ASH 2025 | Race-agnostic AI models maintain accuracy in predicting M-protein levels in multiple myeloma

The artificial intelligence models and machine learning, they are increasingly used in myeloma. Myeloma is one of those diseases that disease volume is based on the number rather than imaging and so forth. So it’s more prone to have an AI model, predictions, tools that we’re going to use in myeloma more and more. We ask a very fundamental question that whether race-agnostic models will perform suboptimally or not. And that’s I think is important because myeloma, along with triple-negative cancer and breast cancer and prostate cancer, they are the three main most common malignancies among black and African Americans. So we ask whether adding race and considering race and ethnic group will change the robustness of the myeloma models or not. The way that we try to answer this question, we try to predict M-spike based on systemic signatures of myeloma. Let’s say kidney function, anemia, protein levels, electrolyte levels, and all of that. Based on that, we try to predict M-spike based on previous M-spikes. And we form two models, with and without race. Race is just self-reported race. And we found out that adding race really does not add predictive power to AI models in this example. That’s very good news because always there is a limitation of AI models if really with the race we’re going to have a change in the robustness of the model. So this is, considering that myeloma is very dominant in the black population, very good news that really if we miss the race we are not having suboptimal models among African Americans.

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