February 4 , 2008
Speech Recognition in the ED — and Beyond
By Laura Gater
For The Record
Vol. 20 No. 3 P. 20
Even in the hectic ED environment, speech recognition technology is helping to decrease costs, increase efficiency, and speed clinical information through the network.
Speech recognition (SR) is making inroads into all areas of hospital life, including the emergency department (ED), an area not usually associated with the technology. One may think that the ED is not an emerging market for SR use, but with ED physicians’ time at a premium, SR is increasingly viewed as a logical option to help ease department workflow problems.
Physicians are finding that although electronic medical records (EMRs) are an excellent way to access and store patients’ information, they are often limited in menu choices and selections. With a good SR program, physicians can dictate a more complete and viable record of patient ED visits, thus maintaining the EMR’s integrity and usefulness without feeling constrained by a drop-down menu.
However, SR must overcome obstacles if it is to become a staple in the ED. “Speech recognition is not being used in emergency rooms in comparison to its use in clinics and with private physician groups. Speech recognition use in hospitals has not made significant inroads because the physicians can dictate for free in the hospital dictation system—where is their motivation?” says Michael Bliss, MA, a national SR consultant and author of Speech Recognition Technology for Health Professions. “Physicians who pay for their own transcription in their private practice and/or clinic appear to have better success because they are motivated to use speech recognition to eliminate transcription costs. People will change their behavior if it impacts them directly.”
Inside the ED
Those physicians fortunate enough to have SR in the ED appreciate its efficiency, ease of use, and applicability to their patients’ EMRs.
“The ED is a busy, noisy environment, very challenging. It represents a broad range of clinical challenges,” explains Nick van Terheyden, MD, chief medical officer for Philips Speech Recognition Systems. “Developing a solution that suits ED was quite a challenge. Radiology was first and easy to adapt. At Sir Mortimer B. Davis Jewish General Hospital [in Montreal], they identified the need to give physicians a tool and provide the technology that actually helps and supports staff in the ED. In testing our SR product in the ED, we were able to capture voice from physicians despite the busy, noisy atmosphere that provided a very dynamic workflow for physicians with flexibility and mobility and integrated with their EMR.”
Mount Carmel St. Ann’s Hospital near Columbus, Ohio, went live with its SR and EMR in 2001. Oren Leidheiser, MD, the hospital’s ED director and chairman, recalls that the “drop-and-click” menu he and his staff used prior to installation didn’t capture unique patient proceedings or the characteristics of ED visits. The software did not offer any method other than typing to capture certain elements of a patient’s visit, which the ED physicians did not have the time to do. Now, however, “Voice recognition allows us to capture unique features of patient encounters,” says Leidheiser.
Some SR software programs provide an emergency medicine-specific vocabulary, which increases accuracy levels, and continuous speech dictation, which enables ED physicians to dictate a patient-specific narrative without being limited by menu choices that are part of an EMR system’s clinical content. In some programs, a voice may trigger predefined text macros, which reduces the number of clicks and pull-up templates.
“We felt that our original SR tool was not efficient, precise, or detailed enough to meet our requirements,” says John Robinson, MD, an ED physician at Allina Hospitals & Clinics in Minneapolis. “We developed an electronic template, which pulled in some information from our EMR and also used Dragon voice recognition software to provide more specific documentation. It’s also very useful in documenting what happened during the ED visit and during the medical decision-making process. In addition, we use Dragon to document MRI, x-ray and other diagnostic and lab tests, ultrasound tests, and responses to treatment and medications.”
More than 100 physicians at Allina’s five hospitals and clinics have been using SR technology integrated with an EMR for more than two years. SR has replaced paper templates and streamlined patient information gathering and sharing.
The integration of SR and EMR technology often depends on the text editor. With Microsoft Word, there is no issue concerning SR working well in the EMR. If an EMR company has developed its own proprietary text editor, sometimes SR functionality becomes a concern. For example, the command of “delete previous three words” may not work in the proprietary text editor as it does in Microsoft Word and DragonPad.
“There may be a 20% or 30% loss of cursor mobility skill sets, depending on how the text editor was designed,” explains Bliss. “There have been some circumstances where the text editor does not accept unique icons that a physician may want to insert in the note. ... Dragon NaturallySpeaking has a dictation box that can be activated and utilized with text editors of the EMRs that do not operate well with cursor mobility and command functionality. The provider dictates in the dictation box and then executes an enter key, and the text is then inserted in the text editor of the EMR. This works; however, it is an extra step of inserting text into the EMR text editor window.”
Experts agree that SR technology is useless if it does not positively influence patient recordkeeping.
“A well-designed speech recognition program and an EMR system and a tightly integrated workflow will ensure that speech recognition is an aid, not a hindrance,” says van Terheyden. “Medicine is now team based. Maximizing the opportunity for communication, including rapid capture and distribution of information to the whole team, is the key to success for high-quality care. Speech is the most natural form of communication and helps medical providers provide a higher quality of care.”
The benefits of SR are numerous, according to Bliss. “Speech recognition provides immediate feedback/notes inserted into the EMR as opposed to waiting 12 or 24 hours from the transcriptionist, then reviewing the note and possibly making changes to its original format, and having the document retyped once again,” he explains. “Speech recognition allows billing to be sent out the same day the patient is seen when used with an EMR because the coding and the notes [billing] are ready to send at the end of the day.
“Some physicians enjoy dictating in front of patients using speech recognition, and it helps to validate the accuracy of what goes inside the record,” he continues. “Some physicians have been very thankful that the patient said, ‘No, Doc, it was the left knee, not the right knee.’ Providers have informed me that patients are very appreciative of knowing what is going inside the record when they use speech recognition and dictate in front of the patient. Some physicians have reported that their notes are more exact, specific, and terse, where once before they would dictate too many superfluous issues. Other physicians have responded that they are billing more accurately and dictating more specific notes on the patient, providing more detail with the use of speech recognition. In many cases, physicians have improved the quality of their dictation because they realize they are poor dictators when using speech recognition. Dictating in a clear, enunciated voice improves recognition accuracy, and it brings to light to the providers that they need to take responsibility for speaking clearly when dictating patient notes.”
Unlike primary care providers who have returning patients with a past medical history that is already documented, ED providers treat patients with a full triage, patient/family medical history, and physical exam. According to Nuance Communications, one third of all hospital admissions typically arrive from the ED, so it is critical that documentation be complete at the time of admission. SR enables physicians to provide more complete notes in a shorter period of time.
For example, Robinson uses the ED computer workstation after seeing a patient to document the patient interview, patient history, examination, and treatment. SR enables him to dictate his notes immediately, without having to write notes to himself or worry about forgetting something.
“My notes are much more accurate and precise the sooner I document everything after seeing a patient,” he says. “SR enables me to include more detailed information, including smaller items that may be important later. Efficiency is absolutely a crucial element because of the time crunch due to having no scheduled appointments in the ED. SR enables me to build a better overall patient record. If that patient returns to an ED of any hospital in our system, my document is there for anyone to review—it’s on the record in real time. Dictation has a lag time of at least several hours or a day before the medical record is available.”
SR is also a crucial element in maximizing reimbursement for each ED visit and reducing denials. The more information that the ED physician provides, the more likely a visit can be coded at an appropriate level rather than relying on clinical documentation menus alone.
The MT’s Role
Many medical transcriptionists (MTs) who have tried SR need to understand the importance of selecting the right SR software, says Bliss. This is just as important for physicians.
“If an MT wishes to use speech recognition to ‘redictate’ as opposed to typing, it will be important for him or her to create separate vocabularies for each physician. An MT who types for a cardiologist and a neurologist can purchase the medical version of speech recognition. The MT can wear an earpiece connected to the recorder and wear a headset for speech recognition redictation. The MT will need to use a separate vocabulary when redictating for the cardiologist and switch vocabularies when redictating for the neurologist. A common mistake of MTs using this type of transcription methodology is that they often make the mistake of redictating inside the same vocabulary for neurology and cardiology,” explains Bliss.
“Recognition accuracy will improve if you build a proprietary word list and grammar model for neurology and build a separate vocabulary and word list for cardiology. The MT must then switch between these two modalities when redictating for the physicians,” he adds. “This methodology is especially useful for the MT who only can achieve 40 or 50 words per minute. Typing is a cognitive motor skill set that requires, at a minimum, 30 hours of instructional practices to reach 50 words per minute. An MT can redictate at 100 correct words per minute using this methodology; therefore, the MT who has problems developing efficient typing skills can double or triple their productivity level using this redictation method.”
The standard and preferred versions of SR software have inferior headsets, according to Bliss. A superior headset is important for noise canceling, which is crucial for improving recognition accuracy. Bliss says MTs cannot get by with SR’s standard version, but they can purchase the preferred version and purchase a high-quality headset. MTs can then build a proprietary word list specific to the medical practice and import those words and terms into the preferred version to achieve significant recognition accuracy.
Embrace SR Technology
“I stopped typing eight years ago. I use speech recognition in Microsoft Word, Excel spreadsheets, PowerPoint, e-mail, etc,” Bliss says. “Many MTs are afraid of speech recognition. I believe they should embrace it. I trained four MTs to utilize the redictation methodology. They asked me to not share with their physicians that they are not typing the medical notes from the practice. The MTs were average with their typing skills and tripled their productivity with the use of speech recognition.”
MTs can also have their clients dictate into digital recorders and then pass it through the SR engine to create the first draft, according to Bliss. In this scenario, he notes, the MT is then a medical editor with a very different skill set.
“Listening to a WAVE file and typing a document from scratch, a cognitive auditory motor skill set, is very different than listening to the voice and watching on the screen for recognition errors,” he says. “Training is a critical element in learning how to use digital recorders and downloading them to the speech recognition engine. I have one physician in Tennessee using this methodology, and he dictates approximately 18 patient records in the morning and 15 patient records in the afternoon. He hands the digital recorder to his staff at lunch, and they download the files to a very fast desktop PC. The office staff opens up the voice profile of the physician and points it to the speech recognition WAVE file that was transferred from the digital recorder. The patient records are then transcribed in one event. The transcription ratio time is three minutes of dictation to one minute of transcribed text, dependent upon the speed of the computer. I trained the two in-house staff persons to be the medical editors for the practice. It took a full day of one-on-one training for them to understand the operation and how to be an efficient medical editor.”
Leidheiser has occasionally encountered words that his SR program doesn’t immediately recognize, but it asks him to spell them or select the right word from a list that it provides. The program then remembers that word in that context and will spell it correctly in the future. He dictates letters or meeting minutes on a Sony Dictaphone, which he later connects to his laptop computer, and the software converts his words to text. He uses Dictaphone to respond to his e-mail rather than typing responses. Leidheiser frequently utilizes macros, such as a standard letter to respond to correspondence. Each of the macros, or letters, is named, and recalled by that name.
“Mount Carmel St. Ann’s Hospital used to spend $450,000 every year for transcription. The goal was to reduce this amount by 75%. We haven’t dictated anything in over five years, and the cost savings are absolutely huge,” says Leidheiser. “We will see SR more and more as technology gets better and better. A lot of people got really turned off four to five years ago.”
The Future of Wireless SR
“Wireless speech recognition is just another methodology for voice capture that has challenges for the level, reliability, and quality of the audio that can be recorded using a wireless device,” explains van Terheyden. “We’ve seen improvement in that technology. People are looking for flexibility. Wireless enables physicians to be more mobile away from hardware that translates speech.”
van Terheyden admits that Philips has “struggled” with developing Bluetooth technology. “The challenge is that Bluetooth wasn’t really designed to capture high-quality audio and transmit it effectively. The quality is insufficient for speech recognition. We’ve customized some Bluetooth headsets to capture good audio. It’s not being used widely—it’s still in its early days,” he says.
— Laura Gater’s medical and business trade articles have been published in Healthcare Traveler, Radiology Today, Corrections Forum, Credit Union BUSINESS, and other national and online publications.