So first of all thank you everyone for the opportunity to talk to you here today. My name is Ron Gilat and this is evidence-based machine learning algorithm to predict failure following cartilage procedures in the knee. This study was done at Rush and I'd like to thank my co-authors and mentors for the help on this study and these are our disclosures which none are relevant for this current study. So the problem is how to manage cartilage injuries in the knee and we know we have quite a few options I'm just gonna run through them because I'm sure all of you know these but the thing is that it's pretty hard for us to choose a specific treatment for a specific patient and therefore a lot of surgeons try to figure out come out with these algorithms that most of them focus mainly on defect size maybe one or two more variables but nothing more and you can see that if you go down the branches of each algorithm you are still stuck with quite a few options so we still don't have an answer on how to choose a specific treatment for a specific patient and so the purpose of this study was to develop machine learning algorithms to predict failure following these procedures to detect the most valuable features associated with failure and then to compare risk of failure of specific patient procedure combinations and so this was a retrospective study minimum two years outcomes failure was defined as revision cartilage surgery and or knee arthroplasty well we collected a bunch of variables preoperative and intraoperative pretty much everything that we can find that we thought might have an effect on outcomes we use 70% of our data for training and 30 for testing we trained and internally validated for machine learning algorithms we assess their performance and their algorithm fidelity using the Lyme analysis this is the workflow the classic workflow of this type of algorithms nicely published by Kyle Kunz from HSS and so what we got is a 1091 patients which make this one of the biggest cartilage cohorts out there we had a mean follow-up of 3.5 years we had 18% failures which is concurrent with prior literature on these procedures and so I'm just gonna run through what we collected we don't have time to run through all of these but we pretty much try to collect and it's not all the data but we try to collect everything that we could now what we got is that the ten important most important features predicting failure following these procedures were symptom duration age BMI lesion grade total lesion area number of previous surgeries number of lesions in the knee gender athletic level and traumatic etiology now these were our four machine learning algorithms and you can see that the random forest algorithm was found to be the best performing algorithm now how can we use these to choose the best treatment options for our patients so this is what is really nice about these tools you can do a patient specific analysis this is an 18 year old male you just plug in the specifics of this patient he had a 25 by 25 LFC lesion and you can see that if you are the surgeon considering OCA or Macy for this patient this patient might be better off with an OCA and so I can show you a ton more examples and you can play with it as much as you want and this is what is really nice about these tools and so what we found out is that machine learning algorithms were accurate in predicting the risk of failure following these procedures we were able to compare specific patient procedure combinations now it's important to understand this study is retrospective it's not externally validated there are a lot of limitations so it's not like any of these machine learning algorithms can replace the surgeon or will replace the surgeon anytime soon but if you are considering one two or three more options then these might help you now anyone who's interested in machine learning in orthopedics this is a really nice editorial in arthroscopy talking about what we know and what we still don't know and how we can move forward with machine learning and use it for better now what you can do in the meantime is just open up chat GPT and this is me asking what's the best treatment for two by two cartilage defect in the MFC and then it's amazing you know it comes back with a response pretty much everything that we talked about but tells me to go and consult a specialized orthopedic surgeon but then I go I am an orthopedic surgeon I'm in the operating room right now and I need to make a quick decision can you please narrow down the option for me and then the AI model apologizes for not being able to see or examine the patient but still narrows down the options for ACI and oats but then I go I'm in a lot of stress right now this is my first time I was not prepared for this surgery and I do not have the instruments for oats can you please choose one and then you see that the bot becomes a bit compassionate also a bit maybe worried for my patient but in the end offers me to do an ACI so definitely we are living at exciting times and thank you all for listening

machine learning algorithm

failure prediction

cartilage procedures

random forest algorithm

orthopedics