Dress Rehearsals Key to Speech Rec Success
By Susan Chapman
For The Record
Vol. 27 No. 10 P. 10
Practice? Yes, we're talking about practice. Find the time and be adaptable, say experts.
Speech recognition software is becoming an increasingly common way for physicians to enter data into the EMR. The proliferation of this new technology raises questions as to how readily providers are adapting to and accepting speech recognition, and how much training they're receiving in order to accurately and efficiently integrate it into the EMR and their daily workflow.
In front-end speech recognition, because the record appears on the computer screen as physicians make their entries, they must be concerned not only with dictating the patient encounter correctly but also with editing and proofreading the resulting document for accuracy.
"While this may eliminate the back-end review process in many cases, it does not necessarily eliminate the need for it," says Jay Vance, CMT, CHP, AHDI-F, production coordinator at Keystrokes Transcription Service. "There are studies that show that front-end speech recognition has a much higher error rate."
In fact, the AHIMA Practice Brief "Speech Recognition in the Electronic Health Record" addresses the accuracy issue, stating, "Editing text, whether done by physician or MT [medical transcription] editor, reduces content errors in patient reports, provided it is done meticulously prior to signing. In current transcription practices, many transcribed reports are not reviewed closely before a signature is applied by the physician. Standards for ensuring accuracy with all documents produced using speech recognition call for third-party editing."
Brina Hollis, CST, MHHS, PhD, CBRS, a health informatics faculty member at Kaplan University, believes physicians who use front-end technology require training to maintain the integrity of the medical record. "With a front-end system, doctors are speaking directly into a microphone. They have the ability to correct errors as they go along," she explains. "However, the physicians are not medical transcriptionists and, at the very least, need to be trained in correcting and editing data appropriately."
Because back-end speech recognition uses medical transcriptionists to edit dictated documents, Vance notes that when physicians use back-end technology, there's little to no training. In fact, sometimes doctors are not even made aware they are using speech recognition technology. "One of the marketing tools that vendors use to sell their technology to institutions is that the physicians don't have to do anything different. They can dictate just like they have been, and our fantastic software will recognize and understand what they say," he says.
While, in general, most vendor training can last anywhere from 15 minutes to four hours, staff at larger institutions sometimes receive guidance over the course of several weeks. "Training for physicians is best handled by department," Hollis says. "The time taken with each department then depends on the number of physicians who need to be trained and what they need to implement in order to meet their needs."
At Chicago's Rush University Medical Center, because there are few replacements available to cover for attending physicians, one-on-one training is provided, says information systems manager Dallas Smith, adding that residents are educated based on when they arrive at the hospital. "Due to high volume turnaround in the middle of the year, residents/fellows are trained as a group but may require a refresher at some point. Residents or fellows who report outside of the turnaround period are provided one-on-one training," he says.
Keystrokes offers one to two hours of one-on-one training to supplement whatever instruction physicians may have received from the speech recognition vendor. As part of this second wave of training, the team reiterates such things as the importance of enunciating clearly. "Accents are not as big of an issue as you might think," Vance says. "Speech recognition 'listens' to several hours of dictation in training mode. During that time, a human transcribes the dictation. Then, that data is fed back into the system, and that system learns as it goes along. The machine can understand what that sound means, accent or not. In this way, accents are not as much of an issue as the physicians dictating clearly and making sounds the same way."
There are certain steps physicians can take to make the technology more effective. Clear, audible enunciation is critical for accuracy, as is reduced background noise. "Those are the kinds of things that we would help the physicians address as well as creating macros or templates," Vance says. "The latter helps doctors avoid dictating long templates or paragraphs. Basically, the physician can say, 'Insert my such and such template,' and the technology will recognize that as a command, scan its database, find the template, and insert it."
However, Anne Cardwell, MD, an assistant professor of diagnostic radiology at Rush University Medical Center, says macros and other "quick keys" can offer a false sense of efficiency. "Part of the resistance facilities experience from physicians has to do with how front-end speech recognition software puts more work in front of us. We think we will save so much time because of things like macros and templates, but then there are errors. We think, 'I have to dictate, and now I have to edit and correct the record, too?' Having editors on the back end makes it really helpful," she says.
The Value of Pilot Programs
Hollis believes that trial runs are vital for successful speech recognition technology adoption. "It's during this pre–go-live phase that we'll have a little extra time to see if there are any challenges with the system, to make sure it's user-friendly, and to assess any challenges users may be having," she says. "We get all excited about going live, but we have to be careful, so that we have a solution to the challenges we face before fully implementing the system."
Pilot programs can offer valuable insights into the level of training physicians may need and their willingness to embrace speech recognition. At Bryan Medical Center, a small pilot program created by HIM manager Leigh Anne Frame, CHDS, AHDI-F, reveals differences in how front- and back-end technologies are implemented.
"Some of the things that we thought were the same are not the same at all," Frame notes. "For instance, there are vast differences in training. When we launched back-end speech recognition, for example, we taught physicians the basic dictation functions. Beyond that, we didn't even tell the doctors that it was a speech recognition technology unless they asked. There was no additional training, and there still is no training other than the dictation basics. Sometimes, I may contact a physician and ask her to slow down or not use her cell phone for dictation, but we had those same kinds of discussions prior to the implementation of back-end speech recognition."
At Bryan Medical Center, three clinicians were selected for training on a front-end system. Frame and her team offered a one-hour session, which included about seven minutes of voice training in which they read and recorded several sentences from a script, instruction on the use of the speech microphones and editing features, and other factors that could influence the quality of the record such as reducing background noise. Following the training, Frame stayed with the clinicians during their first dictation sessions to ensure they were comfortable and everything was working properly.
"Through this program, we were able to customize some of the settings on the clinicians' PCs. It helped with navigation," she says. "Working with vendors, we did some special programming to the clinicians' microphones and added some different buttons and voice commands. Those customizations were pretty big satisfiers for them."
However, what the pilot revealed had little to do with the software. "We learned how difficult it is to schedule time with clinicians in order to train them," Frame says. "Once you do get something scheduled—which often was months down the road—at the last minute they cancel due to patient issues."
Frame and the quality assurance (QA) specialist who oversee the pilot program are rarely at the same location as the clinicians. Consequently, when the clinicians have needed help, it hasn't come immediately. "That immediacy is one of the things that we need to acknowledge and talk through when we begin training other clinicians," Frame says.
After the clinicians completed their first dictation, the reports were reviewed and the process was fine-tuned. "We helped them create macros and voice commands," Frame says. "After they saw how much faster this could make their dictation, they were more willing to take time for the initial setup."
The pilot program also helped Frame address system issues, such as microphone and software compatibility on clinicians' PCs. While the pilot group comprised hospital employees, their PCs were set up differently from the test PC, causing the speech application and microphone to respond poorly. To remedy this issue, the IT department reimaged the clinicians' PCs, delaying the launch and creating further scheduling conflicts.
"Another surprise we encountered was the varied computer skills among this small group, which was something we hadn't taken into consideration at the outset," Frame says. "If a clinician didn't know computer basics, then I had to spend additional time with that individual. That factor made a difference in actually creating the documents themselves and putting them into the EMR. That outcome is something universal—hospitals need to know the level of each clinician's computer skills and not assume anything before training each individual."
The Best Candidates
When it comes to adopting speech recognition, not every physician fits the bill. While those raised in the digital age seem to be more enthusiastic about the technology, that's not always the case.
"What we learned from the pilot was that each clinician has a different working style, personality, and attitude," Frame says. "One clinician said she'd use the technology only as much as she's forced to. Another clinician started using it, going great guns. All of a sudden, in the middle of it, she stopped using it completely. I asked her what was going on, and she said she had encountered a computer problem and stopped using it and went back to dictating like before. It just happened that I noticed it when running management reports and I was able to stay on top of it. However, the fact that clinicians either don't have the time to call or simply won't call is an issue that should be discussed."
Vance says physician comfort level is one of the keys to successful adoption. "In general, younger practitioners are more comfortable with technology and find it less intimidating; the younger physicians are more enthusiastic about trying something new," he says. "For those who have been working for many years, change sometimes doesn't come as easily. So, in our experience, we've found a variety of responses."
As long as practitioners put forth a reasonable effort, Vance says they generally improve their speech recognition proficiency to a significant degree. "It really is the amount of enthusiasm and effort on the doctor's part that impacts the level of productivity," he notes.
Hollis says the ideal candidates are open and adaptable to change, and willing to learn. "As human beings, if everything is going well for us, we may not see the need or value in change," she says. "Having the attitude that change is the only constant, adaptability and willingness to learn are the two main characteristics that will lead to success."
Bryan Medical Center has steps in place as it transitions from pilot to launch. "To roll out to the larger population, we have targeted groups, which we will tackle one at a time," Frame says. "I do have a lot of hospitalists here, plus a huge emergency department. There is some interest from those particular groups for front-end speech recognition."
Because the QA team is small, Frame must determine how to handle training when the software is launched to more clinicians. "At this time, there are no plans to hire additional QA staff for this new initiative, and this will be a challenge," she says. "If we use the medical transcriptionists/editors to help bridge the QA gap, then we run the risk of not meeting our commitment to turnaround times. It's a Catch-22."
Smith says it's important for testing to be completed thoroughly before any new application is rolled out in production. "Also, make sure you can roll back to any version prior to upgrade for any issues that do come up that can't be resolved immediately or to the satisfaction of your customer," he says.
John Weiss, vice president of Scribe Healthcare Technologies, says having plans in place to fall back on an older software version offers users a sense of security. "If the physicians have trouble, they can always go back to what they were doing prior," he notes.
Leaders of the Cause
Physician champions and super users can go a long way toward ensuring speech recognition technology is accepted and used effectively. Frame believes super users benefit their colleagues by being available to help the group when technical personnel are unavailable.
Hollis cites the enormous impact a physician champion can have on the success of a speech recognition project. "We can all be resistant to change. We get into our groove, and that's kind of where we remain. The physicians may see the need for change, but don't necessarily embrace it. Physician champions can take information back from the committee or project group to their peers, talk about the advantages of speech recognition, and address any concerns that they have. That person can get buy-in and support and quell some of the trepidation that fellow doctors may have about speech recognition implementation," she says.
Tips for Successful Adoption
Providing physicians with as much information as possible and evidence of speech recognition's effectiveness can help ensure an implementation becomes a success. To that end, Smith recommends facilities enlist dedicated personnel who can meet with physicians to get a feel for what they are expecting from the software while also informing them of any new functions.
This form of two-way communication is essential, Hollis says. "The physicians are highly educated. Therefore, I can't stress enough how important communication is," she says. "Many vendors don't offer enough training, or maybe organizations don't offer enough training because of a go-live date. Having adequate training so that physicians are 100% comfortable will go a long way to make the process successful."
Frame believes well-trained clinicians, in conjunction with consistent QA support, are a recipe for success. "Speech recognition technology will keep getting better and better in terms of usability and accuracy, and clinicians who are well-informed and understand what the technology can and cannot do are the ones who are doing a terrific job with it," she says.
— Susan Chapman is a Los Angeles-based freelance writer.