Epigenomic liquid biopsy molecular lung subtyping and real-world (RW) patient outcomes in advanced NSCLC (aNSCLC)

Introduction: NSCLC subtypes include adenocarcinoma (LUAD), squamous cell (LUSC), and small cell (SCLC). Mixed histology, limited sampling, and discordant interpretations often delay treatment. To address this, a plasma-based Molecular Lung Subtype Predictor (MLSP) was developed to quantify LUAD, LUSC, and SCLC from circulating hypermethylated DNA (Guardant360 Liquid, Guardant Health). We report concordance between MLSP and histology from test requisition forms (TRF), real-world (RW) outcomes by therapy type, and subtype-specific genomic profiles. Methods: The InfinityAI Data Library links de-identified genomic/epigenomic results with longitudinal claims data. Samples were considered “pure” when ≥90% signal was from one subtype. Cohorts were stratified by MLSP-TRF concordance and first-line therapy post-Guardant360 Liquid: chemotherapy (chemo), immunotherapy (IO), chemo-IO, or targeted therapy. RW time to treatment discontinuation (RW-TTD) and time to next treatment (RW-TTNT) were analyzed using Kaplan-Meier and log-rank tests. Results: Among 8,559 MLSP results, 69.4% were LUAD, 12.4% LUSC, 3.7% SCLC, and 14.6% mixed. Concordance with TRF was 91.6% (LUAD), 75.1% (LUSC), and 56.1% (SCLC). In MLSP-LUAD, Tier 1 mutations included KRAS G12C (11.9%), EGFR ex19del (9.5%), and BRAF V600E (2.1%); PIK3CA E545K (6%) was common in MLSP-LUSC, and RB1 (19%) in MLSP-SCLC. Targetable biomarkers occurred in 43.5% of MLSP-LUAD vs 3.4% LUSC and 2.6% SCLC. In discordant MLSP-TRF cases, genomic profiles favored MLSP predictions. KRAS G12C (7%), PIK3CA E545K (5%), and RB1 (39%) were the most frequent subtype-specific alterations. Targeted therapy improved RW outcomes in MLSP-LUAD; chemo-IO yielded best outcomes in MLSP-LUSC, and IO benefited MLSP-SCLC. Discordant cases treated with IO or targeted therapy showed extended RW-TTNT and RW-TTD, aligning with MLSP-predicted biology. Conclusion: This study demonstrated high concordance between MLS and TRF, which is consistent with previous analyses. The genomics and RW outcomes of discordant cases were more consistent with MLSP-predicted subtype vs. TRF-reported histology. Discordant pts exhibited genomic patterns and RW outcomes more aligned closely with the MLSP-predictions, suggesting MLSP may more accurately represent underlying tumor biology than standard histology.