Rituximab (RTX) is an effective therapy for pediatric steroid-nonresponsive immune thrombocytopenia (ITP), yet persistent challenges remain in predicting early therapeutic responses. To address this, we conducted a prospective cohort study analyzing serial urine samples from 37 steroid-nonresponsive ITP patients (17 RTX responders (RTX-R), 20 non-responders (RTX-NR)) and 38 age- and sex-matched controls using untargeted LC-MS metabolomics. We identified 16 pretreatment and 2 posttreatment metabolites distinguishing RTX-R from RTX-NR. Pretreatment metabolic pathway analysis in RTX-R revealed significant enrichment in folate-mediated one-carbon pool, ether lipid metabolism, porphyrin metabolism, and amino acid metabolism (cysteine/methionine, arginine/proline), and purine pathways. Longitudinal clustering analysis demonstrated RTX-induced metabolic changes in purine metabolism, pantothenate and CoA biosynthesis, beta oxidation of fatty acids, and valine, leucine and isoleucine biosynthesis. A machine learning-derived predictive model incorporating seven metabolites (N6-methyl-2'-deoxyadenosine, dopamine 4-sulfate, DL-homocystine, folinic acid, S-adenosylmethioninamine, lysoPA (P-16:0e/0:0), and coproporphyrinogen III) demonstrated robust stratification of patients eligible for RTX before treatment (AUC 0.85 +/- 0.13). Complementing this, a two-metabolite (N6-methyl-2'-deoxyadenosine and dehydroepiandrosterone) monitoring model was developed to track treatment responsiveness dynamically (AUC 0.83 +/- 0.16). This study elucidates metabolic mechanisms underlying RTX responsiveness and delivers clinically actionable tools to enhance personalized management of pediatric ITP.