Evaluating the prognostic utility of cell-free (cf)DNA tumor fraction (TF) in metastatic castration-resistant prostate cancer (mCRPC).

5052 Background: Accurate prognostication is essential for guiding treatment decisions. cfDNA TF provides a non-invasive quantitative assessment of tumor burden without the need for tissue biopsies, which are notoriously challenging to obtain in mCRPC patients. This study evaluates the utility of cfDNA TF as a prognostic biomarker in mCRPC. Methods: Patients treated for mCRPC at the Dana-Farber Cancer Institute with available plasma were identified. Plasma samples underwent (epi)genomic profiling using the Guardant360 platform. TF was estimated by normalizing cancer-specific differentially methylated regions with appropriately matched control regions within each sample. The association of TF with clinical prognostic biomarkers and overall survival (OS) from the time of plasma collection was assessed. Survival analysis was performed using cox proportional hazards methodology, continuous variables were compared using Mann-Whitney test, and linear correlations were calculated using Pearson correlation coefficient. Decision tree (DT) models were developed to evaluate the benefit of incorporating clinical prognostic biomarkers with TF in predicting OS at 12 months. A 70/30 split was used to separate the training and testing cohorts and hyper-parameters were optimized for the F1 score (harmonic mean of precision and recall). Results: A total of 103 patients with mCRPC were identified (median age at plasma collection: 72; median prior lines of therapy: 3; median mTF: 3.7%). 36%, 34%, and 30% of patients had TF <1%, 1-10%, and >=10% respectively. Compared to patients with TF <1% (median OS: 28 months), OS was significantly shorter among patients with TF 1-10% (median: 16 months; HR=2.6, p=.00032) and >=10% (median: 10 months; HR=7.8, p=6.0x10-12). TF was significantly correlated with max variant allele frequency (mVAF) (rho=0.79 ,p=2.1x10-23), a traditional genomic-based cfDNA measure of tumor burden; however, on multivariable analysis, only TF (HR=5.8,p=2.8x10-9), and not mVAF (HR=0.97,p=0.95), was independently associated with worse OS. Elevated TF was associated with known poor-risk disease features, including visceral metastases, higher ECOG scores, and elevated serum markers (alkaline phosphatase, PSA, LDH). Multivariable analysis identified TF > median as the strongest negatively prognostic marker for OS (HR=3.5,p=.00029). TF alone demonstrated comparable predictive accuracy for 12-month OS (F1: 0.72) to a model including all clinical prognostic markers plus TF (F1: 0.72). Decision tree models identified an optimal TF cutoff of 5.6%, with 88% (53/60) of patients with TF <5.6% alive at 12 months, compared to 36% (11/42) for those with TF ≥5.6%. Conclusions: Our findings suggest that methylated tumor fraction is a robust independent non-invasive prognostic biomarker in mCRPC, outperforming traditional clinical markers and genomic metrics. Future studies should explore the integration of TF into standard prognostic workflows and its potential to guide therapeutic decisions.