[1749] An Objective Appraisal of Computerized Speech Recognition (CSR) in Final Routine Surgical Pathology Reporting
Z Qu, U Sivagnanalingam, B Forcione, U Sivagnanalingam, JS Van Vranken. University of Rochester, Rochester, NY
Background: Recently, there is renewed enthusiasm of computerized speech recognition (CSR) in routine pathology work-flow. Since CSR has created repeated cycles of enthusiasm, frustration, and disappointment in the past ten years, there are diverse opinions on the usefulness or lack of it in routine pathology service. An objective appraisal/evaluation of CSR is eagerly awaited. We recently reported an outcome assessment of its application in surgical pathology with focus on gross examination. As the extension of the effort, this report focuses on the validity of CSR in routine pathology diagnosis reporting/sign-out.
Design: CSR and transcriptionist-assisted documentation (TAD) are compared for 7 parameters (see table). Parallel routine sign-out and simulation experiment are designed for the outcome assessment/appraisal. In the former, 186 surgical cases and 531 surgical cases are included in CSR and TAD, respectively. In CSR simulation arm, prior reports of 76 cases are repeated by three different users. Four participating users include a pathologist with >10-year CSR experience, and three students with >12 hours of general and pathology CSR training. Dragon Naturally Speaking version-9, (Nuance Inc.) is used for CSR. Time is recored by stop-watch. Error rate is recorded manually. Multivariant statistical analysis is used. No built-in macros are used in this study.
Results: For pathologist, CRS seems to require slightly more dictation time, but takes less total time to complete the diagnosis reporting than TAD:
| Parameters | CSR | TAD |
| Case number | 474 | 571 |
| System involved | 1 | 2 |
| Steps Required | 4 | 7 +  1 |
| Dictation Speed (sec /slide) | 15.9 +/- 1.9 | 14.8 +/- 2.2 |
| Efficiency (words/minute) * | 61.80+/-13.8 | 68.88+/-12.06 |
| Accuracy rate | 97.7 +/- 2.2% | 98.7 +/- 1.2% |
| Overall Time (sec/slide) * | 70.4+/-12.3 | 83.0+/-19.2 |
| RTAT | <2 minutes | 2.5 hours |
RTAT: Relative turn around time -- from the final diagnosis rendered by voice input till release of the final report. Overall Time: total time required to complete reporting by pathologist. Transcriptionist's efforts are not included in the result table. * p value <0.05
Conclusions: CSR requires less time overall on pathologist's part in reporting final diagnosis, mainly by eliminating additional steps associated with TAD. The objective assessment provides tangible evidence of the value of CSR in routine pathology diagnosis reporting. More work is needed to complete the appraisal.
Category: Techniques
Wednesday, March 11, 2009 1:00 PM
Poster Session VI # 218, Wednesday Afternoon