After years of waiting to be the next big thing, speech recognition has finally hit the tipping point with excellent accuracy rates, few bugs and smooth implementations. The technology is now used by almost half of the academic and approximately a quarter of private practices in radiology. With clinicians recognizing the need for productivity gains and more detailed and accessible documentation, and administrators interested in patient safety and cost savings, the technology keeps adding more fans.
A few more speech recognition implementations like the one that occurred at Washington Hospital Center in Washington D.C., and the technology might just lose its reputation as being challenging to adopt. Savings in time and money are just some of the many benefits the hospital has experienced in the past year. When the radiology department implemented MedQuist’s SpeechQ for Radiology speech recognition technology last January, it was the culmination of two years of due diligence and preparation.
Director of Radiology Gayle Thompson Smillie, CRA, RT, says it was “a well thought-out process.” Once the decision to implement the technology was made, a team narrowed it down to two vendors and conducted two customer site visits for each vendor.
Room for improvement
The hospital was outsourcing about $450,000 a year in transcription, despite having five full-time, in-house transcriptionists dedicated to radiology. The facility’s 26 radiologists performed 270,000 exams last year at the 900-bed hospital. Average report turn-around time was 38 to 42 hours.
The entire implementation process took about five months. A team from the hospital went to a week-long training class and then served as resources once the facility went live. The PACS administrator became the speech system administrator. The radiologists had been on PACS for several years, Smillie says, and they “had picked up a lot of efficiencies through that. We kept digital dictation as a back up.”
Although some facilities institute a drop-dead, go-live date, eliminating digital dictation, “we absolutely took the opposite stance,” says Smillie. The goal was letting the radiologists learn the new system at their own pace with as much help as they needed. “We offered enough support mechanisms until they all felt comfortable,” she says. The physicians could go back to digital dictation if they felt overwhelmed by speech recognition. For the first couple of months, they used it about 30 percent of the time; after that it was rare.
Today, report turn-around time is about 14 hours. For all ICU and ED cases, the radiologists are required to dictate, self-edit and sign off immediately, Smillie says.
Alegent Health, a healthcare system in Nebraska and southwestern Iowa with nine acute-care hospitals, implemented FusionText powered by SpeechMagic from Philips Speech Recognition Systems and Dolbey about 18 months ago. Although the organization planned on a 9 percent increase in dictation volume, they’ve had 17 percent without having to increase staff. Physicians now include more details in their reports, too.
“Our volume of dictation has just skyrocketed,” says Michael Westcott, MD, chief medical information officer. “We were looking for something to help with that expense.”
Alegent implemented speech recognition enterprise-wide to have the most impact, says Joanna Christensen, operations director, document management. “We’re getting the bang for our buck that we wanted.”
About 300 Alegent physicians are doing back-end recognition and an editor does corrections. Despite some interface troubles for front-end recognition, Westcott expects implementation of that in the radiology department and perhaps the emergency department as well, within weeks. “The goal is for the doctors to see the report, make minor edits and sign off at the same time so they don’t have to handle a document twice.”
About a quarter of Alegent’s transcription volume is going through FusionText—500,000 lines every two weeks. “We’ve been able to increase transcription lines by about 25 percent without adding FTEs because a transcriptionist working as an editor can get through more reports much quicker,” Westcott says.
Westcott says Alegent’s physicians have been very interested in any efforts that can reduce report turn-around times. With back-end speech recognition, report turn-around times went from about two hours to about 30 minutes. The most beneficial improvements are getting dictations done as quickly as possible and now other clinicians can see the reports remotely. Plus, he says physicians now are including more details in their reports. It’s routine these days to have five to seven doctors on a case and “they’re lucky if any of them can read each other’s handwriting,” says Westcott. So, speech technology can certainly help.
Pick up the mike
Despite the benefits, Alegent has weathered some minor storms. Westcott says that a lot of physician documentation doesn’t lend itself well to using templates and macros. In some cases, “you want to use the patient’s own words about the situation that brought them into the hospital or clinic. So, I don’t think transcription will ever go away.”
He adds that the physicians at Alegent also found that the technology wasn’t successful unless they used a speech microphone. The use of phone lines led to uncontrollable background noise and static on the line.
The organization also underestimated how much work the front-end interface would require. “I think we might have applied more resources upfront,” Westcott says.
But, the physicians are pleased with the system’s ability to recognize complex medical terminology and adapt to different styles and accents. For most users, it took just a few hours of actual transcription before the engine got very good, Westcott says.
Christensen advises that you recognize that “speech recognition isn’t a panacea.” For the best results, “you have to do a critical analysis of what is needed in your organization and who needs it.”
Few bugs left
“We’re very, very pleased,” Washington Hospital Center’s Smillie says. The hospital experienced “an extremely fast return on investment—less than one year,” she adds. However, “the financial savings were secondary to the fact that we could provide almost instantaneous results to clinicians. That was really the driver.”
“Sometimes I am awed that it wasn’t more difficult,” Smillie says. She had heard that radiology is the most painful implementation for speech, experiencing about 50 percent pushback from radiologists. She had some pushback, but says the radiologists were ready for the technology.
These kinds of implementation successes will probably only continue, Smillie says. “I really think that [speech recognition] has very few bugs left. We’ve been looking at voice recognition for 10-plus years. Earlier products failed for a reason—the technology just wasn’t there.”