Error Rate 7.4 Percent in Speech Recognition-Assisted Notes
Seven in 100 words in SR-generated documents contain errors; many errors involve clinical information.
HealthDay News — The error rate in speech recognition (SR)-assisted documentation is 7.4 percent, according to a study published online July 6 in JAMA Network Open.
Li Zhou, M.D., Ph.D., from Brigham and Women's Hospital in Boston, and colleagues examined the quality of SR-assisted documentation by analyzing errors at each stage of the SR-assisted dictation process. A random sample of 217 notes dictated by 144 physicians were reviewed. Errors were annotated in the SR engine-generated document (SR), the medical transcriptionist-edited document (MT), and the physician's signed note (SN).
The researchers found that the error rate in SR notes was 7.4 percent; the rate decreased to 0.4 and 0.3 percent after MT and SN, respectively. Errors were identified in 96.3, 58.1, and 42.4 percent of SR notes, MT notes, and SNs, respectively. The most common errors were deletions and insertions (34.7 and 27.0 percent, respectively). Overall, 15.8, 26.9, and 25.9 percent of errors at the SR, MT, and SN stages, respectively, involved clinical information, while 5.7, 8.9, and 6.4 percent, respectively, were clinically significant. Compared with other types, discharge summaries had higher mean SR error rates (8.9 versus 6.6 percent). Lower mean error rates were seen for surgeons' SR notes versus those of other physicians (6.0 versus 8.1 percent).
"Seven in 100 words in SR-generated documents contain errors; many errors involve clinical information," the authors write. "That most errors are corrected before notes are signed demonstrates the importance of manual review, quality assurance, and auditing."
Several authors disclosed financial ties to the medical device and medical technology industries.