Abstract Introduction: CUP is diagnostically complex commonly with treatment delays and poor outcomes. Diagnosis integrates clinical, radiologic, and pathologic data, but limited or exhausted tissue can restrict accuracy. Interrogation of cancer-derived epigenomic signatures provides insights that may support CUP adjudication. The Guardant360 Liquid (Guardant Health, Palo Alto, CA) Molecular Tumor Typing predictor (MTT) applies methylation signatures to predict cancer signal of origin with graded confidence across 14 solid tumors. This study describes MTT predictions and genomic landscapes in CUP vs non-CUP (NCUP) patients (pts) and examines real-world (RW) outcomes in CUP pts predicted to have lung cancer. Using InfinityAI Data Library, a database of claims data, treatment outcomes were examined in selected pts with MTT high confidence predictions. Methods: CUP and NCUP pts with Guardant360 Liquid (LB) testing were identified from InfinityAI Data Library. Pts with high confidence MTT prediction (≥80% confidence) were included (n= 28,712). LB assessed alterations of 700 genes and thousands of differentially methylated regions. RW time to treatment discontinuation (RWTTD) and time to next treatment (RWTTNT) were evaluated as surrogate markers for RWOS and RWPFS, respectively, for CUP pts with MTT-predicted lung cancer vs NCUP pts with lung cancer on test requisition form (TRF). Pts were stratified by first line therapy post LB. Log rank test was used for time-to-event analyses. Results: Of eligible pts, 1,145 had CUP and 27,641 had NCUP recorded on TRF. MTT with high confidence was generated for 74.7% of CUP and 71.7% of NCUP patients. MTT most often predicted lung (29.4%), biliary (15.5%), and gastroesophageal (10.5%) primaries in CUP, whereas lung (22.4%), breast (17.2%), and colon (18.7%) predominated in NCUP. In CUP pts with MTT predicted lung cancer, common alterations included TP53 (69.1%), KRAS (31.7%), EGFR (28.4%), KEAP1 (28.0%), and STK11 (25.1%), while NCUP predicted as lung showed TP53 (77.8%), EGFR (28.7%), KRAS (28.1%), ATM (26.4%), and PIK3CA (12.3%). In addition, there was no statistical difference in KRAS G12C (p0.1813) and EGFR exdel19 (p=0.273) between the 2 cohorts. Treatment outcomes were available for 146 CUP-lung and 9,489 NCUP-lung pts. In the first 4 months (mo) RWTTNT was not statistically significant different, though favored targeted therapy. RWTTD was significantly longer in CUP-lung vs NCUP-lung (p0.0001): median 3.7 vs 4.2 mo for non-targeted therapy, and 3.0 vs 8.0 mo for targeted therapy. Conclusions: This study demonstrated RW pt outcomes after applying a plasma-based epigenomic MTT high confidence predictor. CUP-lung pts with targeted therapy had better survival outcomes vs pts with non-targeted treatment and have unique genomic profiles. This data is encouraging for MTT concordance with patient therapy selection and outcomes. Citation Format: Axel Grothey, Daniel Hintz, Jayati Saha, Sheila R. Solomon, Nicole Zhang, Jack Tung. Novel liquid biopsy epigenomic molecular tumor typing in carcinoma of unknown primary (CUP) abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6516.
Grothey et al. (Fri,) studied this question.