1802RE: [agile-usability] Re: QWERTY, mouse, and novel input
- Dec 8, 2005Ron Vutpakdi wrote:
> Just as an aside: the doctors and psychologists that I know who useAnother interesting example. Audio noting to the chart, whether supported by
> speech to text for dictation do so because it's faster for them to
> dictate reports rather than typing (not that they can't type). They
> can do so while walking around or even just sitting at their desk, but
> speaking is faster than typing.
> In their cases, with a special dictionary and training, the
> recognition is generally better than 95% since the vocabulary used is
> considerably more limited than full speech.
> Many doctors and psychologists still dictate reports/evaluations to a
> phone service which then uses a person to transcribe the reports.
speech-to-text software or human transcription is a must-have function in
modern medical informatics, but that does not mean it is truly efficient or
sufficiently reliable to meet real medical practice objectives. Because of
the high potential for errors (95% accuracy sounds good until you turn it
around: 1 out of 20 words is wrong), transcribed audio does not become part
of the legal patient record until the dictating clinician reviews and signs
off on the transcription. Reviewing for errors and correcting is a somewhat
tedious process and itself quite error prone, particularly as clinicians
typically do so at a later time when the context is no longer fresh in their
heads. Transcribed audio, even after review, correction, and sign-off, has a
significantly higher error rate than directly entered notes and orders.
I don't know if the analysis has been done in medical settings, but in other
contexts, when all activities in the process are taken into
account(including slowed speech, repetition and correction on the fly,
review and editing), the effective total throughput is almost invariably
less than even slow direct keyboard entry. We can process up to about 400
wpm when heard and rapid speech clocks at nearly 200 wpm, although 120-160
is considered tops for persuasive communication. The best commercial
"trained" speech-to-text systems are typically only good to about 100 wpm.
But, users typically find they can spend as much time correcting errors as
dictating (some report as much as 2-3 times). So effective throughput drops
to well within the range of typical typing (30-60 wpm).
That said, it can still be more efficient use of the clinician's time if
notes and orders can be dictated while moving between patients or while
riding the subway. (Although HIPAA compliance may become an issue in the
I think audio notes and orders could actually diminish in use over time, at
least in the short run, because the new generation of clinicians has grown
up with computers. My personal physician does all his own notes and orders
directly into the medical system, typing away at 100+ words/minute. When I
commented, he mentioned growing up with computers and video games, then
added that being a musician also helped!
--Larry Constantine, IDSA
- << Previous post in topic Next post in topic >>