Episode 12: When Patients Show Up Already Diagnosed

Patients used to Google their symptoms. Now some of them feed their MRI report to a chatbot and walk in with a theory.

In this episode of Healthcare Unscripted, Dr. Hunter Greene talks with Dr. Ryan Halvorson about what that means for a surgeon just starting out. Dr. Halvorson is an orthopaedic surgery resident at UCSF. He studied engineering at Harvard before going to medical school at UCSF, and he works with OpenEvidence, the AI platform most people in medicine already know. He's a fan of the technology. He's also very clear about where he stops relying on it.

Dr. AI Is in the Exam Room

Dr. Halvorson doesn't see patients' chatbot habit as a problem. They ask better questions and show up knowing more about their diagnosis and their options. Dr. Greene calls it the jump from Dr. Google to Dr. AI, and it changes the visit. Now the surgeon's job is to walk through the report together: yes, that finding is there, it doesn't matter clinically for you, but this one does. Plenty of the time the answer is that a meniscus tear doesn't need surgery at all, and the conversation turns to conservative care.

Borrowed Experience

For Dr. Halvorson, the most useful thing AI does right now is help him prepare. When he's about to do a procedure he hasn't seen much of, or use a technique he doesn't know well, it can pull up the papers and background in a fraction of the time it used to take to dig through textbooks and PubMed. Dr. Greene compares it to having thousands of attending surgeons looking over your shoulder. Dr. Halvorson says it's a way to build on the shoulders of others.

Dr. Greene pushed on that with a real case. A 73-year-old comes in with shoulder pain, mild to moderate arthritis, and a small to medium rotator cuff tear. Therapy and injections haven't worked, and now he wants to talk about surgery. A cuff repair means six to nine months of recovery and a real chance the cuff won't heal. A reverse shoulder replacement could have him feeling great in two months. Could AI help find the dividing line?

Dr. Halvorson's answer is careful. It can find the relevant papers and show which factors matter, and that feeds a shared decision with the patient. But he won't let it make the call.

"I'm not quite ready to use it as a tool to tell me what to do, but it definitely could help me find resources or find factors that I should consider that have been shown in the literature to help affect this patient's outcome." — Dr. Ryan Halvorson

A Study Built in the Waiting Room

Dr. Halvorson did a study at UCSF last year using the pre-visit questionnaire patients fill out in the waiting room or online. His team built a model to predict, from those answers alone, who would go on to need surgery and who would need advanced imaging like an MRI. No imaging, no physical exam, no clinician notes. For common diagnoses, it predicted the need for surgery with about 80 percent accuracy.

The practical use is triage. Some patients need to see a surgeon quickly. Others might do better starting with a primary care sports medicine doctor who can talk about injections or other options. Dr. Greene knows the problem well: patients turn up at a surgeon's office without ever having tried physical therapy or an injection. Dr. Halvorson describes it as matching patients to a treatment that makes sense for them, with AI as the go-between.

Paperwork First, Operating Room Later

Both doctors expect the biggest changes to happen away from the operating table. Dr. Halvorson is hopeful that AI can take some weight off burnout. Scribes are already saving clinicians time and cutting paperwork, and AI is helping with insurance authorization and other admin work too. There's an odd twist, though. Insurers are using AI to flag and deny claims, while hospitals use AI to write prior authorizations and rebuttals, so the models end up duking it out. Dr. Greene calls it the battle of the bots. Scheduling is harder than it sounds in orthopaedics, he adds, since a new patient consult, a post-op visit, a second post-op and an injection all need different kinds of slots.

In the operating room, Dr. Halvorson sees AI working its way into navigation and robotics, including robotic arthroscopy, where it can label what the scope is looking at. He expects that to move slowly. Surgeons are conservative about new technology, and it will take a lot of proof. Language-heavy tasks are where the technology is strongest today, which is why they're getting the attention first.

Learn It Before You Use It

Dr. Halvorson's advice for young surgeons who haven't started yet is to understand the technology first. You don't need an engineering degree. A few YouTube videos or lectures on how a large language model works will do: what a token is, what it means to turn text into vectors, how a sentence becomes numbers and then becomes new sentences. Once you know that much, you can see the limits and the promise. After that, get involved in a research project that uses it. He's a big believer in learning by doing.

Dr. Greene compares it to learning to do a scope. You get some background, you learn your anatomy, then you get in the room. People who skip that step, he says, are sitting in a Ferrari driving around a parking lot.

AI can find the paper, help you prep for the case, and sort the referral pile. What it can't do is sit across from a patient and work out what they want and expect. That part, Dr. Greene says, is experience, time, and the individual in front of you.

Watch or listen to this episode now: linktr.ee/healthcareunscripted