We'll be talking about what exactly AI is and my experience with it over the last five years. We'll start with the definition of it and followed by some of the opportunities that those who took the time to come to this room may be able to get out of it. So let's just simplify what it is. It's basically AI is essentially just automation and if you want more information, it's the application of math and software to teach computers how to understand, synthesize and generate knowledge in the same way humans do it. So just like real life, knowledge is a function of exposure. So a computer's knowledge is a function of its data. And the I only sees what the mind knows is a common quote that you hear in training that comes from the French philosopher Henry Bergson and it applies to real life situations. So if any of you have children or you remember what it was like learning at a young age, you only learn how to write based on what someone tells you, what cursive looks like and you only learn how to read based on what you're exposed to and it applies similarly to surgical training where if you train with one person for a year, you're going to learn everything from the way that one person taught you or if you see things many different ways, you might have a broader exposure. But things can go wrong. Amazon specifically early deployed an algorithm that was based on their prior experience which was rooted heavily in men as they were trying to recruit more candidates for the job and when they released it too soon, they actually found out when unsupervised, this algorithm was biased against women. And so just going to go through three different avenues of ways you can think about how to use this technology. The first would be if you want to pursue it from an academic perspective and study it, that may be one way. The second is if you're involved with industry, you can talk about specifically ways to leverage your products. And the third is if you ever have an interest in starting a company, that may be a perfect way to delve deeper into analytics and overall empower the physician since we are the front line. And so it's very interesting seeing how AI has evolved since 2018 where it had a series of rejections and many people saying it was pointless. But my luck essentially turned around when Dr. Krebs at Cleveland Clinic was able to afford me and my team a $30,000 supercomputer that was able to process imaging, patient report outcomes, and really make some meaningful strides that allowed me to automate a lot of the work. And so from an academic perspective, I basically thought of these things in terms of big columns. The first being big data where from a sports perspective, we can predict injuries, predict clinical outcomes, and then value-based care, payment models and policy. It's valuable to be able to talk to your administrator if you can predict how much that cost of care is going to be before you make an incision. Second being, sorry, I should say third being imaging analysis, being able to automate x-rays, MRIs, and CAT scans. And so these initial studies basically show that linear regression or logistic regression or linear regression models are not better than AI-based models, and we were able to duplicate that when you have a lot of data. Secondarily, we were able to take multiple different variables from mental health to imaging findings to age and help show what is important in the decision-making of cartilage procedures. Third, when there was a potential bundled payment system for hip fractures, we all know as surgeons here that bundling payments doesn't make sense, but in order to communicate that to our administrators, if you have a sick patient that comes in, the risk rises quite highly if they're very sick, and so you can use this model to convey that point before even doing the surgery. I have an orthoplasty background, so we used implants to basically help us identify what implants are inside the knee just based on an x-ray, and this just shows some of the ways that we can use this technology to identify the implants that are inside the body before you undergo revision, especially if you're in an area where you don't have access to the op notes, and we were able to duplicate this with hip replacement implants. CAT scans, now we're able to basically look at multiple variables that are used for hip arthroscopy from center edge angle to version angle to come up with a solution where we can take multiple variables and really appreciate what matters when it comes to outcomes, and so the list just goes on and on about ways you can take multiple variables to solve a complex problem, and so skipping ahead, the tide really turned with invited reviews from multiple authors from these journals as it basically increased, and now we're at this inflection point where we're probably too enthusiastic about AI, and from these three years, these are the number of articles that basically came out on the subject of AI. A hundred and seventy-eight papers came out. Basically a third of them were narratives or reviews. These are very hard publications to study, so I'd recommend if you want to go into this, you should have a technical person beside you, and this basically is my rules for how to spot research that could be misleading. The first would be you have to assess whether AI is actually needed to answer the question. The second is making sure that you're not just going through a harvest database, but flipping the question around to ask it using machine learning. The third being those authors who equate machine learning to AI and vice versa. The fourth would be saying, here's a study that shows a machine learning or AI shows, proves, or predicts. That's equivalent to saying a t-test shows, proves, or predicts. We don't do that, so we shouldn't expect that. The fifth being where the authors may or may not have provided the full open source code. The sixth would be applying tripod guidelines, which only work for regression modeling, and the seventh would be releasing a model too soon into the world without having external validation. Just my experience with industry, if you have small data and understand the sensors that we carry around on our phones, there's over 50 different sensors on our phone. We basically came together to take sensors from a wearable knee brace with the sensors of a smartphone and use AI to, again, combine those multiple input data to tell you how compliant your patient was with your physical therapy, how many steps they were walking, and then what their problems looked like over a period of time. We were able to shift the paradigm to continuous home monitoring over constant post-op surveillance. As we think about the cost of care, cutting out those visits can be very important. We basically were able to take this data and have some insights that would be actually useful at the clinical level, and right before the pandemic happened, got acquired by Globus, and that was one way that AI was helpful. Third, my big passion project is essentially insurance authorization. We all know that's a huge problem, and so there's a lot of people in medicine who have to work with us to be able to know what cases should be approved and what should not be approved, and the data is very confusing for those. What we're able to do is look at various outcomes and track records of surgeries and say which one has a good rapport. I recommend this book to anyone who is truly interested in this because it flips the narrative and helps everyone understand that AI can actually automate the elements in our healthcare that we don't want to do, and lets doctors be doctors again so we can look patients in the eye and talk to them about their clinical diagnosis and their plan, and obviate the need for 25 clicks just to order Tylenol. In my, again, level 20 expertise, I basically see the big issue in healthcare being the rise of administrators, and the way we can actually think about AI, in my humble opinion, is from the perspective of control. So if you were an administrator whose livelihood depends on performing tasks, who are you going to automate, healthcare providers or administrators? So we made this mistake when it came to business of healthcare beforehand where we basically gave away our business to them, and I think those who actually control the levers for AI will essentially run the future of medicine, so I don't want this to be a self-cannibalizing concession. Third, and finally, we shouldn't be automating physician tasks. We should actually consider automating administrative tasks, and that's where we should dedicate most of our efforts. Thank you.

AI basics

automation

computer knowledge

data importance

future of healthcare