Groundwork Key to Speech Recognition Success
By Susan Chapman
For The Record
Vol. 26 No. 7 P. 14
A healthy dose of preparation can minimize rough patches and increase the odds of long-term productivity.
In an effort to increase productivity while still maintaining quality, find cost-effective transcription solutions, and provide timely documentation, more health care facilities are integrating both front- and back-end speech recognition technology. However, facilities taking this route must understand what implementation entails, including how standard processes and staff positions may be affected and how the system performs after the initial rollout.
In preparation for speech recognition implementation, several factors must be considered. “Something important that I’ve learned over the years is that you need to allow ample time to get everything in order before you start. Don’t rush it,” says Leigh Anne Frame, transcription and HIM manager at Bryan Medical Center in Lincoln, Nebraska. “During our implementation, I felt that if we had an extra week, it would have made our implementation smoother.”
Take the time to understand the process and how key players will be trained, says Frame, who recommends working with staff as far in advance as possible to help them prepare for the transition. “Sometimes that’s hard to do if this is new territory for you,” she says. “Even if you haven’t formalized anything with the vendor, it’s important to talk with those affected about concepts and experiences that other facilities have had to help make the transition successful.”
Also, become familiar with the technology’s capabilities. For example, physicians and other staff members may rely on certain reports and expect the same information in the identical format to be available when the new system debuts. Frame recommends communicating effectively and clearly both with vendors and end users to be certain they understand what is available and to manage expectations.
In facilities that rely on back-end technology, make medical transcriptionists (MTs) central to any implementation, says Jamie Collins, RHIT, CMT, a senior implementation consultant at 3M. “Many times, MTs who are not familiar with speech recognition are not sure what to expect and may feel threatened that their jobs will go away if they embrace the technology,” she says. “It needs to be made very clear to the MTs that they still are the brains behind the documents, as all speech-recognized results must be verified and edited in order to produce quality results. Their roles will change from transcriptionists to transcriptionists/editors, but they are still a vital part of the process.”
Gary C. David, PhD, an associate professor of sociology at Bentley University in Waltham, Massachusetts, notes that most facilities make the mistake of believing that going live with speech recognition technology is the end of the implementation process. “But instead, it’s really halfway,” he says. “Hospitals need to know what the postimplementation plan is. Are you looking at quality and usability for coders and patients? Who will track how these notes fare in audits? These are the kinds of factors that must also be taken into consideration.”
Strategies for getting physicians up to speed vary by specialty, David says. “Are the physicians radiologists, psychiatrists, or anesthesiologists, for example? How the organization prepares its physicians should be different for each field,” he says. “The larger challenge is from that strategic perspective. Are we attempting to implement the technology in a one-size-fits-all manner? Overall, hospitals need to think strategically about what modality is the best fit for which medical practice workflow.”
Occasionally, physicians who are told speech recognition is on the horizon will feel compelled to change their natural speaking habits, which creates problems and stress. “The speech recognition software will recognize a physician’s style as long as he or she is consistent,” says Victoria MacLaren, product manager at SpeechMotion. “So when physicians are encouraged to speak in ways that aren’t natural for them, they actually become inconsistent in their speaking style, and speech recognition engines have a harder time adapting to them. This is especially an issue with back-end systems.”
Lynn Kosegi, director of solutions realization at M*Modal, recommends organizations ease physicians into speech recognition by offering them a hybrid workflow of both front- and back-end technology. “Physicians will embrace the technology much more readily if they know that they won’t be forced into a workflow that does not fit their respective schedules,” she says.
For example, if physicians need documents quickly, front-end speech recognition would be the wise choice. However, they still would have the option of having an MT review and edit the document for later use. “Even if the technology would work for physicians some of the time, they might reject it all of the time if they feel forced to embrace it,” Kosegi says. “But if they have flexibility, then they are much more likely to adopt the solution. It’s a matter of physician education. By showing the physicians the benefits of both front-end and back-end solutions and reassuring them that they will have both options, they are much more likely to be successful with the technology.”
Collins says educating physicians on how best to use the technology to improve turnaround times and documentation and familiarizing them with dictation guidelines can help ease the transition.
MacLaren believes it benefits facilities to pilot the technology prior to implementing it throughout the organization. “They need to observe what the physicians are doing with regard to existing workflow to highlight areas where speech recognition can improve upon existing practices,” she says. “If a decent sample of physicians from across the organization is willing to work with it, they’ll have a leg up at the time of rollouts.”
Conversely, not involving and preparing physicians can create issues, a lesson learned firsthand at Bryan Medical Center. “Our viewpoint was that physicians didn’t have to do anything different, so we didn’t need to take steps to prepare them,” Frame says.
One unexpected obstacle encountered during the rollout required additional steps. “Before we implemented speech recognition, physicians used hardwired dictation stations. With the new technology, it wasn’t feasible to use hardwired dictation stations,” she says. “Because this presented a challenge for some physicians, we had to work with telecommunications to turn regular phones into dictation stations, and some physicians struggled with a learning curve. This issue affected emergency department physicians more than others due to their high patient volume. Had I realized that the new technology would have had such an impact on those particular doctors, I would have spent time with them prior to go-live.”
Although front-end speech recognition should shorten document turnaround times, it’s best not to assume that outcome, Kosegi says. “There is the assumption that if a doctor creates the document himself using front-end technology without transcription, then it will be completed at the time of the patient visit,” she says. “However, the doctor may not have time to complete the document until that night or even on the weekend. Therefore, providing options is the best way to increase adoption, get the financial benefit, and decrease interference between the patient and doctor. In the end, the goal of speech recognition technology, whether front or back end, is to enhance and improve patient care and that should always be our goal.”
Adapting to a New Environment
With front-end speech recognition technology, physicians directly interact with the technology, making logistics a concern. For example, can physicians access the program away from the hospital?
No matter if it’s front or back end, clear speech input is essential. While most speech recognition systems can readily adapt to accents—particularly if the speaker properly enunciates—many systems struggle with rapid and/or slurred speech.
MacLaren says physicians should not have to adapt to the technology. In fact, it should be the opposite, with the platform learning and evolving based on MT and physician edits. “How quickly the software improves is related to the frequency of dictation,” she says. “The software needs to see enough examples before it can improve through any passive learning that may be ongoing. Users can configure recognition results, like dates, or add words directly to the software.”
In Kosegi’s experience, radiologists have had a high success rate with front-end speech recognition. “Radiology is a specialty that got into front end earlier than all others,” she notes, adding that the platform also works well for physicians who use a great deal of narrative when creating notes but is less effective for those who have to complete forms or use automated templates. “However, there are very few instances in which back-end speech recognition wouldn’t be appropriate.”
How to Prepare MTs
MTs must keep pace with technology and ensure their computer skills are up-to-date. If an organization adopts back-end technology and each MT receives the same training, supervisors will gain a greater understanding of who is doing well and who may benefit from additional tutoring. Supervisors then can route challenging projects to the more advanced MTs while offering additional training opportunities to those not yet up to speed.
Frame recommends using training sessions to allay any fears surrounding job security and to transition MTs to back-end editors. “I tried to prepare them by giving them educational activities, visuals of what to expect with speech recognition,” she notes. “I provided exercises to ‘retrain their brains’ to help them make that transition. That was a very important piece of getting speech recognition to come in and, for the most part, it worked. It started helping them think in a different manner than they had before.”
Bryan Medical Center’s transcription team had some trouble adapting to the vendor’s training program, which took place via a two-hour demonstration-type webinar. “Trainers were available by phone, but once the webinar was over, the visual component was no longer available,” Frame says. “That made it difficult for our MTs, who needed both the visual and hands-on aspects of the training. In the future, I’ll ask the vendor specifically how training is conducted and be sure that the training suits our team.”
Rollout and Beyond
While no speech recognition rollout is typical, it is often best if the process starts with the engineering team developing a test system. Schedule a conference call to outline expectations, create a training schedule, and set up dictation testing. Then enroll physicians to pilot the program and select appropriate document types—those that do not require much formatting are ideal. To “train” the system, previously transcribed documents are read into the system. After a comprehensive review, it is time to raise the curtain.
“Once the system goes live, lead MTs should closely monitor productivity and quality,” says Janice Jones, a senior implementation consultant at 3M. “Project leads should be looking for improvement and those who are not getting more accomplished.”
Postimplementation strategies rank high on David’s must-do list. “Hospitals have to continually monitor results and not assume the rollout has been successful using limited data over a short time horizon,” he notes.
Frame’s experience underscores David’s advice. “After almost a year, we are still dealing with an issue of the voice just cutting out completely and missing chunks of information or disconnecting in the middle of dictation,” she says. “It’s related to incompatibility of our phone system’s network configuration with the dictation voice server. Although we tested for a very long time and our telecommunications and IT people were involved, there was no way of predicting that. So we had to search for a solution, which means upgrading our telephone system sooner than planned, an added expense that also takes time.”
To further ensure the technology’s success, Frame suggests working closely with MTs long after implementation. “We communicate a great deal to make sure we have set up consistent rules,” she says. “We send out announcements and do town hall meetings. It’s not a one-week or one-day process. It’s more like three months or longer depending on the size of the facility.”
Collins says MTs must be aware they do not have to accept all speech-generated text. “For example, there may be instances where the speaker is trying to dictate in a noisy environment. This may result in a portion of the dictation having poorly recognized results,” she explains. “The MT needs to know it’s OK to discard the poor results and accept and edit the good results. Help them feel assured that the dictation quality should improve as the system adapts to the physician’s dictation patterns and encourage them to stick with it.”
Best Fits, Less Tantrums
For small practices that may not have the finances for a sophisticated workflow, Kosegi believes front-end technology may work best. “There is a place for both [front and back end] in most facilities, but there is not a particular type of physician or provider that couldn’t make use of front-end speech recognition,” she notes.
However, MacLaren cautions that front-end speech recognition may not be a prudent choice for every department in a large facility. Frame says some small physician offices with front-end speech recognition have staff dedicated to immediate chart editing. In a large hospital, however, such activity would disrupt workflow and be expensive over time.
Where physicians dictate can affect speech recognition’s effectiveness. “What’s happening in the background in terms of background noise makes a huge difference,” Frame says. “Sometimes the voice gets lost with the background noise. Physicians can be anywhere when they dictate. They may work in hospitals but are not necessarily just in a hospital setting.”
Because dictation is not always created in the ideal environment, Kosegi says selecting the appropriate technology is paramount. For patient records to accurately reflect care, it all starts with how data are captured, she points out.
It’s a process that begins with preparation and patience.
— Susan Chapman is a Los Angeles-based writer.