Circulating tumor DNA for characterizing intratumor heterogeneity and outcomes in metastatic colorectal cancer.

3585 Background: Intratumor heterogeneity (ITH) drives CRC resistance, yet tissue-based assessments remain limited. ctDNA captures heterogeneity, though variable tumor fractions confound measurements. We assessed raw vs fraction-adjusted ITH in ctDNA for overall survival (OS) in metastatic CRC (mCRC). Methods: We analyzed two real-world mCRC cohorts from MD Anderson: a 73-gene panel (1,137 patients, 2014-2023; limit of detection (LoD) 0.1%) and a 701-gene panel (150 patients, 2024-2025; LoD 0.5%). Quality filters included ≥ 3 detected variants, MaxVAF > 0.01, and bioinformatic filtering of putative germline variants. ITH was quantified via raw Mutant-Allele Tumor Heterogeneity (rMATH; VAF distribution width) and the Shannon entropy-based, MaxVAF-normalized (VAF/Max_VAF) blood ITH (bITH; multiplies bin entropy by clonality weights). OS was assessed using Cox models by continuous variables and optimal cut-points determined by maximally selected log-rank statistics. A stress test subset (top 50% MaxVAF) was analyzed to decouple findings from analytical noise inherent at low tumor fractions. Results: In the 73-gene cohort (median age 53.2; 91% MSS, 14% BRAF mutated, 64% RAS mutated), MaxVAF was the strongest predictor of worse OS in both univariable (HR 1.32, p < 0.001) and multivariable (MV) models (HR 1.31, p < 0.001). Paradoxically, high rMATH was associated with improved OS (MV HR 0.89, p = 0.014). This protective signal of high heterogeneity was more pronounced in the dichotomized model (MV HR 0.71, p < 0.001) and persisted in the subset of high-burden patients (top 50% MaxVAF) (MV HR 0.69, p < 0.001) and the MSS subset (MV HR 0.68, p = 0.003). In the 701-gene panel cohort, MaxVAF remained the dominant independent predictor (MV HR 2.37, p < 0.001). bITH was not independently prognostic (MV HR 0.98, NS). Conclusions: MaxVAF serves as the primary ctDNA-derived driver of clinical outcomes in mCRC, confirming that ctDNA-inferred ITH metrics are heavily confounded by total tumor shedding. The unexpected association between high rMATH and improved OS suggests that increased clonal diversity, at a fixed tumor burden, may indicate a less evolutionarily pruned or consolidated tumor compared to those that have undergone a selective clonal sweep. These results highlight that while ctDNA can characterize clonal architecture, the clinical interpretation of heterogeneity must be contextualized within the tumor’s total systemic volume and evolutionary state. Cohort Variable (High vs Low) Multivariable HR (95% CI)* p-value 73-gene panel MaxVAF 1.31 (1.20–1.43) <0.001 Raw MATH 0.89 (0.81–0.98) 0.014 bITH 0.98 (0.90–1.08) 0.716 701-gene panel MaxVAF 2.37 (1.52–3.71) <0.001 Raw MATH 0.85 (0.50–1.44) 0.536 bITH 1.23 (0.82–1.87) 0.320 *HR per 1 SD increase. Raw MATH was adjusted for MaxVAF, while bITH was tested without redundant MaxVAF adjustment. All models adjusted for age, sex, MSI, BRAF, and RAS status.