Tumor-of-origin prediction using methylation signals from plasma cell-free DNA (cfDNA): Real-world experience in Asia and the Middle East (AME).

3051 Background: Accurate identification of tumor tissue of origin (TOO) is essential for guiding treatment decisions in advanced cancers; however, tissue biopsies may be limited by accessibility, sample adequacy, or diagnostic delays. Plasma cfDNA-based methylation profiling provides a non-invasive approach for TOO determination and has demonstrated high agreement with clinicopathologic diagnoses in prior studies. However, real-world data from AME remains limited. We evaluated the real-world performance of a methylation-based TOO classifier using plasma cfDNA in this population. Methods: We retrospectively analyzed plasma cfDNA samples from patients with advanced solid tumors tested across AME using the Guardant360 Liquid assay through December 2025. The assay interrogates epigenomic signals from >19,000 methylated regions. The TOO classifier assigns a ranked cancer signal of origin (CSO) based on tumor-specific methylation patterns, reporting a primary CSO prediction and, when confidence is intermediate, a secondary CSO prediction. CSO confidence is quantified algorithmically based on methylation signal strength and classification certainty. Only samples with medium (50–80%) or high (>80%) confidence predictions were included. All tests were ordered as part of routine clinical care, and available clinicopathologic diagnoses (CPD) served as the reference standard. Results: Among 1,782 cfDNA-profiled samples, 1,230 (69%) yielded a CSO prediction with medium or high confidence. The primary CSO prediction matched CPD in 85% of cases overall, increasing to 91% when secondary CSO predictions were included. Tumor-type–specific accuracy varied, with the highest performance observed for colorectal, breast, and prostate cancers (94% each). Agreement with CPD was lower for gastroesophageal (72%) and kidney cancers (68%). Among 114 cancer of unknown primary (CUP) cases, a suspected clinical diagnosis was available for 48. Medium- or high-confidence CSO predictions were generated in 42 (87%). CSO predictions aligned with subsequent clinicopathologic consensus or confirmatory biopsy in 35 (83%), spanning 13 tumor types, most commonly lung cancer (24%). Therapy response evaluation was available for 12 patients, with partial response in 9 (75%) and stable disease in 3 (25%). Conclusions: In this real-world AME cohort, the TOO classifier identified probable tumor tissue of origin in most advanced cancer patients. CSO predictions showed high agreement with CPD across multiple tumor types. This approach demonstrated potential value in resolving CUP cases, with the predicted origin confirmed in 83% of evaluable patients. These findings support the clinical feasibility of methylation-based tumor-of-origin testing in AME and warrant further studies to evaluate its impact on clinical decision-making.