CAR T-cell therapy has revolutionized cancer treatment for B cell and plasma cell malignancies, turning patients’ own immune cells into cancer-fighting superheroes—but even superheroes can cause collateral damage. Enter the phenomenon of immune effector cell-mediated hematotoxicity (ICAHT): where the cure packs a little too much punch, with blood cells caught in the crossfire. ICAHT can lead to serious complications following CAR T-cell therapy, requiring blood transfusions, extra growth factors, and dealing with severe infections—some of which can be fatal.
The CAR-HEMATOTOX model is an established predictive tool designed to assess a patient’s risk of developing severe blood-related side effects (like prolonged low blood counts) after CAR T-cell therapy, ideally helping clinicians identify those who may need closer monitoring or early interventions. In many ways, it’s like an early GPS system and was not specifically developed to predict ICAHT as defined per newer criteria or roadmaps.
A model published in Blood Advances led by Dr. Emily Liang, a hematology and oncology fellow at Fred Hutch and the University of Washington, alongside researchers from Dr. Jordan Gauthier's group in the Clinical Research Division, aims to be an updated and more accurate navigation app to help clinicians steer patients toward better outcomes. Using a cohort of 691 patients that received CAR T-cell therapy for blood cancers at Fred Hutch, Liang and Gauthier have developed two early ICAHT (eICAHT) Prediction Models (eIPM) based on several critical pre- and post-lymphodepletion factors (eIPMpre and eIPMpost).
Using univariate logistic regression, a statistical method that evaluates the impact of individual factors on severe eICAHT risk, Liang and Gauthier assessed 58 variables, finding several that were associated with severe eICAHT. Patients with acute lymphoblastic leukemia (ALL) appear to be at greater risk, along with those who have abnormal blood counts before undergoing lymphodepleting chemotherapy, including low levels of infection-fighting white blood cells (absolute neutrophil count, or ANC) and high levels of lactate dehydrogenase (LDH), an enzyme linked to tissue damage and aggressive disease.
Inflammation also plays a key role, with elevated levels of C-reactive protein (CRP), ferritin (a protein that stores iron but also rises in response to inflammation), and interleukin-6 (IL-6), a molecule involved in immune system activation, all associated with a higher likelihood of severe eICAHT. Additionally, blood clotting abnormalities, measured by D-dimer levels, appear to be another important risk factor. After infusion, patients who experience further increases in inflammatory markers such as CRP, ferritin, and IL-6, as well as worsening blood clotting issues, tend to be at even greater risk. The severity of CAR T-cell therapy side effects, including cytokine release syndrome (CRS)—a widespread inflammatory reaction—and neurotoxicity, which affects brain function, also correlate with severe eICAHT.
Liang and Gauthier took all this information and trained a predictive model using a method called LASSO, which sorts through a pile of information and keeps only the most important pieces. This allows the model to focus on key factors—like disease type (ALL vs. other), pre-lymphodepletion neutrophil count, platelet count, LDH, and ferritin in the case of eIPMpre—while discarding less relevant details, making it more accurate in predicting which patients are at higher risk for severe eICAHT. They also performed a decision curve analysis to test the model’s real-world utility and were pleased to find that using either eIPMpre or eIPMpost in clinical decision-making led to better outcomes for patients—something that’s always the ultimate goal at Fred Hutch!
But what are these clinical decisions being determined? For patients at a higher risk for iCAHT, there are several steps that clinicians can take. They can modify the intensity of the lymphodepletion protocol to avoid excessive suppression of the immune system, or plan for the possibility that the patient might require an infusion of their own stem cells, called an ‘autologous’ transplant. They could also choose to employ additional agents to fight potential infections while the patient is immunocompromised.
The final step was to take this new and improved model out for a spin and apply it to a set of “test” patients. While a majority of the cohort was included in the “training” set that built the model, splitting data into training (70%) and test (30%) sets is a crucial step in building computational models, and it’s done to ensure that the model is not just memorizing the data but is able to generalize well to new, unseen data. When tested, both models accurately predicted severe eICAHT, with strong results showing how well they can distinguish high- from low-risk patients (AUROC of 0.87 for eIPMPre and 0.88 for eIPMPost). The closer the AUROC score is to 1, the better the model is at making accurate predictions.
As for what’s next? Liang is already thinking ahead: ‘While our work takes a significant step forward in the field of CAR T-cell therapy, the big question remains—can our model, built on data from Fred Hutch patients, be applied to those at other institutions? We’re eager to test its power by validating it with external datasets.’ This next phase will be crucial in expanding the reach of the model and ensuring it can make a lasting impact on patient care, not just at Fred Hutch but worldwide.
The spotlighted work was supported by the National Institutes of Health, the Fundación Espanola de Hematología y Hematerapia, and Swim Across America.
UW/Fred Hutch Cancer Consortium members Drs. Qian Vicky Wu, Mohamed L. Sorror, Rahul Banerjee, Andrew J. Cowan, Ajay K. Gopal, Mazyar Shadman, Alexandre V. Hirayama, Brian G. Till, Folashade Otegbete, Ryan D. Cassaday, Filippo Milano, Cameron J. Turtle, David G. Maloney and Jordan Gauthier contributed to this work.
Liang EC, Huang JJ, Portuguese AJ, Ortiz-Maldonado V, Albittar A, Wuliji N, Basom R, Jeon Y, Wu Q, Torkelson A, Kirchmeier DR, Chutnik A, Pender BS, Sorror ML, Hill JA, Kopmar NE, Banerjee R, Cowan AJ, Green DJ, Gopal AK, Poh C, Shadman M, Hirayama AV, Till BG, Kimble EL, Iovino L, Chapuis AG, Otegbeye F, Cassaday RD, Milano F, Turtle CJ, Maloney DG, Gauthier J. Development and validation of predictive models of early immune effector cell associated hematotoxicity (eIPMs). Blood Advances. doi: https://doi.org/10.1182/bloodadvances.2024014455.