Systematic identification of site-associated blood-based signature patterns across 29 infection types: a large-scale analysis of 455,530 patients toward precision infection medicine

Accurate infection diagnosis remains challenging due to overlapping clinical presentations and lack of site-associated biomarker profiling. This study represents the first comprehensive investigation to systematically characterize site-associated blood-based signature patterns across multiple anatomical systems and infection types. This retrospective study analyzed 455,530 infection patients and 55,010 healthy controls. Infections were classified into 29 site-associated types across five anatomical systems. Propensity score matching and multiple imputation were applied. Beyond conventional differential analysis, we constructed multivariable logistic regression and random forest models using 12 hematological parameters plus age and sex to evaluate the combined discriminatory performance for infection versus control, and explored two-site classification (respiratory vs. digestive). Respiratory infections were most prevalent, followed by urogenital. Site-associated hematologic patterns emerged: digestive infections showed elevated neutrophil (NEUT) counts; respiratory infections exhibited increased hematocrit (HCT) levels; fallopian infections demonstrated decreased platelet-to-lymphocyte ratio and elevated basophil counts. In predictive modeling, random forest achieved system-level AUROCs of 0.928 (dermatological), 0.953 (digestive), 0.894 (head-neck-orofacial), 0.965 (respiratory), and 0.901 (urogenital), consistently outperforming single biomarkers and logistic regression. In the exploratory two-site classification, random forest yielded an AUROC of 0.795, indicating moderate discriminative ability. Principal component analysis revealed separation between infection patients and controls. This largest-scale study provides a comprehensive characterization of site-associated blood-based signatures across anatomically defined infection types. The identified hematologic patterns, while reflecting non-specific systemic inflammatory responses, offer auxiliary information for infection site classification. Combined models, particularly random forest, improved discrimination but remain exploratory; these findings support the potential of routine blood tests as probabilistic adjuncts, not standalone diagnostics, in precision infection medicine.