#89: AI in Documentation with Dennis Morrison

CE Courses
, , , , , , , , ,
1

Take the Quiz to get 1 CEU

Listen: Apple | Spotify | YouTube
Full Course Details: Public course page
Learn more about our guest: Dennis P. Morrison, PhD

In 50 years, you will look back on your occupational therapy practice and see a clear line in the sand between pre-AI augmented documentation and today.

Ambient AI augmented documentation is going to:

:white_check_mark:Make documenting QUICKER
:white_check_mark:Make your notes HIGHER QUALITY
:white_check_mark:And, transform how you interact with clients

And, if we harness this technology correctly, it will improve​:clap: client :clap:outcomes. :clap:

Today, we’ll look at an example of the research that is already being published about how an ambient artificial intelligence tool can improve clinical documentation.

Then next week we will welcome to the podcast, Dennis Morrison, PhD, a clinician who has specialized in consulting with AI documentation startups. Together we’ll walk through the state of AI-augmented documentation and the questions of the MANY tools that are coming to the market for OT.

:white_check_mark: Agenda

Intro and breakdown and analysis of journal article

Discussion on practical implications for OTs

  • 00:13:25 How Dennis became an AI and documentation consultant
  • 00:15:25 Is Dennis hopeful or weary about the tech?
  • 00:21:39 What do OTs need know about how AI documentation tools work
  • 00:31:19 Article Impressions
  • 00:37:38 How do we differentiate one AI tool from another?
  • 00:46:09 What questions should we ask about privacy?
  • 00:50:39 What questions should we ask about EHR integration?
  • 00:57:53 How will these AI tools improve patient outcomes?
  • 01:03:26 How do we push back against increasing pressure to increase caseloads?
  • 01:07:02 How do we position ourselves as leaders in this movement?

:white_check_mark: Supplemental Resources

:white_check_mark: Article Review

Read Full Text: Use of an ambient artificial intelligence tool to improve quality of clinical documentation
Journal: Future Healthcare Journal
Year Published: 2024

In 50 years, you will look back on your occupational therapy practice and see a clear line in the sand between pre-AI documentation and the AI-augmented documentation we’re starting to see today.

Documentation augmented by ambient AI is going to:

:white_check_mark: Make documenting QUICKER,
:white_check_mark: Make your notes HIGHER QUALITY, and
:white_check_mark: Transform how you interact with clients.

And, if we harness this technology correctly, it will improve​:clap: client :clap:outcomes.:clap:

Today, we’ll look at an example of research that is already showing how ambient artificial intelligence tools improve clinical documentation.

And next week, we will welcome to the podcast Dennis Morrison, PhD, a clinician who provides specialized consulting services to AI documentation startups. Together, we’ll walk through the current state of AI-augmented documentation and answer common questions about the MANY tools coming to market for OT.

Let’s dive in…

Quick intro to AI in documentation

As we all know, electronic health records (EHRs) have contributed to an increased administrative workload for clinicians—ultimately leading to higher rates of burnout.

Large language models (LLMs)—the building blocks of generative AI—have the potential to improve the clinician documentation experience. The challenge, though, is developing AI tools that:

  • Integrate with existing EMRs and EHRs.
  • Fit into established clinical workflows.
  • Meet all information governance and security requirements.

Which leads us to this paper…

What was the intent of this research?

This study aimed to evaluate the clinical utility of an end-to-end ambient AI tool in documenting a clinical consultation.

What were the researchers’ methods?

The study involved an AI tool specifically designed to ambiently capture real-world clinical consultation audio and then summarize it in a clinical note and letter.

First, a speech-to-text transcript was created; then, the tool used an LLM (specifically, GPT-4) to generate the note and letter based on the transcript. The LLM was given prompts to follow—for example, it was prompted to use a standardized note template and a specific style of writing.

Here’s a visual of this process:

Who participated in this study, and how was it carried out?

8 experienced clinicians—including 1 occupational therapist—carried out simulated consultations. The clinicians were all from the Great Ormond Street Hospital in the UK.

Pairs of actors playing patients and their caregivers rotated between the 8 simulated consultations.

Clinicians had 20 minutes to complete each consultation and produce the associated clinical note and letter.

Each clinician participated in:

  • 3 “control rotations” where they used the EHR as they normally would in practice, and
  • 3 “intervention rotations” where the consultation was conducted with the AI tool.

When using the AI tool, clinicians reviewed the generated documentation, made edits, and transferred their final notes into the EHR.

The clinicians were given a 10-minute training on the AI tool before the intervention rotations.

How was the quality and efficiency of documentation assessed?

To quantitatively assess documentation quality, 2 independent clinicians scored each note and letter using the Sheffield Assessment Instrument for Letters (SAIL).

To gather subjective data about the clinician experience, clinicians were asked to rate their experience with both the EHR and the AI tool using the NASA Task Load Index.

Patient-actors were also asked to rate their experience using a bespoke (i.e., customized) list of Likert questions.

To measure the impact of each documentation method on clinician-client interaction, the following data was collected:

  • Time spent on in-room consultation
  • Time spent actually conversing with the patient-actors

Lastly, focus groups were conducted with each set of clinicians to further discuss their experience.

What were their results?

23 EHR-only consultations were completed, and 24 consultations were completed using the AI tool.

Notes for 6 of the EHR-only consultations were not completed because the clinicians ran out of time. Notes for 5 of the AI-supported consultations were not completed (2 due to technical error, and 3 due to human error).

Quality of documentation

The SAIL scores were higher for the letters created with the AI tool, indicating a more than twofold increase in quality.

One interesting thing to note is that the AI tool functioned well in scenarios with multiple speakers (e.g., the patient, caregiver, and clinician). Some patient-actors were specifically chosen for their strong accents. Additionally, some consultations were carried out in a loud environment where multiple consultations were happening in the same open room.

Efficiency

Clinicians spent 26.3% less time in the consultation room when using the AI tools versus the EHR only. The time spent actually conversing with the client was statistically the same for both tech setups.

Subjective experience

Feedback gathered from clinician questionnaires showed the AI tool improved their experience and reduced computer disruption during consultations.

For example, 100% of clinicians reported being able to give the client their full attention when using the AI tool—versus 66% when using the EHR alone.

On the NASA Task Load Index, the AI tool showed a perceived improvement on 5 out of 6 metrics.

Patients also reported an improved experience during consultations where the AI tool was used, with 87% strongly agreeing that the clinician gave them their full attention—versus 75% for EHR-only consultations.

Lastly, a word cloud was generated from the conversations that occurred during the clinician focus groups. The top 3 words were:

“POTENTIAL”
“QUICK”
“AMAZING”

Author discussion/conclusion

The results of this study demonstrate that the use of ambient AI technology has the potential to significantly improve note quality beyond what is possible with standard EHRs.

It’s important to note that there was no clinically significant erroneous content identified in the ai-assisted notes.

Also, it is hypothesized that the efficiency gains achieved with AI are due to reduced task-switching. Even speech recognition documentation solutions—which allow clinicians to compose notes verbally—require task-switching, theoretically reducing efficiency.

Based on all of these findings, further studies should evaluate the use of AI documentation tools in a real-world clinical context across disciplines and settings.

Ep89-AOTA-Badge

:white_check_mark: Review %open%

After earning your certificate, please consider taking a moment to answer the following three questions:

1.) On a scale of 1-5 please indicate the degree to which the learning objectives were met.

  • 1
  • 2
  • 3
  • 4
  • 5
0 voters

2.) Please rate our instructors on a scale of 1-5.

  • 1
  • 2
  • 3
  • 4
  • 5
0 voters

3.) On a scale of 1-5, please rate the learning environment of the Club.

  • 1
  • 2
  • 3
  • 4
  • 5
0 voters

Please share any other feedback below! Including, ideas for future programming, and most importantly, how you feel this podcast will impact your practice!

3

Could this improve insurance reimbursements?

4

Christine26, if the documentation quality is stronger than it appears to me there is the potential to improve insurance reimbursement.

5

Sarah,
What a great study that used occupational therapy. I am excited about the positive results of stronger documentation, which may lead to better reimbursement. But, from an OT lens, I appreciated that there was less disruption while documenting. I struggle when I visit my health care professionals or take an elderly parent to their health care provider, and much of the contact is to the back of the health care provider because they are documenting as we go. I know the need for productivity, but empathy, collaboration, and relationship with the health care provider are impaired with this approach.
Additionally, I have watched minimal interaction and personal experience with my mother-in-law, who has low vision and is hard of hearing due to the need to document. The health care provider is missing the facial interaction and cues that the client offers all ‘yeses’ and minimal details because the client will not say I did not hear you. All that is to support the use of AI if it will allow healthcare providers to go back to providing care during interactions and focus less on documentation.
I also thought the 26% increased productivity was an exciting, surprising outcome. The word cloud words were exciting: “POTENTIAL”
“QUICK” and “AMAZING!
Finally, AI is great, but I highlighted the line on page 4 of the article that states, " it remains the clinician’s professional responsibility to check documents before final approval."
In academics, we have discussed the need for students in the healthcare profession to have the skills to discern correct information from incorrect information, which the article called minor and major hallucinations. (This is an Interesting label for incorrect information.)
I look forward to reading other individual’s thoughts on the article.

6

I agree , hopefully it will work. It has been challenging to write a good note for low functional level patients.

7

I am so excited about the integration of AI into clinical spaces. As a PhD student involved with educating/training OTD students at my university, I get so jazzed when I see appropriate spaces of AI being used to improve care. I had a recent encounter with my own PCP and my after visit summary had a line that stated “AVS developed through use of AI-generated notetaker” … or something like that. I know that in my clinical practice we often use “dot phrases” in Epic as a way to speed up documentation - this just feels like an adjacent approach to address the same problem.

Even though I’m so excited about AI, my fear is two-fold. One, obviously this was done at one hospital and admittedly as I know nothing about the dmeographics of this hospital/community, I fear that until language learning models are more advanced, we introduce opportunities for bias and non-affirming care. I know that part of the improving LLMs is by using LLMs, but that’s one fear.
Two, as an educator, I would love to see this introduced into our OT curriculums: how are we training students to use these sort of tools (because they exist and will CONTINUE to exist) but also still develop critical thinking and reflective skills as a future clinician? I think we need to see this same study replicated and adapted (1) in different hospital settings or outpatient or inpatient or schools or whatever! and (2) with OT students.

1 Like
8

I love your comments, Sherry, about the ways in which use of AI like this not only improves documentation and “speed” of care but also potentially an opportunity to improve quality of care, with those face to face interactions and whatnot. I know doctors are already very aware of the perception that they spend 10-minutes/client and they try and move very quickly, because they do, but without placing blame on these providers with extremely high productivity standards, can we prioritize opportunities (like using AI) to improve quality of care?

9

@allison5 I’ve missed you!So glad to see you here on this topic. :heart: I can’t believe you’ve already encountered a AI-generated visit summary. Truly the future is now.

I am not an educator, but I do keep thinking: if our students are only meeting ACOTE standards, they will not be prepared for practice, because healthcare is changing so quickly.

I think I’m going to make this AI episode my free offering, because I think it is so important to disseminate this information. I hope you have a chance to listen to the episode when it comes out. I’d love to hear your thoughts.

2 Likes
10

I am so excited by this use of emerging technology. I have 2 concerns that came up: privacy and workload. We need to make sure the system we use is HIPAA-compliant. Also, I fear that companies may use this against providers, increasing our caseloads and/or expecting higher productivity, reducing the positive impacts of this tool.

11

Task switching has certainly been the curse of point-of-service documentation with our current EMRs. I can’t wait for that curse to be lifted!

12

:point_up: This is why I believe OTs need be involved in implementing these tools. They should be used to improve the therapist and patient experience. NOT increase burnout.

2 Likes
13

@allison5 I had to share this study I saw last night.

I was pretty blown away :exploding_head: that this study showed that when an AI documentation tool was used, patients attended, on average, 67% more sessions compared to those whose therapy sessions were conducted without the tool.

2 Likes
14

I’m telling ya - AI is the future. It’s understandable that we as clinicians and academics are hesitant and nervous but (1) forms of AI have been around for 20+ years, and (2) we gotta hop on board to figure this out.

Thanks for sharing, Sarah!

2 Likes
15

@allison5 I’m literally feeling like we are in space race to figure out how to be the profession that goes all in adopting and leveraging this tech…

1 Like
16

Gah! To be missed by Sarah Lyon - I feel so supported!

Yes! At first I was like, “wait, hmmm. They never shared that.” I think if my provider was like, “hey, by the way - I am using this tool to help me provide quality care” I would have had absolutely zero qualms. But it was a cool experience overall!

What a great statement - our OT student education needs to meet and exceed ACOTE standards. Truly our healthcare and even our standards are always changing. I’ve started to see “AI in OT” or other healthcare settings at conferences so it’s definitely upon us.

Can’t wait to listen!

1 Like
17

Thank you so much for such a timely post! It can be overwhelming but also exciting to learn and embrace the evolving world of AI and its application to therapy. I have been exploring Magic School AI and am trying to learn as much as possible to keep up. Thank you for the great article and discussion, I will listen to the Podcast soon, looking forward to it!

1 Like
18

Hi @Pollywallace! I’m so glad you mentioned Magic School AI. I’ve been wondering what form this AI tech is going to take for our school therapists, keep me posted if you learn anything!

And, I’m so excited for you to listen to the episode!!