Abstract A038: Performance of classification models using limited genomic data in cases of newly diagnosed large B cell lymphoma

Abstract Introduction: LymphGen and DLBClass assign cases of newly diagnosed large B cell lymphoma (LBCL) into genetic subgroups; however, performance using limited genomic data (lim) is unknown. Methods: Cases included in the discovery cohort had published genetic subgroup assignment by LymphGen and DLBClass using comprehensive genomic data (comp) and available mutation analysis data reported in Schmitz, et al (PMID 29641966) and Chapuy, et al (PMID 29713087) and were compared to a validation cohort of cases from the University of Pennsylvania that underwent mutation analysis via clinical laboratory assays. Genes common to these assays and the algorithms were included in lim: B2M, BCL2, BCL6, BRAF, CARD11, CD79B, CDKN2A, CIITA, CREBBP, CXCR4, DDX3X, EP300, ETV6, EZH2, GNA13, ID3, IRF4, KLF2, KLHL6, KMT2D, KRAS, MAP2K1, MEF2B, MYD88, NFKBIA, NFKBIE, NOTCH1, NOTCH2, PIK3CG, PIM1, PRDM1, PTEN, RHOA, SF3B1, SOCS1, STAT3, TBL1XR1, TET2, TNFAIP3, TNFRSF14 and TP53. Results: For cases from the discovery cohort (n=535), LymphGen assignment using comp, attributing cases classified with multiple subtypes to each individual subtype, was A53 13%, BN2 16%, EZB 19%, MCD 15%, N1 3%, Other 36% and ST2 6%. DLBClass assignment using comp was C1 20%, C2 23%, C3 16%, C4 14% and C5 in 27%, with confidence ≥0.7 for 75% of cases. As compared to assignment using comp, assignment using lim demonstrated sensitivity/positive predictive value (PPV) by LymphGen of 0.21/0.95 for BN2, 0.75/0.68 for EZB, 0.85/0.72 for MCD, 0.82/0.93 for N1, 0.82/0.58 for Other and 0.55/0.57 for ST2; by DLBClass, 0.40/0.66 for C1, 0.13/0.83 for C2, 0.92/0.35 for C3, 0.44/0.57 for C4 and 0.75/0.68 for C5, with confidence ≥0.7 for 52% of cases. Of note, use of lim resulted in assignment to Other by LymphGen for 50% of cases. For the clinical analysis sub-cohort (n=261), the proportion of cases achieving 24 month disease free survival (DFS24) by R-IPI score was 0.93 for 1, 0.79 for 2 and 0.58 for 3 (p<0.001). The proportion of cases achieving DFS24 when comparing genetic subgroup assignment by comp vs lim for LymphGen was 0.70 vs 0.63 (p=0.69) for BN2, 0.79 vs 0.79 (p=1.00) for EZB, 0.64 vs 0.69 (p=0.79) for MCD, 0.25 vs 0.50 (p=1.00) for N1, 0.75 vs 0.71 (p=0.54) for Other, 0.79 vs 0.93 (p=0.35) for ST2; for DLBClass, 0.68 vs 0.82 (p=0.26) for C1, 0.79 vs 0.75 (p=1.00) for C2, 0.75 vs 0.77 (p=0.83) for C3, 0.82 vs 0.75 (p=0.56) for C4 and 0.64 vs 0.62 (p=0.87) for C5. For cases from the validation cohort (n=114), there was no statistically significant difference in the distribution of R-IPI score or genetic subgroup assignment, nor the proportion of cases achieving DFS24 by R-IPI score or genetic subgroup assignment, when compared to discovery cohort cases using lim. Conclusions: Use of lim allows for high confidence assignment of a specific genetic subgroup by LymphGen or DLBClass in ∼50% of newly-diagnosed LBCL cases, yielding variable sensitivity/PPV for assignment using comp, but similar DFS24, the latter of which was reproducible in an independent cohort. Citation Format: Daniel J. Landsburg, Jennifer JD. Morrissette, Caren Gentile, Sunita D. Nasta, Jakub Svoboda, Elise A. Chong, Stefan K. Barta, Jordan S. Carter, Stephen J. Schuster, Colin J. Thomas, Michael R. Cook, Emily B. Tomasulo, Ashkan Bigdeli, Eitan Halper-Stromberg, Salvatore F. Priore. Performance of classification models using limited genomic data in cases of newly diagnosed large B cell lymphoma [abstract]. In: Proceedings of the Fifth AACR International Meeting on Advances in Malignant Lymphoma: From Discovery to Clinical Impact; 2026 Jun 24-27; Philadelphia, PA. Philadelphia (PA): AACR; Blood Cancer Discov 2026;7(3_Suppl):Abstract nr A038.