Inflammation is increasingly recognized as a critical factor in acute myeloid leukemia (AML) pathogenesis. We performed blood-based proteomic profiling of 251 inflammatory proteins in 543 newly diagnosed AML patients. Using a machine learning model, we derived an eight-protein prognostic score termed Leukemia Inflammatory Risk Score (LIRS). Individual proteins were evaluated in multivariable cox models and model performance was assessed by cumulative concordance index. Findings were validated in internal and external cohorts across two institutions. Blood-based LIRS significantly outperformed the European LeukemiaNet (ELN) 2022 risk model and was independently prognostic of overall survival after accounting for known clinical and molecular prognostic factors. OSMR was uniquely identified as the strongest independent predictor of survival, early mortality, and induction chemotherapy response, and further validated in an independent assay. These blood-based biomarkers could have significant clinical implications for risk stratification and prognostication in patients with newly diagnosed AML.