What Are the Latest Artificial Intelligence Advancements for Myeloma?
What Are the Latest Artificial Intelligence Advancements for Myeloma? from Patient Empowerment Network on Vimeo.
What artificial intelligence advancements have emerged for myeloma? Expert Dr. Ola Landgren from University of Miami Sylvester Comprehensive Cancer Center discusses the IRMMa prediction model for myeloma care, factors that go into the IRMMaa model, and potential AI advancements for myeloma.
Download Guide | Descargar Guía
See More from START HERE Myeloma
Related Programs:
|
|
|
Transcript:
Lisa Hatfield:
So that kind of leads to the next question that is really an exciting area. I know it’s not necessarily new, but newer is artificial intelligence. And I know I was reading an article about one of, that you and your colleagues have worked on a newer project and I don’t know if you pronounce it IRMMa or not, but using these large databases to help predict I think, it’s the response of treatment in some patients. So can you talk about that a little bit and tell us about that development and what developments are exciting with artificial intelligence in cancer, in particular myeloma?
Dr. Ola Landgren:
Yeah. So you mentioned the study we just published. We published a model that we call IRMMa and that stands for individual risk prediction for patients with multiple myeloma. So what we were thinking was at the current time, all the existing models are pretty much providing the average patient’s predicted outcome. So think about it is like it’s a probability measure.
So you say, if I take this about therapy, what’s the predicted average outcome for patients that take this therapy, say, five years later? So on average, say 70 percent of patients are free from progression. That sounds pretty good. The problem is that you don’t know if you are in the group, 70 percent group that didn’t progress or if you’re in the 30 percent that did progress.
So where are you as an individual? So it’s almost like looking at the weather app on your phone. If it says it’s a 70 percent probability of sunshine and then you go outside and it’s raining, it’s because it didn’t say that it’s 100 percent probability of sunshine. So if you think about another situation would be, say, in a GYN clinic, if a woman were to come and ask the doctor, am I pregnant? Yes or no? You couldn’t say it’s 70 percent probability. You would say, yes, you’re pregnant or not pregnant.
So for myeloma, we have for a long time been living in these weather report systems where we say 70 percent or 30 percent. And we want to go in the other direction of the pregnancy test, where we actually can say for someone with this particular disease profile, with this treatment, this is where this is going to take us. We worked on this project for almost four years, and we worked with a lot of other groups around the world that have a lot of data. And they have graciously agreed to collaborate with us and share their data sets. The beauty with this collaboration, there are many beauties of it, but one of them is that people don’t treat patients the same way.
And that actually has allowed us to say for patients that have a particular biological or genomic makeup, if you’re treated this way or that way or the other way or a fourth way and so forth, which of these different treatments would make patients have the longest progression and overall survival? So if you have a large database, you can actually ask those questions. So you can say that you profile individual patients in full detail and you put them in detailed buckets instead of grouping everybody together.
And now if you add a new case, if a new patient is being added and you say, which bucket would this individual fit? Well, this is the right biological bucket. You can then use this database to say out of all the different treatment options, which treatment option would last the longest, which would give the best overall survival? Other questions you could ask is also, for example, you have a patient with a certain biological workup or makeup. And you say, if I treat with these drugs, will the addition of, say, transplant, will that prolong progression for his survival?
And you can go into the database and the computer will then say, I have these many patients that have this genomic makeup and these many people that were treated with this treatment with transplant versus the same treatment without transplant. There was no difference in their progression or overall survival. So then the computer would say, it doesn’t add any clinical benefit, but there could be another makeup where the answer is opposite, but transplant actually would provide longer progression for his survival. I think the whole field of medicine is probably going to go more and more in this direction.
So what we want to do is to expand the number of cases. So we are asking other groups around the world, if they have data sets with thousands of patients, they could be added to this database and we could then have more and more detailed information on sub-types of disease and more and more treatment. So it will be better as we train it with larger data sets. The model is built as an open interface so we can import new data. And that’s also important because the treatments will continue to change. So we, for example, say I have a patient that has this genetic makeup. I was thinking of using a bispecific antibody for the newly diagnosed setting.
How is that going to work? The computer will say, I don’t know, because we don’t have any patients like that in the database because that’s not the data, type of data that currently exists from larger studies. But let’s say in the future, if there were datasets like that, you could ask the computer and the computer will tell you what the database finds as the answer. But if you go for another combination, if that’s in the database, it would answer that too. That is where I think the field is going.
And lastly, I would say we are also using these types of technologies to evaluate the biopsies, the material. We work with the HealthTree Foundation on a large project where we are trying to use computational models to get out a lot of the biological data out of the biopsies and also to predict outcomes. So I think artificial intelligence is going to come in so many different areas in the myeloma field and probably in many, many other fields in medicine.
Lisa Hatfield:
Yeah, that’s wonderful. So if you have a newly diagnosed patient coming in to see you, do you use this model and explain to them what came back to you? Or is it right now just collecting the data for this dataset?
Dr. Ola Landgren:
So at the current time, we have to be cautious. We cannot promise things that we cannot deliver. So we have clearly said this is a research tool. It was just published less than a month ago in the Journal of Clinical Oncology. It is publicly available. The paper is available. Anyone can go there and download the paper and anyone can also in this paper find there’s a website. You can actually see the database as well. And there is a lot of corresponding material online on how to interpret. So for now, it is a research tool. But I think it’s possible in the future that we would start considering using it. And if other people find it useful, maybe they will do that, too. But for now, it is a research tool.
Lisa Hatfield:
As a patient, I would be very intrigued with that and what might come back. Just like you said, it’s a tool to maybe help identify new treatments or whatever. I did ask ChatGPT if there’s a cure for myeloma, if there will be one in the next 10 years. I didn’t really love the answer. It was a little bit vague. But yeah, I like looking into AI a little bit and ChatGPT and all that. So thank you so much for that overview.
Dr. Ola Landgren:
This is exactly like ChatGPT. It works the same way, but it’s only centered around multiple myeloma. For now, the way we have done it is that we have to start somewhere. As I told you, it’s a four-year work effort with a lot of people. We have like 10, 15 people working day and night on this project. So we started on the newly diagnosed patients. But we intend to scale it up. We intend to build in a lot of new features. And, of course, we want to add more datasets to it. And last thing I want to say to you, I find it very, very fascinating how you can, as a human being, as a researcher, you can ask the computer, and it will give you answers back that you didn’t think about yourself.
So you talk about ChatGPT. So we are using our model. We can have the model looking at biopsies. We can ask the computer, what is this biopsy? What’s going on? And the computer will say this and this and this genetic feature is going on. And then we ask the question, how did you conclude that? And then the computer will say, look here. So it would then label areas in the biopsy and say, I looked here. It doesn’t yet tell us what it found, but it tells us where it found it or where it looked to come to its conclusion.
So when it finds the right conclusion, we are looking in these areas to see what’s going on in these areas and how does it look different from other areas or in other samples. So having a dialogue with the computer can give us new insight. It’s almost like taking a young kid and you go out for a walk and you look and you see a lot of buildings and the kid looks down and looks and finds a little flower on the ground. And you say, oh, my gosh, I missed that one. The kid would not miss it. The computer is the same way.
Lisa Hatfield:
Yeah, that’s a great analogy, too. And I think we could have a 10-hour conversation about that, particularly with myeloma, because it’s so complicated, complex. So I hope we can in the future again talk about that.
Share Your Feedback:
Create your own user feedback survey
