Can you hear me now? Voice recognition is becoming the new standard

As the heaviest "dictators" in healthcare, radiologists are most likely to welcome voice recognition for dictating reports. Thorough planning, ample support, and improved efficiency and turnaround times have facilities talking up voice recognition's capabilities. Long looked at as a futuristic application, for many the future is now.


In on the ground floor

While some hospitals and imaging centers have made the switch recently or are in the process, Next Generation Radiology in Great Neck, N.Y., integrated voice recognition into its RIS-PACS when it opened its doors in 1998. Medical Director David Katz, MD, wanted to bypass the need for transcription altogether. The practice went through some trial-and-error but has been getting near 100 percent recognition with the Dragon NaturallySpeaking solution from Nuance for the past two years. Director of Information Systems Daniel Castaldo calls the product "exceptional. We've tried and tested other products and this offers the highest level of speech recognition."

Voice recognition for the practice's nine radiologists at four locations is second nature, Castaldo says. "Economically it works. It's definitely the way to go." The physicians can have their reports on a web site for referring clinicians - usually within 22 minutes of the study being completed. The efficiency means that the report is usually complete before the patient can even finish changing clothes and leave the practice. There have been occasions where the report results have led to the need for further imaging - and with the patient still at the practice, those additional scans could proceed right away rather than having to schedule another appointment.

Radiologists new to the practice, yet in an older age range, can find voice recognition and the PACS software a bit more challenging, Castaldo admits. But after their first few days of adjusting, "they love it," he says. They quickly appreciate the vast difference in turnaround time. The editing usually takes seconds as opposed to sending out dictation to a transcriptionist, getting the report back perhaps days later, and still needing to read through and edit the documentation.


Getting started

At United Hospital System, which serves southeast Wisconsin and northern Illinois, PACS Manager Jeff Sculuca is in the process of preparing to add voice recognition in the radiology department. PACS was installed about three years ago with little success. Now, he's working on replacing that system. Once the staff has had some time to adapt to the new PACS, Sculuca plans to integrate voice some time in 2006.

He's getting a mixed response from radiologists about voice recognition while he researches and evaluates vendors. Some physicians are hesitant and some are excited. With that in mind, he's planning for a system that allows the radiologists to use voice recognition but also fall back on transcription if they prefer it or if the voice system goes down. Those using voice recognition will have their reports dictated on the back end. Sculuca hopes to find a transcription vendor that can guarantee a maximum six-hour turnaround time 24 hours a day, seven days a week.

Once Sculuca is further along with the new PACS installation, he plans to schedule on-site demonstrations of voice recognition solutions for the radiologists. They can explore the various capabilities over the course of about a week and provide their feedback. "I know that these systems are learned over a period of time," he says. "One week may not be enough time, but it's enough to give some sense of which system is easier to use."


Patience pays off

Jean Plummer, manager of radiology informatics at Baylor Health Care System in Waco, Texas, has been watching the advances in voice recognition capabilities since 1997. In 2000, her organization decided to first pilot speech recognition within its radiology department. The goal was to improve turnaround time and reduce transcription costs. With an ongoing shortage of experienced, qualified transcriptionists across the country, Baylor was paying high rates for outsourced transcription. They also wanted to ensure that the voice system was fully integrated into PACS so that radiologists could pull up exams and start dictating reports right away.

Voice recognition has been in use at a small imaging center since May 2003 and throughout the health system since January 2004. The organization deployed PACS and made sure the staff was comfortable with that before introducing voice. Now, radiologists dictate as they view images. Some portion of the reports are self-corrected, Plummer says. After dictating, the radiologist hits the "done" button on the microphone. The voice system automatically converts the recording to a text file. That text appears on the PACS screen, the radiologist reads it, makes necessary changes, signs off, and then the report is immediately sent to other caregivers involved in that patient's care.

The dictation can be sent to the editors' queue for reading and correcting. "That's up to the physician," says Plummer. "I hear them saying that if it's a very long, complex report they prefer to send it off to an editor because of the time it takes the physician to review it." They'll also send off dictation when they are particularly busy. For example, at 4 a.m., almost every patient in the ICU gets a chest x-ray. Radiologists may send that first bolus of reports in the morning to an editor and then go back to self-correct mode. The editor service is outsourced and runs 24 hours a day, seven days a week. Baylor's contract, however, ensures that no work is done offshore.

From the beginning, Baylor's radiologists were very engaged in the move to voice recognition, Plummer says. In a specialty driven by technology, "they're very familiar with technology, so they are much more receptive to looking at what technology can do for them. That doesn't mean that they all love computers, but they understand the value it brings to their work."

A project champion assessed the group's greatest needs. Plummer and others worked closely with Dictaphone to ensure that enough advance testing was performed, and that enough specialists and trainers were on site for the go live. Each person received about two hours of training and the transition from the old system took a little over two weeks. "Both the vendor's resources and our internal resources were in the reading rooms for first few weeks, helping to reinforce the functionality that perhaps the physicians weren't as familiar with," says Plummer. "Despite all of the planning, there's always fear of the unknown and resistance to change. There were some people who were very apprehensive and some very excited." Of the 75 radiology residents and fellows, the majority now wouldn't give up voice, she says.


Turnaround up, frustration down

Very quickly after the go-live date, Baylor saw a turnaround time drop of two to three days down to four to five hours. Plummer also monitors how many completed exams don't have a final report. This used to be a big problem, and with missing dictation, someone had to redo the work. Now, it's rare to have something fall through the cracks, she reports. Plus, the radiologists are self-correcting their reports about 80 percent of the time immediately after the exam which means they are better able to review the information while it's still fresh in their mind rather than trying to remember the specifics three days later. Now, one-third of reports are signed within five minutes of dictation, she says.

The system has a lot of checks and balances, such as auditing tools Plummer can use to ensure that when a patient is imaged, that exam quickly has a corresponding report.

Before voice recognition, the radiologists were frustrated by transcription delays. They were getting calls from referring physicians looking for results. They also sometimes received well over 50 reports at a time to review. "They would have to carve out time to do that," Plummer says. That time often came evenings at home. "I think they felt out of control. They weren't able to provide the kind of care they wanted to provide."

The key was having PACS, she says, which made images immediately available. It gave them control of their own work and cut their dependence on outside resources. As a result, Baylor administration is pleased with voice recognition, says Plummer. "If the physicians are happy, they are happy."


Mandate reaps rewards

The University of Southern California Hospital first adopted voice recognition in April 2004, recalls Rasu B. Shrestha, MD, informatics director in the department of radiology. Shrestha and the department chairman have long been voice recognition proponents. The timing became right when the radiology group opened an outpatient imaging center.

The goal for the new center was being filmless and paperless with a full-blown RIS-PACS integrated with voice recognition. After deciding on a PACS from Siemens Medical Solutions, Shrestha tested and compared two voice packages that would integrate well with PACS. After one voice miscue, USC adopted Dictaphone's PowerScribe in January 2005.

The imaging center's radiologists were required to use voice recognition. Shrestha believes this no-option mandate contributed to their success. Another advantage was the new center itself as it wasn't trying to implement a new system in an older, established hospital. Shrestha also didn't offer the option of onsite editors. "I'm not a big fan [of onsite editors]. I think it defies the purpose of voice recognition."

In his experience at other facilities, having onsite editors prevented radiologists from taking the extra step to dictate clearly and the other habitual changes needed to suit voice recognition. This rule only applied to the imaging center. Because it was new, working at the center was considered a "prize" for a lot of the faculty. Shrestha "sold" time working there as an opportunity to experiment with new technology. Although some physicians expressed concerns at first that they would be doing transcriptionists' or secretarial work, demos and training sessions alleviated many of those worries.

The department chair was one of several champions behind the project who led by example. "It's really important to have champions who are comfortable with the technology," Shrestha says. "Others can see it working for them, which pushes the cause forward."

The chairman actually did the first demonstration of the system. Then, the group moved to individual sessions for more advanced training. This was particularly helpful to those with strong verbal accents. Shrestha expected them to face more challenges, but they actually latched on to the system, he reports. "They made a conscious effort because they were aware of their accents. We found that the system picked up on the various types of speaking of people with heavy accents better than rotating transcriptionists had in the past." Once they were used to the system, there were fewer errors than there had been with transcriptionists. Shrestha made every effort to spread the word on that kind of positive feedback.

Another contributor to USC's success was a phased-approach roll out. The facility went live with CR and CT with orthopedics first and then phased-in more departments. Shrestha divided the radiologists up by subspecialties so they could teach each other as they were phased in. It took about four months for all the radiologists in the facility to be comfortable with the system, no matter who was on duty. "We were certainly happy with that."

Aside from comfort with the system, the radiologists saw a dramatic drop in report turnaround time, reports Shrestha. It's gone from three to seven days down to just a few minutes up to an hour. That was one aspect of improved quality of patient care Shrestha and the rest of the team were aiming for. Another was avoiding the delay between dictation and final sign-off that is part of the traditional report dictation model. When a radiologist gets a dictated report back a few days later, he or she doesn't have studies or images to check against. "They're kind of in the dark. They have to go by memory," he says. That leads to errors, which is exactly what he wanted to curtail. "We saw that as bad patient care which could lead to misfindings. We made it clear that with voice recognition that would not be the case. The radiologists can sign off on their reports while the images are still open."


Steps to success

Shrestha also attributes USC's success to a good relationship with its vendor. For example, one or two radiologists had trouble with their voice files. Shrestha set up follow-up training for them and Dictaphone conducted high-level analysis of the voice files and made recommendations.

He also recommends a process allowing your facility's IT team to learn the system well so they can take ownership. If the IT group views the system as being under their purview, they will take the initiative to help with clinician issues, he says. Plummer's IT department even developed a physicians' portal, providing them with access to images and reports from any location.

Most important is well-planned training before you go live, as you go live, and after you go live, Shrestha stresses. Plummer says it is important to remember that this is one of the few applications installed in the healthcare environment in which physicians are the primary users. You must assess their needs and determine how it will fit into the ways they work, she says. They should participate in the evaluation process, she advises.

One more added benefit, according to Plummer, is the effect on radiology technologists. "As PACS and voice become more of the norm, [technologists] are looking for organizations that have invested in tools that help them do their job," she says. She anticipates the voice recognition system to aid in radiologist recruitment and retention in the future.