Ansell, Stephen M.

e13622 Background: Chimeric antigen receptor T-cell (CAR-T) therapy has revolutionized treatment of hematologic malignancies; however, severe toxicities like cytokine release syndrome (CRS) remain a limitation to broader utilization. Using a large dataset of patients treated with CAR-T, we developed a machine learning (ML) model leveraging multi-model patient characteristics to predict the risk of CRS onset with clinically meaningful lead times. Methods: A prospectively maintained dataset of CAR-T patients treated at Mayo Clinic was used. Baseline disease/patients characteristics, labs, ECG, echocardiogram, and vital signs (including data from remote patient monitoring program) were collected. A 24-hour rolling window was used to generate daily averages, maximum, minimum, and deltas for each parameter. ECG and echocardiogram data were forward-imputed; vital signs and labs were not imputed. Each day-post-infusion served as a potential event (CRS onset vs. no CRS). A gradient boosting classifier was trained and tested on a subset of patients (n = 351) using 1,000 bootstrap iterations, evaluating 24-, 48-, and 72-hour lead times. Shapley-based feature importance guided simplification to an 11-feature model, externally validated on an independent cohort (n = 51). Models’ performance assessed using the area under the receiver operating characteristic curve (AUC). Results: A total of 402 patients treated with CAR-T were examined. Median age was 64 years and 36% were female. Primary diagnosis was B-cell lymphoma (n = 240) or multiple myeloma (n = 162). A total of 226,332 vital sign measurements and 372,871 lab values, as well as ECG and echocardiogram data, were collected for 30 days post-infusion. Overall, 79.8% of patients experienced CRS within 30 days. The full ML model demonstrated AUCs of 0.90 (95% CI 0.86–0.93), 0.83 (95% CI 0.77–0.88), and 0.82 (95% CI 0.77–0.87) for 24-, 48-, and 72-hour predictions of CRS, respectively. Including “day of CRS onset post CAR-T” as a feature improved the 72-hour AUC to 0.87 (95% CI 0.83–0.90). A model restricted to vital signs alone showed AUCs of 0.88, 0.75, and 0.71 for 24-, 48-, and 72-hour horizons. An 11-feature model—focusing on temperature, blood pressure, fibrinogen, CRP, serum sodium level, QRS interval, and lymphocyte counts—maintained high accuracy at 24 hours (AUC 0.90, 95% CI 0.86–0.93) and was validated externally using an independent dataset with an AUC of 0.92. Conclusions: Our ML model provides a reliable approach for predicting if/when CRS will occur, with up to three days of lead time before clinical manifestation, offering an opportunity for proactive management. The robust performance of a parsimonious 11-feature model supports practical adoption across clinical settings and real-time risk stratification to enhance patient safety. Prospective validation and integration into clinical workflows will be critical to translating these findings into practice.

e13738 Background: Persistent racial/ethnic and regional disparities in neoplasm mortality remain significant public health challenges. This study evaluates long-term trends in neoplasm mortality by race/ethnicity at the county level in the United States and examines the relationship between hospital density and mortality. Methods: Using data from the Global Burden of Disease Study (2000-2019), we analyzed age-standardized mortality rates (ASR) for neoplasms by ethnicity/raceHispanic (all races) and non-Hispanic American Indian or Alaska Native (AIAN), Asian or Pacific Islander (API), Black, and White. A joinpoint regression analysis was used to identify significant changes in the mortality trends, summarized as average annual percent change (AAPC). Correlation analysis evaluated the relationship between county-level hospital density and neoplasm mortality. Results: Neoplasm mortality rates declined across all racial/ethnic groups from 2000 to 2019, but disparities persisted. ASR per 100,000 population decreased from 240.5 (95% UI: 214.3–269.9) to 203.0 (180.8–228.3) for AIAN, 149.5 (144.9–153.9) to 115.3 (111.7–118.7) for API, 314.5 (312.1–317.0) to 218.7 (217.1–220.3) for Black, 168.4 (164.6–172.0) to 133.8 (131.0–136.4) for Latino, and 241.2 (240.5–242.0) to 187.5 (186.9–188.2) for White. AAPC from 2000 to 2019 ranged from -0.94% (AIAN) to -1.90% (Black). Gender-specific analyses revealed similar trends, with males showing slightly greater declines. Southeast states experienced slower declines compared to the Northeast. Some counties exhibited worsening mortality trends, particularly for AIAN, API, and Latino. The county-level hospital density (table) showed weak but statistically significant correlations in 2019 for AIAN (r = -0.35; p < 0.0001), White (r = -0.15; p < 0.0001), and Black (r = -0.18; p < 0.0001). No inverse correlation was observed for API (r = 0.07; p = 0.071) and Latino (r = 0.06; p = 0.012). Conclusions: While significant progress has been made in reducing neoplasm mortality, racial/ethnic and regional disparities persist, and health care access appears to contribute to this at least in part. Future analyses should evaluate mortality by cancer site and also account for broader clinical and public health infrastructure, as well as social determinants of health to better understand and address underlying disparities. The correlation between hospital density and age-standardized mortality rate from neoplasms by race/ethnicity in 2019. Correlation coefficient with mortality and hospital density AIAN API Black Latino White 2019 -0.35(-0.43 – -0.27) 0.069(-0.0058–0.14) -0.18(-0.23 –-0.13) 0.064(0.014–0.11) -0.15(-0.19 – -0.12)

7075 Background: Hepatosplenic T-cell lymphoma (HSTCL) is a rare, aggressive peripheral T-cell lymphoma arising primarily from γδ T-cells. It carries a poor prognosis and resists conventional chemotherapy. Most reports on HSTCL are case-based. This study comprehensively analyzes a large cohort, evaluating treatment strategies and survival outcomes. Methods: This retrospective study included patients (pts) with pathologically confirmed HSTCL diagnosed between 2000-2024, consecutively seen at Mayo Clinic MN. Clinical, pathological, genomic, and treatment-related data were extracted when available. Descriptive statistics were used to summarize baseline characteristics. Time-to-event analyses, including Kaplan-Meier estimates, median overall survival (OS), and survival time estimates were conducted from the date of diagnosis. Results: A total of 20 patients with newly diagnosed HSTCL were included, with a median age of 57 years (range: 35-71). The cohort was predominantly male (70%) and non-Hispanic (93%). Molecular data was available for five patients, revealing abnormalities in STAT5B, MLL3 deletion, TP53, EZH2, TERT, and NF1 E291D . The median follow-up was 27.6 months (m) with a median OS of 17.6 m (95% CI: 11.2 - NA). First-line treatment was anthracycline-based in 68% of pts and non-anthracycline-based in 32%, with higher response rates in the latter group (50% vs.83%). Although non-anthracycline regimens showed a trend toward improved 3-year OS (100% vs. 29%) the difference was not statistically significant (p = 0.16). Achieving a complete response to first-line therapy was also associated with a trend towards a better 3-year OS compared to refractory disease (80% vs. 50%, p = 0.17). Most pts (79%) underwent hematopoietic stem cell transplant (HSCT), primarily allogeneic, with only one receiving autologous HSCT. First-line therapy before HSCT was evenly distributed between anthracycline (55%) and non-anthracycline (45%) regimens. HSCT recipients had significantly higher 3-year OS than non-recipients (83% vs. 33%, p = 0.017). Notably, the patient with a TP53 mutation has remained in remission for over a year post-allogeneic HSCT. Conclusions: HSTCL predominantly affects younger pts, with nearly half dying within a year. Allogeneic HSCT, rarely used in other NHL subtypes, improved survival. Non-anthracycline regimens and achieving CR trended toward better outcomes. Our study, leveraging a sizable cohort, highlights the need for targeted research and novel therapies to improve HSTCL management. Summary of survival outcomes in HSTCL. Characteristics Median OS (y) 3-Year OS (95% CI) 1.48 [0.93- NA] 46% [0.26 - 0.81] Transplant Transplant NA [NA - NA] 83% [0.58 - 1.00] No Transplant 1.02 [0.93 - NA] 33% [0.07 - 1.00] Treatment Anthracycline Based 1.02 [0.36 - NA] 29% [0.11 - 0.73] Non-Anthracycline Based 4.61 [NA- NA] 100% [1.00 - 1.00]

7073 Background: The Optimized HLH Inflammatory (OHI) index (Zoref-Lorenz et al., Blood , 2022), based on serum soluble CD25 (sCD25) and ferritin, is a prognostic tool identifying inflammation in the hemophagocytic lymphohistiocytosis (HLH) spectrum and early mortality risk. Prior studies were retrospective and enriched with HLH-suspected cases. We hypothesized that the OHI index would predict outcomes in unselected diffuse large B-cell lymphoma (DLBCL). Methods: 670 newly diagnosed DLBCL patients from a prospective cohort (2002–2015) were analyzed: 335 for discovery and 335 for validation. Pre-treatment serum sCD25 and ferritin levels were quantified via ELISA, and we evaluated original thresholds (sCD25 ≥ 3,900 U/mL; ferritin ≥ 1,000 ng/mL) and developed DLBCL-optimized thresholds (using receiver operator curves) for predicting 500-day mortality. We also evaluated event-free survival (EFS), EFS at 24 months (EFS24), and overall survival (OS). Results: The median age was 64, and 51% were male. Using original thresholds, 4.2% (n=14) of patients were OHI+, with a 5.6-fold higher risk of 500-day mortality (95% CI 1.6–17; p<0.001). Optimized thresholds identified 23.6% (n=78) as OHI+ with a 9.2-fold higher 500-day mortality risk (95% CI 4.0–24; p<0.001). Optimized OHI+ also had a higher risk of EFS24 failure and inferior EFS and OS (Table). OHI+ patients had higher rates of B symptoms (35% vs. 20%; p=0.027), elevated LDH (76% vs. 34%; p<0.001), worse performance status (ECOG ≥2: 38% vs. 12%; p<0.001), and advanced-stage disease (Stage IV: 57% vs 38%; p<0.001), though IPI scores (3–5) were not significantly different (44% vs 36%; p=0.28). Adjusting for age and IPI, OHI+ independently predicted 500-day mortality (OR=5.0; CI 1.9–13; p=0.001), EFS24 failure (OR=3.2; CI 1.5–6.5; p=0.0018), and inferior long-term EFS (HR=2.2; CI 1.1–4.3; p=0.030) and OS (HR=2.0; CI 1.0–4.0; p=0.040) at a median follow up of 9.8 years in living patients. In validation, optimized OHI predicted 500-day mortality, EFS24 failure, and long-term OS and EFS (Table). Cytokine profiling revealed elevated inflammatory markers in OHI+ patients, including IL1a (p=0.008), CCL2 (p=0.03), CXCL10 (p<0.001), and CXCL9 (p<0.001), reflecting systemic hyperinflammation. Conclusions: The OHI index is a powerful predictor of early and long-term outcomes in DLBCL patients. Optimized thresholds identify a larger OHI+ group, highlighting hyperinflammation’s critical role in poor outcomes. These findings support its use in routine management and clinical trial design for novel therapies. Odds ratios or hazard ratios and p-values for optimized OHI+ and DLBCL prognosis. Outcome Discovery Validation OR for 500-day mortality 9.2; p<0.001 4.0; p=0.002 OR for EFS24 failure 5.2; p <0.001 6.3; p<0.001 HR for OS 2.3; p<0.001 1.8; p=0.006 HR for EFS 2.4; p<0.001 2.5; p<0.001

7052 Background: Secondary central nervous system lymphoma (SCNSL) is a rare but serious complication that occurs in 2-10% of patients (pts) with diffuse large B-cell lymphoma (DLBCL). The CNS-International Prognostic Index (CNS-IPI) is a validated tool used to assess the risk of CNS relapse. However, with advancements in molecular profiling and treatments, the predictors of SCNSL, including the predictive ability of CNS-IPI combined with other molecular markers and treatment regimens, have become uncertain. This study evaluates new prognostic markers for CNS relapse in DLBCL, focusing on cell-of-origin (COO), gene rearrangements, and treatment types. Methods: DLBCL pts enrolled from 2002-2015 in the Mayo Clinic and University of Iowa Lymphoma Molecular Epidemiology Resource were prospectively followed for incidence of CNS relapse. Pts with primary CNS lymphoma or CNS involvement at diagnosis, and pts treated with high-dose methotrexate containing regimens were excluded. COO classification (GCB vs non-GCB) was determined using Hans algorithm and/or NanoString data. Multivariate Cox regression models estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for predictors of SCNSL. Survival analysis was performed from time of DLBCL diagnosis using Cox proportional hazards model. Results: Among 1278 newly diagnosed DLBCL pts included (median age: 63 years; 43.6% female), SCNSL occurred in 46 (3.6%) pts. Pts with SCNSL were more likely to have intermediate/high CNS-IPI scores (80.4%) than pts without SCNSL (63.9%, p=0.02). The 3 yr SCNSL incidence was 0.016 (0.008-0.0033) in low CNS-IPI vs 0.043 (0.031-0.0060) in high CNS-IPI. Those with SCNSL were significantly more likely at baseline to have non-GCB COO (56.3% vs. 36.5%, p=0.02), and advanced stage (76.1% vs. 59.9%, p=0.03) compared to those without SCNSL. The incidence of SCNSL was not influenced by MYC, BCL2, or BCL6 rearrangements, nor by treatment with R-CHOP, R-EPOCH, or other immunochemotherapy regimens. MVA showed that intermediate CNS-IPI (HR 2.32, 95% CI 1.09-4.92, p=0.029), high CNS-IPI (HR 3.59, 95% CI 1.49-8.62, p=0.004), and non-GCB COO (HR 2.19, 95% CI 1.08-4.42, p=0.0029) were independently associated with increased SCNSL risk. After a median follow-up of 12 years (IQR 9–15), overall survival (OS) was 52.1% for the cohort, with OS significantly worse for SCNSL pts (17.4% vs. 53.4%, p<0.001). Conclusions: This large cohort with extensive follow-up highlights the role of COO as an additional prognostic factor for SCNSL while reinforcing CNS-IPI’s prognostic value. Non-GCB COO pts had a higher risk of SCNSL even when adjusted by CNS-IPI risk. Despite advancements in management, SCNSL pts continue to have poor outcomes. Strategies to reduce the incidence of CNS relapse in high-risk pts require further investigation. MER CNS relapse hazard ratios (95% CI). Characteristics Hazard Ratio 95% CI P-Value CNS-IPI Group 0-1 Low Ref Ref Ref 2-3 Intermediate 2.32 1.09 - 4.92 0.029 4+ High 3.59 1.49 - 8.62 0.004 Cell of Origin* GCB Ref Ref Ref non-GCB 2.19 1.08 - 4.42 0.029 Double Hit* non-DHL Ref Ref Ref DHL 1.08 0.14 - 8.3 0.941 Not Done/Missing 1.38 0.76 - 2.51 0.297 Double Expressor* Negative Ref Ref Ref Positive 2.36 0.71 - 7.85 0.162 Not Done/Missing 1.91 0.89 - 4.08 0.097 IC Group* R-CHOP R-EPOCH Ref 1.63 Ref 0.71 - 3.74 Ref 0.249 Other IC 1.18 0.57 - 2.46 0.660 Any EN Involvement* No Ref Ref Ref Yes 0.98 0.51 - 1.88 0.946 Albumin Group* >=Normal Ref Ref Ref <Normal 1.70 0.75 - 3.85 0.201 Head & Neck Involvement* No Ref Ref Ref Yes 0.35 0.048 - 2.63 0.310 *Adjusted by CNS-IPI risk group.

7032 Background: The impact of immune cell counts pre LD and during immune reconstitution on CAR T therapy outcomes is poorly understood. We investigated the association between CD4, CD8 and NK cell counts and CAR T therapy outcomes in patients with LBCL. Methods: Retrospective study of R/R LBCL patients who received CAR T cells between 2016-24 at Mayo Clinic, Rochester, were included in this analysis. Peripheral blood CD4, CD8 and NK counts were measured pre LD and at M1 post CAR T infusion. The receiver operating characteristic (ROC) curve was used to determine the optimal cutoff for pre LD and M1 immune cells to predict patients who were alive and in remission at 6 months (M6) post CAR T infusion. Patients who progressed or were lost to follow up prior to day 30 were excluded from M1 ROC analysis. Results: Of 140 patients, 81 were alive and in remission at M6 (Group A), while 59 had relapse or death (Group B). Axicabtagene ciloleucel was the CAR T product given in 81% of patients (114/140). Median pre LD CD4 counts were significantly lower in Group B compared to group A (143 vs 280 cells/µL, p = 0.001). No significant difference was observed in median pre LD CD8 counts (205 vs 221 cells/µL, p = 0.8). ROC analysis identified an optimal pre LD CD4 count of 124.5 cells/µL for M6 alive+remission. Lower pre LD CD4 counts (<124.5) predicted worse progression free survival (PFS) in univariate (HR = 3.02, 95% confidence interval [CI]: 1.73–5.27, p< 0.01) and multivariable analysis (MVA) adjusted for IPI and the number of prior lines (aHR = 2.54, 95% CI: 1.39–4.62, p< 0.01).Lower pre LD CD4 counts also predicted inferior overall survival (OS) in MVA (aHR = 2.27, 95% CI: 1.08–4.77, p = 0.03). (Table 1) On day 30 landmark analysis, ROC identified M1 optimal CD4 count of ≥ 99.5 cells/µL to be associated with a trend toward superior PFS (P=0.09), but not OS (p=0.90) in MVA. Median pre LD NK cell counts were significantly lower in Group B (73 vs 98 cells/µL, p = 0.04). ROC analysis identified an optimal pre LD NK count of 151 cells/µL for M6 alive + remission. Lower pre LD NK counts (<151) predicted worse PFS, both on univariate (HR = 4.17, 95% CI: 1.29–13.47, p = 0.02) and MVA (aHR = 4.64, p< 0.01). The lower pre LD NK cell count group had worse OS in MVA (aHR = 5.69, 95% CI: 1.16–28.1, p = 0.01). (Table 1) On D30 landmark analysis, M1 NK cell count of ≥ 128.5 cells/µL was associated with a trend toward superior PFS (P=0.08), but not OS (p=0.30) in MVA. Conclusions: Pre LD CD4 and NK cell counts are significantly associated with PFS and OS post CAR T therapy in patients with R/R LBCL. Pre-treatment immune subset levels may identify patients at higher risk of relapse or death after CAR-T therapy. Groups Median PFS 2 Year PFS Median OS 2 Year OS CD4 ≥124.5 cells/µL (N=75) NR 58% 3.31 years 67% CD4 <124.5 cells/µL (N=30) 2.2 months 22% 1.59 years 45% NK ≥151 cells/µL (N=17) NR 77% NR 84% NK <151 cells/µL (N=82) 10.6 months 44% 3.31 years 62%