Immune checkpoint inhibitors (ICIs) show limited efficacy in non-small cell lung cancer (NSCLC), highlighting the need for predictive biomarkers. Here we prospectively analysed serial plasma samples from 34 ICI-treated advanced NSCLC patients (plus 30 validation samples) using the Olink Immuno-Oncology panel. We assessed dynamic proteomic changes associated with ICI efficacy and immune-related adverse events (irAEs), and developed a prognostic model. Following ICIs, 42/92 proteins significantly elevated upon ICI treatment (p < 0.05). Baseline levels of CD28, CXCL10, and TNFSF14, and increased CD40L post-treatment, correlated with inferior response. Baseline IL-4, IL-13 and increased GZMA post-treatment were associated with irAE occurrence. Using LASSO-Cox regression, we established an Immunosuppressive Signature of Combined Resistance Elements (I-SCORE) model based on eight plasma proteins (CCL23, ARG1, CD83, ADA, CXCL10, TNFSF14, CD28, GZMA). I-SCORE demonstrated strong predictive power for overall survival (12-month AUC = 0.94), progression-free survival (12-month AUC = 0.75), and treatment response (AUC = 0.62). Furthermore, a high I-SCORE was demonstrated to reflect an inflammatory and immunosuppressive phenotype, showing positive linear relationships with plasma IL-6, IL-8, IL-10, and monocyte count, and negative relationships with IL-33 and active T-cell proportion. Our study identifies I-SCORE, derived from plasma proteomics, as a promising integrated biomarker for predicting ICI outcomes in NSCLC. It suggests targeting specific proteins or the associated immunosuppressive microenvironment might enhance immunotherapy efficacy.