Improving Prognostication in Colon Cancer (MARBLE)

The MARBLE study is a prospective, multicentric, single-arm exploratory trial designed to integrate novel tissue- and blood-based prognostic modalities to improve risk stratification for patients with Stage II and III colon cancer. The study evaluates three primary prognostic modalities:

• Pathological Tumor-Node-Metastasis (pTNM) staging
• Machine Learning (ML) Histopathology: A validated graph neural network model analyzes digitized hematoxylin and eosin (H&E) stained slides to generate survival predictions and predict regional gene expression independent of pTNM stage
• Multi-modal ctDNA Assay: An in-house circulating tumor DNA (ctDNA) assay that moves beyond mutation detection by profiling epigenomic markers and genome-wide fragmentomic patterns in cell-free DNA (cfDNA) to enhance sensitivity for minimal residual disease (MRD).

Study Design and Workflow Participants are stratified post-surgery into four treatment-related groups (A-D) that reflect real-world standard-of-care (SoC) pathways rather than randomized arms:

• Group A: Low-risk Stage II (no adjuvant chemotherapy).
• Group B/C: Intermediate to high-risk Stage II/III (3-6 months of adjuvant chemotherapy).
• Group D: High-risk Stage III (6 months of chemotherapy), featuring intensified longitudinal sampling every 6 months for 36 months post-treatment to capture MRD dynamics.

Statistical Approach The primary analysis involves comparing mono-modal and multi-modal risk models to evaluate their prognostic performance for Time to Recurrence (TTR).

Technical evaluation will utilize Cox proportional hazards models, assessing discrimination via ΔC-index, model fit through likelihood ratio testing, and clinical utility using Net Reclassification Improvement (NRI). The study also quantifies the lead time between molecular (ctDNA) and radiological recurrence.