Strictly Speaking, Voice Recognition Technology Works

Flying cars. Teleportation. A cure for the common cold. Voice recognition.
The future always seems to never arrive. But get ready to start checking that last one off your list.

I've been following efforts to get voice recognition going since, well, Captain Kirk spoke to the Starship Enterprise's computer back in the original Star Trek TV series in the 1960s.

Now, there's a hospital in Chicago that would recognize everyone on that bridge from Kirk to Mr. Sulu and even Ensign Chekov.

At Advocate Illinois Masonic Medical Center, a 410-bed hospital south of Wrigley Field, the future has arrived. Eighteen months into a full rollout of voice recognition–powered technology, physicians and other staff have replaced their paper notes with 25,000 voice- and keyboard-generated notes per month.

Equally powered-up is Adem Arslani, MS, RN, director of information systems and clinical informatics at Advocate Illinois Masonic. He had plenty to say (naturally) when we spoke last week.

Voice recognition turns out to need a few things that weren't around in the 1960s when IBM first started attempting it, using the entire power of a mainframe. Those things, in decreasing order of importance, are: lots of CPU processing power, really good microphones with noise cancellation technology for noisy hospital environments, and software that knows when to suggest something and when to get out of the way. The final, nontechnological, component is commitment from the CEO on down to make it work.

Today the technology is on such a roll at Advocate Illinois Masonic that the CEO makes time to gently remind those straggling physicians still dictating their notes for expensive transcription that there is a better way that costs less, and it's running at every desktop in the facility.

The transition started, as it often does, with a crisis, in the form of the Joint Commission, which a while back visited Advocate Illinois Masonic because of some issues about illegible paper documentation.

With an imperative to improve documentation quality and drive adoption of the hospital's electronic medical record, Arslani and his team selected a leading voice recognition technology and set about creating numerous templates and macros to implement Advocate Illinois Masonic's physician workflow at the service level.

The base technology and add-ons customize the experience for physicians and organize all their commands and templates in a way that can be accessed from the handheld microphone they use for voice input, Arslani says.

"The first thing the physician will do is push a button … so they can see all their templates," he says. "They just select a template or speak into it to generate that template right into the note."

Arslani's got it even worse than the bridge of the Enterprise, with 27 different languages spoken at Illinois Masonic. But the software is up to the task. "The great thing is it actually compensates for heavy accents and the accuracy is quite impressive," he says.

When the project began, Advocate Illinois Masonic spent between $40,000 to $45,000 a month in transcription costs, and employed 6 ½ full-time-equivalent internal in-house transcriptionists. Today, the costs have plummeted to $3,000 to $8,000 a month and the FTEs are gone.

Only 30 to 40 physicians still dictate for transcription, and these are physicians who don't practice frequently at the hospital. By July, to entice these stragglers to join the voice recognition revolution, Arslani's team will let them perform voice input even from their home computers, using ordinary PC microphones which are sufficient in the quieter home environment.

Other incentives will help get stragglers there as well. Templates in the software can prompt physicians to comply with various regulations, Arslani says. For instance, all procedure notes must include the following statement: "The integrity of all instruments and equipment used on the patient during the procedure remained intact after use." All the provider has to do is use a command to insert the sentence into the note.

Now, no system is perfect, and if speech has an Achilles' heel, it's probably the ambiguity built into the English language. "Sometimes it picks up a different word, and we even show that in our demonstrations," Arslani says. "Down in medical records, they pick these things up and get right to the physician. After a while, physicians are cognizant of that. But I've got to tell you, [overall] it makes the doctor much more efficient. From the quality of the documentation, it is far superior than the written note."

That difference is one of quantity as well as quality. Physicians do five times as much documentation as they did on paper systems, Arslani says. "Before, we were only seeing maybe one note per patient day," he says. "Now we're seeing 3 ½ to 4 ½ notes per patient day. That's incredible!"

The voice recognition tools make EMR adoption essentially a done deal. Advocate Illinois Masonic attested to Stage 1 of Meaningful Use last September, and is on track for Stage 2, Arslani says.

"We've worked awfully hard these past 3 ½ years to get where we are, and I have to tell you our demand for site visits has gone through the roof," he says. A CIO from Australia who recently visited pronounced it the highlight of his site visit, he adds.

If you want resource-intensive voice recognition software, get ready to roll out current PC hardware if what you have is aged. Arslani recommends PCs with 2 gigabytes of RAM minimum, even better with 4 gigabytes. A 100 megabit-per-second local area network is a must. Advocate Illinois Masonic opted for 1 gigabit-per-second for each desktop. These days, that level of throughput doesn't cost a lot more.

Part of what triggered my interest in voice recognition was the popular column I wrote last week on whether EMRs are killing the traditional narrative. We're at a dangerous point in EMR adoption where the prevalence of older point-and-click EMR interfaces provide digitized healthcare—but at the cost of a lost narration and even repetitive globs of text cut-and-pasted into EMRs by rushed physicians.

Talking to Arslani, I see hope that incorporating voice recognition into the solution can help with both problems. Physicians freed from the keyboard leave demonstrably richer and lengthier notes in EMRs. And the narrative flow, though different than what came before on paper, isn't lost, just evolved.

Still, even after 50 years, voice recognition is in its early days. Systems still have to be trained and are best used with the kind of templates described here. The current state of the art doesn't mean we can leave voice mails that some computer will understand. So more work remains. It will be interesting to see where voice recognition is in another few years, in healthcare and in society in general.