Machine learning reveals an interleukin-33-associated immuno-metabolic axis defining recovery endotypes in severe combat trauma.

Severe combat trauma is associated with long-lasting systemic changes, yet standard anatomical injury scales may not adequately capture biological differences in patient recovery. To explore this variability, we analyzed the systemic molecular profiles of 53 combat trauma survivors during the rehabilitation phase and compared them to those of 46 uninjured active-duty military controls. Using a multidimensional biomarker panel (IL-33, ST2, TGF-β1, CTGF, galectin-3, heparan sulfate, NT-proBNP, and annexin A5) alongside routine clinical parameters, we applied unsupervised clustering and Random Forest machine learning to identify distinct patient endotypes. Our results showed that patients grouped into distinct molecular phenotypes, independent of their initial structural injury patterns. While localized structural injuries and associated infections (e.g., limb amputations) significantly influenced pro-fibrotic and pro-coagulant pathways, the overarching systemic features defining the machine-learning endotypes closely linked to an immuno-metabolic axis. This axis is characterized by elevated IL-33, endothelial glycocalyx shedding, and markers of metabolic, hepatic, and hematological strain (hemoglobin, white blood cells, AST, ALT, and total protein) rather than classical fibrotic markers. Notably, while circulating IL-33 levels varied significantly among patients, its soluble receptor, ST2, remained stable. These findings suggest that high-risk trauma patients may experience ongoing, ligand-driven systemic exhaustion. Integrating molecular endotyping into military medicine could improve clinical monitoring and guide targeted rehabilitation strategies.