Factors affecting error rate in speech recognition dictation

Use of speech recognition software has greatly expedited radiology report turnaround time and has transformed the way that reports are authored, but at what cost? The subject of errors made using this medium is seldom discussed, so this ambitious study of almost 18,000 radiology reports is refreshing in its size and candor.

This analysis by Zombor Zoltani, a fourth-year medical student and faculty member at the University of Maryland in Baltimore, however, is disturbing and worthy of every radiologist's attention. Only 31.6% of the randomly selected reports were error free at a large academic medical center, representing 5% of three years of output.

Translating this into numbers, mistakes and errors were identified in more than 12,000 reports, with 15% (2,709 reports) containing serious errors that could have potentially altered clinical meaning.

What types of errors are generated? Errors in sentence content, word omissions, inappropriately substituted words, unintelligible phrasing, and punctuation and format errors.

Are they negligible? No. This session will candidly tell all.