Collection 8.

Rapid point of care tests for respiratory infections are associated with high rates of false negative results which can drive empiric, and potentially inappropriate, antibiotic use. Because infectious pathogens alter VOC composition, unique VOC signatures in biospecimens hold the potential to discriminate bacterial and viral infections from uninfected controls. One approach for rapid identification of respiratory pathogens is the electronic nose (e-nose), a sensor device that uses artificial intelligence to recognize disease-specific patterns in VOC profiles of gaseous mixtures. In this preclinical proof of concept study, we tested the validity of an e-nose to discriminate PCR-confirmed cases of infection with three viral pathogens (SARS-CoV-2, RSV, influenza A) from uninfected controls using nasopharyngeal test swab media. Using exploratory factor analysis, the e-nose discriminated both influenza A and SAR-CoV-2 from uninfected controls. To assess sensitivity and specificity, we applied factor analysis-based threshold values and obtained high levels of sensitivity (96.30%) and specificity (90.62%) for influenza A and more modest levels for SARS-CoV-2 (sensitivity=75%, specificity=68.57%). We did not apply threshold values to RSV samples because the e-nose sensors showed low discriminatory power for that pathogen. Our findings support proof of concept of the validity of the e-nose to discriminate common viral respiratory pathogens. Our use of binary thresholds for influenza A, which are easily adapted to point-of-care settings, yielded superior sensitivity results and comparable specificity results when compared to rapid tests. We recommend that future studies apply our analytic approach to samples of human breath to determine if these findings can be replicated or improved.