Assessment of accuracy of detection dog signaling behavior for the diagnosis of SARS-CoV-2 infection: A Canadian study

BackgroundDogs trained to metabolomics detection can identify pathological changes through their refined smelling sense. During the COVID-19 pandemic, studies worldwide evaluated Detection Dog Signaling Behavior (DDSB) for SARS-CoV-2. However, most statistical approaches failed to account for key sources of bias, potentially distorting performance estimates. This study aimed to estimate DDSB accuracy for SARS-CoV-2 infection in a Canadian population while assessing the impact of selected sources of bias on performance estimates. MethodsParticipants attending the COVID-19 assessment clinic at St. Josephs Health Centre, Toronto, were recruited between October and December 2021. Each provided a nasopharyngeal swab for reverse transcription-polymerase chain reaction (RT-PCR) testing and three sweat samples for canine detection. Three dogs were trained to detect SARS-CoV-2 in sweat samples. Validation sessions were video recorded and independently reviewed by two blinded observers. DDSB diagnostic accuracy was estimated against RT-PCR, evaluating the impact of ignoring its imperfect accuracy and repeated sniffing of the same samples during validation. ResultsAmong 2,358 participants (mean age: 34.7 years; 55.7% female), 437 contributed to training. Validation tests included 146 unique participants (25 RT-PCR positive, 121 negative). Assuming RT-PCR was imperfect, DDSB posterior median sensitivity ranged from 67% (95% credible interval [CrI]: 29%-97%) to 78% (95% CrI: 41%-99%), and specificity from 67% (95% CrI: 53%-79%) to 77% (95% CrI: 65%-87%) across the three dogs. Assuming RT-PCR was perfect, sensitivity decreased by 7.9% to 9%, while specificity remained unchanged. Including repeated positive samples without adjustment did not affect specificity estimates but overestimated sensitivity by 7.9% to 13.4% (imperfect RT-PCR) and 11.4% to 18.3% (perfect RT-PCR). ConclusionDDSB shows potential as a non-invasive screening tool for SARS-CoV-2 infection. Our results highlight the challenges of designing such studies and the need for standardized training and validation procedures to ensure the reliability and validity of DDSB.