A Predictive ML Model For Cancer Early Relapse

Value Proposition

Diffuse large B-cell lymphoma (DLBCL) is an aggressive and very prevalent subtype of non-Hodgkin’s lymphoma which accounts for 4% of all cancers in the US. Presently, clinicians have access to multiple CAR T cell therapies to treat patients. One such therapy, axi-cel, was recently approved for second-line treatment for patients with early relapse or refractory disease after first-line therapy. However, the trial data also revealed that nearly 60% of the patients treated had some kind of treatment failure and subsequent relapse, with most of these occurring within a year of treatment. There is an unmet need for clinically interpretable strategies that identify patients susceptible to early relapse and severe toxicity to further guide their treatment journey.

Technology Description

UCSF investigators developed a clinician-facing, machine learning (ML) model that identifies patients at higher risk of developing early (within six months of therapy administration) cancer relapse following axicabtagene ciloleucel (axi-cel) CAR T cell therapy. This ML model can assist clinicians in identifying patients who may benefit from additional treatments to further extend CAR T cell therapy, thus leading to better health outcomes. The model has been trained on real-world data from the University of California Health Systems, encompassing 416 adult patients. While this model was developed for usage in patients receiving axi-cel, it could be adapted to patients receiving other CAR T cell therapies.

Related Materials

• Wang, M., Komanduri, K. V., Datta, D., Patel, A., Whitaker, B., Belov, A., Rubin, B., Rohit Vashist, Rodriguez-Monguio, R., Cinar, P., Anderson, S. A., & Butte, A. J. (2025). An AI model classifies risks of early relapse post–CAR T-cell therapy in a multicenter real-world population with DLBCL. Blood Advances, 9(22), 5837–5852.

Data Availability