Comparing Speech Recognition Software to Transcription Services

Accuracy of Dictation Methods For EHR Charting

A widely publicized report by the U.S. Institute of Medicine, To Err Is Human: Building a Safer Health System, concluded that up to 98,000 people die each year as a result of preventable medical errors.

To Err is Human : building a safer health system / Linda T. Kohn, Janet M. Corrigan, and Molla S. Donaldson, editors.Updated on 06/25/2020
ISBN 0-309-06837-1

A significant portion of these critical events may be traced to critical inaccuracies in medical documentation. The smallest mistake, such as switching two medications that sound alike, adding a zero or misplacing a decimal point can have lethal consequences for the patient, and lead to millions of dollars of malpractice exposure for physicians and medical entities. No one would argue that the quality of medical records is directly related to the quality of care, and conversely, a higher error rate in medical documentation should have a higher rate of medical complications.

EHR Optimization and EHR Clinical Documentation Using Dictation Methods

To compound clinical documentation efforts, physicians are faced with an average of almost six hours of data entry time per day toiling at EHR medical documentation using keyboard and mouse on their computers. The average physician types at 30 words per minute. Very few can type as fast as dictation rates of 100 words per minute. It is only natural that physicians would prefer to dictate their medical documentation. It is important to know if different methods of documentation result in different error rates.

One method of documentation using dictation involves the use of front-end speech recognition software with a microphone tethered to a computer. When the clinician activates the microphone for the speech recognition program, dictates and the speech is almost immediately translated into text where it appears on the computer screen where the doctor places the cursor in the electronic health record. After initiating the program, the doctor has to navigate to the appropriate place in the electronic record and dictate the text for that section.  Best practice would be to correct each section before moving to the next to minimize the time necessary to navigate back to each section again. An alternative method of dictation involves the use of a medical transcriptionist with automated insertion of the transcribed text into the appropriate sections of the EHR. Dictation does not have to be in front of the computer and can take place on a smartphone, digital recorder, tablet or with telephone service as commonly used at the hospital. Of paramount importance is the comparison of the accuracy of these two methods.

Accuracy of Speech Recognition vs. Transcription

Virtually every study performed comparing the accuracy of speech recognition versus transcription reveals that speech recognition has a higher error rate and therefore increased patient risk and greater malpractice exposure than transcription. A systematic review of published clinical studies by Hodgson, et. al. (Page 4, #7) revealed speech recognition accuracy in 10 studies to range from 88.90 to 96.00% and appears to improve only 0.03% per year. This is in contrast to vendors’ claims of 99% accuracy. In contrast, the mean number of errors per report increased using speech recognition, compared to dictation and transcription. In fact, the astonishingly high accuracy of 99.6% was found in numerous studies for dictation and transcription. Accuracy can automatically be measured on a transcription company’s platform, comparing typed to edited documents and edited to documents checked and signed off by doctors. With a good quality control process in place and continuous quality improvement coupled with feedback for transcriptionists, near-perfection can be achieved with dictation and transcription. In a subsequent 2017 study by Hodgson (Page 4, #8) increased errors were observed using speech recognition. Over four times the number of errors were observed using speech recognition over keyboard and mouse. He also concluded that, “For clinical documentation, speech recognition was slower and increased the risk of documentation errors, including errors with the potential to cause clinical harm, compared with keyboard and mouse. A multi-center clinical study performed at Harvard Pilgrim, Guisinger and _____ documented over a 7% error rate with speech recognition. Goss, et. al. (Page, 6, #17) found that 15% of ER notes contained one or more critical errors, potentially leading to miscommunication that could affect patient care. Transcription. accuracy was measured in the emergency room by Zick and Olsen (Page 11) and found to have 99.7% accuracy rate.

Speech Recognition vs. Transcription in Radiology

Although speech recognition can offer faster turnaround time for radiology reports, Hodgson, et. al  revealed speech recognition accuracy, and numerous other authors document that there may be a trade-off with lower accuracy. In spite of the case of radiology adoption of speech recognition as the trophy use of the technology, nearly all radiologists surveyed in  a Brown University study expressed dissatisfaction with voice recognition, with feelings of frustration and increased fatigue. “In summary, in non-academic settings, utilizing radiologists as transcriptionists results in more error-ridden Radiology reports and increased cost compared with conventional transcription services.” (Page 6, reference Page 5.) Motyer et.al., (Page 8) found the occurrence of errors in 4.2% of reports with the potential to alter the report interpretation and patient management. Ringler, (Page 9) found occurrences of material errors in 1.9% of reports that could alter interpretation of the report. In another radiology study, du Toit, et. al., (Page 9) found the occurrence of clinically significant error rates of 9.6% for speech recognition services versus 2.3% for dictation and transcription. Basma, et. al., (Page 10) found Radiology reports generated with speech recognition were eight times as likely to contain major errors as reports from transcription.

Limitations of Speech Recognition with dictation

Aside from critical errors affecting patient safety, tertiary care physicians’ reputations and referral business ride on the quality of their work and are largely derived from the quality of their reports. Speech recognition requires complicated commands to format and correct grammar, including punctuation that must be dictated or added. Human transcriptionists handle these deficiencies automatically and can be especially useful for Physicians with English as a second language.  These foreign doctors may not want verbatim transcription, but may want grammar and phrase corrections of their dictations performed by experienced transcriptionists that can derive their intent. (Page, 13 #35) Chang, et. al, (Page 10) found the occurrence of 5% nonsense phrases in Radiology reports.

Speech Recognition Lacks Contextual Understanding

Speech recognition users can optimize efficiency using commands for navigation, correction and insertion of macros. Misrecognition of these may leave the microphone on or off, wasting time, but human transcriptionists routinely deal with these issues. It is quite common that a doctor will want to go back and add to a section of the medical record, delete or correct something. This requires extensive mouse clicking or navigation and correction commands that are routinely handled by medical transcriptionists. With speech recognition, words are misunderstood but not misspelled. There is no contextual understanding by the software that a human transcriptionist would have, given their understanding and familiarity with cases, drugs, anatomical findings, testing and the like, on a level commensurate with many physicians.

Physicians Do Not Want to Correct Documents

Although physicians clearly have the ability to correct their own errors, this apparently is not being done. Numerous studies have referenced above had documented that they are not taking the time and effort to do so. Numerous studies document that the time required to edit text is about twice the time needed to dictate. (Page 13, references 27, 29, 45, 56.) In fact, 50% of physicians who try speech, recognition abandon it, and the main reason for 70% of the users was the time required to correct errors. (Reference 27.) The disturbingly high error rate of speech recognition software is actually easily correctable with the implementation of transcriptionist review. The JAMA study saw error rates fall to just 0.4% when the documentation was reviewed by a transcriptionist. Physicians have made it clear that they do not want to perform the function of a transcriptionist.

With speech recognition software, the doctor, must complete a review. Anyone who has tried to edit their own work knows that it’s easy to let mistakes fall through the cracks, but physicians can’t afford to create documentation with even a few mistakes when resulting errors can be disastrous. There’s no replacing the human touch when it comes to proofing critical documentation, and quality medical transcription provides it.

Other Considerations in Choosing a Dictation Solution

With improved accuracy comes less risk of malpractice suits and better protection against audits, both benefits that can deliver better peace of mind to practices of all sizes and specialties. Other articles will address the advantages of dictation and transcription over speech recognition for EHR, documentation. While the dictation solution is the most accurate method for EHR documentation, it also is the easiest to implement, and requires no IT support or upfront investment. Training is as simple as a dictation service in the hospital and the advantage of smartphones allows integration with the patient schedule to assure a perfect lock with the appropriate EHR encounter.

Clinicians will appreciate transcription solutions that require little training, ease of use, with low cognitive effort, that fits into their workflow. They will appreciate the high accuracy and professional documents that generate more referrals with rich, readable, human health story. The significant Improvement in efficiency can alleviate the EHR documentation burden, the leading cause of physician burnout.

Hospital administrators and practice managers benefit from immediate return on investment with transcription and EHR documentation solutions, with higher accurate accuracy of documents, than scribes, without the HR hassles. Cloud-based transcription solutions require no software to install and can be implemented and can often be implemented in one day. The productivity impact on workflow removes bottlenecks and can allow physicians to be more productive, with higher satisfaction. Detailed notes can support higher levels of billing and defend against malpractice and with audits. Cloud-based transcription services are complementary to EHRs and can function even when an EHR is down, and provide a separate repository for clinical documentation, in the event of malware or outages. Productivity gains are easy to measure and risk-free, with free trials. Rapid turnaround of items such as discharge summaries within two hours can save the cost of a day’s hospitalization.

Security of Dictation Solutions

Technologists can evaluate reliability of technology by the history of companies’ outages and downtime that can be weeks even for market leaders. HIPAA compliance and controls should be verified, and independent third-party audits are the golden standard for this. Audit logs and encryption should be able to demonstrate that transcriptionists are located in the United States and verify each access of a record. Speech recognition that is commercially available may not be HIPAA compliant or accurate enough for medical needs.  Even locally run speech recognition sends information in the background to vendors for assorted purposes. Employees are always the weak link and background checks of employees can be reassuring. Utilizing secure cloud-based platforms and meeting HIPAA or HITRUST compliance keeps data safe. When all else fails, cyber- and HIPAA Insurance help us to sleep better at night.

Business Impact

An easy to use, efficient and highly accurate EHR documentation solution with transcription can have a huge business impact. Inherently, better records generate better medical care and document the excellence of any healthcare institution. Transcription offers instant time to value and does not require additional IT personnel, scribes, HR loads, HIPAA and security training and can be implemented immediately. Small practices can self-enroll and start the same day. Many EHRs are pre-integrated. Highly detailed and accurate medical records should reduce billing errors and provide support for higher levels of coding. With automated EHR insertion, cost is lower than scribes and provides higher accuracy, is comparable to transcription cost and is a lower cost than scribes that have to be hired, educated in HIPAA privacy practices, familiarized with medical terminology and EHR usage. The return on investment is best understood by the time savings physicians have with dictation and transcription over use of speech recognition, scribes or keyboard and mouse.  Physicians are the highest-paid people in the healthcare system and they have opportunity costs of over $10/minute. A pilot test is a low-risk, high-reward approach to evaluating a dictation, transcription, and EHR documentation solution, and avoids the expensive and lengthy evaluation process that healthcare entities traditionally take in implementing technology. This agile approach is much more efficient and cost-effective, with little risk.

References:
1. To Err is Human : building a safer health system / Linda T. Kohn, Janet M. Corrigan, and Molla S. Donaldson, editors. ISBN 0-309-06837-1

2. Arun Kalava, Sapna Ravindranath, Inessa Bronshteyn, Ripudaman S Munjal, Joseph Schianodicola, Joel M Yarmush. Typing Skills of Physicians in Training. J Grad Med Educ. 2014 Mar;6(1):155-7. doi: 10.4300/JGME-D-13-00164.1.PMID: 24701328

3. David R Kaufman, Barbara Sheehan, Peter Stetson, Ashish R Bhatt, Adele I Field, Chirag Patel, James Mark Maisel. Natural Language Processing-Enabled and Conventional Data Capture Methods for Input to Electronic Health Records: A Comparative Usability StudyJMIR Med Inform. 2016 Oct 28;4(4):e35. doi: 10.2196/medinform.5544.

4. Hodgson T, Coiera E. Risks and benefits of speech recognition for clinical documentation: a systematic review. J Am Med Inform Assoc. 2016 Apr;23(e1):e169-79. doi: 10.1093/jamia/ocv152. Epub 2015 Nov 17.2

5. Maamoun M Al-Aynati 1, Katherine A Chorneyko. Comparison of Voice-Automated Transcription and Human Transcription in Generating Pathology Reports. Arch Pathol Lab Med. 2003 Jun;127(6):721-5. doi: 10.1043/1543-2165(2003)127<721:COVTAH>2.0.CO;2.

6. R G Zick , J Olsen. Voice Recognition Software Versus a Traditional Transcription Service for Physician Charting in the ED. Am J Emerg Med. 2001 Jul;19(4):295-8. doi: 10.1053/ajem.2001.24487.

7. Hodgson T, Magrabi F, Coiera E. Efficiency and safety of speech recognition for documentation in the electronic health record. J Am Med Inform Assoc. 2017 Nov 1;24(6):1127-1133. doi: 10.1093/jamia/ocx073.

8. Li Zhou, Suzanne V Blackley , Leigh Kowalski, Raymond Doan, Warren W Acker, Adam B Landman, Evgeni Kontrient, David Mack, Marie Meteer, David W Bates, Foster R Goss. Analysis of Errors in Dictated Clinical Documents Assisted by Speech Recognition Software and Professional Transcriptionists. JAMA Netw Open. 2018 Jul; 1(3): e180530. Published online 2018 Jul 6. doi: 10.1001/jamanetworkopen.2018.0530

9. Tobias Hodgson, Enrico Coiera. Risks and Benefits of Speech Recognition for Clinical Documentation: A Systematic Review. J Am Med Inform Assoc. 2016 Apr;23(e1):e169-79. Doi: 10.1093/jamia/ocv152. Epub 2015 Nov 17.

10. John A Pezzullo 1, Glenn A Tung, Jeffrey M Rogg, Lawrence M Davis, Jeffrey M Brody, William W., Mayo-Smith. Voice Recognition Dictation: Radiologist as Transcriptionist. J Digit Imaging. 2008 Dec;21(4):384-9. doi: 10.1007/s10278-007-9039-2.

11. Motyer R E, Liddy S, Torreggiani W C, Buckley O. Frequency and analysis of non-clinical errors made in radiology reports using the National Integrated Medical Imaging System voice recognition dictation software. Ir J Med Sci. 2016;185(04):921–927.

12. Ringler M D, Goss B C, Bartholmai B J. Syntactic and semantic errors in radiology reports associated with speech recognition software. Stud Health Technol Inform. 2015;216:922.

13. du Toit J, Hattingh R, Pitcher R. The accuracy of radiology speech recognition reports in a multilingual South African teaching hospital. BMC Med Imaging. 2015;15:8.

14. Basma S, Lord B, Jacks L M, Rizk M, Scaranelo A M. Error rates in breast imaging reports: comparison of automatic speech recognition and dictation transcription. AJR Am J Roentgenol. 2011;197(04):923–927.

15. Pezzullo J A, Tung G A, Rogg J M, Davis L M, Brody J M, Mayo-Smith W W. Voice recognition dictation: radiologist as transcriptionist. J Digit Imaging. 2008;21(04):384–389.

16. Pappas JN, Bissett GS, Ravin CE. Satisfaction survey from 77 radiologists in a single large academic radiology department using a voice recognition dictation system—a one-year experience. Radiology. 2001;218:611.

17. Yaa A. Kumah-Crystal,1 Claude J. Pirtle,1 Harrison M. Whyte,2 Edward S. Goode,2 Shilo H. Anders,1,3 and Christoph U. Lehmann. Electronic Health Record Interactions through Voice: A ReviewAppl Clin Inform. 2018 Jul; 9(3): 541–552.Published online 2018 Jul 18. doi: 10.1055/s-0038-1666844.PMCID: PMC6051768.PMID: 30040113.

18. Chang C A, Strahan R, Jolley D. Non-clinical errors using voice recognition dictation software for radiology reports: a retrospective audit. J Digit Imaging. 2011;24(04):724–728.

19. Hoyt R, Yoshihashi A. AHIMA.Speech Recognition in the Electronic Health Record AHIMA Practice Brief; 2013.

20. Hoyt R, Yoshihashi A. Lessons learned from implementation of voice recognition for documentation in the military electronic health record system.
Perspect Health Inf Manag. 2010 Jan 1; 7():1e.

21. A JAMA study saw error rates fall to just 0.4% when the documentation was reviewed by a transcriptionist. Physicians have made it clear that they do not want to perform the function of a transcriptionist.