I read an article today about artificial intelligence (AI) in primary care¹ — and now I feel frustrated. Maybe writing about it will help me calm down and express what I'm thinking. (It's also a Monday, and my day in the office was super busy, so that might have something to do with my reaction, but I digress.)

The author begins with a good premise: AI-enhanced electronic health records (EHRs) could support primary care physicians. Great start. Even if you don't believe in the power of AI, anything that potentially saves us time and makes our day better is worth considering, right? The author lays a foundation of how AI can already help in the detection of diseases or complications using visual tests (chest x-rays, mammograms, retinal photos, etc.) while the things proceduralists do (cystoscopies, cardiac catheterizations, etc.) will always require a skilled human technician.

OK, so far so good, I guess. The promise of AI is heavily techy and specialist focused, so I'm waiting for the article to deliver the hook — something along the lines of “and none of that matters if your doctor can't see you, so here's how AI would help in family medicine.”

That's not what came next. This did: “Although the arrival of AI may thus be good news for specialists, primary care physicians (PCPs), who function as critical gatekeepers to these specialists, have been left in an uncomfortable limbo.”

Wait, what? We function as critical gatekeepers for specialists? Really? That's weird, because not one patient today asked me to “open the gate” and let them visit a specialist. I don't know how it's done in Los Angeles or Cambridge, but here in Akron, we manage our patients. We are not bouncers at the door, letting people into some exclusive specialist club. We are comprehensivists.^2,3

The author goes on to point out several theoretical examples of what AI could do for us in primary care, like scanning discharge summaries looking for diagnoses or reading a lab report to identify thrombocytopenia while cross-checking the medication list for possible causes. Great ideas. Cool technology. But is that what slows us down during the day? No.

One of the biggest time wasters in my practice is prior authorization. I recently had to get prior auth for a patient on dulaglutide who has been on it for years and now has an A1C of 6.2%, thanks in part to the medication. If AI is going to help me, it's going to have to help with that: “Hey, AI. Write a letter to the insurance company telling them to please let the patient stay on the medication they're already taking and paying for because it's working” or “Hey, AI. Call the insurance company, stay on hold for 35 minutes, and explain to them that the patient and I know exactly what we're doing.”

And then there's the patient who showed up 15 minutes late (through no fault of her own — we've all had car trouble), and now I'm at least 15 minutes behind. Can AI help with that? What about the patient who needs me to order an MRI of his back before the orthopedist will see him because “their office said the doctor doesn't have time to order tests”? And let's not forget the seven care gaps I need to close on the next patient, including documenting again why they can't be on a statin, so as not offend the SUPD police.⁴

I'm probably being a little critical. The article does add to the shared pool of knowledge, includes a nice scenario about identifying a patient's ocular cicatricial pemphigoid as a possible cause of a hepatic adenoma, and suggests some ways things could be better.

But the big question to ask is not “How do we free primary care from the bonds of the EHR using AI?” but “Why are we bound to the non-clinical workload of primary care?” The EHR is just one slice of that pie. When AI can help solve the bigger problem for us in primary care, I'm all ears. Until then, it is just one more tool in the toolbox.