Clinical Practice - Health Informatics in the Cloud - page 45

Fig. 4.1 M*Modal Recognizes Clinical Concepts in Free Text and Codes them into SNOMED
and LOINC as shown in Fig. 4.1 . Here again the system has been trained over time
as humans corrected coding mistakes or manually coded those things the computer
could not code. The specific accuracy levels are highly dependent on (a) noise lev-
els, (b) medical specialty, (c) amount of similar data previously seen, and (d) the
subset of SNOMED codes of interest (e.g. allergies,, diseases, symptoms, proce-
dures). Very specific clinical concepts, such as smoking history, that are normally
expressed in one of a few possible ways can be more accurately identified than con-
cepts that can have a large number of possible expressions, such as the patient's
problem. High levels of accuracy are possible. For example, the company says it has
97- 100% speech understanding accuracy for radiologists dictating on high quality
microphones in a quiet image reading room.
However, the overall accuracy is sufficient that some clients of the company use
the technology to code already digital text information . They submit it from their
system over the Internet to the Fluency for Coding service that identifies the clinical
concepts and puts them into the appropriate structured nomenclature, most typically
SNOMED CT. Some software is installed locally so that this process works better
when there is suboptimal Internet connectivity.
When I first visited the company in 2008 I asked why the technology was not
integrated into EMR systems. Back then the company didn't feel it was robust
enough. Today, it apparently is, because quite a few EMRs now have M*Modal
technology “built in”. Again, in all such cases, it is still a web service so the central
system continues to learn and further develop from all users.
So voice recognition seems to have arrived as a practical method of collecting
and structuring clinical data. To see that in action we'll next look at a leading EMR
from a company near my base in Atlanta, and, so far as I know, the first to integrate
M*Modal with an EMR.
Greenway Medical Technologies was, according to its web site, founded in
1998 by nearly 100 physicians, clinicians, practice administrators, hospital execu-
tives and community leaders with the goal of developing a long-term solution to