"Previous studies have shown that voice recognition improves the time
required to make reports available to referring physicians, which makes
the trade-off in radiologist time worthwhile," said Dr. Amit Mehta,
a radiologist at Massachusetts General and Harvard Medical School.
When MGH implemented the voice recognition system four years ago,
the technology was known to increase dictation time by two- or three-fold,
Mehta said. Healthcare systems have been under pressure to adopt new
technology to increase service while decreasing costs, however. Voice
recognition provided one answer to that pressure.
Mehta reported findings on 200 consecutive imaging studies performed
over a four-day period at an MGH satellite outpatient clinic. Most of
the images were plain film, with a few ultrasound studies included.
Radiologists first dictated reports using the IBM MedSpeak voice recognition
system.
The same 200 studies were reintroduced into the workflow a month later,
under the assumption that the radiologists would not recall their original
reports. The radiologists then dictated reports for each study using
the conventional dictation system. A majority of the studies proved
to be normal, which is associated with reduced dictation time, Mehta
said.
The time required to dictate a report using voice recognition averaged
44.3 seconds, compared to 28 seconds with the legacy dictation system,
an increase of 36.8%.
"It takes radiologists longer to prepare reports with voice recognition,
and our volume of studies has increased substantially. Yet, most radiologists
are still able to leave work at the usual time," Mehta said. "We wanted
to see if we could find an explanation for how radiologists compensated
for the extra time spent on reports."
Working on the hypothesis that voice recognition had changed the way
radiologists prepare reports, Mehta and his associates evaluated the
length of the 200 reports prepared by the voice recognition and legacy
dictation systems. They found that the report length averaged 36 words
with both systems. Accuracy also was comparable.
The composition of the reports differed between systems, however.
With the legacy system, the body averaged 27 words and the impression
nine words. In contrast, the body of the report was shorter with voice
recognition (21 words) and the impression was longer (15 words).
"Our hypothesis is that the difference has to do with dictation continuity,"
Mehta said. "With the MedSpeak system, if you are able to talk continuously,
the accuracy is much better than if you are constantly stopping and
then continuing to speak. When radiologists dictate the body of a report,
they often are looking at the exam and start and stop talking frequently
as they look at the film. By the time they reach the end of the body,
they have usually formulated what they want to say in the impression,
so speech is more continuous."
Future studies of the voice recognition system will focus on effects
on patient care, Mehta said. In particular, he and his associates will
try to determine whether certain types of information might be omitted
to accommodate voice recognition. As an example, he cited numbers associated
with measurements, which often cause problems for the system.
"I think voice technology will continue to improve, leading to better
accuracy and efficiency," Mehta said. "Other advances, such as incorporating
voice recognition into the PACS systems, also should lead to increased
efficiency."