Risk Stratification of Dengue Cases Requiring Hospitalization

Abstract

Dengue pathogenesis involves immune-driven inflammation that contributes to severe disease progression. This study assessed a machine learning model to identify a minimal, yet highly predictive biomarker set, aiming to support clinical decision-making and patient triage. A total of 48 inflammatory mediators were quantified from plasma samples collected at admission from confirmed dengue patients, classified as either dengue without warning signs (DF) or dengue with warning signs/severe dengue (DWS/SD). A random forest approach was applied to identify the most predictive biomarkers associated with disease severity requiring hospitalization, based on admission-time variables. Among the 48 immune mediators, 43 were differentially expressed in dengue patients versus healthy controls, and 26 showed significant differences between DF and DWS/SD cases. Lymphocyte counts negatively correlated with IL-1RA, while liver enzymes showed positive correlations with HGF and SCGF-beta; platelet counts also negatively correlated with these markers. Key severity-associated markers included HGF, TNF-beta, MIP-1-beta, and SCGF-beta. A model incorporating these markers and fever duration achieved nearly 80% accuracy in distinguishing DWS/SD from DF cases, independent of clinical examination. The findings suggest that targeted cytokine profiling may guide early hospitalization decisions and ease healthcare burdens in dengue-endemic regions.