Dr Spiegel recaps his Trailblazer address at ACG on the prospect of developing a 'blended intelligence' that accesses the wealth of information available through artificial intelligence with the human wisdom, experience, and abilities of physicians to diagnose, treat, and manage conditions to achieve optimal outcomes for patients. 

Brennan Spiegel, MD MSHS, is director of Health Services Research and holds the George and Dorothy Gourrich Chair in Digital Health Ethics at at Cedars-Sinai Medical Center in Los Angeles, California.

AI-Assisted Diagnosis and Gastroenterology Training: ACG Trailblazer Talk

• At the American College of Gastroenterology meeting, Brennan Spiegel, MD, described “blended intelligence”: AI identifies patterns in clinical records while clinicians assess patients’ symptoms, circumstances, and treatment choices.
• At Cedars-Sinai, an AI chart review flagged possible eosinophilic disease in a 45-year-old man presenting for colorectal cancer screening. A clinician elicited previously unreported swallowing difficulty; upper endoscopy found eosinophilic esophagitis.
• Spiegel announced ACG’s VISTA GI initiative, which will use digital standardized patients and AI-assisted scoring to assess history-taking, empathy, clinical reasoning, and diagnosis in gastroenterology training.

Hey, I'm Dr. Brennan Spiegel. I'm director of health services research at Cedars-Sinai in Los Angeles. And today at the American College of Gastroenterology meeting, I gave the Trailblazer Talk. And the focus of today's talk, the title was Why Doctors Will Not Be Replaced: The Future of Humanistic Medicine in the Era of AI. And really what I wanted to talk about today was what's going on with AI generally? How do we get here? And what does a beneficent future look like for doctors, whether we're talking about GI or beyond GI? And that was the discussion.

And so I started by giving a little bit of a history of how technology has so exponentially grown in a way that's just mind boggling and how we got to this point today where computer intelligence is now exceeding, far exceeding, human intelligence. But I introduced this idea of blended intelligence.

This is what I call BI. This is where we allow computers to bring us access to knowledge, but also allow us as humans to bring wisdom. And those aren't exactly the same thing. We live in this world. We don't live in some computational realm of information. That's where AI lives. AI lives in some platonic space of information that does not look like the world we're in here. We live in a world where we engage with human beings and we look them in the eye and we have to understand what's happening in their lives, their biopsychosocial illness experience. We do physical exams, we do procedures, we have to communicate and collaborate. And we don't have robots doing that right now. And I hope we never do. I don't want a robot that does a rectal exam. We're going to need to figure out how to use AI to help us do better doctoring in this world.

And that's what BI is about. So I talked about some examples of how we're using BI at Cedars-Sinai where I work, where we use AI to look through a chart, for example, and it'll find patterns to identify potential conditions that may have been overlooked. And I talked about a case where a patient came to me for colon cancer screening and the AI saw a pattern suggestive of eosinophilic esophagitis. And that's interesting because nowhere in the chart did anyone mention that. Nowhere in the chart was there a discussion about swallowing problems or abdominal pain or diarrhea, nothing, nothing at all.

So how did the AI even come up with that? So I talked about how it did. It looked through the chart and it found a pattern. It noticed that in the CBC, in the white count differential, the eosinophil fraction was just a little bit higher than it should be. In a patient who has a history of asthma, who's a 45 year old man, the computer said, "You might want to think about eosinophilic conditions." Although it had low pretest probability, it wasn't that confident because it didn't have access to interview the patient. So the computer said, "Why don't you go ahead and ask this patient? Do they have trouble swallowing?"

And I did. And I went and talked to the patient and he said, "Yeah, I can't swallow very well." And I said, "Have you ever told anybody that?" He said, "No, I've never told anybody. I didn't think it was that big a deal. I think it only happens when I drink alcohol." I said, "Well, I don't think alcohol does that. I think we need to look at this." And sure enough, we did a colonoscopy because he was there for colon cancer screening, but then we did an upper endoscopy and found evidence of rip roaring eosinophilic esophagitis.

We treated him with therapeutics. He's now feeling better. He can swallow much easier. Would we have diagnosed that without the heads up display of AI? Not sure. He had been seen over a hundred times before that and no one had thought about that condition. He hadn't brought it up and no one had noticed this pattern. So the AI did it.

And this is an example of blended intelligence. We still need to confirm in real life what's going on in somebody's life. We still need to communicate and collaborate and talk about treatment plans and whether this is a good idea or not. The AI is not really doing that. Could it? Could a chatbot do that? Could. But humans kind of want to talk to humans too, particularly when they're sick, when they're vulnerable. So we have to get better as doctors at looking other doctors and the other patients in the eye, at taking a history, doing a physical exam, going back to those basics because the AI is going to run circles around us when it comes to knowing things, multiple choice questions.

And that brings me to an announcement that I made today during this talk for what we're calling VISTA. It's a new program. This is a virtual immersive simulation for training in gastroenterology. We're calling it VISTA GI. And this is a new announcement, a new initiative through the American College of Gastroenterology to change how we are evaluating and training young doctors, even old doctors. We always test on multiple choice questions. Who knows the most? Guess who does great on multiple choice questions? AI. AI can do better than all of us, period. So is that the best way now to tell if somebody's going to be a good doctor? Clearly not. It never was. What we're going to do now with AI, with AI's help, is to super-skill doctors by going back to the basics, by giving people access to a digital standardized patient that you talk to through a screen.

And we can set up these cases to be increasingly complicated and then we can score fellows and trainees on how well did they take a history? Did they ask the right questions? Did they show evidence of empathy while they were talking? Did they pick up on the subtle clinical clues that were displaying through this digital standardized patient? Did they ask for the right physical findings, the right labs? Did they come up with the right differential diagnosis? Did they find the diagnosis if there is one? And how can we score all of that? AIs can help us do that so we can have standards for each level of training of what we expect.

So rather than just administering a multiple choice question each year, we administer a multidimensional AI scored experience that reflects what we value in great doctors. And that's not answering multiple choice questions. That's being great at talking to people, at shared decision making, at making diagnoses, at picking up the slightest clinical clue that others would miss. And that's what we have to be great at in this era is going back to the basics. So that's what I talked about in my lecture today and it's been really gratifying to hear some of the feedback.

And I think all of us right now in this era need to think about how can we use AI for good? There's so much concern appropriately about how AI could end civilization, that it could change civilization and tear it at the seams. And we have to of course put guardrails around it to make sure that doesn't happen. We don't need super intelligence at this point. The AI we already have is extraordinary and we need to start figuring out how can we use it right now to improve the lives of the patients and the communities we serve.

So that's what I talked about today and I hope that is a helpful summary.