2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Saturday, December 12, 2026
05:30 p.m. - 07:30 p.m.
MCCNO - Hall F
Access, Representation, Population Health, Technology and Procedures, Imaging, Treatment Considerations, Combination Therapy, Supportive Care, Therapy Sequence, Biological Therapies, Antibody Therapy, Bispecific Antibody Therapy, Chimeric Antigen Receptor, (CAR)-T Cell Therapies, Immunotherapy, Monoclonal Antibody Therapy, Non-Biological Therapies, Chemotherapy, Pharmacology, Radiation Therapy, Workforce, Adverse Events, Artificial Intelligence (AI), Generative AI, Natural Language Processing, Biological Processes, Immune Mechanism, Molecular Biology, Multi-Systemic Interactions, Clinical Practice (Health Services and Quality), Clinical Guidelines, Health Economics, Patient Safety, Practice Models, Quality Improvement, Quality of Care, Systems-Based Hematology, Diseases, Infectious Diseases, Viral, Bacterial, Lymphoid Malignancies, Lymphoid Leukemias, Lymphomas, Plasma Cell Disorders, Myeloid Malignancies, Acute Myeloid Malignancies, Chronic Myeloid Malignancies, Measurable Residual Disease, ALL, CLL, Hodgkin Lymphoma, Non-Hodgkin Lymphoma, B Cell Lymphoma, T Cell Lymphoma, Indolent Lymphoma, Aggressive Lymphoma, Research, Clinical Research, Clinical Trials, Epidemiology, Health Outcomes Research, Health Disparities Research, Patient-Reported Outcomes, Real-World Evidence, Registries, Survivorship, Study Population, Human, Adult, Elderly, Pediatric, Young Adult
PAPER

2766 The CAR-T funnel: AI-assisted mapping of patient attrition and barriers across the second-line CAR-T pathway in relapsed/refractory diffuse large B-cell lymphoma

Nilanjan Ghosh, Eric Meadows, PhD2, Ahmed Sawas, MD3, Danni Zhao, PhD, MSc4, Julie Wu5, Danielle Janiszewski, PharmD6, Kim Saverno, PhD, RPh7
1 Levine Cancer Institute Atrium Health, Charlotte, North Carolina, United States
2 Flatiron Health, New York, New York, United States
3 Flatiron Health, New York, New York, United States
4 Flatiron Health, New York, New York, United States
5 Incyte Corporation, Wilmington, Delaware, United States
6 Incyte Corporation, Wilmington, Delaware, United States
7 Incyte Corporation, Wilmington, Delaware, United States

Introduction: CD19-directed chimeric antigen receptor T-cell (CAR-T) therapies are approved for relapsed/refractory diffuse large B-cell lymphoma (R/R DLBCL), yet real-world utilization in the second-line (2L) setting remains limited. Prior studies examining CAR-T treatment decision-making have been constrained by the effort required to extract nuanced information from clinical notes. Large language models (LLMs) enable scalable analysis of unstructured clinical data, providing an opportunity to better understand barriers to CAR-T use in R/R DLBCL. This study used LLMs to characterize the real-world 2L CAR-T treatment pathway, sequential attrition, and document barriers to CAR-T utilization among patients with DLBCL following first-line (1L) treatment failure.

Methods: A retrospective cohort study was conducted among US patients diagnosed with DLBCL using the nationwide Flatiron Health electronic health record (EHR)–derived deidentified database. The Flatiron Health database is a longitudinal database, comprising patient-level structured and unstructured data, curated via technology-enabled abstraction. Eligible patients consisted of US adults diagnosed with DLBCL on/after Jan 1, 2021, who had R/R disease confirmed by manual abstraction on/after April 1, 2022, following 1L systemic treatment. The date of the R/R event was the index date and patients were required to have at least 60 days of follow-up after index. For each eligible patient, a clinician-informed LLM extracted documented discussion about CAR-T with the primary oncologist to derive CAR-T referral, evaluation, apheresis, and infusion information. Bridging or holding therapy in the 2L immediately followed by CAR-T was considered 2L CAR-T. Among patients with documented CAR-T discussion who did not receive CAR-T in 2L, the LLM extracted documented reasons for non-receipt. Reasons were classified using a clinician-developed predefined taxonomy of patient-, disease-, treatment-, and healthcare system–related factors; multiple barriers could be recorded per patient. Expert human abstractors validated LLM outputs in an independent sample. Clinical and demographic characteristics were summarized and compared between patients who received CAR-T in 2L and those who received other systemic therapy in 2L using nominal p-values (<0.05).

Results: A total of 577 patients (56.5% male; median [IQR] age at index, 68 [59, 77] years) met eligibility criteria after failing 1L treatment. Most patients (60.7%) originated from community practices. Within the 2L treatment decision window, 20.6% (n=119) received CAR-T, 51.6% (n=298) received other (non–CAR-T) systemic treatment, and 27.7% (n=160) had not yet received any 2L therapy. Patients who received CAR-T in 2L (n=119) were younger than those who received non–CAR-T treatment in the 2L (n=298; median age at index, 62 vs 68 years; p<0.001) but had similar Eastern Cooperative Oncology Group performance status ([ECOG PS] 0/1 at index: 52.1% vs 43.6%; P=0.078). Attrition occurred sequentially throughout the 2L treatment pathway. Of 577 eligible patients, 62.7% (n=362) had a documented discussion regarding CAR-T, 49.4% (n=285) were referred, 39.0% (n=225) underwent evaluation, 23.2% (n=134) completed apheresis, and 20.6% (n=119) received CAR-T infusion. Among the 243 patients with a documented CAR-T discussion and who did not receive CAR-T infusion in 2L, 91 (37.4%) were actively pursuing CAR-T. Among the remaining 152 patients, the most common barriers to CAR-T receipt were comorbidities (22.4%, n=34), patient preference (19.1%, n=29), poor ECOG PS (17.8%, n=27), rapid disease progression (14.5%, n=22), and financial considerations (9.9%, n=15).

Conclusion: To our knowledge, this is the first multicenter study using LLMs to characterize sequential attrition across the 2L CAR-T treatment pathway and identify barriers to CAR-T utilization in R/R DLBCL. Although CAR-T was discussed for most patients, only ~1/5 received CAR-T in 2L. Clinical barriers, including comorbidities, poor ECOG PS, and rapid disease progression, as well as non-clinical barriers, such as patient preference and financial considerations, contributed to non-receipt of CAR-T. These findings highlight unmet needs among patients who did not receive CAR-T and the necessity for alternative treatment options. LLM-based analysis of unstructured clinical notes provides scalable insights into treatment decision-making that cannot be captured in structured EHR data alone.

Nilanjan Ghosh: AbbVie - Research Funding; Perpromene - Research Funding; AstraZeneca - Research Funding; Genentech/Roche - Research Funding; Galapagos - Consultancy (Includes expert testimony); Genmab - Consultancy (Includes expert testimony); Merck - Consultancy (Includes expert testimony); BMS - Consultancy (Includes expert testimony); Genentech/Roche - Consultancy (Includes expert testimony); Pfizer - Consultancy (Includes expert testimony); AstraZeneca - Consultancy (Includes expert testimony); AbbVie - Consultancy (Includes expert testimony); ADC Therapeutics - Consultancy (Includes expert testimony); Incyte - Consultancy (Includes expert testimony)