Machine Learning-Enabled Blood Test Predicts Immunotherapy Response in Lymphoma Patients

However, over half of non-Hodgkin lymphoma (NHL) patients who fail to respond to conventional treatments also experience relapse or disease progression within six months after undergoing CAR T therapy. In response, a new tool using machine learning has been developed to predict how well an NHL patient might respond to CAR T-cell therapy prior to starting the treatment.

Called InflaMix (Inflammation Mixture Model), this innovative tool was developed by researchers at City of Hope (Duarte, CA, USA) to evaluate inflammation, which is considered a potential cause of CAR T failure, by testing for various blood biomarkers in 149 patients with NHL. Using machine learning, a form of artificial intelligence that analyzes data through algorithms to identify patterns and draw conclusions, the model was able to identify an inflammatory biomarker through a set of blood tests that are not typically used in standard clinical practice. By examining the inflammatory signature identified by InflaMix, the researchers found a significant association with an increased risk of CAR T treatment failure, including a higher risk of death or relapse. Notably, InflaMix is an unsupervised model, meaning it was trained without prior knowledge of clinical outcomes.

The research team noted that the machine learning model is highly adaptable, showing good performance even when using just six commonly available blood tests—tests that are typically evaluated for lymphoma patients—to assess InflaMix's functionality with less data. This is an important feature because it suggests that the test could be accessible to a broad range of lymphoma patients. To validate their initial findings, the researchers studied three independent cohorts, comprising 688 NHL patients with diverse clinical characteristics and disease subtypes who had received different CAR T products. Moving forward, the team plans to explore whether the inflammation identified by InflaMix directly impacts CAR T-cell function and to investigate the underlying sources of this inflammation.

“These studies demonstrate that by using machine learning and blood tests, we could develop a highly reliable tool that can help predict who will respond well to CAR T cell therapy,” said Marcel van den Brink, M.D., Ph.D., president of City of Hope Los Angeles and City of Hope National Medical Center, and a senior author of the paper published in Nature Medicine. “InflaMix could be used to reliably identify patients who are about to be treated with CAR T and are at high risk for the treatment not working. By identifying these patients, doctors may be able to design new clinical trials that can boost the effectiveness of CAR T with additional treatment strategies.”

Related Links:
City of Hope