Monick, Sarah

e18609 Background: Anemia is a frequent complication of myelofibrosis (MF) and is embedded in widely used prognostic systems, yet contemporary real-world estimates of its association with long-term survival and leukemic transformation remain limited. We evaluated the impact of baseline anemia (hemoglobin [Hb] <10 g/dL) on overall survival (OS) and risk of transformation to acute myeloid leukemia (AML) in adults with MF. Methods: We performed a retrospective cohort study using the TriNetX Research Network, identifying adults (≥18 years) with myelofibrosis and stratifying them by anemia status at index (Hb <10 g/dL). Patients with the outcome of interest prior to each analytic window were excluded. To minimize confounding, cohorts were 1:1 propensity score–matched using nearest-neighbor matching with a caliper of 0.1 pooled standard deviations of the logit of the propensity score, adjusting for demographic factors, comorbidities, iron supplementation, leukocyte count, and platelet count. Balance was assessed using standardized mean differences <0.1. The primary outcome was all-cause mortality and the secondary outcome was transformation to AML, assessed at 1, 3, and 5 years. Risk estimates were calculated, and survival analyses were performed using Kaplan–Meier curves and Cox proportional hazards models, with statistical significance defined as a two-sided p<0.05. Results: After matching, 18,504 patients with MF were identified, of which 51.6% were females and 48.3% were males. The mortality rate in patients with anemia was higher than in those without at years 1, 3, and 5: 21.4%, 33.2%, and 39%, respectively, versus 5.4%, 12%, and 16.9%, respectively. The prevalence of transformation to AML was also higher in patients with anemia at 3.7%, 4.9%, and 5.3% at years 1, 3, and 5, respectively, versus 0.1%, 0.3%, and 0.4% in patients without anemia at those same time points. The HR for mortality in MF patients with anemia when compared to those without anemia at years 1, 3, 5 was statistically significant at 4.57 (95% CI 4.14-5.05, log-rank p<0.01), 3.51 (95% CI 3.28-3.76, log-rank p<0.01), and 3.04 (95% CI 2.87-3.23, log-rank p<0.01) respectively. The HR for transformation to AML in patients with MF with anemia versus those without anemia was statistically significant at years 1, 3 and 5, with HRs of 34.96 (95% CI 19.17-63.75, log-rank p<0.01) 20.75 (95% CI 13.97-30.83, log-rank p<0.01) and 17.07 (95% CI 12.11-24.07, log-rank p<0.01), respectively. Conclusions: In this large, propensity-matched real-world EHR analysis, baseline Hb <10 g/dL was associated with substantially worse OS and a markedly higher risk of AML transformation through 5 years. These findings support anemia as a pragmatic risk marker to inform risk-adapted monitoring, supportive care, and therapeutic decision-making in MF.

e18608 Background: Myelofibrosis (MF) is a myeloproliferative neoplasm (MPN) that is characterized by bone marrow fibrosis and inefficient hematopoiesis, leading to extramedullary hematopoiesis. Leukocytosis, a common disease manifestation and a minor criterion for the diagnosis of MF, has been associated with poorer clinical outcomes. This study evaluates the long term impact of leukocytosis on clinical outcomes at different time points in the real-world population of patients with MF. Methods: We conducted a retrospective cohort study using de-identified EHR data from the TriNetX Global Federated Health Research Network. Adults (≥18 years) with myelofibrosis were identified using ICD-10 and ICD-O-3 codes and stratified into leukocytosis (WBC ≥25.0 ×10³/µL; L+) and non-leukocytosis (L−) cohorts at the index diagnosis. Patients with prior or current hematopoietic stem cell transplantation were excluded. Cohorts were propensity score–matched 1:1 for demographics, comorbidities, and baseline laboratory values. Outcomes at 1, 3, and 5 years included all-cause mortality, pulmonary embolism, venous thromboembolism, and transformation to acute myeloid leukemia, with hazard ratios estimated using Cox proportional hazards models. Results: After matching, 13237 patients with MF were identified, of whom 52.12% were male, and 47.8% were female. The mean age in L+ and L- was 66.5 +/- 16.6 years and 68.1 +/- 15.5 years, respectively. The mean hemoglobin in L+ was 10.9+/- 2.8 g/dL, and in L- was 12.9+/- 2.3 g/dL. The mean platelet count in L+ was 303.4+/- 294.9 and in L- was 239.5+/- 151.4 ×10³/µL. HR for all-cause mortality in L+ when compared to L- at years 1, 3, and 5 were statistically significant at 4.079 (95% CI 3.691-4.508), 3.215 (95% CI 2.982-3.466), and 2.787 (95% CI 2.608-2.978), respectively, with log-rank p < 0.01. The HR for PE in L+ when compared to L- was statistically significant at years 1, 3, and 5 at 2.014 (95% CI 1.455-2.787), 1.883 (95% CI 1.451-2.787) and 1.659 (95% CI 1.318-2.088) with log-rank p<0.01. HR for VTE in L+ when compared to L- was statistically significant at years 1, 3, and 5 at 2.501 (95% CI 1.937-3 .229), 1.802 (95% CI 1.491-2.178), and 1.561 (95% CI 1.32-1.847) with log-rank p<0.01. Risk of transformation to AML in L+ when compared to L- at years 1, 3, and 5 was significant with HRs being 9.806 (95% CI 6.575-14.627), 8.823 (95% CI 6.334-12.29), 9.371 (95% CI 6.813-12.891) with log-rank p<0.01. Conclusions: In this large real-world, propensity-matched cohort, leukocytosis in myelofibrosis was associated with significantly higher risks of all-cause mortality, thromboembolic events, and transformation to acute myeloid leukemia across 1-, 3-, and 5-year time points. These findings underscore leukocytosis as a powerful adverse prognostic marker in MF and highlight the need for closer surveillance and risk-adapted therapeutic strategies in this high-risk population.

e18605 Background: Patients with myelofibrosis (MF), especially those with marked leukocytosis, face increased risk for thrombotic complications such as pulmonary embolism (PE) and venous thromboembolism (VTE). Hydroxyurea (HU) is often used to control blood counts. However, evidence for HU reducing thrombotic events in MF is limited, and contemporary real-world data on its impact in MF populations with significant leukocytosis remain sparse. Methods: We conducted a retrospective cohort study using the TriNetX research network. Adult patients with MF and leukocytosis (white blood cell count ≥25×10⁹/L) were identified. Patients with prior allogeneic hematopoietic stem cell transplantation were excluded. HU-exposed patients were compared with non-exposed patients using 1:1 propensity score matching. Matching criteria included age, sex, race, comorbidities, and baseline laboratory values when available. The study included 3,590 patients (1,795 per cohort). Primary outcomes were time to PE and VTE at 1, 3, and 5 years. Hazard ratios (HR) with 95% confidence intervals (CI) were estimated using Cox proportional hazards models. Results: In the propensity-matched cohort, HU exposure was not associated with a statistically significant reduction in thrombotic events. For PE, hazard ratios were numerically higher in the HU-exposed group at 1 year (1.5% vs 1.0%; HR 1.40, 95% CI 0.76–2.56; p=0.278) and 3 years (2.2% vs 1.8%; HR 1.29, 95% CI 0.80–2.08; p=0.289), with a modest increase observed at 5 years (2.8% vs 1.9%; HR 1.57, 95% CI 1.01–2.45; p=0.04). For VTE, no statistically significant differences were observed between HU-exposed and non-exposed patients at any time point (1 year: HR 0.95, 95% CI 0.63–1.44; p=0.821; 3 years: HR 0.97, 95% CI 0.69–1.36; p=0.85; 5 years: HR 1.14, 95% CI 0.84–1.57; p=0.40). Overall, PE and VTE event rates remained low and comparable between cohorts. Conclusions: In this large real-world cohort of patients with myelofibrosis and significant leukocytosis, hydroxyurea exposure was not associated with a statistically significant reduction in thrombotic events, including PE or VTE, over short- or long-term follow-up. Thrombotic event rates were low and generally comparable between hydroxyurea-exposed and non-exposed patients, with no consistent pattern suggesting a protective effect of cytoreductive therapy. Although a modest increase in PE risk was observed at 5 years, this finding was isolated and lacked a consistent temporal pattern, limiting its clinical interpretability. These findings suggest that the role of hydroxyurea in modifying thrombotic risk among MF patients with leukocytosis remains uncertain and highlight the need for individualized treatment decisions and further studies to better define its impact on thrombotic outcomes in this population.

7041 Background: CD20×CD3 bispecific antibodies and CD19-directed CAR T-cell therapies are increasingly used in relapsed/refractory DLBCL. CAR T therapy is associated with higher rates of CRS,ICANS, and cytopenias, whereas bispecific antibodies are often perceived as less toxic and more outpatient-friendly. Methods: We conducted a retrospective cohort study using the TriNetX Global Collaborative Network. Adult patients with DLBCL treated with bispecific antibodies (glofitamab, epcoritamab) were compared with those receiving commercial CAR T-cell therapy (tisa, liso, axi). Propensity score matching was performed on demographics, comorbidities, baseline laboratory values, and corticosteroid exposure, yielding 468 patients per cohort. Outcomes included CRS, ICANS, severe neutropenia (ANC <500/µL), and thrombocytopenia (platelets <50×10³/µL), as well as hospitalization and ICU admission at 1, 3, 6, and 12 months. Overall survival (OS) was assessed using Kaplan–Meier analysis with log-rank testing. Results: After matching, CAR T therapy was associated with significantly higher immune-mediated toxicity. CRS occurred more frequently with CAR T at 1 month (37.4% vs 22.4%), 3 months (39.3% vs 27.6%), 6 months (40.6% vs 28.8%), and 12 months (41.2% vs 29.7%) (all p<0.001). ICANS was also more common with CAR T across all timepoints (1 month: 10.5% vs 4.9%; 12 months: 12.6% vs 7.5%). Severe neutropenia and thrombocytopenia were markedly more frequent in the CAR T cohort, with neutropenia affecting 66.7% vs 12.8% at 1 month and 71.6% vs 32.7% at 12 months (all p<0.001). Despite higher toxicity, healthcare utilization was comparable. Hospitalization rates were similar between groups at 1 and 3 months (61.8% vs 66.7% and 67.1% vs 75.9%), and ICU admission rates did not differ meaningfully at any timepoint. Importantly, overall survival consistently favored CAR T therapy. Survival differences emerged early and persisted over time, with lower mortality in the CAR T cohort at 1 month (2.1% vs 9.0%), 3 months (7.7% vs 18.6%), 6 months (15.8% vs 26.7%), and 12 months (26.5% vs 35.3%), corresponding to significantly improved OS by log-rank testing at all timepoints (p<0.001). Conclusions: In this large, real-world, propensity-matched analysis of relapsed/refractory DLBCL, CAR T-cell therapy was associated with substantially higher rates of CRS, ICANS, and severe cytopenias compared with bispecific antibodies. However, these toxicities did not translate into higher hospitalization or ICU utilization. Despite a less favorable early toxicity profile, CAR T therapy conferred a consistent and clinically meaningful survival advantage from early to late follow-up. Early toxicity alone should not preclude CAR T referral in appropriate candidates, underscoring the need for optimized patient selection, sequencing strategies, and supportive care.

1617 Background: Large language models (LLMs) are increasingly evaluated for oncology decision support, yet guideline concordance and hallucinations remain variable across cancer types and tasks. Retrieval-augmented generation (RAG) may improve safety by constraining the evidence base used during generation. Prior oncology LLM studies report discordance with guideline-based care, but few describe a reproducible health-system “vetting mechanism” that couples guideline-only retrieval with clinician-scored safety rubrics across multiple oncology domains. Methods: We assembled 316 de-identified oncology cases in a tumor-board template: medical oncology (breast, GI, GU, leukemia; n=216), gynecologic oncology (n=50), and neuro-oncology/CNS metastatic cancer (n=50). Each case was run on three systems: baseline GPT-4 (O1), NCCN-guideline anchored RAG (O2; guideline-constrained retrieval), and a literature-anchored system (OpenEvidence; O3). Oncologists independently scored outputs using a modified Generative Performance Score (mGPS; −1 to +1) combining guideline concordance with hallucination penalties; readability/rationality was scored on a 1–5 Likert scale. Inter-rater reliability was assessed (κ =.63). This workflow was designed as a pre-deployment model credentialing step for health-system governance (source policy → case bank → clinician-scored acceptance criteria). Results: Across all 316 cases (case-averaged), O2 achieved the highest mean total mGPS (0.75) compared with O1 (0.43) and O3 (0.59). In the medical oncology cohort (n=216), O2 outperformed O1 and O3 (mean mGPS 0.72 vs 0.35 and 0.54), with higher guideline concordance (0.85 vs 0.68 and 0.75) and lower hallucination penalties (−0.09 vs −0.25 and −0.19); readability was also highest for O2 (4.17 vs 3.22 and 3.86). In gynecologic oncology (n=50), O2 again performed best (mGPS 0.827) relative to O3 (0.697) and O1 (0.647). In the CNS cohort (n=50), mean composite mGPS favored O2 (0.82) over O3 (0.68) and O1 (0.55), while readability remained stable across models (3.55–3.73). Severe hallucination/unsafe recommendation rate: O1 [26%], O2 [5%], O3 [17%] (pre-specified definition; e.g., major fabricated citation or guideline-discordant harmful recommendation). Conclusions: In a 316-case, multi-domain clinician-scored evaluation, limiting retrieval to oncology guideline sources improved safety-aligned performance relative to unconstrained generation or literature-only anchoring. A guideline-constrained RAG configuration paired with clinician-scored rubrics and tail-risk reporting can serve as a practical health-system vetting mechanism for governance, re-validation, and change control of LLM decision support, including in resource-limited settings where reliable guideline-consistent recommendations are essential.

Abstract Guideline-concordant breast cancer treatment selection is increasingly complex due to rapid evolution in biomarker-driven therapies. Large language models (LLMs), such as GPT-4, offer a potential tool to support clinical decision-making. However, hallucinations and deviation from evidence-based guidelines remain significant concerns. This study evaluated whether retrieval-augmented generation (RAG) using National Comprehensive Cancer Network (NCCN) guidelines improves GPT-4 performance in treatment recommendation tasks. We developed a retrieval-augmented GPT-4 agent (RAG-GPT) by indexing 2025 NCCN Breast Cancer Guidelines into a vector database using sentence embeddings. Forty clinical vignettes were created from published case reports to reflect diverse scenarios, including HER2-low, triple-negative, and BRCA-mutated disease across all stages. Two models—a baseline GPT-4 (without augmentation) and the RAG-GPT—were prompted with structured templates requesting treatment plans, rationale, clinical trial options, and citations.Outputs were independently scored by two blinded oncology-trained reviewers using the modified Generative AI Performance Score (mG-PS). The mG-PS scored outputs for 2 categories, guideline concordance (Gold standard = 1.0, Acceptable = 0.5, Non-concordant = 0.0) and hallucination errors (Severe = -1.0, Moderate = -0.5, Mild = -0.25, None = 0.0). Guideline Concordance was deemed based on NCCN guidelines and what an oncologist would due based on using the guidelines for each case. Hallucinations are a term used for falsehoods in Artificial Intelligence, like made-up facts or studies, incorrect information, improper doses, or inappropriate or wrong treatments. To ensure consistency and minimize inter-rater variability, reviewers were provided with comprehensive scoring guidelines and detailed instructions for each evaluation metric. Scores were normalized between -1.0 and 1.0 for improved evaluation of tools. Readability and rationale clarity were rated on a 5-point Likert scale. Inter-rater reliability was assessed using Cohen’s kappa. The RAG-GPT demonstrated a significantly higher mean mG-PS score (0.599 ± 0.102, 95% CI: 0.498-0.702) compared to the baseline GPT-4, which not only scored lower but had a negative mean (-0.21 ± 0.143, 95% CI: -0.353 to -0.067). For the mG-PS Cohen’s Kappa test resulted in a score of .69 with substantial aggreament. An independent sample t-test showed a difference (t(78) = 29.1, p &lt; 0.0001). Severe hallucinations were also absent in RAG-GPT outputs (0/40) but occurred in 8/40 baseline GPT-4 outputs. The evaluation of readability and rationality also showed notable differences, with the augmented LLM scoring significantly higher (M = 4.35, SD = 0.66) compared to the naïve LLM (M = 3.05, SD = 1.06). An independent samples t-test confirmed this difference t(65.3) = 6.58, p &lt; 0.0001, further supporting the benefit of augmented Large Language models. Cohen’s Kappa test resulted in a score of .76 for a substantial agreement for readability and rational clarity scale. Limitations include small sample size; further fine-tuning and larger real-world validation studies remain necessary. Linking GPT-4 to NCCN guidelines via retrieval-augmented generation significantly improves treatment accuracy and reduces hallucination severity in breast cancer care scenarios. While further refinement is needed, the reduced incidence of severe hallucinations, increase in guideline-concordant recommendations, and enhanced readability findings support further development and prospective validation of guideline-anchored LLMs for clinical use. Citation Format: C. Yost, A. Aseem, B. Callas, S. Monick, L. Bonilla, R. Nguyen, R. Galamaga, S. Osborn, Y. Kumar, N. Ertz-Archambault. Retrieval-augmented GPT-4 improves NCCN-concordant breast cancer treatment recommendations [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-06-29.

Abstract Secondary central nervous system (CNS) large B-cell lymphoma (SCNSL) occurs in the de novo setting, as a CNS-isolated relapse, or synchronous (concomitant CNS and systemic) relapse. SCNSL is a devastating event without therapeutic consensus. Thus, we aimed to evaluate treatment outcomes in an international cohort. Progression-free survival (PFS), overall survival (OS), and cumulative incidence of relapse (CIR, estimated using competing-risk models) were reported. Prognostic factors were identified in a 6-month landmark multivariate analysis. Outcomes after thiotepa autologous stem cell transplant (ASCT) and chimeric antigen receptor (CAR) T-cell therapy (CAR-T) delivered at relapse were compared after propensity score matching (PSM). A total of 1139 patients were included in the analysis (de novo: 537; relapsed SCNSL: 602). Two-year PFS estimates were 40.4%, 43.9%, and 16.2% for de novo SCNSL, CNS-isolated relapse, and synchronous relapse, respectively. Patients with CNS-isolated relapse demonstrated low rates of systemic recurrence (24-month CIR, 6%). Thiotepa-ASCT correlated with longer survival in de novo SCNSL (PFS: hazard ratio [HR], 0.57; P = .005; and OS: HR, 0.62; P = .023) and CNS-isolated relapses (PFS: HR, 0.55; P = .002; and OS: HR, 0.39; P&lt; .0001). ASCT (thiotepa or no thiotepa) also associated with improved survival in synchronous relapses (PFS: HR, 0.57; P = .023; and OS: HR, 0.48; P = .019). Higher survival with thiotepa-ASCT than CAR-T was observed after PSM (PFS: HR, 0.45; P = .005 and OS: HR, 0.41; P = .014). These data support thiotepa-ASCT in eligible patients, particularly de novo disease and CNS-isolated relapses. CNS-isolated relapse was infrequently associated with systemic recurrence, supporting treatment regimens adopted from primary CNS lymphoma.

Abstract While the use of NCCN guidelines is essential in lymphoma care, adherence among oncologists remains highly variable. Real-world concordance ranges from 65–80% overall, dropping to as low as 11–46% for certain lymphoma subtypes. Although 88% of clinicians report using NCCN guidelines for Hodgkin lymphoma, adherence to PET-adapted approaches is inconsistent. This gap highlights a critical need for tools that support guideline-concordant care.Large language models (LLMs) such as GPT–4 promise decision support, yet hallucinations and divergence from evidence-based care remain obstacles. We assessed whether retrieval-augmented generation (RAG) with National Comprehensive Cancer Network (NCCN) guidelines improves GPT–4 performance in lymphoma treatment recommendation tasks. To address this, we developed a RAG-enhanced GPT-4 agent (RAG-GPT) by indexing the 2025 NCCN Lymphoma Guidelines into a vector database. Fifty clinical vignettes spanning diverse lymphoma subtypes were compiled from published case reports; this sample provides 80 % power (α = 0.05) to detect a mean score difference of ≥ 0.18 on the modified Generative AI Performance Score (mG-PS). Guideline concordance was assessed based on NCCN guidelines and clinical judgment reflecting real-world oncological decision-making. Scores were normalized between -1.0 and 1.0 for consistent evaluation. Readability and rationale clarity were rated on a 5-point Likert scale. Baseline GPT-4 and RAG-GPT generated treatment plans via standardized prompts. An oncologist scored all outputs with mG-PS, which rates NCCN concordance (0.0–1.0) and penalizes hallucinations (-0.25 to –1.0). RAG-GPT achieved a higher mean mG-PS (0.32) than baseline GPT–4 (0.10) (t(98)=-2.11, p=0.038; mean score difference ΔmG-PS=0.22). Importantly, RAG-GPT eliminated severe hallucinations, reinforcing confidence in the safety of its treatment reccomendations. Fully concordant “gold-standard” plans were produced in 46 % of vignettes with RAG-GPT versus 22 % with baseline. Readability scores favored RAG-GPT modestly without statistical significance. Lymphoma's bimodal age distribution, heterogeneity, and lack of standardized treatment algorithms, coupled with regimen toxicity and the need to account for patient frailty and comorbidities, create complex clinical scenarios that constrain LLM applicability for nuanced decision-making. Non-Hodgkin lymphoma, particularly relapsed/refractory, transformed disease, or rare subtypes, lacks a defined standard of care, leading reviewers to rely on clinical judgment or published case series beyond NCCN guidelines. This underscores a key limitation of LLM-supported tools: their performance is highly dependent on the presence of clear algorithmic consensus within a given disease. Linking GPT-4 to NCCN guidelines via RAG significantly improved treatment accuracy and reduced hallucination severity in lymphoma, though performance remained lower than prior RAG studies in cancers like breast and CNS suggesting a greater impact on hallucination reduction than on retrieval of a consistent gold standard regimen. These findings support continued development and prospective validation of guideline-anchored LLMs for clinical use.

Adolescents and young adults (AYA; ages 15–39) with acute leukemia have significant unmet supportive care needs. This pilot study integrated same-day palliative care (PC) into an AYA leukemia clinic, measuring symptom burden (SB) with validated symptom assessment tools at baseline, 3, and 6 months. Among 31 patients, moderate-severe symptoms included tiredness, anxiety, and poor appetite. AYAs experienced high SB and psychological distress, regardless of care phase or sociodemographics. With PC integration, tiredness and depression improved significantly; no symptoms worsened. Half of the PC visits led to management changes. Early integration of symptom-focused PC was feasible, value-adding, and well-received by patients.