into structured documentation for purposes of computer-assisted coding (CAC), clinical documentation improvement (CDI), quality metrics and analytics.
The latter alliance, Brown observed, appears to be as much related to business as to technology. 3M earlier entered a strategic alliance with Nuance, he noted, but was "surprised" when Nuance acquired the CAC product of Quadramed, a 3M competitor. So now, he speculated, 3M felt free to partner with M*Modal.
Providers aren't just interested in NLP because of coding, Brown pointed out. "A lot of people are looking at how to drive clinical documentation improvement within departments and across enterprises. They want to improve the format, the way it's done, and the accuracy of the reporting. Also, they want to get the data into a structure so it can be leveraged for analytics."
Not that coding has gone away as an issue for healthcare providers. "There's this big October 2014 date with ICD-10, so CAC and ICD-10 prep is a hot topic," Brown said. "But I don't think CAC is going to be the be-all and end-all for coding. It's a way to improve it, but you'll still need editors to check it and determine which codes need to be there."
As large healthcare providers test the limits, many smaller groups question the value. Also in the new, all-digital Big Data Analytics issue of InformationWeek Healthcare: Ask these six questions about natural language processing before you buy. (Free with registration.)