Background: Non-resectable pancreatic cancer patients have a wide range of median time for overall survival (OS). Currently there is a lack of diagnostic tools to predict patient outcome at diagnosis. KRAS mutations are present in the vast majority of pancreatic tumors. The study objective was to determine if quantitative baseline and longitudinal monitoring of KRAS mutations from plasma circulating tumor DNA (ctDNA) could be used to stratify patients for predicting outcome. Methods: Plasma was prospectively collected from the Danish BIOPAC study for non-resectable pancreatic cancer patients undergoing treatment with gemcitabine or FOLFIRINOX. Feasibility of monitoring ctDNA KRAS mutations was assessed in 10 patients with long OS (median 493 days; range 360-1031) and 10 patients with short OS (median 66 days; range 21-136). KRAS G12A/C/D/R/S/V, and G13D mutations were PCR enriched, sequenced by massively parallel deep sequencing, quantitated and standardized by reporting number of copies detected per 10⁵ genome equivalents (GE). Results: In a pilot study of 20 patients, all 18 patients with evaluable DNA had detectable KRAS mutations. Of 18 patients, 12 had baseline plus longitudinal time points (7 short, 5 long OS). Mutant KRAS copies were higher for short OS (median=994; range 0-34305 copies/10⁵ GE) vs. with long OS (median 196; range, 34-278 copies/10⁵ GE). Longitudinally, KRAS mutation levels remained mostly low with long OS (last time point median 204; range 8-873 copies/10⁵ GE) vs. short OS where levels increased or remained high (median 2363; range 71-47919 copies/10⁵ GE). Identical KRAS mutations were consistently detected for a given patient with short OS. However, long OS patients had variable KRAS mutations in longitudinal analysis. Conclusions: High levels of ctDNA KRAS mutations at diagnosis and post-treatment elevation of KRAS mutations were more associated with short OS. Different levels of KRAS mutation at diagnosis may predict patient outcome and could reflect distinct underlying tumor biology. Expansion of this prospective-retrospective biomarker cohort will be reported.