How clinical documentation became an AI battleground

Health systems are turning to artificial intelligence to easily bring patient-doctor conversations into the electronic health record.

Health systems are buying ambient AI documentation solutions, technology that takes a recording of a doctor-patient conversation and turns it into usable clinical notes. The goal is to give doctors more face-time with patients rather than staring at a computer screen while in the exam room, taking notes.

The past few months has produced a flurry of partnerships between health systems and vendors of the technology Startup company Abridge said Thursday it was working with University of Chicago Medicine. Abridge also won business from Sacramento, California-based Sutter Health in March. Group purchasing company Premier Inc. and York, Pennsylvania-based WellSpan Health recently signed deals with ambient clinical documentation companies.

“I think the widespread rollout of the electronic health record has brought a much more complete record to each provider at the point of care, but at times it's almost an overwhelming amount of information,” said Dr. Hal Baker, Wellspan's senior vice president and chief digital and chief information officer. “[These AI documentation products] take away cognitive burden and allow caregivers to more fully mentally engage with the patient and spend less time focusing on the computer.”

Here's what to know about the technology and its implementation.

What is ambient AI clinical documentation?

Ambient AI applications rely on external devices or smartphones to record conversations between patients and clinicians. The providers then upload the conversation to a server and it is run through a generative AI model to produce a transcript for clinicians will review.

Some ambient AI solutions will compile critical parts of the conversation and embed them directly into a health systems’ EHR. Traditional vendor transcription solutions relied on clinicians to do that step manually.

Since OpenAI's ChatGPT burst onto the scene, generative AI has democratized the sector and opened the door for new entrants. Some startups use multiple existing enterprise large language AI models to generate the output. Other companies say their products are more complicated than having doctors simply use a version of ChatGPT that's compliant with the Health Insurance Portability and Accountability Act of 1996.

"Anyone can use an off-the-shelf [AI] model, and sort of duct tape different models together, potentially to create some semblance of a solution," said Abridge co-founder and CEO Dr. Shiv Rao. 

Does ambient AI clinical documentation help with burnout?

The goal of the products is to minimize "pajama time,” the after-hours work many clinicians spend putting information into the EHR and answering patient questions via email.

Outside studies evaluating vendor solutions have been limited because it's a new product category. Stanford Medicine, in Palo Alto, California, ran a preliminary study in the fall of 2023 using Nuance's generative AI documentation solution and found 78% of physicians who used the technology said it expedited clinical note-taking. A study from researchers at Oakland, California-based Kaiser Permanente and Permanente Medical Group published in February in the New England Journal of Medicine Catalyst found the technology saved time. 

Where does the technology fall short?

Ambient AI technology isn't foolproof. Clinicians must review drafts before they are inputted into the EHR or sent to patients and that takes time. 

The Kaiser Permanente and Permanente Medical Group researchers were unable to verify edits the AI made to the doctor-patient conversation such as when it omitted parts of the conversation that the model deemed unrelated to patient care.

The technology also is not available for all specialties and many vendors do not offer services for non-English speakers. 

Who are the big companies offering ambient AI clinical documentation?

Nuance is one of the industry's most established clinical documentation companies. The company was acquired by Microsoft in March 2022 for $19.7 billion. In March 2023, it added OpenAI’s ChatGPT successor GPT-4 to its product offerings, bringing in generative AI capabilities. 

Abridge raised $150 million in Series C funding in February. When the deal was announced, Rao said the cash infusion would help it expand and fuel its research and development efforts.  

Both Nuance and Abridge are participating in a recently announced program from Epic that makes a small group of vendors co-developers with the electronic health record giant.  

Amazon entered the space in July through its Amazon Web Services subsidiary when it announced it was launching a product called AWS HealthScribe. Amazon Web Services is building the speech recognition and generative AI product with startup DeepScribe. 

What other vendors are in the space? 

Augmedix went public in 2021. It, along with Meditech, have inked partnerships with Nashville-based HCA Healthcare and Google.  

Nabla took in $24 million in a January Series B funding round. The company, which automates some functions of clinical note-taking, counts Children’s Hospital Los Angeles and the physician-led Permanente Medical Group as among its customers.

Ambience announced a $70 million Series B funding round in February with an investment from OpenAI’s Startup Fund and venture capital firm Kleiner Perkins. Suki and Fabric are two other startups.