Effective triage of dengue patients early in the disease course for in- or out-patient management would be useful for optimal healthcare resource utilization while minimizing poor clinical outcome due to delayed intervention. Yet, early prognosis of severe dengue is hampered by the heterogeneity in clinical presentation and routine hematological and biochemical measurements in dengue patients that collectively correlates poorly with eventual clinical outcome. Herein, untargeted liquid-chromatography mass spectrometry metabolomics of serum from patients with dengue fever (DF) and dengue hemorrhagic fever (DHF) in the febrile phase (<96 h) was used to globally probe the serum metabolome to uncover early prognostic biomarkers of DHF. We identified 20 metabolites that are differentially enriched (p<0.05, fold change >1.5) in the serum, among which are two products of tryptophan metabolism-serotonin and kynurenine. Serotonin, involved in platelet aggregation and activation decreased significantly, whereas kynurenine, an immunomodulator, increased significantly in patients with DHF, consistent with thrombocytopenia and immunopathology in severe dengue. To sensitively and accurately evaluate serotonin levels as prognostic biomarkers, we implemented stable-isotope dilution mass spectrometry and used convalescence samples as their own controls. DHF serotonin was significantly 1.98 fold lower in febrile compared to convalescence phase, and significantly 1.76 fold lower compared to DF in the febrile phase of illness. Thus, serotonin alone provided good prognostic utility (Area Under Curve, AUC of serotonin = 0.8). Additionally, immune mediators associated with DHF may further increase the predictive ability than just serotonin alone. Nine cytokines, including IFN-γ, IL-1β, IL-4, IL-8, G-CSF, MIP-1β, FGF basic, TNFα and RANTES were significantly different between DF and DHF, among which IFN-γ ranked top by multivariate statistics. Combining serotonin and IFN-γ improved the prognosis performance (AUC = 0.92, sensitivity = 77.8%, specificity = 95.8%), suggesting this duplex panel as accurate metrics for the early prognosis of DHF.