Can AI Really Document a Patient Encounter?

By Erica E. Remer, MD, CCDS

I have accepted that it is inevitable that artificial intelligence, or AI, is going to become an integral part of our healthcare process. The question is how is it going to be used, and will it improve patient care? You know that I do not believe documentation is a burden. I believe it is part of the job and a responsibility.

Providers shouldn’t be documenting just to get it out of the way; they must recognize that it can enhance or detract from patient care.

I recently read an article in Medscape titled “We Asked Doctors Using AI Scribes: Just How Good Are They?” by Lorraine L. Janeczko, MPH. It points out that you need to get trained on the use of the hardware and the software, and it will likely take a while for you to feel comfortable to use it for all patients.

I wonder how the AI picks up on all the nonverbal cues and unsaid thoughts of the clinician. I am the course director for a course on medical documentation (Intensive Course in Medical Documentation: Clinical, Legal and Economic Implications for Healthcare Providers); 80 percent of the attendees are mandated to take it by their medical boards.

I suggest the potential use of a scribe (although I am thinking more in terms of a human one) as a tool to bring full attention back to the patient and to make the practitioner more time-efficient.

When I talk about this, I envision that there needs to be some alteration in how you talk with the patient, being cognizant that the scribe, human or AI, is passively listening. For instance, I recommend that when clinicians examine the patient, they verbalize aloud what they are doing and seeing. I suspect that patients would like to hear most of this, because most of the physical exam is usually normal.

But what if you notice a scary mole or palpate an ominous mass? If you don’t mention it, the scribe won’t know it is there, but if you do note it out loud, it could alarm the patient. The provider may prefer to discuss it when they are talking about the plan.

The discussion of the medical decision-making (MDM) has to be much more robust if the clinician is counting on the scribe/AI to document it for them. Their professional fee may be based on the MDM if they are not doing time-based billing. Do they need to articulate everything with the patient on the front end, though, or do they need to manually augment the documentation after it is available?

It seems as though it could be difficult to keep track of what needs to be added. But we should keep in mind with Open Notes that the patient can review the documentation anyway, so whatever we think is relevant and should be in the note, they are going to be privy to it.

There were a couple other fascinating points in this article. A surprising one to me was that “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.” It is possible that your state law does mandate it, however. I think it is best practice to let the patient know regardless of the regulations, and give them the opportunity to opt out of AI scribing.

The other key point stems from the observation that if you asked five providers to document the same patient encounter, you’d get five different notes, and it might be hard to judge the “accuracy” of the AI’s note. This piqued my interest because I am constantly trying to make my medical documentation course even better, and an idea we recently came up with was to have our attendees view a simulated patient encounter, compose a note, and discuss it in small breakout sessions. After assessing the quality of their documentation, we plan to demonstrate how they would level-set the evaluation and management service based on their notes. Similar to the fact that we recognize that we are going to get multiple different versions of the same encounter, depending on how the encounter unfolds, the AI rendering may need significant editing to represent what the clinician experienced.

The most important recommendation I have to make is that however a document is composed, be it via voice recognition, by a trainee, or by a human or computer scribe, it is incumbent upon the provider to read, edit, and revise it to ensure it accurately describes the patient encounter. And, seeing how badly we are at performing this task with copying and pasting, I am worried.

I am not ready for HAL 9000 to do everyone’s documentation quite yet.

Programming note: Listen to Dr. Erica Remer every Tuesday when she cohosts Talk Ten Tuesdays with Chuck Buck, 10 Eastern.