Outcome Assessment of Computerized Speech Recognition To Validate its Application in Surgical Pathology
Lanjing Zhang ; Mount Sinai Medical Center;
Content:
Although computerized speech recognition (CSR) has been reportedly matured into a practically useful tool in medical documentation, widely diverse opinions on usefulness or lack of it in pathology service still remain. This appears to stem from the lack of evaluation/appraisal based on objective outcome measurement. This study is designed to evaluate the utility and validity of CSR in routine surgical pathology service.
Technology:
CSR is tested and established. Four trained users are recruited. Documentation time (DC, i.e., the time required to complete the documentation task) for gross examination, and reporting time (i.e., the time required to review and sign out the case) by CRS and transcriptionist-mediated dictation (TMD) are compared. The efficacy of CSR is validated by time, effort (and accuracy) compared to those of TMD.
Design:
For gross examination, the DC for a total of 83 commonly encountered surgical specimens by CSR (44) and TMD (39) are collated. No grossing templates are used for CSR. The DC by different methods for similar specimen types is compared. The documentation adequacy is measured by the presence of key words against a third-partys gross examination manual. The reporting times of 445 biopsy cases by these two methods are separately collected and compared. Unpaired two-tail Students t test is used in statistical analysis.
Results:
CSR and TMD require equal DC (n=18 vs. 6, Mean±SD=13.20±3.98 vs. 12.20±6.02 sec/word) for large resection specimen. Interestingly, more or equal DC for small specimens is required by CSR compared to that by TMD (n=26 vs. 33, Mean±SD =5.12±2.13 vs. 2.95±1.50 sec/word, p<0.01). The mean reporting time of biopsy cases by CSR (n=201) was 338.8±75.4 sec/case or 77.3±20.2 sec/slide, and that by TMD (n=244) was 345.1±81.9 sec/case or 80.7±18.3 sec/slide. No statistical significance was established between reporting time by CSR and TMD (p>0.05). A documentation accuracy of more than 95% was reached by either CSR or TMD.
Conclusion:
The efficacy and utility of CSR depend on the types of surgical specimens to be documented. For gross examination, CSR appears to require equal or more DC than TMD to complete small resection specimens (Mean, 5.12 vs. 2.95 sec/word), while equal DC to complete large ones (Mean, 13.20 vs. 12.20 sec/word). In reporting biopsy cases, CSR takes same or even less time of TMD (Mean, 77.3 vs. 80.7 sec/slide). In these two settings, CSR may significantly shorten report turn around time and cut down the cost by eliminating transcription step by office staff.
