Background: Currently available circulating cell-free DNA (cfDNA) assays require deep-targeted sequencing to detect cancer-specific mutations at low mutant allele frequency (MAF) levels in the blood. Recently, we developed a tumor-agnostic, mutation-independent approach that utilizes low-coverage whole genome sequencing called DELFI (DNA evaluation of fragments for early interception) Tumor Fraction (DELFI-TF), a model designed to predict plasma tumor fractions based on genome-wide fragmentation-related features. Here, we report the results of DELFI-TF applied to a prospective cohort of patients with metastatic colorectal cancer (mCRC). Methods: Overall, 692 longitudinal plasma samples collected from 153 initially treatment-naive mCRC patients participating in the phase III CAIRO5 study (NCT02162563) were sequenced at low coverage and used for training and cross-validation. In patients with tumor-tissue-proven RAS/BRAF mutations, the tumor fractions were quantified as the cfDNA MAF of the RAS/BRAF variant measured by droplet digital PCR (ddPCR). Using fragment-sequencing statistics, a Bayesian regression model was trained against the MAF of the tumor-specific driver RAS/BRAF variant in all longitudinal timepoints to generate DELFI-TF scores. Changes in longitudinal DELFI-TF scores during first-line therapy (DELFI-TF slopes) were examined to predict treatment response and survival outcomes. Results: The DELFI-TF scores were strongly correlated with RAS/BRAF cfDNA MAF measured by ddPCR (Pearson, r= 0.85, p< 0.001). Baseline DELFI-TF correlated with dimensions of liver metastases reported in CT scans (Pearson, r= 0.49, p< 0.001) as well as clinical response, with pre-treatment levels significantly lower in patients with a later-confirmed partial or complete response (Wilcoxon, p< 0.05). Patients with low or negative DELFI-TF slopes presented with longer progression-free survival in the overall study population (13.4 months vs 10.4 months, HR = 2.03, 95% CI 1.25 to 3.32, Log-rank p< 0.01) and in patients who experienced durable clinical benefit (16.7 months vs 13.3 months, HR = 2.24, 95% CI 1.1 to 4.55, Log-rankp= 0.023). Patients with low or negative DELFI-TF slopes also experienced significantly longer overall survival (59.4 months vs 29.1 months, HR = 3.05, 95% CI 1.58-5.90, Log-rankp< 0.001). Tissue-informed focal and arm-level copy number changes were detected 4-12 weeks after liver metastases resection. Most patients detected as having a molecular relapse were diagnosed earlier than clinical recurrences identified by conventional imaging. Conclusions: DELFI-TF demonstrates the ability to use cfDNA fragmentomes to estimate cfDNA tumor burden with performance comparable to standard approaches for treatment response monitoring and clinical outcome prediction.