Dual-Scale StaphAIR: Predictive Modeling for the Diagnosis of S. aureus Infection via Simultaneous Detection and...

Staphylococcus aureus causes life-threatening bacterial infections. Current diagnostics for bone infection consist of invasive sample collection and lengthy microbial culture; available blood tests are not effective in the orthopedic context. To address this gap, we report an immune biomarker panel for the simultaneous measurement of cytokines and S. aureus-specific antibodies in serum and evaluate predictive machine learning models for the diagnosis of S. aureus infection. The approach relies on a newly developed "dual scale" 25-plex Arrayed Imaging Reflectometry (AIR) immunoassay able to simultaneously quantify 6 cytokines and 19 S. aureus-specific antibodies on a single platform. The dual-scale StaphAIR assay was used to quantify 429 individual serum samples collected from patients hospitalized with culture-confirmed S. aureus infection (infected) and from healthy patients undergoing elective surgery (control). Multivariate logistic regression was used to identify combinations of predictors with the highest diagnostic potential. We found that the most important single predictor was IL-6 (AUC 0.85), and the highest-performing logistic regression combination was IL-6 + IL-10 + IL-17A + TNFα + IsdB (AUC 0.92). Patient demographic data (age, body mass index, race, gender, smoking status) and laboratory data (hemoglobin A1c and albumin) were assessed, and only albumin was found to be a potential confounder. Then, predictive machine learning models were trained and validated (5-fold cross-validation) using the dual-scale StaphAIR data with and without albumin as a predictor and evaluated using Receiver Operating Characteristic and Area Under the Curve (ROC/AUC) analysis. The neural boosted model performed the best on the validation folds with and without albumin level as a variable (AUC 0.975 with albumin, 0.914 without). Predictive models using combinations of immune biomarkers in serum demonstrated excellent diagnostic utility for S. aureus infection, although further external validation with an independent data set is recommended. We anticipate that this approach will also have significant value in diagnosis of other bacterial or viral infections.