Adapting AI writers is key to reducing editing time for doctors

Excessive clinical documentation is a widespread problem that is causing physician burnout and even impacting patient care – as evidenced recently by an American Medical Informatics Association survey that found widespread frustration among physicians and nurses with the burden of charting electronic health records, and the time and effort required to complete the necessary documentation.

There are technologies that can help. Many healthcare systems are using natural language processing tools, generative artificial intelligence, and environmental writers to help their physicians with EHR documentation tasks.

But these tools are not plug-and-play. To work effectively, they must be adapted to the specific needs of their clinical end users.

Dr. Dean Dalili, chief medical officer at DeepScribe, says customization with environmental writers is an essential feature, especially for medical specialties, because the ability to personalize note-taking reduces the time doctors spend editing automated notes.

“Less time is spent in a physical interface” creating formatting and other individual note preferences, and more time with patients, he said.

In this question and answer session, Dalili also discussed ambient intelligence and its benefits for enterprises. The feature compares note-taking against coding standards and creates a reporting structure to assess the quality of clinical visit notes at various points of care.

Q. How can clinical documentation improve the healthcare experience and reduce burnout?

A. To understand the impact of ambient AI documentation, consider the previous standards. Clinical documentation was paper-based and then moved to EHRs as part of (federal) policy.

That transition has been good for quality and safety, especially in the area of ​​medication safety, and doctors gain a better understanding of the patient’s context when making a decision.

The problem, however, is that EHRs worsen healthcare delivery because physicians are often more visually engaged with their computers than with their patients.

EPDs contribute to burnout – which is actually a phenomenon of high mental strain. That’s part of being a doctor, but it gets worse when you’re constantly switching between different types of work and translating an encounter into some kind of visual output in documentation format via a keyboard.

Clinical AI documentation allows the healthcare provider to connect directly with the patient and have a normal conversation. That conversation becomes the source of information for which the software produces structured documentation that is comprehensive. In some ways more comprehensive than relying on the provider’s memory.

The technology is always listening and can sometimes catch details that a provider forgets or is not focused on. Those details form the clinical documentation, resulting in a better quality patient experience and better note quality.

Q. Why are customizable AI scribe tools important for specialty healthcare organizations?

A. Customization is important for every doctor, but especially for specialists. As doctors, we fall into rhythms.

A one-size-fits-all writing solution that is going to produce a note with only a standard structure will not suit most providers. There are nuances to how healthcare providers choose to record information – these may include elements in the subjective history portion of the note, portions of the physical examination, or in the assessment and plan portion in which the provider groups together portions of the treatment plan at each clinical assessment.

There are also formatting elements of the note that every doctor likes. If you are a geriatrician, you may want to call patients Mr. or Mrs.; If you are a pediatrician, you don’t want to refer to a young patient that way and may want to use a first name only.

Having customization options helps create a note that is more likely to mimic what the provider prefers, is used to, or has written in the past. This is important because if the output matches the provider’s documentation preferences, it requires less editing.

Think about the value proposition of any AI writing: you don’t have to spend time having a conversation and then spend more time documenting the conversation. But if you still need to edit that note to make it look the way you want, you’ll still have a lot of extra work. That is why customization in specializations is important.

Specialty-specific workflows are recorded differently than during a normal primary care visit. As such, there are several areas of focus and details that providers will want to capture. It is important to set up a format that is not in a generic, one-size-fits-all form that applies to some, but to tailor the output and structure so that the AI ​​listens for parts of the visit that are unique to that specialty.

In oncology, for example, there is usually a very long summary of the data that determines a patient’s diagnosis, as well as all the data elements that went into identifying the problem. In addition, specific elements of the plan may be unique to cancer treatments – related not only to medical therapy, but also to social support, nutrition, and other issues. The note for orthopedics could look very different and focus on a physical examination of the musculoskeletal system, imaging and so on.

Q. How is the technology different from other AI writers available in the market?

A. First, we use a unique large language model that contains historical data from clinical visits, coded by live writers, which helps create structured data elements. We train our LLM – unlike an LLM like ChatGPT4, which is trained on the internet.

If you use medical information to refine your LLM, you are more likely to get accurate medical-related results. If you train the LLM on the entire internet, you get extra noise that can echo in the content.

DeepScribe has the largest source of training data based on user notes used to produce highly accurate documentation. This helps build trust and adoption and minimizes the time providers spend on rework or editing.

The second differentiator is that the tool offers more than 50 different customization elements, allowing providers to produce work closer to what they would create from scratch, for a wide range of specialties and users.

The third key differentiator is a new category called ambient intelligence, which includes functionality beyond just writing.

This is where the patient conversation can be applied to any form of structured data, whether it is a coding standard or a clinical quality standard. From there, the AI ​​can determine whether the conversation met that encryption standard or not.

This intelligence also allows us to create a reporting structure in which an enterprise can see at a glance how physicians are performing across a broad spectrum of providers. It is the ability to help determine high-quality clinical content at the point of care and also assess whether or not that content has been delivered.

Q. Can DeepScribe be integrated into a physician’s existing workflows and tech stack?

A. With ambient AI, the workflow is fundamentally different and requires some adjustment: the clinician must express findings in a way they may not have done in the past.

Providers can’t simply say, “This doesn’t look good.” Instead, they should say, “Your left knee looks swollen.” So there is a certain degree of specificity in the language that healthcare providers must adapt to. This level of detail ensures that the AI ​​listens more robustly.

The degree of integration depends on the EHR and DeepScribe has integrations with Epic, athenahealth, eClinicalWorks and over a hundred other EHRs, but if a physician wants to connect their own EHR, DeepScribe can also integrate via an API.

Andrea Fox is editor-in-chief of Healthcare IT News.
Email: afox@himss.org

Healthcare IT News is a HIMSS Media publication.