Current approaches to assessing risk in patients with multiple myeloma rely on clinical variables such as patient age and stage, but these models generally do not predict outcome well. A new approach developed by Cleveland Clinic researchers surpasses these older models by combining clinical staging, genomics, and machine learning to more accurately predict survival.
advertisement
Cleveland Clinic is a nonprofit academic medical center. Advertising on our site helps support our mission. We do not endorse any non-Cleveland Clinic products or services. policy
The model represents a step toward precision medicine for patients with multiple myeloma and can help oncologists more accurately target treatment based on an individual patient’s disease and risk, said lead author Shahzad Raza, M.D., a hematology-oncologist at the Cleveland Clinic Cancer Institute. It could also be a tool to help researchers better understand multiple myeloma itself, how it develops, and why some patients still have much worse outcomes than others.
“We have made great strides in the past few years, and treatments for multiple myeloma have improved dramatically, but despite these advances, not everyone will have good outcomes,” Dr. Raza said. “We are missing something, and the answer lies in understanding the basics.”
Identifying “very high risk” patients
Multiple myeloma accounts for approximately 10% to 15% of blood cancers. Although recent advances in treatment have allowed some patients to survive years or even decades with the disease, a minority of patients still have significantly worse outcomes.
Previous studies have shown that mutations in a gene called TP53 are found in about 5% of newly diagnosed cancer patients and 25% of patients with advanced cancer, making them more likely to have a malignant disease with rapid progression, resistance to treatment, and worse survival rates. But there is variation within this group, with some “very high-risk” patients experiencing significantly worse outcomes, Raza noted.
“The disease behaves very differently between the two high-risk patients,” he says. “So why is this happening? And how can we accurately predict the outcome in an artificial intelligence setting? That’s the goal of our model.”
The new model was developed by Sriram Subramanian, a high school student working in Dr. Raza’s lab. This grew out of an effort to find better tools to identify these patients and determine a more accurate prognosis.
“Two people with the exact same diagnosis can have completely different outcomes. We wanted to see if machine learning could find patterns in gene expression that explain why,” Subramanian says. “When the six genes we discovered matched how long patients lived, and even how cancer cells responded to different drugs, we knew this model could help doctors choose treatments.”
The research team analyzed data from a previous Multiple Myeloma Research Foundation treatment response study involving a total of about 753 patients, 36 of whom carried TP53 mutations that increased their risk. The team then used machine learning to compare gene expression between the two groups and identified signatures in six genes that could help predict how a patient’s disease would progress.
“It’s not just the TP53,” he explains. To develop the new model, the team combined factors such as clinical staging, TP53 mutations, and six gene signatures. We then used the CoxBoost machine learning algorithm to predict outcomes with 18% higher accuracy than current risk stratification approaches.
Application to drug sensitivity, basic research
The researchers then used this model to investigate drug sensitivity, applying different drugs to cells from patients with different risk levels. Although further studies are needed to verify the findings, the researchers found that certain drugs were more effective on cells from high-risk patients, while others were more effective on cells from low-risk groups.
“It’s a great concept,” Raza said. “This means this model could be incorporated into human trials to potentially discover drugs that are more effective for individual patients, enabling the move towards precision oncology.”
“Our model is a more accurate predictor of survival,” Dr. Raza added. This model could be another tool to study not only prognosis but also the pathways of multiple myeloma and the mechanisms by which multiple myeloma becomes highly progressive in some patients. “We’re interested in understanding why, even though we’re doing everything possible, some patients still have a poor prognosis.”
The project, “Machine learning (ML)-based six-gene signature for risk stratification and therapeutic target identification in multiple myeloma,” was presented at the American Society of Clinical Oncology annual meeting.
