As treatment options for multiple myeloma continue to expand across the disease continuum, efforts and tools to improve risk stratification and treatment selection will be imperative toward driving personalized medicine approaches for this patient population, according to C. Ola Landgren, MD, PhD.
“I strongly believe that integrating the biology of the disease and the clinical variables and [using] that information to propose a better treatment is the way the field will be going,” Landgren said in an interview with OncLive® during Multiple Myeloma Awareness Month, observed annually in March. “Precision medicine is here to stay, and that’s going to drive all the good outcomes in the future for patients with myeloma.”
At the Sylvester Comprehensive Cancer Center at the University of Miami in Florida, Landgren and colleagues are developing CORAL-histomorph, which is a novel, next-generation machine-learning model designed to predict genomic abnormalities from routine histopathology bone marrow biopsy images.1,2 The model is also intended to integrate clinical and treatment data to provide treatment guidance on an individual patient level. Although it remains a research tool rather than a current resource for clinical practice, Landgren and colleagues have explained that CORAL-histomorph can accurately predict genomic subgroups based on histopathology slides, and the scalable architecture allows for continuous databased expansion, which could extend to other tumor types.1
In the interview, Landgren explained the need for improved risk-stratification strategies in multiple myeloma, how tools like CORAL-histomorph could aid in determining individual risk and optimal treatment approaches, and the continuing goal of bringing personalized therapy to this patient population.
Landgren is a professor, chief of the Division of Myeloma in the Department of Medicine, director of the Sylvester Myeloma Institute, co-leader of the Translational and Clinical Oncology Program, and the Paul J. DiMare Endowed Chair in Immunotherapy at the Miller School of Medicine at the University of Miami and the Sylvester Comprehensive Cancer Center.
OncLive: What are the current approaches for risk stratification in multiple myeloma, and what is the importance of continuing to refine this process?
Landgren: Patients with multiple myeloma are risk stratified using different types of prognostic models that have been around for a long time. The first models came in the 1970s, and we have seen updates in the [decades since]. [The University of Miami] has been involved in the development of many of these models.
Current models put patients in categories of being lower risk, intermediate risk, or higher risk for various outcomes, but these are really risks within a group. Patients are not individualized to determine their risk. It’s the group risk we’re talking about. That means a patient can be in the low-risk group, but some [of those low-risk] patients [could still have] adverse outcomes; [conversely,] patients could be in the higher-risk group but not have an adverse outcome. Because you don’t know if [a given patient] is going to have the adverse outcome, it ends up being a probability game where [patients] have lower-risk, intermediate-risk, or high-risk [disease] on a group level.
What we have worked on more recently is to develop individualized prediction models, and I think that’s the direction where the whole field of medicine is going.
AI-Based Risk-Stratification Tools in Multiple Myeloma
- A novel AI model, CORAL-histomorph, can predict genomic abnormalities from bone marrow biopsy images, offering a potential new approach to risk stratification in multiple myeloma.
- Unlike traditional group-based models, CORAL integrates biological, clinical, and treatment data to generate individualized risk predictions and therapy recommendations.
- Although still a research tool, the model highlights how AI-driven precision medicine could improve treatment selection and outcomes in multiple myeloma care.
Researchers at Sylvester Comprehensive Cancer Center at the University of Miami have developed CORAL to help predict individual patient outcomes and guide treatment decisions. How was this model designed?
At the Sylvester Myeloma Institute at the University of Miami, we have worked for a long time developing individualized prediction models for outcomes. More recently, we disclosed and showed outcomes from our CORAL model. We have a CORAL-histomorph model, and we have other types of CORAL models. They take into account all the factors that include disease biology, clinical data from the individual patient, and treatment data. When we know all these variables and we know what the clinical outcome is, we train the model with thousands of cases. Now, when we show the model a patient with certain biological and clinical features, we could ask: Which would be the best treatment to give to this patient? Or we could say: We are planning on giving this treatment; what would be the predicted outcome, and do you have better suggestions? The model is helping us with decision-making.
In the most recent version of the CORAL-histomorph model, we have also integrated what people refer to as multi-modal components of the model. What we have realized is that most people have clinical data; you can capture treatment data for the majority of our patients. However, we may not have all the details of the disease. In order to do full [disease] characterization [via] whole genome sequencing or other types of transcriptomic analysis. Many times, those [data] are not available. A patient may have been diagnosed and worked up somewhere else.
With the CORAL-histomorph model, we take pictures of the bone marrow biopsy from the initial diagnosis, and we have built advanced models that can analyze these pictures and translate that information into the language of all these sequencing technologies.
How are you applying the CORAL tool in myeloma practice?
The CORAL-histomorph model we have developed is a new model. We have just started presenting it at different meetings. We have worked on it for several years, and we have shown that we can take a picture of a bone marrow biopsy that’s been obtained at diagnosis, and if we have the information from fluorescence in situ hybridization [FISH] or sequencing, we have trained the model to [see] what [different presentations] looks like.
Now, with thousands of cases [taught to the model], if we show a picture from the bone marrow of a case where the computer does not know the FISH, the cytogenetics, or sequencing [results], we can ask the computer, and it can then compare the pictures with all those pictures that it has been trained on, when we have that information. We have already found that the prediction [rate] is over 90% correct. In more than 90% of cases, it will, for example, tell us if there is an 11;14 translocation.
We can do a lot of other things [with CORAL]. We can predict the subtype of the disease. Also, by integrating [several] variables, [the computer can] say: Based on what I see, this treatment would be the best, because from all I know, compared with all these other cases that looked like this, if the patient were treated with all these different therapies, this therapy would give the best projected long-term outcome.
We could integrate [this model] for suggestions on [myeloma] management. It’s not yet implemented in clinical care, and I don’t think it will replace doctors, but this will be an important tool going forward. Right now, it’s a research tool, but we are considering this as a tool to enhance and disseminate all the knowledge we have at expert centers to maybe centers where [oncologists] don’t see as many patients [with myeloma].
As the multiple myeloma treatment arsenal continues to grow and provide patients with different options across various lines of therapy, what is the importance of further individualizing risk stratification and treatment decision-making?
The [multiple myeloma] field has come a long way. We have developed many new treatments and strategies for patients, both in the newly diagnosed and relapsed settings. Still, we do not have a curative treatment, for example. We also have patients who don’t do as well with these new, fantastic treatments that the majority of patients can benefit from.
In order to solve those problems and move to the next level, we need to think about the details of the individual cases. Prediction of individual cases takes us back to precision medicine. Instead of giving all the treatment to every patient, where we probably could over-treat some patients, maybe we can give less. For those patients who don’t do as well with the best therapies we have, maybe we need to change those treatments. Maybe we need to add some other drugs. What are those drugs? What are the mechanisms of those drugs that we need? Precision medicine will be the tool.
References
- Rajanna AR, Hurtado Martinez JA, Durante M, et al. Individualized treatment risk stratification using histopathology-based genomics prediction in multiple myeloma: A multicenter study in 1429 participants. Blood. 2025;146(suppl 1):4008.
- Myeloma: how AI is redrawing the map of cancer care. News release. Miller School of Medicine. December 19, 2025. Accessed March 30, 2026. https://news.med.miami.edu/a-new-lens-on-multiple-myeloma/
