Voice-Enabled AI App May Reduce EHR Documentation Time

Ken Terry

November 20, 2019

A new effort to enlist artificial intelligence (AI) to improve the usability of electronic health records (EHRs) and reduce the administrative burden on physicians is starting to bear fruit, but at least one expert considers the effort to be "shortsighted."

A year into the American Academy of Family Physicians' (AAFP) 42-month Health IT Innovation Project, a voice-enabled AI assistant called Suki has cut EHR documentation time by more than 50% in some pilot family practices, Steven Waldren, MD, AAFP vice president and chief medical informatics officer, told Medscape Medical News.

Leading EHR vendors are developing similar enhancements to their products, Waldren said. "There's a significant amount of [research and development] in this space. We've had conversations with Epic and Cerner, and both are working on AI-powered, voice-enabled clinical assistants for documentation and order entry. Those EHR products are a couple of years out, potentially."

The AAFP is also holding discussions with several third-party software vendors, he noted. But so far, the only one that is actually conducting a pilot with AAFP member practices is Suki (also the name of the firm). Several medical practices across the country are participating in the experiment, which began in July.

Suki has already launched its product commercially with Ascension and some other healthcare organizations, Punit Soni, the company's chief executive officer, told Medscape Medical News. But the firm also welcomes the opportunity to pilot the product in independent family practices, he said, and has learned a great deal from the test.

Although Soni is a former Google executive and Suki recently signed a machine learning agreement with the tech giant, company spokesman Nicholas Samonas told Medscape Medical News that it has nothing to do with a controversial Google foray into healthcare involving Ascension called "Project Nightingale."

That research partnership has drawn national attention because of allegations that patients' personal health information could be improperly shared with Google.

Beyond Speech Recognition

Suki's downloadable app generates text-based documentation with simple voice commands. Using either the Fast Healthcare Interoperability Resources (FHIR) standard or a custom interface, Suki can access the data in an EHR, add notes or other inputs at a doctor's direction, and return the additions to the EHR.

The difference between Suki and conventional speech recognition is that the voice-enabled digital assistant uses AI to understand commands and create blocks of text based on a doctor's past documentation patterns. Voice-recognition programs like Dragon and M*Modal, in contrast, mainly convert speech into text, Soni said. While they have some workflow features, they don't use machine learning.

With a voice-recognition system, he said, doctors typically go to the place in the EHR where they want text inserted and dictate their note. Then the transcribed text appears, to be corrected in real time or later.

In contrast, Suki automatically navigates the EHR to deposit the transcribed text in the right place. Moreover, the AI software creates macros from its understanding of how the physician has documented a particular concept in the past.

"When you say, 'I did a normal review of systems' in Suki, it recalls this is part of a doctor–patient encounter, and based on the last thousand patient encounters the doctor has done, it has machine learned what a normal review of systems for this doctor is, and it generates a review of systems that may be half a page long based on what it's seen. Then it puts that into the right place in the [EHR]," Soni said.

Alternatively, the physician can say he did a normal review of systems except for two changes, and the digital assistant will adjust for that, he added.

Because Suki responds to commands rather than just transcribe words, Soni said, its accuracy rate is much higher than that of speech-recognition programs alone. "When you use speech recognition systems and natural language processing systems together, you can get a Siri or an Alexa for doctors," said Soni, who used to lead Google's mobile apps group and worked on its first voice-enabled digital assistant.

John Abraham, MD, a Tulsa, Oklahoma, family doctor who is part of the AAFP pilot, used Dragon when he was a hospitalist and said Suki is more accurate. For example, Dragon would often get patients' names wrong and he'd have to correct them.

I got into family medicine to connect with patients, and I'm doing that in a meaningful way. I'm not on my keyboard the whole time in most visits. I'm looking at my patients and talking to them. Dr John Abraham, Tulsa family physician who is part of the AAFP pilot

"With Suki, I dictate straight through and I get my note back and it's cleaned up, without me having to correct things like patient names, which are pulled from the chart. AI cleans up the notes and puts them in for me. It saves 10 seconds here and 15 seconds there and it adds up," he told Medscape Medical News.

Orders Are Next

Suki has created a prototype for automated order entry, although that isn't included in the AAFP pilot. Primary care physicians order the same medications 90% of the time, Soni noted, so Suki learns those orders and the dosages they normally contain. "When a doctor says 'prescribe doxycycline,' it automatically pulls up doxycycline at the regular dosage, makes sure it matches the diagnosis and does other things. When you say 'order it,' Suki pushes the prescription into the EHR. When you sign the note, you sign the order."

In contrast, he noted, EHRs typically require doctors to click six times off four dropdown lists and then correct the dosage if it's wrong. "This is a simple but smart enhancement that you can do with one voice command."

Also in the works is the ability for Suki to populate certain discrete data fields in EHRs. To do so, its natural language processing must reliably understand certain medical concepts, and it must also be able to write back to EHR databases that can recognize FHIR-based data.

Some FHIR-enabled EHRs don't allow writeback because of technical and security challenges, Waldren said. Soni agreed, but added, "With Epic and other EHRs, we sometimes use FHIR-based APIs [application programming interfaces] or write interfaces. Compared to three years ago, when we had to write custom interfaces to each EHR, it has become easier to plug into EHRs."

Abraham, the Oklahoma family doctor in the AAFP pilot, said that besides the time savings from using Suki, it has reduced his anxiety during patient exams and has increased his quality of life. "I got into family medicine to connect with patients, and I'm doing that in a meaningful way. I'm not on my keyboard the whole time in most visits. I'm looking at my patients and talking to them. I either dictate in front of them or when I step out. It's not just the logistics, but also the fulfillment in work."

Waldren agreed that reducing this administrative burden can improve doctor-patient interactions and physicians' quality of life. Currently, he noted, "Some docs are spending two to three hours a night and five to eight hours on weekends, just doing charting and inbox management."

One goal of the AAFP's Health IT Innovation Project, Waldren added, "is to get companies like Suki to focus on primary care." It's easier for them to work with large healthcare organizations, he noted, but primary care doctors in small and medium-sized practices badly need technological assistance to reduce burnout.

"Shortsighted" Approach

Not everyone thinks applications like Suki represent a big advance. Peter Basch, MD, a longtime information technology advisor to the American College of Physicians and senior director for IT quality and safety, research, and national health IT at MedStar Health, told Medscape Medical News that the digital assistant is a "shortsighted" approach that's tackling "yesterday's problem."

With the Centers for Medicare and Medicaid Services set to scale back E/M documentation requirements, Basch noted, "creating the verbose notes we have today" will no longer be necessary.

So just making it easier to generate text and insert it in the right place will no longer confer much of an advantage, he said.

"Focusing on a digital assistant rather than how you're managing the patient is the wrong way to go. As you start to pull away the regulatory burdens and have the EHR resolve the simple administrative burdens in the background, you can think about having the EHR help everybody — the doctor, the care team, the patient — to do a better job with less and to break out of the tyranny of what are currently documentation-based charges and the checkbox mentality."

Suki's Google Involvement

A recent Wall Street Journal article raised questions about Google's Project Nightingale, which will apply AI analytics to patient data from Ascension, the country's second-largest health system, as Ascension moves its data into the Google cloud.

Because Google will reportedly have access to identifiable patient information, the announcement of the partnership generated blowback from the public and politicians. Soon after the first report appeared, the Office for Civil Rights in the Department of Health and Human Services launched a federal inquiry, the Wall Street Journal reported.

Google made a similar deal with the Mayo Clinic in September, and it has research partnerships with the health systems of Stanford University, the University of Chicago, and the University of California at San Francisco, according to Modern Healthcare .

A Google spokesperson told the Wall Street Journal that Google's work with Ascension "adheres to industry-wide regulations (including HIPAA) regarding patient data, and comes with strict guidance on data privacy, security, and usage."

Suki's new deal with Google is to use its machine learning techniques to develop and improve clinical language models. Under that arrangement, Suki spokesman Samonas said, "Individual data is protected from disclosure to Google."

Asked about the Suki-Google connection, the AAFP's Waldren said, "For our project, Suki enters into a business associate agreement with the medical practice and therefore Suki must comply with the privacy and security provisions of HIPAA. HIPAA allows for covered entities to share data with other covered entities or their business associates for treatment, payment, or operations.

"Suki is supporting the operations of the practice, so it's no different than a healthcare practice/hospital that is using a clearinghouse to help process claims or an EHR company to provide their IT system."

Waldren added that he sees a bigger threat to patient information from other types of companies.

"The most pressing privacy issues in AI are for those companies that are not bound by HIPAA because they are not a covered entity nor a business associate of a covered entity. For these companies, as long as they disclose to the patient how they plan to use their data (which is normally part of a long End User License Agreement that the patient can click to agree to), then they can use the data however they wish with limited restrictions," he said.

Follow Medscape on FacebookTwitterInstagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.