Commentary

Video

Although AI Tools Are Advancing, Expert Says There Is a Need for More Tools Post-Diagnosis in Myeloma

Although Madabhushi said there are emerging tools for clinical decision support, he said there is a greater need for tools further downstream in the post-diagnosis stages.

In an interview with Pharmacy Times, keynote speaker at the International Myeloma Society 2024 Annual Meeting Anant Madabhushi, PhD, executive director of the Emory Empathetic AI for Health Institute, discussed how he has seen artificial intelligence (AI) advance in the myeloma space. Although Madabhushi said there are emerging tools for clinical decision support, he said there is a greater need for tools further downstream in the post-diagnosis stages.

Q: Where do you see the greatest potential for AI in multiple myeloma?

Anant Madabhushi, PhD: Again, I will give some initial thoughts based on my very, very preliminary appreciation of this disease. I suspect my perspective will probably change dramatically after the meeting, after I've learned a little bit more, but I will give you a preview about some of the work that we're doing that I'll be showcasing at the meeting, that I'm very excited about. Now, that enthusiasm may be dampened at the meeting, and perhaps it may be that we have a potential solution to a problem that doesn't exist, but one of the things that our group has been doing has been looking at imaging data and finding with AI that we seem to be able to predict the onset of multiple myeloma in patients who don't have the disease but will likely develop the disease in a 10-year time frame. And we generated some data, we validated it on a small number of patients, and the results look very, very promising. And to me, that ability to—even in advance of the disease developing—be able to predict that seems like a very compelling opportunity. Again, I add the qualification that at the meeting I might discover that all my enthusiasm was perhaps misplaced, and that we have a solution to a problem that doesn't exist. But it seems to me that the ability to identify onset of the disease means that clinicians can start to think about intervention strategies perhaps years in advance, before manifestation of the disease itself. And it seems like there should be something we can do about that, to potentially make some alternative choices, or make some decisions to change the trajectory of the disease. So, that's something that I'm looking forward to sharing with the audience and certainly getting their perspectives.

Q: How could AI be of particular use in treatment selection, particularly as the body of available treatments grows?

Madabhushi: That's a great question. I think this is an area that our group has been very, very focused on, and I'll be sharing with the audience the work that we're doing, particularly in the context of lung cancer, where we've been developing AI algorithms to analyze routine CT scans in patients with lung cancer and predicting, in advance of the treatment, whether somebody's going to respond to chemotherapy or immunotherapy or CDK4/6 inhibitors in the context of metastatic breast cancer. So, I'm really excited to share some of that data with the audience and try to get their perspectives. I'm fairly convinced that there is a big role for AI to play in treatment selection. You know, like you correctly said, the armory, so to speak, of drugs and treatment regimens continues to grow, which is a good thing, but it also comes with complications. It comes with more questions about the appropriate treatment strategy, and given that we have this choice, which is good, we also need the tools to help make the informed choices. And I think that's really one of the areas that I think we can do better, right? I think the community can do better, and this is something that I really look forward to this meeting, to exploring, to seeing how some of the tools that we developed in lung and breast cancer potentially might help with treatment selection and outcome prediction in the context of multiple myeloma.

Q: How close are we to actual implementation of these tools? Are they already being used in practice?

Madabhushi: I think it's evolving. I think that there are some tools that are already in play in the radiology space, in the ophthalmology space, and the cardiology space. We do have some AI tools. I believe the FDA has so far approved 700 to 800 of these so-called clinical decision support tools or AI machine learning tools, so we are seeing a lot more tools. I will argue, though, that there's not been a lot in the context of post-diagnosis further downstream. So, looking at, you have the disease, now how do you treat the patient? How do you predict outcome? I think there's a lot less in that particular space, and going forward, I suspect that there's going to be a lot more innovation that should be focused and will be focused in that particular area. So, in terms of implementation, really, there's a lot of AI work that's happening in the diagnostic realm that is getting integrated into clinical workflows further downstream. There's less of that, but hopefully we'll start to see some of these tools entering the clinical continuum and the clinical workflow to address some of those downstream questions about treatment response and adverse events and outcome prediction. Hopefully, we're about maybe 3 to 5 years away.

Q: What key challenges have you identified with the use of AI in myeloma, and how are these being addressed?

It's 2-fold. I mean, no.1 is implementation, and I think that we know how to do implementation, or at least we're learning more and more about how to do implementation of AI. The flip side to that is that we know that implementation without adoption is largely irrelevant, right? Because ultimately, the clinicians need to be able to trust these algorithms. You could have FDA approval and that doesn't matter. You know, a clinician may not want to actually use those algorithms if she or he does not feel that there is a modicum of trust with regard to how these algorithms are working and really trusting the outcome prediction. So, I would argue that implementation is only one side of the coin. The other piece is engendering trust in our clinicians with regard to what the AI tools are providing so that they are willing to now actually adopt and employ these technologies as part of their workup and management of patients.

Related Videos
3 KOLs are featured in this series.
3 KOLs are featured in this series.
Patient receiving cancer care from provider -- Image credit: Rido | stock.adobe.com
3 KOLs are featured in this series.