AI is the hottest word in healthcare today, promising to revolutionize the landscape and suddenly make everything brighter, easier, and faster than ever.
Stop me if you’re a clinician who has heard this story before. Wasn’t that supposed to be what happened with EMR? You know, the same EMR that more than half of PCPs think needs a complete overhaul, and detracts from their professional satisfaction? The EMR that 75% of clinicians cite as the major source of their burnout? Clearly, these doctors just don’t know what’s good for them in terms of tech. Adding another good mandatory training module ought to sort them out.
All joking aside, let’s assume for a minute that “this time is different”. I happen to think this might actually be the case, but I’m a tech junkie, a population health administrator, and an EMR optimization nut. I’ve seen some great things come out of tech, and I remain cautiously optimistic that AI could actually make our lives as clinicians easier. There are 5 areas I’d like to see AI implemented to make a huge splash – hopefully sooner rather than later.
- Automating Billing and Coding
- This is my number one wish for AI, as soon as it can come. Billing guidelines are opaque, vary by payer, and change rapidly. I am definitively not a biller. I have no desire to be. While I have a good grasp of it after a few years as a clinician, I wish I didn’t have to expend so much effort or energy just to get paid for my work. Automation of billing and coding based on the note I actually write would save a bundle of time for me on a daily basis.
- Deep Search
- Anyone who has ever tried to look for a colonoscopy report from “Outside Hospital” knows where I’m going with this one. Health systems aren’t interoperable. They don’t share information well, and even if you can find the information you’re looking for, it’s with great effort and usually involves calling someone up on the phone, having a copy of a record faxed, and then uploading it unceremoniously into a non-searchable “documents” section of the EMR. If I can get an AI to reliably search for a document and extract this information for me when I need it with a prompt, it’ll save me and my staff countless hours of chasing down records from God knows where.
- Patient Triage
- Controversial? Absolutely. But one of the most challenging workflow issues in primary care is determining the need for an appointment versus telling someone to go to the ED. The usual pattern looks something like this: patient calls in with an issue relayed to a staff member without clinical training (receptionist), a message is sent to the physician (who is seeing patients), the physician calls back (or doesn’t) before the end of the day, and an appointment is (or is not) offered based on the snippet that was relayed through this game of telephone. You can’t tell me that using an AI to prompt the next destination for care and the need for emergent treatment is going to be worse than this current system. I want my patients with substernal crushing chest pain to go to the ER when they have that chest pain, not 3 hours later when I get the message that they had it this morning. AI agents like Ada Health can do this already – and if it gets patients faster care for emergency conditions, I’m all for it.
- Appointment Scheduling and Care Navigation
- Being inundated with calls about referrals, scheduling issues, procedural questions like where to go for labs, imaging, or a flu shot are all par for the course in primary care. Can they be solved by workflow modifications or delegation to clinical staff? Sure. Is this the most valuable use of staff time? Probably not. AI could offer a digitized version of this process to help patients self-troubleshoot while waiting for a callback. Online scheduling already exists, and it’s in wide use by most of my patients. They love the ability to plop themselves into an appointment slot without waiting for a confirmation. Why not give patients a tool that allows them to just search for the information they need in a trusted way?
- Followup Instructions
- EMRs by and large have failed to deliver on the value of an “after visit summary,” which is usually just a glorified list of medications (frequently incorrect), meaningless ICD codes, and templated garbage that may or may not be relevant to the patient’s medical condition. I’d love to have an AI generate a real patient summary based on what I’m actually saying during the visit – removing any ambiguity about what the next steps they should take in care are.
Notice I’ve left out the most common use case I see for AI, ambient listening technology. My personal experience with this to date has been underwhelming. Most clinicians my age are actually pretty tech savvy. We can type quickly and use templates, macros, and keyboard shortcuts to get our actual time writing down what the patient says done rather quickly. Don’t believe it? Here’s a study showing that ambient listening technology was really pretty minimally impactful on documentation burden, and the authors note that this was in a group of individuals who were optimistic about the tech pre-study.
I won’t sneer completely at saving 2 minutes per note, but realistically 2 minutes isn’t moving the needle much for me. Ambient listening, at least the way I’ve seen it offered so far, also doesn’t solve my biggest time suck, which is trying to select the appropriate codes for billing and insurance purposes – which is just an abstraction, again, from the note I’ve already written based on a searchable dataset, and should be ripe for AI to target.
I think there’s also something to be made of the fact that note-writing often helps me to solidify my thought process. There’s some research to back this claim up as well (which was stronger for handwriting than typing, but that’s a different topic entirely). I might suggest that when doctors say they want help with the burden of “documentation” they’re not necessarily talking about speech-to-text, but everything else that comes along with closing the note and taking care of the patient. If AI can help to close this gap and make care for my patients safer, more efficient, and more accurate, I’m all for it.