Real-world analysis of epigenomic molecular tumor-type prediction for biliary tract cancer in CUP.

Abstract

4161 Background: Biliary cancers (BTC) are often diagnosed at advanced stages and can present as metastatic liver tumors, making determination of the primary site difficult, leading to frequent cancer of unknown primary (CUP) designations. Conventional histopathology or imaging may misclassify tumor origin, yet accurate distinction is essential because therapies differ for BTC and CUP. A novel liquid biopsy (LB) molecular tumor type (MTT) classifier using epigenomic signatures predicts tumor origin across 14 solid tumors with graded confidence. This study evaluated concordance, genomic features, and real-world outcomes of high-confidence MTT-BTC predictions in patients (pts) with BTC or CUP using the Infinity AI, a clinical-genomics and epigenomics database. Methods: Guardant360 LB results showing high-confidence (>80%) MTT predictions for BTC in pts with BTC or CUP listed on the test requisition form (TRF) were identified. Infinity AI links de-identified clinical, genomic, epigenomic, and administrative claims The LB assay interrogated >700 genes and thousands of methylation regions. First-line (1L) therapies were categorized as chemotherapy (chemo), immunotherapy (IO), or chemo-IO. Real-world time to treatment discontinuation (rwTTD) and time to next treatment (rwTTNT) were compared between TRF-BTC and TRF-CUP cohorts using log-rank tests. Results: Among pts with high-confidence MTT-BTC predictions (TRF-BTC n=410; TRF-CUP n=56), MTT accuracy for TRF-BTC was 51.7%. Female proportion was 58% (TRF-BTC) and 64% (TRF-CUP), respectively. Median ages for TRF-BTC and TRF-CUP pts were 68 years, respectively. The most frequent oncogenic mutations in TRF-BTC were TP53 (34), ARID1A (8), KRAS (7), and IDH1 (6); none had FGFR2 . In TRF-CUP, common mutations included TP53 (3), ARID1A (2), and IDH2 (2). For TRF-BTC pts, rwTTD was longer with 1L IO+chemo vs IO or chemo alone (7.9 vs 4.3 vs 3.4 mo; p<0.0001). TRF-CUP pts showed similar, but non-significant trends. rwTTNT was longer for TRF-BTC pts receiving IO (p=0.00026) and trended higher for TRF-CUP (p=0.069). Conclusions: Guardant360 LB epigenomic–based MTT classifier consistently predicted biliary origin in pts labeled as BTC and in a subset labeled as CUP, with broadly concordant genomic features and real-world treatment patterns. CUP pts with high-confidence MTT-BTC predictions demonstrated outcomes that approximated those of TRF-BTC, supporting biologic similarity and suggesting that some CUP cases may represent occult BTC. These findings support consideration of MTT as a tool to refine uncertain diagnoses, guide BTC-aligned therapy selection (including targetable alterations), and improve clinical trial matching. Larger cohorts with longer follow-up and prospective validation are needed to define the impact of MTT-guided management on patient outcomes.