ABSTRACT:Castleman disease (CD) is a heterogeneous group of lymphoproliferative disorders anatomically classified by distribution (unicentric CD [UCD], oligocentric CD [oligoCD], or multicentric CD [MCD]). Human herpes virus 8-negative MCD is called idiopathic MCD (iMCD), which includes clinical subtypes with varying phenotypes and responses: TAFRO (thrombocytopenia, anasarca, fever, renal dysfunction and/or reticulin fibrosis, and organomegaly), IPL (idiopathic plasmacytic lymphadenopathy), and not otherwise specified. OligoCD has recently emerged as an intermediate form between UCD and MCD, with unclear clinical behavior, and IPL has not been validated in a Western cohort. We retrospectively analyzed 217 patients with CD evaluated at our institution between January 2004 and August 2024. Survival probabilities were compared using log-rank tests. Overall, 57% had UCD, 20% had oligoCD, and 23% had iMCD. Patients with oligoCD and iMCD more frequently exhibited systemic symptoms than those with UCD. Patients with oligoCD and iMCD had significantly shorter event-free survival (EFS) of 8.9 and 2.3 years, respectively, than those with UCD (not reached; P< .001 and P< .02). Among iMCD subtypes, iMCD-IPL demonstrated a longer EFS than iMCD-TAFRO (P = .02), with no deaths during the follow-up periods. The results validate oligoCD and iMCD-IPL as new subtypes in this Western cohort. The survival outcome in oligoCD was intermediate between UCD and iMCD, with only a subset of patients with oligoCD requiring systemic therapy. iMCD-IPL had a favorable survival outcome in this US cohort, similar to what has been reported in non-Western countries. Tailoring treatment strategies to disease subtypes and vigilant monitoring of oligoCD for progression may improve survival outcomes in CD.

Abstract Large Language Models (LLM) are being widely adopted into the medical field for their impressive ability to analyze and summarize large amounts of text data. These models enable clinicians and researchers to extract meaningful insights from complex datasets and may assist with decision making. Here we present our workflows and application of LLM for the interpretation and summarization of clinical data related to CAR-T cell therapy. Using an LLM (Gemini 2.5 pro) in the Google Cloud Computing (GCP) environment, two applications were developed to analyze CAR-T cell therapy clinical data: 1) extracting and summarizing CRS and ICANS event-related data to streamline the compliance team workflow, and 2) identifying features available at time of CAR-T infusion able to classify patients into high- or low-monitoring needs 14 days post CAR-T. Patient data (vitals, labs, hematology notes, and EKGs) were extracted from the electronic medical record (EMR) using Google BigQuery into SQL tables in GCP. For each application, relevant data fields were retrieved, formatted into JSON objects, and embedded in the LLM prompt for context-aware processing. Both applications have undergone iterative prompt engineering after analyzing the LLM output against ground truth data in the EMR. For the application extracting and summarizing CRS and ICANS events, the LLM was optimized on Mayo Clinic Rochester data. When compared to the IEC compliance database, the LLM achieved 100% accuracy and F1 score for CRS events, and 96% accuracy and 82% F1 score for ICANS events. The match rate for CRS and ICANS grades were 83% and 89% respectively. We applied the same LLM to Mayo Clinic Arizona (MCA) and Mayo Clinic Florida (MCF) cases, who document clinical notes and toxicity flowsheet differently, and achieved an accuracy (MCA: 93%, MCF:93%) and F1 score (MCA: 96%, MCF: 96%) for CRS and accuracy (MCA: 81%, MCF:82%) and F1 score (MCA: 76%, MCF: 75%) for ICANS. LLM was able to capture events missed by manual reviews. For most of the discrepancies where compliance team final adjudication is needed, LLM will be updated to flag discrepancies for review by the compliance team. For the application related to the monitoring needs 14 days post CAR-T, the LLM identified 5 categories predictive of high or low monitoring needs post CAR-T infusion (disease status, inflammatory and tumor burden markers, hematologic status, renal function and performance status). Our model achieved a sensitivity of 85.7%, specificity of 23.8%, and F1 score of 65.5% for our first cohort, compared with data extracted by the LLM from the EMR. For our second cohort, with demographics statistically similar to cohort 1 and using the same 5 categories, the LLM achieved a sensitivity of 83.3%, a specificity of 28.6%, and F1 score of 65.4%. Our workflow and applications of LLM’s provide examples and guidance to others interested in applying LLM’s for clinical and research applications. Citation Format: Emmanuel Contreras Guzman, Matthew Jankowski, Andre De Menezes Silva Corraes, Malvika Gupta, Monica L. Shaw, Madiha Iqbal, Talal Hilal, Saurabh Chhabra, Ricardo Daniel Parrondo, Jody K. Mclean, Kim R. Riester, Kayla Joseph, Melinda Tan, Holly Ross, Cleyonia Barnett, Sylvia Carter, Semy Girmay, Rachel Wolan, Milana Ramsey, Christian Downhour, Kristy Morgan, Shae Sibley, Erica Rushing, Lucy Holmes, Allison Burgstahler, Stephen M. Ansell, Hassan Alkhateeb, Matthew Hathcock, Ramona Bruno, Allison C. Rosenthal, Hemant Murthy, Patrick B. Johnston, Jonas Paludo, Yi Lin. Applications of large language models to CAR-T cell therapy clinical data using Google cloud computing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2740.

The NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Hodgkin lymphoma provide expert, multidisciplinary recommendations for the diagnosis and treatment of Hodgkin lymphoma. The panel convenes annually to update recommendations based on a review of recently published clinical trials. This selection from the NCCN Guidelines for Hodgkin Lymphoma focuses on the management of newly diagnosed and relapsed/refractory classic Hodgkin lymphoma in adults aged 18 to 60 years.

Background CAR-T therapy improves outcomes in relapsed/refractory lymphoma and multiple myeloma, yet clinicians lack tools that deliver individualized, real-time toxicity forecasts after infusion. We recast prediction as time-dependent risk modeling of event processes, treating outcomes such as CRS and ICANS as phenomena that unfold over time rather than single static events. Methods We assembled a retrospective cohort of adults treated with CAR-T. Five data modalities consisting of vital signs, laboratory tests, ECG, echocardiography, and PET-CT radiomics were included. Longitudinal courses from Day 0–30 were transformed into person-time tables and partitioned into 6-hour intervals. Each row represented a discrete observation window paired with contemporaneous measurements, allowing time-varying covariates to update as physiology changed and accommodating censoring when observation ended before an event. Pooled-logistic discrete-time hazard models (XGBoost) estimated daily hazards for first CRS ≥ 1, first ICANS ≥ 2, and the need for ≥ 2 tocilizumab doses within 48 hours of fever. Model outputs included time-resolved hazard trajectories and corresponding cumulative event probabilities. Performance was summarized by per-day dynamic AUCs. Iterative feature reduction honed ∼400 candidate predictors to 15–25 contributory features per endpoint. Results Across 476 infused patients (ages 21–89; mean 62; 63% male) with lymphoma (n=270) or multiple myeloma (n=206), CAR-T infusions occurred from June 23, 2016 until June 3, 2024. Clinically, CRS ≥ 1 occurred in 377/476 (79.2%), ICANS ≥ 2 in 103/476 (21.6%), and ≥ 2 tocilizumab doses were used within 48 hours in 70/476 (14.7%). Time-dependent modeling produced accurate daily risk forecasts throughout the first month; integrated dynamic AUCs across days 1, 3, 5, 7, 10, and 15 were 0.86 for CRS ≥ 1, 0.85 for ICANS ≥ 2, and 0.87 for the need for two tocilizumab doses within 48 hours. The ICANS and tocilizumab models ultimately retained features from all five modalities, whereas the CRS model primarily required laboratory and vital-sign data. Conclusions By modeling hazards rather than only survival probabilities, this multimodal, post-infusion strategy shifts from static prognostication to dynamic risk forecasting. The framework recalibrates in real time as new vitals, labs, ECG, echocardiography, and radiomics data accrue, generating time-resolved hazard curves and cumulative event probabilities any time within 30 days post-infusion. These efforts may support individualized counseling, adaptive monitoring, and early-warning strategies.

We report correlative circulating tumor DNA (ctDNA) analyses from TRANSFORM (ClinicalTrials.gov identifier: NCT03575351) evaluating lisocabtagene maraleucel (liso-cel) versus standard of care (salvage immunochemotherapy, high-dose chemotherapy, autologous stem cell transplantation [ASCT]) in second-line large B-cell lymphoma (LBCL). ctDNA association with efficacy was investigated at predefined time points (random assignment, day 43, day 64, and day 126 [3 months after liso-cel, approximately 2 months after ASCT]) for 136 patients using ultrasensitive PhasED-Seq. ctDNA clearance (measurable residual disease [MRD]neg) predicted longer event-free survival (EFS) at all time points in both arms, with significantly more liso-cel-treated patients achieving MRDneg. Liso-cel demonstrated superior outcomes versus ASCT, including longer EFS, progression-free survival (PFS), and duration of response among patients in complete response (CR) and MRDneg. ctDNA re-emergence in patients with CR after ASCT confirmed its potential in predicting relapse. MRDneg remained significantly associated with EFS after adjusting for positron emission tomography (PET) response, while interaction testing revealed a significant interaction between PET status and treatment arm for EFS. Liso-cel achieved deeper, more durable molecular clearance by ctDNA, consistent with superior EFS and PFS versus ASCT for second-line LBCL treatment. ctDNA-MRD provided prognostic value beyond PET, supporting its role as a complementary biomarker for treatment response and relapse prediction.

Anti-CD19 chimeric antigen receptor T-cell (CAR-T) therapy is a standard treatment option in relapsed or refractory (RR) large B-cell lymphoma (LBCL), including older patients. However, immune effector cell-associated neurotoxicity syndrome (ICANS)/neurotoxicity is frequent and can be debilitating, especially in older patients, and incidence and risk factors are not well defined in such patients. This retrospective multicenter study included older patients (≥65 yr) with RR LBCL treated with commercial CAR-T therapy between December 2017 and April 2023. The primary study outcome included 30-d cumulative estimates of ICANS, using deaths/progressions as competing risks. A total of 224 patients were included; among them, 131 (58%) received axicabtagene ciloleucel, 36 (16%) were treated with tisagenlecleucel, and 57 (25%) had lisocabtagene maraleucel. The median age at CAR-T therapy was 71 yr (range 65 to 89), and 26 (12%) were ≥80 yr old. The 30-d estimates for all grades and grades ≥3 ICANS were 50.9% (95% CI 44.7 to 57.9) and 29.8% (95% CI 24.0 to 37.0), respectively. Age at CAR-T therapy did not impact incidence, onset, and duration of ICANS. In contrast, ECOG PS scale of ≥2, an elevated lactate dehydrogenase level, and the use of axicabtagene ciloleucel were associated with a higher incidence of grade ≥3 ICANS. Age at CAR-T therapy did not impact the risks of ICANS in older patients with RR LBCL. Comprehensive neurocognitive and frailty assessments and appropriate candidate selections are crucial.

Chimeric antigen receptor T cells directed against CD-19 are highly effective therapies for various subtypes of B-cell non-Hodgkin lymphoma (NHL). We have studied the impact of patient-specific lymphocyte growth kinetics after lymphoid depletion chemotherapy on both long-term outcomes and toxicity in patients with diffuse large B-cell non-Hodgkin lymphoma (DLBCL). The growth of the lymphocyte populations early after chimeric antigen receptor T-cell (CAR-T) therapy is well described by an exponential function and the growth rate of the cells appears to be one of the most critical determinants of overall survival in these patients. The maximum lymphocyte peak and the area under curve for lymphocytes within the first 4 weeks after CAR-T therapy had no impact on survival. Receiver operator characteristic analysis determined that a replication rate ≥0.3876/day is associated with superior rates of complete response and durable responses. This threshold is associated with higher grades of cytokine release syndrome and the need for tocilizumab. However, patients with this threshold have a median overall survival of 3.04 years compared to 1.035 years for patients with slower lymphocyte replication (p = 0.0206). The major determinant of depth of response and durability of response to CAR-T in NHL is the speed of lymphocyte expansion.

1503 Background: CAR T-cell therapy has revolutionized treatment outcomes in hematological malignancies, but has traditionally required inpatient post-infusion monitoring, which is associated with significant cost and time toxicity. Outpatient CAR T infusion and monitoring has been the standard approach at Mayo Clinic site in Rochester, MN, with demonstrated feasibility and safety (Bansal et al., ASCO 2023). We present data comparing healthcare utilization between CAR T recipients who underwent outpatient remote monitoring and those with traditional inpatient monitoring. Methods: Electronic medical records were retrospectively analyzed for 293 patients who received CAR T-cell therapy for a hematological malignancy at any of the 3 Mayo Clinic sites (MN, AZ, FL) between 2020 and 2024 . Healthcare utilization was compared between patients who received outpatient CAR T with remote patient monitoring in Rochester, MN (N = 125) and those who received inpatient CAR T in the Arizona (n = 90) and Florida (n = 78) sites. Per protocol, patients in the inpatient group underwent mandatory inpatient monitoring for 7 days, with extension per provider discretion. Data was analyzed through Chi-Square and Wilcoxon tests. P values < 0.05 were considered statistically significant. Results: There was no significant difference in baseline sex, age, ethnicity, LDH, platelets, or neutrophils among the 293 patients. Diagnoses included lymphoma (n = 175), multiple myeloma (n = 106), and B-ALL (n = 12). The most common CAR T-cell products used were axicabtagene ciloleucel (n = 138) idecabtagene vicleucel (n = 56), and ciltacabtagene autoleucel (n = 50). Median hospital days were lower in the outpatient group vs the inpatient group at 30 days (4.4 vs 13.3, P < 0.001) and 90 days (4.6 vs 13.6, P < 0.001). ICU admission occurred in 1 outpatient patient ( < 1%) vs 18 inpatient patients (11%) in the first 30 days (P < 0.001). In the first 30 days, 22% of patients in the outpatient group did not require hospitalization, while 56%, 15%, and 6% required 1, 2, and 3 hospitalizations, respectively. Similarly, only 11 (8.8%) patients in the outpatient group had ≥1 ED visit in the first 30 days compared with 32 (20%) pts in the inpatient group (P = 0.009). At 30 days, there was no difference in patient portal use, with 58% of patients using the portal at least once and a median of 1 message in both groups. Within the first 30 days, 53% and 2% of patients in the outpatient group required 1 and 2 outpatient visits, respectively, vs 29% and 2% in the inpatient group (P < 0.001). There was no difference in 30-day mortality (1.6% outpatient vs 2.4% in inpatient) between the 2 groups (P = 1.00) Conclusions: Outpatient CAR T-cell therapy monitoring was associated with significantly fewer hospital days without increased ED visits, ICU stays, portal use, or 30-day mortality. This demonstrates lower healthcare utilization for outpatient CAR-T therapy.

Toxicities from CAR-T cell therapy are complex, severe, and often require multidisciplinary management. To standardize care, our institution established a Serious Adverse Event Oversight Committee (SAEOC), comprising specialists from hematology, neurology, nephrology, infectious diseases, intensive care, cardiology, and pharmacy. Once activated, the committee convenes virtually twice daily until SAE resolution or deactivation. We retrospectively examined outcomes in patients with macrophage activation syndrome/ hemophagocytic lymphohistiocytosis (MAS/HLH) treated before and after SAEOC implementation in December 2022. Provider perceptions were evaluated through a survey.The SAEOC was activated for 19 patients, most frequently for MAS-like syndromes (n=7), MAS/HLH (n=4), and high-grade ICANS (n=14). Additional triggers included CD8+ T-cell lymphoproliferative disorder, delayed IEC neurotoxicity and enterocolitis, massive pulmonary embolism, and diffuse alveolar hemorrhage.To assess outcomes, we focused on a more homogeneous subgroup of patients with MAS/HLH, comparing all patients treated prior to SAEOC implementation (n=8; 3 axi-cel, 1 tisa-cel, 1 ide-cel, 1 cilta-cel, 2 investigational CAR-Ts) with those treated after (n=6; 3 axi-cel, 1 brexu-cel, 2 cilta-cel). In the post-SAEOC cohort, 30- and 100-day treatment-related mortality were numerically lower, compared with the pre-SAEOC cohort (Table 1). Although these differences did not reach statistical significance, the findings may suggest a potential benefit of structured multidisciplinary oversight, acknowledging the limitations of small sample size and treatment-era differences.A total of 21 providers completed surveys (Figure 1), with ≥80% reporting perceived improvements in timeliness of interventions, consistency and comprehensiveness of management, interdisciplinary communication, and confidence in SAE management (Figure 2). Respondents valued SAEOC contributions to management of complex cases such as IEC-parkinsonism, MAS/HLH, CNS infection, and multiorgan involvement. SAEOC implementation also contributed to refinement of institutional standard operating practices. Reported challenges included workload and meeting frequency. One respondent provided uniformly negative Likert ratings but described the intervention as highly beneficial in free text; this discordant response was kept but interpreted cautiously. Suggested improvements included streamlining activation, reducing meeting frequency when appropriate, formal recognition of provider effort, and greater incorporation of evidence-based frameworks.The SAEOC represents a model for early multidisciplinary engagement and provides a practical framework for harmonizing CAR-T toxicity management, particularly for complex cases.

The NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Hodgkin lymphoma provide expert, multidisciplinary recommendations for the diagnosis and treatment of Hodgkin lymphoma. The panel convenes annually to update recommendations based on a review of recently published clinical trials. This selection from the NCCN Guidelines for Hodgkin Lymphoma focuses on the management of newly diagnosed and relapsed/refractory classic Hodgkin lymphoma in adults aged 18 to 60 years.

Introduction Reporting patient outcomes by CAR-T treatment centers is part of the accreditation requirement under the FACT IEC Program standard. CIBMTR audits IEC programs annually and requires the reported data to be >97% accurate. Currently, this is a time-consuming and manual process, making it difficult to sustain with the growing CAR-T indications and patient volume. Objective The objective of this study is to develop a Large Language Model (LLM) to automate the abstraction from the medical record and apply guidelines to define and grade toxicities after CAR-T therapy. Methods Clinical data from patients treated with CAR-T between January 2018 and June 2025, across all three Mayo Clinic sites, Minnesota (MCR), Florida (MCF) and Arizona (MCA) were used. Using LLM (Gemini 2.5 pro), we created a prompt by embedding both the clinical data and the ASTCT guideline for CRS and ICANS grading to generate output for whether or not CRS or ICANS event occurred, and if occurred, dates of occurrence, grade and medications used for management, as well as contextual justification for the output. The CRS and ICANS outcome reports generated by the LLM were then compared to the historical database. Any discrepancies identified were then reviewed manually. Results LLM was optimized using MCR data, showing 100% accuracy and F1 score, and 96% accuracy and 82% F1 score, for identifying CRS and ICANS events respectively, compared to the database. The match rates for CRS and ICANS grades were 83% and 89% respectively (Figure 1). MCA and MCF use different processes for clinical notes and toxicities flowsheet documentation. Applying the same LLM to those two sites yielded a lower rate of accuracy and F1 score for CRS and ICANS (Figure 1). In general, confusion matrix scores are higher for CRS events than for ICANS.Review of the discrepancies between LLM and database identified three main causes for discrepancies between LLM and database: A) Missed information by manual review for the historical database that was correctly identified and categorized by the LLM; B) correct categorization of event by LLM using objective data that was incorrectly graded by clinical team or attributed by the clinical team to other etiologies; C) LLM incorrect (Figure 2). For the first discrepancy, LLM use can improve the accuracy of the data capture for the compliance team. For the second, LLM will be updated with output to alert compliance team of discrepancy between objective data and clinical notes for manual review and final adjudication. For the third, additional review with clinical teams will be done to determine the best action to improve accuracy that may include updates to LLM prompts and clinical documentation process. Conclusion Our study demonstrates that LLMs can significantly reduce manual work to abstract and categorize CAR-T toxicities, while identifying opportunities for process improvement and quality review for the compliance team.

162 patients with relapsed secondary central nervous system lymphoma (R-SCNSL) (median age, 65 years; male, 59.9%) including central nervous system (CNS)-only (n = 120) and concomitant CNS/systemic relapse (n = 42) were retrospectively analyzed. Overall, 21.9% of patients were classified as high risk according to the CNS International Prognostic Index (CNS-IPI). Several biological and clinical features were significantly associated with leptomeningeal involvement, including double- or triple-hit (DHL/THL) status, MYC rearrangement, negative BCL6 expression by IHC, bone marrow involvement, and concomitant R-SCNSL. Multivariable analysis showed that leptomeningeal involvement independently predicted inferior OS and was associated with a 98% increase in the hazard of death compared with parenchymal relapse (HR = 1.98, 95% CI: 1.20-3.26, p = 0.008). Based on anatomical localization, R-SCNSL was classified into four subtypes: parenchymal-only involvement (parenchymal-CNS [P-CNS], 68/42%) and parenchymal involvement plus systemic relapse (parenchymal-concomitant [P-concomitant], 17/10.5%); and leptomeningeal with or without parenchymal involvement (leptomeningeal-CNS [LM-CNS], 52/32.1%) and leptomeningeal with systemic relapse (leptomeningeal-concomitant [LM-concomitant], 25/15.4%). This anatomical classification significantly impacted OS and PFS (p < 0.001). Two-year OS and PFS were 58.2% and 29.1% for P-CNS, 32.4% and 17.7% for P-concomitant, 22% and 13.9% for LM-CNS, and 7.1% and 0% for LM-concomitant, respectively. ASCT showed a trend toward improved survival among patients with a response (CR/PR) in a 4-month landmark analysis. These findings support the clinical application of the anatomical classification in the management of R-SCNSL.

Abstract Introduction: Therapeutic outcomes in B-cell non-Hodgkin’s lymphoma (B-NHL) have substantially improved with the introduction of chimeric antigen receptor T-cell (CAR-T) therapy. Yet, persistent immune dysfunction and underlying metabolic dysregulation remain major obstacles to achieving durable clinical responses in B-NHL. We hypothesize that mitochondrial pathways play a critical role in mediating anti-tumor immunity during CAR-T therapy. Methods: Peripheral blood mononuclear cells were collected from healthy controls (CNTRL, n=5) and patients with advanced stage lymphoma (LYM) who received FDA approved CAR-T (n=32). Samples were collected before lymphodepletion (BL), at peak CAR-T expansion (PK) and one-month post-infusion (M1). Single-cell RNA sequencing was used to interrogate differential mitochondrial gene expression. Samples were obtained from LYM patients with complete remission for >6 months (CR), primary refractory (PD1), or relapsed (PD2) disease. Results: Compared to CNTRLs, LYM patients exhibited reduced expression of cytochrome c oxidase. At BL, CR patients demonstrated broader mitochondrial differential gene expression across all T-cell subsets relative to PD1 and PD2 and showed higher expression of MT-ATP-6, MT-ATP-8, and MT-CO3. Following CAR-T infusion, global mitochondrial gene expression increased across patients corresponding with expansion of CD8 effector T cells (Tem tumor circulating) and activated peripheral memory T cells (Tpm). Certain mitochondrial genes, such as MT-CO1, were durably expressed across LYM patients at BL, PK, and M1 in peripheral and central memory T cells (Tpm, Tcm). At PK, cytochrome c oxidase and ATP synthase genes were upregulated in CR Tpm and Tcm. At M1, MT-ATP-6, MT-ATP-8, MT-CO3 were significantly upregulated in CD8 memory precursor effector T-cells (MPECs) of CR patients compared to both PD1 and PD2. While CR and PD2 patients appeared clinically responsive at M1, CR samples maintained significantly higher MT-ATP-8 expression across T cell subsets (CD8 MPECS, Tpm, Tcm) compared to PD2 (p values 0.008; <0.0001; <0.0001). Conclusions: We report a temporal shift in immunometabolism with CAR-T therapy in LYM. Clinical responders exhibited sustained mitochondrial activity characterized by persistent expression of ATP synthase in CD8 MPECs and cytochrome c oxidase in Tpm and Tcm. In sum, mitochondrial programming may distinguish effective versus dysfunctional T cells and identify targets for therapeutic development. Citation Format: Jacqueline Turner, Panwen Wang, Patrizia Mondello, Melinda Tan, Christoph Schaefers, Chen Wu, Andre de Menezes Silva Corraes, Kevin Regan, Zuoyi Shao, Ma Audrey, Arushi Khurana, Nora N. Benanni, Yucai Wang, Paul Hampel, Saad J. Kenderian, Jonas Paludo, Urshila Durani, Patrick B. Johnston, Jose Caetano Villasboas, Stephen M. Ansell, Haidong Dong, Ying Li, Zeng Hu, Yi Lin. Mitochondrial transcriptome profiles associated with clinical responses to CAR-T therapy in aggressive lymphoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1213.

Abstract Chimeric antigen receptor (CAR) T-cell therapy achieves high response rates in relapsed/refractory aggressive B-cell non-Hodgkin lymphoma, yet only 40% of patients achieve durable remission. Identifying biological programs linked to long-term response remains critical. As metabolic fitness underlies T-cell persistence and effector function, we profiled metabolic pathways associated with clinical outcomes following commercial CD19 CAR T-cell therapy. Single-cell RNA sequencing of peripheral blood mononuclear cells collected at baseline (BL), peak CAR-T expansion (PK), and one-month post-infusion (M1) was analyzed using CellRanger v7.0.1, immunopipe. Seurat v4.3.0 was applied for unsupervised clustering to delineate cell subsets based on top differentially expressed genes. Patients were categorized as durable complete remission (CR ≥6 months; n = 16), primary refractory (PD1; n = 4), or relapse after initial response (PD2; n = 12). Gene-set enrichment analysis defined metabolic pathway activity across T-cell, monocyte, dendritic cell (DC), and natural killer (NK) subsets. Oxidative phosphorylation (OXPHOS) emerged as the dominant metabolic program distinguishing clinical outcomes. At BL and PK, OXPHOS was consistently enriched in PD1 relative to CR across T-cell, monocyte, and DC subsets, suggesting early oxidative activation in non-responders. By M1, this pattern inverted, with higher OXPHOS activity in CR. PD2 largely paralleled PD1, but several subsets including CD8 T central memory, classical monocytes TGFβ, intermediate monocytes CD38, monocytic myeloid-derived suppressor cell (mMDSC) HIF1A, and mMDSC SIRPA, showed OXPHOS enrichment in CR at BL and/or PK that persisted through M1. In contrast, DC and NK subsets exhibited the opposite pattern: OXPHOS was enriched in CR at PK (conventional DC 2 (cDC2), plasmacytoid DC (pDC), NK, proliferating NK), but shifted toward enrichment in PD2 at M1 (cDC2, NK, NK CD56bright). Glycolysis (GLY) followed a similar trajectory in monocytes and NK cells, with enrichment in PD1/PD2 at BL and PK, followed by enrichment in CR at M1. In T-cells, PD1 maintained GLY enrichment from BL through PK, with no significant differences observed at M1. The inositol-phosphate metabolism pathway showed a more static pattern and was consistently enriched in multiple PD2 effector and memory T-cell subsets without reversal at M1. Taken together, these data reveal distinct, lineage-specific metabolic signatures that differentiate durable remission from early and late progression. Dynamic OXPHOS and GLY programming, characterized by lower activity early and enhanced activity at M1 in CR, may reflect adaptive metabolic programming that supports sustained antitumor immunity. Immune-metabolic profiling may therefore serve as a biomarker of response and highlight actionable metabolic pathways to enhance CAR T-cell durability. Citation Format: Melinda S.Y. Tan, Panwen Wang, Patrizia Mondello, Jacqueline Turner, Andre de Menezes Silva Corraes, Chen Wu, Zuoyi Shao, Kevin Regan, Ma Audrey, Arushi Khurana, Nora N. Benanni, Yucai Wang, Paul Hampel, Jonas Paludo, Saad J. Kenderian, Urshila Durani, Patrick B. Johnston, Jose Caetano Villasboas, Stephen M. Ansell, Ying Li, Haidong Dong, Hu Zeng, Yi Lin. Metabolic pathway signatures defining response to CD19 CAR T-Cell therapy in aggressive B-cell non-hodgkin lymphoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3278.

Introduction Chimeric Antigen Receptor T-cell (CAR-T) therapy has changed the landscape of cancer treatment since the first approval by the FDA in 2017. With increasing experience in management of acute toxicities such as Cytokine Release Syndrome (CRS) and Immune Effector Cell-Associated Neurotoxicity Syndrome (ICANS), the FDA re-evaluated the requirement for Risk Evaluation and Mitigation Strategies (REMS) for CD19- and BCMA-targeted CAR-T therapies and withdrew this requirement on June 2025, decreasing the recommended stay near treatment sites for patients from 1 month to 2 weeks. This change has emphasized the need for improved transition of care between CAR-T centers and local providers as patients' recovery and needs for supportive care vary beyond the initial post-infusion period of two weeks. We assessed if artificial intelligence (AI) multimodal large language model (LLM) could predict which patients would need higher levels of monitoring from days 14-30 post-infusion, based on pre-leukapheresis data. Methods 362 patients treated with FDA-approved CAR-T at Mayo Clinic Rochester from December 2022 through June 2025 has been analyzed. Patients were characterized as high monitoring needs if during days 14-30 after infusion they were readmitted to hospital for recurrent CRS/ICANS or needed more than 1 transfusion per week (RBC or platelets). Others were categorized as low monitoring needs. Information from clinical notes, lab tests, vital signs, ECGs and echocardiograms from their CAR-T evaluation was analyzed by the LLM (Gemini) for features predictive of later high monitoring needs. Results Among 362 patients, 84 met criteria for classification as high monitoring and 84 as low monitoring controls. These 168 were then evenly split into training and test groups. The LLM identified five feature categories prominently associated with needs for monitoring in these patients’ post-treatment, including (1) disease status (bulky or CNS disease), (2) indicators of inflammation/tumor burden (CRP, LDH), (3) hematologic status (platelets, ANC), (4) renal function, and (5) performance status. Using thresholds such as platelets < 10⁹/L, ANC <1 × 10 9/L, creatinine clearance <30mL/minute, patients were classified as at high monitoring need if high risk feature was identified in category 1 or any category combinations. In the training cohort, the LLM model was found to have achieved an 85.7% sensitivity, 23.8% specificity, F1 score of 65.5%. In the test cohort, performance was consistent and found to have 83.3 % sensitivity 28.6% specificity, F1 score of 65.4% Conclusion The current proof of concept study suggests that LLMs can analyze clinical data at the time of evaluation for CAR-T eligibility to predict for supportive care needs post treatment. Further attention will be devoted to methods of discrimination to improve specificity and balanced accuracy of the classification systems.

e19008 Background: Cytokine release syndrome (CRS) is typically reversible in CAR-T therapy, while hyperinflammatory syndromes like macrophage activation-like syndrome (MAS-L) and high-grade immune effector cell-associated hemophagocytic syndrome (IEC-HS) are associated with significant morbidity and mortality. Early interventions are critical for improving outcomes. We aim to analyze clinical cytokine assays and cellular phenotypes early in hyperinflammatory syndromes to identify profiles for severe cases to inform management decisions. Methods: This study included 113 CAR-T patients (pt) (September 2019-March 2024) with cytokine profiles (CP) performed at the second tocilizumab dose or with suspected hyperinflammatory states (CP1), followed by assessments at 24 hours (CP24) and 48 hours (CP48). 52 pts consented to blood (PB) and bone marrow (BM) cell immune phenotyping at pre-treatment (pre-LD) and day 1 for PB. Results: Among the 113 pts, CRS pts (N=74), when compared to MAS-L(N=19) & IEC-HS (N=20) at CP1, had lower levels of TNF (median in pg/mL; 31, 62.9, 47.7) , IL-18 (794, 1353, 2011), and MCP-1 (689, 2722, 2485.5). While MAS-L and IEC-HS had similar CP1 profiles, CP24 and CP48 differed between them: TNF & MCP-1 levels decreased in MAS-L but remained elevated or increased in IEC-HS. IL-18 was stable in MAS-L but increased steadily daily in IEC-HS. Infections were more frequent in IEC-HS (CRS: 43%; MAS-L: 58%; IEC-HS: 85%; p=0.002). Notably, patients who died with persistent E. faecium bacteremia had higher IL-18 in CP48. In the 52 pts with cell phenotypes available by flow, 33 had CRS, 18 MAS-L and 8 IEC-HS. The correlation between cell phenotypes and cytokine levels showed that elevated TNF at CP1 correlated with increased non-classical monocytes in pre-LD BM and higher NK and NK-T cells in pre-LD PB. Similarly, elevated IL-18 CP48 levels correlated with increased CD8 T-cells, NK-T cells, and intermediate monocytes in pre-LD PB, as well as increased NK cells at both pre-LD and day 1 (Table). Conclusions: We report distinct cytokine profiles in IEC-HS early in the inflammatory course post-CAR-T therapy. These elevated cytokines correlate with specific immune cell subsets, including BM-resident monocytes, intermediate monocytes with heightened cytokine trafficking and inflammatory capacity. This study demonstrates the feasibility for identifying potential biomarkers for early risk stratification and targeted interventions. Cytokine Sample source Immune cells R P value TNF CP1 BM pre-LD Non-classical monocytes (CD14loCD16+) 0.38 0.025 PB pre-LD Monocytes (All) 0.47 0.022 NK cells (CD56+CD16+) 0.34 0.024 NK-T cells (CD3+CD56+CD16-) 0.29 0.04 IL-18 CP48 PB pre-LD CD8+ T-cells 0.59 0.0093 NK-T cells (CD3+CD56+CD16-) 0.53 0.023 Intermediate monocytes (CD14+CD16+) 0.58 0.012 PB pre-LD NK cells (CD56+CD16+) 0.56 0.49 0.015 0.024