Quantitative assessment of tumor shedding using AI-derived radiographic tumor burden and quantified liquid biopsy in advanced gastrointestinal cancers.

3063 Background: Radiographic imaging remains the clinical standard for assessing cancer treatment response; however, it lacks sensitivity to detect early biological shifts and spatially heterogeneous tumor behavior. Methylated circulating tumor DNA (mctDNA) provides a complementary molecular measure of disease burden, yet the quantitative relationship between radiographic tumor burden and DNA shedding remains poorly characterized. In this cohort study, we evaluated whether early changes in shed rate predict progression-free survival (PFS) and overall survival (OS) in patients with advanced gastrointestinal (GI) cancers. Methods: Patients with advanced GI malignancies receiving systemic therapies underwent paired longitudinal radiographic and liquid biopsy assessments. Radiographic tumor burden was quantified (OSCAR AI platform). Tumor contours were independently blinded reviewed by two radiologists. Up to 15 target lesions per organ were selected, and the total tumor burden was calculated analogous to longitudinal RECIST assessment. Tumor Methylation Score (TMS) from the Northstar Response assay quantified mctDNA. Tumor shed rate (TSR) was defined as the ratio of TMS to the AI-derived radiographic tumor volume and calculated at baseline and at the first on-treatment landmark using the closest-paired CT scan and blood drawn within 120 days. Fold-change in TSR from baseline was compared with PFS and OS. Undetectable tumor volume cases were excluded. Results: Forty patients were evaluable, with both baseline and landmark timepoints available within the 120-day window. TSR ranged from a min of 0 methylated molecules per mm 3 (m4) to a max of 1.0 m4 with a median of 0.06 m4. TSR significantly differed between tumor types (p=0.013, ANOVA), with CRC having the highest median shed rate (0.13 m4) and EGA and pancreatic having the lowest median shed rates (0 m4). TMS was significantly associated with tumor volume at baseline (Rsq=0.15,p=0.01, Deming) and on-treatment (Rsq=0.6, p<0.001, Deming). An increased TSR was associated with OS (HR=5.8, 95% CI 1.7-19.2, <0.01) and PFS (HR=3.5, 1.4-8.6, <0.01). Conclusions: TSR represents a novel, multidimensional integration of AI-driven volumetrics and liquid biopsy. Our findings suggest that an increasing shed rate is a predictor of PFS and OS, potentially serving as a lead-time indicator of tumor-host homeostatic decompensation. Future analysis will determine if TSR captures critical transitions in tumor biology that precede and perhaps drive radiographically detectable progression.