AI in Surgery and Medicine: The Good and Bad
Every time you visit a doctor next year, both you and the doctor will be AI-enabled. Here are a few of the good and bad parts to that.
You, as a patient, should not be required to remember all your medical history or even all the details of your most recent injury or illness. As Siri listens to your iPhone requests, other voice agents and data monitoring programs are following your every text, email, voice call, picture taken (posted or not), website visited, and real-time health status, including your Fitbit and smartwatch info. Even your emotions are better known to your digital avatar than you realize.
All of your previous medical visits and surgeries can now be compiled (almost 50% of Americans have a health record in a single electronic medical record company called EPIC). Many will not fear this invasion of privacy, but embrace it as the most accurate health status information resource that any doctor could have ever dreamed of. When you sit with your doctor, your voice agent will remind you both of any details you may have passed over.
Your doctor will have the world’s medical knowledge accessible through his or her AI-enabled dictation monitoring system. Currently, as I listen to my patient’s history, an AI dictation system completes the sentences, places them in the medical record, and prompts me for missing information. That basic feature will quickly expand, providing differential diagnoses for each patient complaint. I will have the data from thousands (if not millions) of cases just like the one in front of me to help form the treatment plan.
What could go wrong? Well, one current joke is that a patient unfortunately undergoes a leg amputation after a car accident. “Don’t worry,” his doctor says. “The next time you go to a doctor, your leg will be back on.”
And indeed it will – at least as far as the medical record is concerned. When the patient goes for an eye exam, for example, the eye doctor will examine his eyes—but has no time to examine the rest of the patient’s body. So she checks the “all normal” boxes under the review of other systems that all electronic medical records require. Now, according to the EPIC record, the patient still has that leg!
Multiply this oversight by thousands of patients, and it’s easy to see how the entire digital medical record becomes filled with erroneous exam data. Any AI trained on stored medical records will be confronted with a garbage in/garbage out problem. Massive conflicts may lead to a high likelihood of poor suggestions, bad diagnoses, and unwise surgical recommendations.
It is not only data that AI can’t differentiate between. It is also the variability each patient presents with. I see patients every day with bone-on-bone arthritis who have no pain at all, and other patients with a small amount of cartilage loss who can barely walk without pain. Some patients get dramatic relief from a PRP injection lasting months to years, while others have only a brief response.
I’m always trying to improve both the technique and the outcome. There are no two patients whose injuries and anatomy, combined with their sports goals, are exactly alike. Therefore—if patients want excellence—there is no “standard” procedure. Every “simple” surgical case I deal with, I approach with the goal of performing the surgery better this time than the last. And while robotics has made my joint replacement surgery far more precise, the need for human intervention, judgment, and experience has gone up rather than down. For example, the electronic data presented on the screen in the OR reflects the motion of the knee joint but doesn’t reflect the actual motion of that patient’s body—which also affects decisions about placement of the implants to optimize tightness of the ligaments.
So be prepared to be overwhelmed with data and underwhelmed by how much it has made your life and your doctor’s life better. That said, we can all be both better informed and much better at our jobs—as long as we don’t put all of our diagnostic eggs in the AI basket.
