Can Trained Dogs Sniff Out COVID-Positive Travelers?

Four trained dogs detected people who were positive for SARS-CoV-2 with similar accuracy to PCR tests, according to a randomized trial and real-life study at a busy international airport earlier in the pandemic.

In the validation trial, the dogs had 92% accuracy, with 92% sensitivity (95% CI 89-94) and 91% specificity (95% CI 89-93) compared to an RT-PCR test, reported Anu Kantele, MD, PhD, of the University of Helsinki in Finland, and colleagues.

In a real-life setting, where the dogs sniffed skin swabs from airline passengers, the animals correctly identified 98.7% of swabs negative on PCR, the authors wrote in BMJ Global Health.

The authors noted that while scent detection dogs have been trained to detect samples from hospitalized COVID patients, this approach is "particularly appealing" for screening COVID-infected individuals via mass transit.

Kantele's group trained the dogs, validated the dogs' accuracy for detecting SARS-CoV-2, and then set up a real-life prospective trial in the Helsinki airport to test the dogs' ability to sniff out the virus as accurately as the nasopharyngeal swab.

The dogs -- Silja, Rele, Kosti, and E.T. -- ranged in age from 4 to 8 years, two males and two females. Three dogs were Labrador retrievers, two had worked with "dangerous goods," one with narcotics, and one with canine cancer. Two dogs would sit when discovering a positive sample, while one would paw and one would "nose freeze + one paw."

The dogs were trained to detect samples through "cans" -- picking out the SARS-CoV-2 positive sample from the negative sample or from confounding samples, such as other viruses, cancer, and diabetes. Once they achieved more than 80% success in detecting SARS-CoV-2, they moved on to the validation study. There were 420 skin swab samples, 306 negative and 114 positive. Each dog was exposed to 140 "scent tracks."

Kantele's team noted only "minor variation between the dogs," although they noted that performance was "less accurate" for the Alpha variant (36% for Alpha variant vs 89% for wild-type virus).

The dogs then moved on to the real-life cohort of 303 samples from airport passengers. They identified 296 of 303 samples correctly and 296 of 300 RT-PCR negative samples. There was one positive case total, which the dogs did not identify. However, due to the low prevalence, Kantele's group also provided 155 novel RT-PCR "spike" samples, and the dogs identified 98.7% of them as positive.

A post hoc analysis indicated that if these 155 samples were added into the 303 samples from the real-life study, the dogs would have had a sensitivity of 97% (95% CI 92-99) and a specificity of 99% (95% CI 96-100).

They added that their study collected samples from a wide variety of patients, both asymptomatic and symptomatic individuals, children and seniors, as well as those with non-communicable diseases. The authors also randomly included tracks with no positive samples, which "mimics better the real-life situation in low-prevalence settings," they wrote.

While they noted that the only variable tied to failure to identify positive COVID samples was the Alpha variant, the dogs had only been trained to detect the wild-type virus, and were still able to pick up the Alpha variant in some cases.

"Our preliminary observations suggest that dogs primed with one virus type can in a few hours be retrained to detect its variants," they wrote.

Limitations to the data include that scent dogs previously trained to detect other substances may mark them because they recognize the scent, while the handler would record them as positive for COVID. Age of the samples also varied and there was a low number of positive samples in the validation test.

Comment