Meaningful Implementation of Computerized Speech Recognition (CSR) in Routine Pathology Reporting Must Be Determined by Its Accuracy and Efficacy.
Zhenhong Qu, Cathrine Jose, Margaret Huynh, Brittany Forcione, Urmila Sivagnanalingam, Shoanae Thomas, Kendra Alexaner. William Beaumont Hospital, Royal Oak, MI; University of Rochester Medical Center, NY
Background: Effective implementation of computerized speech recognition (CSR) has been hampered by the lack of assessment of its effectiveness and shortcomings. We previously presented preliminary results of our objective outcome measurement of CSR. As the final part of our study, we stratify the measurements, and based on the data, propose recommendation for adopting this technology in surgical pathology.
Design: Voice dictation of 2,873 routine surgical pathology cases was simultaneously captured by CSR and routine transcriptionist-assisted documentation (TAD). Five additional participants whose native language is English also used CSR to capture the accession numbers and patient names of these cases. Fifty-two diagnostic templates (including “Note” in some) representative of routinely encountered disease spectrum were repeated in CSR by the participants. The accuracy in each report component (table) by CSR was compared to those by TAD and/or the templates. Dragon Naturally Speaking (v9) by Nuance installed in Dell Precision desktop computers with Microsoft Window-XP was used for the CSR.
Results: The accuracy of CSR for different components in pathology report varies greatly. Highest accuracy rate for the main diagnosis components requires 5-8% reduction in dictation speed from that for TAD.
|Report Component||N =||N of CSR Participant||Accuracy rate (%) *||Comment|
|Diagnosis||2,873||1||91.6-94.4||Dictation speed as that for TAD|
|52||5||93.7-97.8||Dict.speed to maximize accuracy|
|Report Format||2,873||1||0.71-2.2||Format pre-defined in CSR|
|52||5||4.5-7.9||No pre-defined format|