A real-world analysis of ctDNA methylation-based histology prediction and associated clinical characteristics in non–small cell lung cancer (NSCLC).

Abstract

8539 Background: The treatment of NSCLC differs by histologic subtype, though traditional histologic review faces several challenges including insufficient tissue and intratumoral heterogeneity. The circulating tumor DNA (ctDNA) assay Guardant360 Liquid (G360) now includes a molecular tumor type (TT) and lung subtype (ST) predictor based on DNA methylation patterns. This study aimed to evaluate the concordance between G360 TT/ST prediction and tissue-based histology in a real-world sample of patients (pts) with NSCLC. A comparative analysis was conducted to understand differences in clinical characteristics between pts with and without G360 TT/ST prediction. Methods: This retrospective study included pts with NSCLC who underwent G360 testing from May to October of 2025. G360 predicted TT and ST (proportion of adenocarcinoma (AD), squamous cell carcinoma (SQ), and small cell carcinoma (SC)). Dominant TT was defined as the top cancer of origin prediction. Dominant ST was defined as a proportion >80%. G360 TT/ST was compared to tissue histology to determine concordance using Cohen’s kappa coefficient. Demographic and clinical characteristics were collected, and statistical differences (p<0.05) between pts with and without TT/ST prediction were evaluated. Results: Of 145 pts who underwent a combined 176 G360 assays, 47 (32%) had assays predicting TT/ST, including 9 pts with TT prediction only and 38 with both. Of these 38 pts, 32 had tissue-proven AD, 3 had SQ, and 3 had poorly differentiated NSCLC. The overall accuracy of G360 TT prediction was 100% (47/47). The accuracy of ST prediction, excluding the 3 pts with poorly differentiated NSCLC, was 94% (33/35, k=0.72), with 2 discordant cases. One case was a pt with AD whose G360 at diagnosis predicted AD, but 4 months later predicted a SC component (40%). Both pts with discordant subtyping died within 4 months of their G360 result. Pts with TT/ST prediction were statistically younger, had more advanced disease, higher extra-thoracic metastatic burden, and higher ctDNA tumor fraction (TF) than pts without TT/ST prediction (Table 1). Conclusions: In this real-world analysis, G360 TT/ST prediction demonstrated good concordance with tissue-based histology, though prediction was more likely in pts with more advanced disease and higher TF. In the future, ctDNA-based TT/ST prediction may be a useful noninvasive tool to confirm primary lung cancer, classify histologic subtype, and identify histologic transformation, particularly during periods of disease progression, leading to better therapeutic decisions. Baseline pt characteristics. (+) G360 TT/ST prediction (n=47) (-) G360 TT/ST prediction (n=98) P -value Median age (IQR) 68 (64-77) 71 (60-74) 0.03 Female (%) 27 (57.4) 60 (61.2) 0.66 Stage III/IV (%) 46 (97.9) 67 (68.4) <0.0001 Extra-thoracic metastases (%) 35 (74.5) 20 (20.4) <0.0001 TF (%) 3.9 0.7 0.01