Sniffer dogs performance is stable over time in detecting COVID-19 positive samples and agrees with the rapid antigen test in the field.

Rapid antigen diagnostic (RAD) tests have been developed for the identification of the SARS-CoV-2 infection. However, they require nasopharyngeal or nasal swab, which is invasive, uncomfortable, and aerosolising. The use of saliva test was also proposed but has not yet been validated. Trained dogs may efficiently smell the presence of SARS-CoV-2 in biological samples of infected people, but further validation is needed both in laboratory and in field. The present study aimed to (1) assess and validate the stability over a specific time period of COVID-19 detection in humans' armpit sweat by trained dogs thanks to a double-blind laboratory test-retest design, and (2) assess this ability when sniffing people directly. Dogs were not trained to discriminate against other infections. For all dogs (n. 3), the laboratory test on 360 samples yielded 93% sensitivity and 99% specificity, an 88% agreement with the Rt-PCR, and a moderate to strong test-retest correlation. When sniffing people directly (n. 97), dogs' (n. 5) overall sensitivity (89%) and specificity (95%) were significantly above chance level. An almost perfect agreement with RAD results was found (kappa 0.83, SE 0.05, p = 0.001). Therefore, sniffer dogs met appropriate criteria (e.g., repeatability) and WHO's target product profiles for COVID-19 diagnostics and produced very promising results in laboratory and field settings, respectively. These findings support the idea that biodetection dogs could help reduce the spread of the virus in high-risk environments, including airports, schools, and public transport.