e20563 Background: Immune checkpoint inhibitors (ICIs) targeting programmed cell death 1 (PD-1), and its ligand (PD-L1) are the mainstay of treatment for metastatic non-small cell lung cancer (NSCLC) in patients lacking targetable driver mutations. However, optimal predictive biomarkers are lacking. We used a plasma cell-free DNA (cfDNA) 5-hydroxymethylcytosine (5hmC) signature-based predictive model to assess treatment response and potentially select patients eligible for ICI monotherapy. Methods: 40 plasma samples were collected from adult patients with stage III or IV lung cancer at Houston Methodist Hospital between 2019 and 2022. These were divided into a training (n = 24) and validation (n = 16) set. By identifying genes associated with progression free survival (PFS), a 16-gene signature was used to develop a 5hmC predictive model. A weighted predictive score (wp-score) based on the model was calculated for each patient and grouped into low and high wp-scores. Results: Low wp-scores were associated with higher objective response rate(ORR) in both training and validation set and better correlated with response rates compared to PD-L1. In the training set, PFS was longer in low wp-scores compared to high wp-scores (median 12.3 versus 3.0 months; p = 3.5x10-6; hazard ratio (HR) 4.6x10-10). This was also seen in the validation set (median 7.6 versus 1.8 months p = .0012; HR 0.12). Contrarily, high ( > 1%) or low ( < 1%) PD-L1 was not associated with PFS (median 6.8 versus 6.0 months; p = .40; HR 0.70; 95% CI, 0.30–1.60) or overall survival (OS) (median 10.1 versus 12.0 months; p = .38; HR 0.56; 95% CI, 0.15–2.10). 15 out of the 40 total patients received ICI monotherapy. Among these, low wp-scores (n = 8) were significantly associated with longer PFS compared to high wp-scores (n = 7), (median 19.6 versus 2.8 months; p = .00073; HR: 0.059). OS although longer, was not significant among the two groups (median 38.1 versus 10.5 months; p = .12). Conclusions: Our study provides a proof of concept that cfDNA 5hmC signature can be used to predict response to ICI therapy and select patients appropriate for monotherapy. It is a more sensitive and specific biomarker to PD-L1. Larger studies are underway to further validate cfDNA 5hmC based-signature predictive model. [Table: see text]