Speech Recognition Meets the Web
By Julie Knudson
For The Record
Vol. 28 No. 7 P. 16
Flexibility and efficiency come together in the documentation process.
Speech recognition technologies are being adopted by an increasing number of hospital systems. These platforms give physicians a new way to document patient encounters while providing health care organizations with a more effective solution for getting critical data into the EMR system with as little lag time as possible. Among the various software flavors, web-based speech recognition tools are becoming more popular as hospitals look for opportunities to remain flexible in a world where hardware and other capital costs are often difficult to justify.
It's important to note, however, that downstream changes are triggered whenever a web-based speech recognition program is launched. Providers may need to modify how they manage the documentation process, from finding the best times to document patient encounters to carving out locations for dictation that are quiet and convenient. For a hospital's implementation to achieve maximum performance, the team must understand where challenges lurk and what type of support physicians need most.
Documentation Speed, Length Impacted
A study published last year by Markus Vogel, a physician in the department of general pediatrics, neonatology, and pediatric cardiology at University Children's Hospital in Düsseldorf, Germany, shed new light on how web-based speech recognition platforms are shaping physicians' documentation trends and habits. Utilizing a browser window augmented with a Java script that calculated and checked every interaction the physician user had with the window, including mouse clicks and arrow keys as well as text input, the project examined where and how speech recognition tools impacted documentation practices and performance, including whether documentation speed was improved and length increased.
Results of the study were eye opening. Even though speed measurements included corrections because the system used in the study wasn't customized or trained for a particular user's input patterns, Vogel says, "We could see that, using the speech recognition program, documentation speed increased by 26%."
That figure is particularly surprising due to the variety of specialties involved, such as surgery and pediatrics, that were using the study's straight-off-the-shelf medical speech recognition system. "Of course, errors were occurring, so those had to be corrected, and the speed increase includes every correction being done by the physician," Vogel says.
In addition to a marked increase in speed, Vogel says, "The second thing that surprised me a bit was that the length of the documentation increased." Participants who had only their keyboard to document encounters typed an average of 356 characters per report. "Those using speech recognition entered 649 characters on average," Vogel says, an increase of 80%.
Together, he says the results "are really a baseline showing what happens if you have no adaptation or training." If a hospital conducted the typical training—speech patterns and speed, dialects, the spelling of often-used terms—Vogel expects normal usage would reflect even larger speed and character count increases.
John Lee, MD, chief medical information officer at Edward-Elmhurst Health in Naperville, Illinois, was part of a speech recognition implementation in the emergency department in 2010 for a platform that wasn't web based but still had a measurable impact on documentation length, speed, and physician engagement. In the previous workflow, physicians documented following patient encounters, an inefficient process on several fronts. "Very often, by the time things had closed, there were so many other things going on that we couldn't really document as well contemporaneously," Lee recalls. "I typically ended up having anywhere between one to two dozen charts at the end of every shift that I had to dictate because I just didn't have time."
When the switch was made to a speech recognition solution, the process changed. "I could document my initial impression from my physical exam just minutes after I did the initial encounter," Lee says. The note could be opened and primary thoughts gathered, with the option to go back and fill in additional details and provide better clarity on the encounter once the patient visit had been completed. Results from the implementation told a tale of increased productivity for physicians and improved documentation. "The workflow ended up providing much more accurate documentation, and it also saved me a lot of time," Lee says.
Rush University Medical Center in Chicago also looked to speech recognition technology for help in improving documentation efforts and addressing physician workflow concerns. Kamlesh Patel, MBA, CPHIMS, PMP, a systems analyst at Rush, says the team was surveyed prior to launch and the results were dismaying. "Twenty-five percent of our physicians were dissatisfied with the quality of clinical documentation, and 75% were dissatisfied with the amount of time it took to complete their documentation," he says.
Overall, physicians at Rush were having more difficulty documenting within the EMR system than they had expected. The organization decided to launch an eight-week pilot program with a front-end speech recognition platform to help smooth out the wrinkles.
The results from the pilot were telling: Patel says more than 87% of participants felt they spent less time on clinical documentation while using the speech recognition system. That was just the start of the findings, however. "Seventy-five percent felt the quality of the documentation improved," Patel says, noting that much of the documentation included in the pilot program was lengthy. "We're talking about longer consults, history of present illness, assessment and plan, and things of that nature."
Interestingly, some participating physicians reported they found shorter pieces of documentation were faster to type rather than run through speech recognition.
Using speech recognition to complete documentation nearer to the patient encounter may result in more detailed notes and may be complemented by additional clarifications directly from the patient. "You're no longer doing your task and then waiting until hours later to come back to it," says Jonathon Dreyer, director of cloud and mobile solutions marketing at Nuance Communications. "The time lapse between when the actual activity happens and when it gets documented is compressed, and the richness of the details are retained as part of that."
Some physicians choose to begin documenting the patient encounter while it is still happening, a practice Dreyer says may result in increased patient engagement. "We're hearing from physicians that tell us patients are actually helping to add clarity to the information as it's being documented," a scenario that wouldn't typically happen as part of conventional transcription workflows, he says.
Physicians shared their feedback with the speech recognition trial run at Rush directly with the physician champion who helped shepherd the pilot program. As the pilot was nearing its end, the system's usefulness and its impact on providers became apparent.
Patel says several physicians were reluctant to relinquish their access to the speech recognition program when the eight weeks were over. "They said, 'We don't want to give up the license because we aren't here until 10 o'clock at night anymore. We're done at a decent time and we can go home and spend time with our families,'" he says.
One of web-based speech recognition's major contributions to physician satisfaction is its flexibility. Traditional desktop stations have been joined by a broad field of devices that can be used to interact with the hospital's systems, opening the door to more fluid technology interactions for physicians. "A tethered microphone almost feels like it's tying you down," says Paula Pasquinelli, vice president of implementation and adoption services at M*Modal. Being able to use a mobile phone as an input device, for example, addresses that age-old restriction. "Smartphone usage with front-end speech has been a big physician satisfier," she says.
Device flexibility, in turn, opens additional doors for providers. "We've seen a lot of physician satisfaction because they want to get out of the hospital, they want to get out of the clinic," Pasquinelli says. "A lot of these providers want to have the flexibility to be able to finish their documentation at home." Accessing the hospital's EMR in a virtual environment via a mobile device—and retaining good performance from the platform—supports the ability to complete documentation at a time and place that suits physicians' busy schedules.
"There has been a general shift in trends of how physicians are just inherently more and more mobile," Dreyer says. "They're in more places, on more devices, doing more things at all these different times in the day." And though smartphones and tablets are an increasingly popular option for documentation, many sit along desktop terminals for a long time, he says. Web-based speech recognition tools take that into account and are designed to offer a consistent experience no matter where or how the physician wants to document. "As providers move between devices, it becomes device form-factor agnostic," Dreyer says. "That really contributes to speeding the input of that information."
Access to speech recognition remains limited in many regions. It requires an investment that some hospitals aren't in a position to justify, as well as buy-in and ongoing support from IT, which can be a difficult prospect in organizations running lean technology teams.
Nevertheless, participants in Vogel's study were grateful for the usefulness the platform provided. "They were loving that they had it," he says. Although some providers encountered challenges navigating the structure of the study's particular software offering, Vogel says, "The majority were very happy with it. What they really missed after the study was that the platform wasn't available anymore."
One challenge hospitals are likely to encounter is getting physicians up to speed on the new workflows created once web-based speech recognition becomes an established part of the process. Pasquinelli says the specific hurdles providers face can vary from one EMR to the next. "Some EMRs are pretty straightforward," she explains. "Getting to certain notes, accessing your inbox—the workflows are not that complex." Others, however, may not integrate as cleanly with the selected speech recognition solution.
Hospitals want to make users' early experiences positive, but Pasquinelli says, "In every case, even with more simplistic EMRs, we find that going back in and doing optimization with the physicians—more personalization—is really necessary." She refers to this follow-up support as "one-on-one, at-the-elbow attention," and stresses its importance in long-term performance of the platform and provider satisfaction.
Some workflow effects may be unexpected, originating with parties other than the providers using the technology. Lee says his team's implementation created ripples far and wide, leading to several curious developments. "Once we started doing this and word got around the hospital, we received feedback from the inpatient doctors and nurses that they knew we were documenting contemporaneously," he says. For example, when a bed request was received, clinicians in other parts of the hospital would quickly see what the emergency physicians had already documented. "That way, they could get an initial sense of what was coming even before I had talked to them," Lee says.
Challenges can be more stressful when the evaluation period is relatively short, as it was during the eight-week pilot program at Rush. "Provider satisfaction wasn't very high initially," Patel says. "It was due to growing pains—primarily learning the system and the appropriate use of it."
Input from the physician champion helped, a development Patel says was an important factor in the improved documentation quality and physician satisfaction results the team ultimately achieved. "We basically tag teamed," he explains of the work he and the physician champion did during the pilot. "You need to have some credible buy-in and also some expertise to help guide a successful pilot."
Someone other than a physician can try to outline the potential benefits of speech recognition, but Patel says, "If a physician champion who has gone through some of these pilots before goes in and says, 'Let's try this specific application,' it makes a vast difference."
The ability of an organization's infrastructure to support the selected application plays a key role in a successful outcome, Patel says. "There's nothing more frustrating to a physician than an application that is kludgy or slow on response, or that just doesn't function because the infrastructure doesn't support the necessary overhead," he says.
Resources such as early training on the platform as well as a help desk that can handle postimplementation difficulties can boost the chances of a project's success. "Have someone the physician can call if they have a question at any time," Patel says.
Broader technology issues also must be considered as hospitals choose, configure, and implement speech recognition tools. Web-based platforms in particular have requirements with the potential to directly affect the providers' ability to use the system. "All the data processing for this type of voice recognition is done through the internet," Lee explains.
Not only will physicians want to have a robust connection on their end—whether that's work, home, or somewhere in between—but the hospital also must pay close attention to its bandwidth and internet reliability. "If you have an internet outage or server problem with your voice recognition provider, then you're out of luck," Lee says, adding that redundancy, from internet pathways to server resources and vendor connection points, should be considered up front.
In Vogel's study, participants received only a small amount of training—speaking a short training clip—with the speech recognition system. "It wasn't the scope of reading the text so much as it was knowing the system, getting used to it, and understanding how to interact with it," Vogel says.
Although minimal, it may be a realistic picture of the level of early-stage training many speech recognition implementations can actually support. "Especially [for] those university hospitals that are large, it can be very difficult to reach every physician and take the time to do extensive training," Vogel says.
Rolling out a new platform may be most successful when physicians are given the option to evaluate where and how it will impact their workflow. "I think if the hospital is thinking of adopting speech recognition, it's very important to give it to the physicians who want it and who can judge what the advantages are for them," says Vogel, who suggests minimizing workflow disruption wherever possible. "I think the workflow they had before shouldn't be changed too much unless the organization can prove the benefit of the altered workflow. Otherwise, it's just more technology without any clinical usage."
— Julie Knudson is a freelance writer based in Seattle.