MEDICAL MESSENGER ARTICLES

Using Speech Recognition with Your EHR System

October 25, 2010

By Daniel Casciato
medicalofficetoday.com

Dr. Steve McCullough is a proponent of using speech recognition software with an electronic health record (EHR) system for two reasonsóit saves him time and money.

In 2005, his practice, Western Kentucky Kidney Specialists, began using an EHR and by early 2006 integrated it with speech recognition software from Dragon Medical so the physicians could dictate notes using speech recognition. Previously, each of the practiceís 12 physicians spent nearly $12,000 annually for transcription costs.

“This was a shared expense,” says McCullough, one of the partners of the Paducah, Ky.-based practice. “Even though some of the physicians were high utilizers, this expense was split evenly among the partners.”

After transitioning to Dragon, the practice changed its cost paradigmótranscription services became a line item expense, prompting eight physicians to switch to speech recognition, collectively saving the practice nearly a quarter of a million dollars.

The time savings is just as dramatic, adds Dr. Andres Jimenez, CEO of Las Vegas-based Implementhit.com, Inc., which helps practices make better use of technology, such as electronic medical records systems. 

“Notes would have to remain un-finalized until the audio file from the physicians is transmitted to the transcriptionist, the audio transcribed, the text transmitted back to the physician for review, verification for accuracy and signature,” Jimenez notes.  “Although in modern times that loop was typically completed in 24 to 48 hours, it was wrought for the potential for error on behalf of the transcriptionist, and additional liability exposure for the physician who would have to quickly review and signoff on sometimes 60+ transcriptions from a single day.” 

Additionally, itís nearly impossible for the physician to ensure the transcriptionist is accurate in every sentence of a transcription of an audio, spoken at about 100 words per minute, and the separation of time also makes this task difficult. 

“Voice recognition changes all of that,” says Jimenez. “Physicians can dictate and immediately see the text appear in their EHR. They can immediately review the text for accuracy, and sign off on the text while the patientís history was fresh in their mind.  The immediate impact on time and money is apparent as well as the long term impact by reducing liability exposure.” 

Slow adoption of speech recognition software

Several issues are still holding back speech recognition. First is the issue of accuracy; it was only until recently that voice recognition software has improved its accuracy rate. 

“Because locating, highlighting, and correcting individual errors are tedious, even a five to 10 percent error rate in accuracy would eliminate the speed benefit of voice recognition,” explains Jimenez. “This was the edge provided by human transcription that in recent times via computer hardware advances has been eliminated.” 

Jimenez says that voice recognition software like Dragon or Microsoft Vistaís speech recognition (free with Windows Vista) program, receive over 1MBs of data provided by the microphone and rapidly process the data to produce transcription accuracy rates as high as 97 to 98 percent.

“Even more impressive is those accuracy rates can be achieved with less than 1 hour of voice training with the application,” he says. “The major obstacle today for more widespread adoption however was that voice recognition did not produce discrete data, or codified data that could in turn initiate alerts or warnings, or help recommend appropriate billing levels for an encounter.” 

As a result, the text is meaningless from a database and connectivity point of view, and that wonít support the industry needs since an EHR needs to collect as much discrete data as possible to allow sharing patient information between two different doctors caring for the same patients using two different EHRs. 

“This is a center point of health information exchanges and by making it easy to share information, fewer tests are repeated because results couldnít be found, less morbidity occurs because even the most trivial of medical tests come with potential complications, and doctors can make better informed decisions on a patientís care because they have all of the information they need at their disposal,” Jimenez adds. “Over the last several years the capture of discrete data has been made possible. Natural language processing software takes the text generated from voice recognition and codifies it, exporting discrete data on the other side.”

McCullough believes that speech recognition has been slow to catch on for the same reasons why EHR hasnít quite caught on. “Prior to some of these mandates, and the initiating drive force from the economic stimulus package, physicians, overall, are people who are fearfully reluctant to change,” he says. “When you get into a routine, itís very difficult to change that attitude and to change that work paradigm.”

Dr. Robert Turner of the Kelsey Seybold Clinic in Houston says the biggest problem is the learning curve at the beginning. “You have to train the program to your way of speaking,” he says. “To become slick and fast with it, you need to do some studying in learn how to use it.”

Also, for those physicians who are not technologically sophisticated, they could struggle with it, Turner adds. “If you use computers all the time, some of the things that are involved in speech recognition software are intuitive,” he notes. “But if you donít, then youíre not intuitive, and you have to learn them all the commands. That means slowing yourself down; lessening your productivity; seeing fewer patientsóof course, that could be problematic.” 

Maximizing speech recognition in EHRs

If you plan to purchase speech recognition software, a good microphone is essential to maximize its effectiveness. A high quality microphone will ensure better accuracy and get you better results.

“Mics are a huge issueóthe software is only as good as your microphone,” says Jimenez. “If youíre using a package like Dragon, we suggest you use the ones they recommend on their site.”

Jimenez says that some vendors will also offer wireless connection, a plus for those physicians using a Tablet PC.

If you want to speed up even the most menial tasks pertaining to EHR, McCullough recommends taking advantage of the macro features that you can build with your software.

“Our IT person helped us create keystroke macros so when you finished a document you can use a voice command, like ësign document,” and the Dragon software would go through the steps to click the sign button, type in your password, and click enter, and the document was finished,” McCullough says. “Little things like that were added to speed things up. I also created document macros, where I can say ëultrasound report,í and the software would print out a document that had certain variables that I can go into and edit, and I donít have to dictate an entire note from scratch.”

McCullough also urges users to take their time with the software after you open it from the box and to try to use it as often as possible. Each software package comes with its own set of simple training tools. Typically, youíd have to spend about 10 to 15 minutes reading from pre-packaged texts to create your own customized audio file.

“The more you use the software, the better it will become,” says McCullough.  

Choosing a speech recognition program

While Dragon may dominate the speech recognition market, practices still like to have a choice. When looking at various software packages, the most important feature for McCullough was user-friendliness. 

“You cannot have a layer of complexity,” McCullough asserts. “When you are working in the office you need a particular workflow. If you have to stop what you are doing to troubleshoot problems, itís incredibly aggravating and ruins the workflow and that product will not work for you.”

No matter what product you ultimately choose, McCullough urges practices to give speech recognition a try. “If any physicians are somewhat computer savvy and are comfortable with computers, but the bullet and try it,” he says. “At about $1,600 per license for a package like Dragon, when you look at that expense, itís a drop in the bucket compared to other expenses you have.”

Turner says that the technology will only get better.

“Voice recognition software is not going away,” he says. “The difference between four years ago and today is like night and day. The improvement from five years ago is immense. My presumption is that the sophistication of the software will continue to improve going forward, and the progress four to five years from now will be just as significant.