ASH 2025 | Real-time multimodal AI for early risk prediction in BCMA CAR-T–treated myeloma

I’m really excited to be presenting this work. The goal of this research is very simple. How can we do better in helping us understand upfront which patients are going to do really well and which patients are going to not do so well with CAR T-cell therapy in particular. It’s a big deal for patients. It’s an investment for them and their families and their time. And so we’re really trying to utilize these new techniques we have available, so artificial intelligence, or AI, is giving us a whole new set of tools to perform really great science. And so by leveraging some really great AI techniques, we’re able to harness the data that we collect from lots of these patients as part of routine care, so not just their treatment history but all the routine lab investigations that we do as part of our standard of care. 

And in addition, what we found was if we start with just the clinical information that we have, so the patient’s history, how they are, and their routine lab tests, we can really identify groups of patients who’ve got very different outcomes. But when you layer in on top of that information that we can leverage out of their PET scans, so a measurement called metabolic tumor volume, or MTV, which basically is a sort of global sum of all the areas that are lighting up on the PET scan. If you layer that in, you improve the discriminatory power of the model, and what that means is you improve the way that the model can understand which patients are going to do really well and which patients are not going to do so well. And then if you add into that a serum biomarker that’s increasingly available called soluble BCMA, so basically measures how much of those little bits of BCMA are floating around in the bloodstream, if you layer those three things on together, you really improve the discriminatory potential. 

And the goal of that is to really help us understand, okay, you are the kind of patient who’s going to do incredibly well and have a very good outcome from this CAR T-cell therapy versus you are the kind of patient who is going to be at high risk of early treatment failure and therefore maybe we should treat you in a more different way, maybe enhance our monitoring or add in alternative therapies or think about clinical trial options, we’ve got all these great trials looking at dual-targeting CARs or combination therapies. So that’s what we’re really trying to help understand. All of this information that we use to build a model, it’s all being pulled from data that we have before the patient gets treated. So when you’ve still got the opportunity to make a decision, and ideally, we’re going to try and leverage that to help guide physician decision-making in the future when we feel like we’ve got something super robust that we can share with the community.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.