Andrea Partida, DO, an obstetrician and gynecologist in Enid, Oklahoma, loves her new assistant. The 15 or 20 minutes she used to spend on documentation for each patient visit is now 3. The 2 to 3 hours she’d spend charting outside clinic hours is maybe 1.
All that time saved allows her to see two to five more patients a day, provide better care to each patient, and get more involved in hospital leadership at Integris Health, where she works.
“I have a better work-life balance with my family,” Partida says. “I leave work at work and get home earlier.”You’ve probably figured out the plot twist: Partida’s assistant is not a person ― it’s artificial intelligence(AI). Partida uses IRIS, a tool from Onpoint Healthcare Partners, part of a fast-growing niche of AI medical scribes designed to automate onerous data entry. The evolution of generative AI ― specifically, large language models, such as ChatGPT ― has led to a rapid explosion of these tools. Other companies in the space include Abridge, Ambience Healthcare, Augmedix, DeepScribe, Nuance (part of Microsoft), and Suki. The newest kid on the block, Amazon Web Services, announced the launch of HealthScribe in July.
These tools ― some of which are already on the market, with more on the way ― record patient visits and generate notes for treatment and billing. Earlier iterations combine AI with offsite human scribes who provide quality control. But more and more are fully automated, no human required. Some also offer video recording and foreign language translation.
The promise is alluring: Ease your workload and reclaim hours in your day so you can spend more time with patients or try that “work-life balance” thing you’ve heard so much about. But do these tools fulfill that promise? According to Partida and other doctors Medscape spoke with, the answer is a resounding yes.
A Tech Solution for a Tech Problem
“I believe a lot of doctors see patients for free. They get paid to do paperwork,” says Anthony J. Mazzarelli, MD, JD, MBE, co-president and CEO of Cooper University Health Care, in Camden, New Jersey. Indeed, for every hour US clinicians spend with their patients, they may spend 2 more hours documenting in electronic health records (EHRs), estimates show. About half of doctors, especially those in primary care, report feeling burned out, and some 42% say they want to quit clinical practice.
Enter AI scribes.
“The holy grail in medicine right now is improving burnout while also maintaining or improving productivity and quality,” says Patricia Garcia, MD, the associate clinical information officer for ambulatory care at Stanford Health Care. “These ambient digital scribes have the potential to do just that.”
While anyone can buy these products, their use has been mostly limited to pilot programs and early adopters so far, says Garcia, who has been helping to pilot Nuance’s digital scribe, DAX, at Stanford. But that’s expected to change quickly. “I don’t think the time horizon is a decade,” Garcia says. “I think within a matter of 2 or 3 years, these tools will be pervasive throughout healthcare.”
Since introducing these tools at Cooper, “our doctors’ paperwork burden is significantly lighter,” says Mazzarelli, who decides which technologies Cooper should invest in and who monitors their results. In Cooper studies, physicians who used DAX more than half the time spent 43% less time working on notes.
“They spend more time connecting with their patients, talking with them, and looking them in the eye,” Mazzarelli says. That, in turn, seems to improve patient outcomes, reduce doctor burnout and turnover, and lower costs.
The AI scribes, by virtue of eliminating the distraction of note taking, also allow doctors to give their full attention to the patient. “The patient relationship is the most important aspect of medicine,” says RaulAyala, MD, MHCM, a family medicine physician at Adventist Health, in Hanford, California, who uses Augmedix. The digital scribe “helps us strengthen that relationship.”
What’s It Like to Use an AI Medical Scribe?
The scribes feature hardware (typically a smartphone or tablet) and software built on automatic speech recognition, natural language processing, and machine learning. Download an app to your device, and you’re ready to go. Use it to record in-person or telehealth visits.
In the first week, a company may help train you to use the hardware and software. You’ll likely start by using it for a few patient visits per day, ramping up gradually. Partida says she was comfortable using the system for all her patients in 6 weeks.
Each day, Partida logs in to a dedicated smartphone or tablet, opens the app, and reviews her schedule, including details she needs to prepare for each patient. At the start of each patient visit, Partida taps the app icon to begin recording and lays the device nearby. She can pause as needed. At the end of the visit, she taps the icon again to stop recording.
The AI listens, creates the note, and updates relevant data in the EHR. The note includes patient problems, assessment, treatment plan, patient history, orders, and tasks for staff, along with medications, referrals, and preauthorizations. A human scribe, who is also a physician, reviews the information for accuracy and edits it as needed. By the next morning, the data are ready for Partida to review.
Fully automated versions can generate notes much faster. Jack Shilling, MD, MBA, an orthopedic surgeon at Cooper University Health Care, in Voorhees, New Jersey, uses DAX. A new feature called DAX Express ― which uses OpenAI’s GPT-4 but no humans ― provides him with a draft of his clinical notes in just seconds.
How Accurate Are AI Notes?
The accuracy of those notes remains an open question, Garcia says ― mostly because accuracy can be hard to define.
“If you asked five docs to write a note based on the same patient encounter, you’d get five different notes,” Garcia says. “That makes it hard to assess these technologies in a scientifically rigorous way.”
Still, the onus is on the physician to review the notes and edit them as needed, Garcia says. How light or heavy those edits are can depend on your unique preferences. Shilling says he may need to lightly edit transcripts of his conversations with patients. “When someone tells me how long their knee hurts, slight variability in their transcribed words is tolerable,” he says. But for some things ― such as physical exam notes and x-ray readings ― he dictates directly into the device, speaking at a closer range and being less conversational, more exact in his speech.
Should You Let Patients Know They’re Being Recorded?
The federal Health Insurance Portability and Accountability Act (HIPAA) does not require providers to inform patients that their face-to-face conversations are being recorded, says Daniel Lebovic, JD, corporate legal counsel at Compliancy Group, in Greenlawn, New York, a company that helps providers adhere to HIPAA rules.
But make sure you know the laws in your state and the policies at your healthcare practice. State laws may require providers to inform patients and to get patients’ consent in advance of being recorded.
All the doctors Medscape spoke to said their patients are informed that they’ll be recorded and that they can opt out if they wish.
How Much Do AI Scribes Cost?
As the marketplace for these tools expands, companies are offering more products and services at different price points that target a range of organizations, from large healthcare systems to small private practices. Price models vary, says Garcia. Some are based on the number of users, others on the number of notes, and still others on minutes.
Amazon’s HealthScribe is priced at 10 cents per minute. For 1000 consultation transcripts per month, with each call averaging 15 minutes, it would take 15,000 minutes at a total cost of £1500 for the month.
In general, the rapidly growing competition in this space could mean prices become more affordable, Garcia says. “It’s good that so many are getting into this game, because that means the price will come down and it will be a lot more accessible to everybody.”
Credits
Lead Image: Westend61/Getty Images
Medscape Medical News 2023 WebMD, LLC
We Asked Doctors Using AI Scribes: Just How Good Are They? – Medscape – August 07, 2023
Sources
Andrea Partida, DO, an obstetrician and gynecologist at Integris Health in Enid, Oklahoma
Anthony J. Mazzarelli, MD, JD, MBE, co-president and CEO of Cooper University Health Care in Camden, New Jersey
Patricia Garcia, MD, the associate clinical information officer for ambulatory care at Stanford Health Care
Raul Ayala, MD, MHCM, a family medicine physician at Adventist Health in Hanford, California
Jack Shilling, MD, MBA, an orthopaedic surgeon at Cooper University Health Care in Voorhees, New Jersey
Daniel Lebovic, JD, corporate legal counsel at Compliancy Group in Greenlawn, New York
JAMA Intern Med. 2022. “Medical Documentation Burden Among US Office-Based Physicians in 2019: A National Study.” doi:10.1001/jamainternmed.2022.0372
JAMA Netw Open. 2019. “Association of Electronic Health Record Design and Use Factors With Clinician Stress and Burnout.” doi:10.1001/jamanetworkopen.2019.9609