Hathcock, Matthew

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.

Abstract Background Hypomagnesemia has been correlated with inferior outcomes in patients with large B cell lymphoma (LBCL) undergoing stem cell transplants. As T-cell and myeloid cell dysfunction have been associated with low magnesium conditions, we investigated whether serum magnesium (Mg) levels could predict clinical outcomes in LBCL patients who received chimeric antigen receptor T-cell therapy. Methods Patients with LBCL who received axi-cel under the ZUMA-1 trial or as FDA approved therapy at Mayo Clinic were examined. Serum samples were obtained at specified time points and cytokine analysis was performed. Single cell RNA sequencing was performed on peripheral blood mononuclear cells. The Student T-test, Kruskal Wallis, or Fisher's Exact Tests were used to compare differences in demographics across Mg levels. Survival curves were plotted using the Kaplan-Meier methodology and compared using the Wilcoxon test. Results We found that hypomagnesemia before lymphodepletion chemotherapy predicted for inferior progression-free and overall survival in the pivotal study ZUMA-1 (NCT02348216). These results were validated in an independent cohort of LBCL patients receiving axicabtagene ciloleucel (axi-cel) at Mayo Clinic. Hypomagnesemia correlated with increased inflammatory serum markers and cytokine levels including ferritin, IL-6, IL1Ra, IL-8, and MIP1a. scRNAseq analysis unveiled altered immune interactions between monocytes and T cells with a concordant immune suppressive transcriptome. Conclusions Hypomagnesemia at the time of CAR-T infusion associates with an unfavorable inflammatory profile and decreased response and survival in LBCL patients receiving axi-cel. These findings suggest a potentially actionable prognostic factor for patients with large cell lymphoma undergoing CAR-T.

Abstract Unirradiated patients with relapsed/refractory (R/R) B-cell non-Hodgkin lymphoma (NHL) who undergo anti-CD19 chimeric antigen receptor T-cell therapy (CART) have a predominant localized pattern of relapse, the significance of which is heightened in individuals with limited/localized disease before CART. This study reports on the outcomes of patients with R/R NHL and limited (<5 involved sites) disease bridged with or without radiotherapy. A multicenter retrospective review of 150 patients with R/R NHL who received CART with <5 disease sites before leukapheresis was performed. Bridging treatment, if any, was administered between leukapheresis and CART infusion. Study end points included relapse-free survival (RFS), event-free survival (EFS), and overall survival. Before CART infusion, 48 patients (32%) received bridging radiotherapy (BRT), and 102 (68%) did not. The median follow-up was 21 months. After CART infusion, BRT patients had higher objective response (92% vs 78%; P = .046) and sustained complete response rates (54% vs 33%; P = .015). Local relapse in sites present before CART was lower in the BRT group (21% vs 46%; P = .003). BRT patients had improved 2-year RFS (53% vs 44%; P = .023) and 2-year EFS (37% vs 34%; P = .039) compared with patients who did not receive BRT. The impact of BRT was most prominent in patients who had ≤2 pre-CART involved disease sites, with 2-year RFS of 62% in patients who received BRT compared with 42% in those who did not (P = .002). BRT before CART for patients with limited (<5 involved disease sites) R/R NHL improves response rate, local control, RFS, and EFS without causing significant toxicities.

Introduction While chimeric antigen receptor (CAR)-T cell therapy has transformed the management of hematologic malignancies, it requires frequent monitoring for acute toxicities, almost universally achieved through mandatory hospitalizations, therefore utilizing a significant amount of healthcare resources. At Mayo Clinic in Rochester, MN, our standard practice is administering CAR-T cell therapy in the outpatient setting through a multi-disciplinary hospital-based outpatient (HBO) unit and remote patient monitoring (RPM) program. Our CAR-T RPM program consists of in-home, electronic health record-integrated technology to monitor vital signs and neurologic symptoms for 30 days post CAR-T infusion. Here we report the healthcare utilization outcomes of CAR-T patients managed in the outpatient setting enrolled in the RPM program. Methods Patients undergoing FDA-approved CAR-T cell treatment in the outpatient setting and participating in the CAR-T RPM program at Mayo Clinic Rochester were included in this IRB-approved study. Patient characteristics, healthcare utilization metrics, inpatient interventions, and patient satisfaction data regarding the RPM program were reviewed. Length of stay (LOS) includes any hospitalizations for which the admission date falls within the RPM monitoring period (CAR-T cell infusion to day +30), even if the hospitalization continued past day +30. Results A total of 123 patients participated in the CAR-T RPM program and underwent treatment in the outpatient setting. Median age at CAR-T infusion was 63 years (range 22 - 83), and most patients were male (n=78, 63%). Non-Hodgkin lymphoma was the most common diagnosis (n=74, 60%), followed by multiple myeloma (n=47, 38%), ALL (n=1, 1%) and other (n=1, 1%). During the PRM monitoring period, 19 (16%) patients were managed entirely in the outpatient setting. The remaining 104 (84%) patients required at least one hospitalization, with most patients (n=70, 57%) hospitalized only once [26 (21%) patients were hospitalized twice, and 7 (6%) patients were hospitalized 3 times, and 1 (1%) patient hospitalized 4 times during the monitoring period]. When considering the first hospitalization post CAR-T infusion, the median LOS was 6 days (range 2 - 47). When considering all hospitalizations post CAR-T, the median LOS was 8 days (range 2 - 47). Only 7 (6%) patients required an ICU stay, with a mean ICU stay of 0.2 days (standard deviation 0.9). Cytokine release syndrome (CRS) was noted in 92 (75%) of patients at a median of 3 days (range 0 - 18) post CAR-T infusion. All patients with CRS received tocilizumab, at a median of 4 days post CAR-T infusion. Steroids for CRS were administered in 63 (68%) of these patients. Neurotoxicity was reported in 45 (37%) patients at a median of 7 days (range 1 - 19) post CAR-T infusion. During the RPM monitoring period, patients were seen in our HBO unit with a median of 18 outpatient visits (range 4 - 30). Patient survey data, obtained at the end of the RPM monitoring period, were available from 37 patients. The majority of patients felt that the RPM program helped them feel comfortable managing their health at home, were able to reach a team member, and receive an appropriate response promptly. The majority of patients felt comfortable using the RPM equipment and interacting with providers virtually. Overall, 83% of patients were satisfied with the RPM program and 6% were neither satisfied nor dissatisfied. When asked about the reason for disappointment, technical difficulties/issues with the devices were the themes mentioned by the few patients (11%) who were unsatisfied with the RPM program. Conclusion: Despite the complexities and frequent monitoring associated with CAR-T cell therapy, outpatient CAR-T cell treatment is a feasible model of care. As the standard practice at our institution, the CAR-T RPM program was part of the management of nearly all patients undergoing an FDA-approved CAR-T product in the outpatient setting. RPM in the acute setting post CAR-T infusion is a well-received practice initiative by the majority of patients, including a high level of comfort utilizing the RPM equipment and virtual interaction with the RPM providers.

8024 Background: Minimal residual disease (MRD) is prognostic for survival in multiple myeloma (MM). MRD is usually assessed in the bone marrow (BM) by flow cytometry or NGS. MRD by FDG PET/CT provides a global representation of the tumor burden, including response assessment of extramedullary disease. We examined MRD status in MM pts using both BM and PET/CT at Mayo Clinic. Methods: Medical records were reviewed retrospectively for MM pts who received CAR-T between 4/2018 and 12/2022. All PET/CT scans were assessed by radiologists. BM MRD was assessed by flow cytometry with a sensitivity of 10-5. Progression-free survival (PFS) and overall survival (OS) were analyzed by Kaplan-Meier method. Results: Among the 157 CAR-T pts, including 89 pts who received FDA-approved CAR-T, median age was 63 years, 59% (92/157) were males, 42% (66/157) had high risk cytogenetics with median of 5 prior lines of therapy, 36% (57/157) had plasmacytoma. Incidence of CRS was 83%, 5% grade ≥ 3 CRS. Incidence of ICANS was 19%, 4% grade ≥ 3 ICANS. CR/sCR rate was 37%. One hundred and thirty-seven pts (87%) had evaluable BM at month (mo) 1, 85% (117/137) were BM MRDneg at mo 1. Baseline demographics were comparable between the two groups except age, % BM plasma cells and use of bridging therapy. Among the MRDneg pts, CR/sCR rate was 44% (51/117) and 95% (111/117) had sFLC below normal. At median follow-up of 13.8 months, median PFS among pts with BM MRDneg at 1 mo was 12 mo (95% CI: 11, 30) vs 3 mo (95%CI: 2, 7) for BM MRDpos (p<0.001). Median OS among pts with BM MRDneg at 1 mo was 34 mo (95% CI: 24, NR) vs 22 mo (95%CI: 7, NA) for BM MRDpos (p<0.01). At month 1, 112 pts had both BM and PET/CT assessments available, 64/112(57%) were both BM MRDneg/PET MRDneg, 38/112 (34%) were MRDneg for either BM or PET and 10/112 (9%) were positive for both BM and PET (MRDpos/PET MRDpos). Baseline demographics except age were comparable between the 2 groups (Table). Rate of conversion from MRDpos to MRDneg was low. Rate of sustained BM and PET MRDneg at mo 12 was 44% (8/18). The median PFS and OS for BM MRDneg/PET MRDneg was significantly longer as compared to others (Table). Conclusions: Achieving FLC below normal, BM MRDneg and PET MRDneg at 1 month, regardless of IMWG response at that time, is prognostic for both PFS and OS in MM pts receiving CAR-T. Failure to achieve any of these confers poor prognosis. [Table: see text]

AbstractBackgroundIt is unknown whether serum procalcitonin (PCT) concentration monitoring can differentiate between bacterial infection or cytokine release syndrome (CRS) when chimeric antigen receptor T‐cell (CAR‐T) recipients present with a constellation of signs and symptoms that may represent both complications.ObjectiveThe objective of the study was to assess the utility of serum PCT concentrations as a biomarker of bacterial infection in CAR‐T recipients.Study designThis single‐center, retrospective, medical record review evaluated patients prescribed CAR‐T therapy until death or 30 days after infusion. Logistic regression modeling determined the association between elevated serum PCT concentrations within 48 h of fever onset and microbiologically confirmed infection. Secondary outcomes included clinically suspected infection, CAR‐T toxicity rates, and broad‐spectrum antibiotic usage. Predictive performance of PCT was assessed by area under the receiver operating characteristic curve (AUC).ResultsThe 98 included patients were a median age of 63 (IQR: 55–69) years old, 47 (48%) were male, and 87 (89%) were Caucasian. Baseline demographics and clinical characteristics were similar between patients with and without a bacterial infection. Serum PCT >0.4 ng/mL within 48 h of fever was significantly associated with a microbiologically confirmed bacterial infection (OR: 2.75 [95% CI: 1.02–7.39], p = 0.045). Median PCT values in patients with and without confirmed infections were 0.40 ng/mL (IQR: 0.26, 0.74) and 0.26 ng/mL (IQR: 0.13, 0.47), respectively. The AUC for PCT to predict bacterial infection was 0.62 (95% CI 0.48–0.76). All patients experienced CRS of some grade, with no difference in CRS severity based on elevated PCT. Broad‐spectrum antibiotics were used for a median of 45% and 23% of days in those with and without confirmed infection, respectively (p = 0.075).ConclusionElevated serum PCT concentrations above 0.4 ng/mL at time of first fever after CAR‐T infusion was significantly associated with confirmed bacterial infection. Furthermore, rigorous, prospective studies should validate our findings and evaluate serial PCT measurements to optimize antimicrobial use after CAR‐T therapy.