Generative AI scribing tools: Considerations for implementation in veterinary practice

When we started writing this article on generative AI scribing tools, we thought, let’s start this story with an example of generative AI.  

We searched “What is AI scribing in veterinary medicine?” Without fail, at the top of the search results page it read: “AI scribing in veterinary medicine is a software tool that uses artificial intelligence to create medical records from conversations between veterinarians and pet owners.”

Not bad. In a simple sense that answer is the power of generative AI. It took lots of information, in this case from across the web, and summarized it.  

While that answer gives us a short and simple definition, generative AI is much more nuanced. This is especially true regarding scribing tools in veterinary medicine. The tools, just like the companies and their policies vary. So, what should you know about AI scribing tools and where they might fit in your practice? 

Understanding how AI scribing looks in a practice 

First off, how AI scribing looks in practice varies from practice to practice. There are different internal use cases and different processes or procedures teams may follow. You’re probably familiar with the ability to do “voice to text” or dictation on your phone (super helpful when you’ve got a lot to say…). That is what some scribing has looked like in recent years: Audio files or live audio taken and written out.  However, full appointment generative scribing is the newest iteration of how scribing can look in practice.  

So, what is it?  

“This is typically the full appointment. It can be 20 to 40 minutes, well up to two hours at times,” said Gary Peters, the CEO and Co-Founder of PupPilot, a generative AI infrastructure for veterinary medicine. “There’s a heavy level of summarization that’s taking place through the AI. What it’s doing is trying to match and format the note into the style that the doctor is expecting.” 

Along with his work at PupPilot, which focuses on working with software engineering and product teams to help accurately produce and use generative AI in veterinary medicine, Peters also heads the Veterinary Innovation Council’s (VIC) subcommittee on scribing.  

The summarization that Peters mentions is exactly what generative AI is – it’s taking possibly an hour of audio or recorded text and putting it into a digestible format instead of having to go back and read a full hour of transcript. It’s important to note though that with any level of summarization, keeping an eye out for errors from the tool is important.  

This brings us to something to keep an eye out for: False positives and false negatives. 

What are false positives and false negatives in AI scribing?  

Accuracy inside of scribing is critical, Peters mentions, so knowing some of the risks that occur within a generative AI scribing tool is important.  

“From an AI lens, we think [in terms like] false positives and false negatives, and sometimes we call this AI over–interpretation or AI under–interpretation,” Peters adds. “If the doctor says we’re ruling out diabetes, does diabetes show up on the medical record?” 

A false positive is when the AI tool identifies something as a “threat” when it’s not. In the diabetes example, it could be the veterinarian saying that diabetes is not present, but the AI tool marks it as present or writes down something relating to diabetes. A similar error happens for false negatives, where the tool fails to pick up something that is an actual issue. In that same example, maybe the veterinarian says we’ve got to monitor for diabetes, and the tool does not make note of it.  

“What we are really wanting to assess here is how well the AI scribe is responding to what the doctor says,” Peters notes. “And the reality is we usually don’t just see things in black and white. What we usually find is a lot of gray.”  

The need for a human touch

That’s where, in the case there’s a chance for the false positive or negative, the AI essentially must interpret if “unlikely” means that it shouldn’t be added to the medical record (or if it should).  

“There’s a lot of clinical intuition that goes into how to interpret what a doctor is saying. And these AI systems do not inherently have that,” Peters adds. 

Peters critically notes that as an industry, practices should not just assume these tools are outputting information that is 100% accurate. There’s an important role for doctors to still review and look for anything that may have been misconstrued by the AI.  

According to Peters, “the nice thing about scribing is you don’t actually have that high of a dependency. The doctor goes in, does the appointment, they come out, and they’re looking at this medical record of the appointment they just did.”

As Peters alludes to, the goal of the AI in the above scenario, and really the goal of a generative AI scribing tool in general, is that the workload of the doctor is lessened. It doesn’t replace medical expertise or human attention to detail. 

Questions to ask when looking for an AI scribing software 

Let’s say your practice is ready to look for some sort of AI scribing tool to implement. After you and your team have figured out how this tool is going to be used, what’s next? For starters, having a list of questions ready for the companies you’re looking into is important. 

 “How do they handle their false positives?” Peters offers as an example. On a similar note: “How do they handle their false negatives? How do they turn those dials to determine how sensitive the AI should be to what you’re saying? All these AI systems operate differently, and they’re all built slightly differently,” he adds.  

Although there are many nuances with how your practice might use a scribing tool, what the tool does, and what company makes it, there are three things to look for at a high level according to Peters. 

Look at the team 

Make sure the team has AI professionals on it. In veterinary medicine, it’s also important to see if the team has veterinary professionals on it as well. “These are all graduate-level degrees,” he noted referring to DVMs and something like an applied mathematics degree. “Ask and ensure that the teams that you’re working with have those types of backgrounds.” 

Accuracy of the AI tool 

“You as the practitioner are liable and reliable for that final medical record,” Peters noted. With that said, making sure you understand how accurate the tool is, and where there might be room for error in the tool is critical to how you interact with it.  

Data privacy and security 

Make sure the data privacy policy is in line with what you’re looking for. Many companies set internal policies of who and how they can access your data. Ask that question to the companies you’re looking at. “You should just be able to ask the company directly, what is your policy on X, Y, and Z when it comes to privacy?” Peters says. Additionally, understanding how the product is secure, and how your information is protected.  

“The questions are there just so that you make sure that you have proper alignment with the scribing company that you choose to use,” he added.  

Watch the full interview with Peters from our interview at VMX here:  

Photo credit: Visual Generation via iStock / Getty Images Plus 

Disclaimer: Trends content is meant to inform, educate, and inspire by providing an array of diverse viewpoints. Any content published should not be viewed as an official stance, position, or endorsement by the American Animal Hospital Association (AAHA) or its Board of Directors.