COVID-sniffing Dogs One Year Later: New Studies Show Promise
Summary: The canine nose is the most sophisticated tool we have for detecting odors in real-life environments. As soon as the pandemic hit, researchers started looking at whether dogs could be trained to detect if a person is infected with the COVID-19 virus. One year after these initial studies, where are we at with the science and implementation now? This article discusses some recent research that suggests dogs could become a valuable assistance in our continuing fight against the coronavirus.
Since the start of the COVID-19 pandemic, researchers have been working on ways to identify COVID cases, screen people for the virus, and attempt to more effectively reduce transmission. These efforts have been inspired in part by past by research that uses dogs to screen people for medical concerns including diabetes,1 prostate cancer,2 malaria,3 and Clostridium difficile.4
There have been a variety of studies showing that dogs can detect COVID with high accuracy in a lab setting.5,6 There are, however, large caveats with the research available from February 2021, detailed in the February 2021 IAABC Foundation Journal. Namely, that dogs were primarily trained using swabs from hospitalized patients, not presymptomatic or asymptomatic cases. Furthermore, all testing and trialing has been done with the swabs themselves, not with the dogs inspecting humans. Since one of the best uses for the dogs is rapid screening of crowds (like sports stadiums, airports, or concerts), there clearly was a long way to go as of February 2021.
The promise and importance of detection dogs remains to this day for several reasons. Per a paper in Nature,7 COVID-sniffing dogs routinely outperform the “gold standard” rRT-PCR tests in sensitivity (ability to detect disease). Dogs have an 89% sensitivity whereas rRT-PCR tests averaged 73% sensitivity. According to the same Nature paper, dogs perform similarly to rRT-PCR for specificity (ability to correctly eliminate negative cases) with dogs at 99% and rRT-PCR at 95%.
According to Our World in Data, as of November 22, 2021, 42.5% of the world’s population is fully vaccinated against COVID; but only 3% of individuals in low-income countries are fully vaccinated.8 It is not expected that the entire world’s population will be vaccinated until at least 2023. Furthermore, the rRT-PCR test that is the gold standard for detecting COVID-19 is insensitive in the first five days of infection. Lateral flow antigen tests complement rRT-PCR tests well but alone are not a perfect tool for rapid screening.9
Here, I aim to investigate whether further work has been done to take the success of dogs from the lab into the field.
In general, I will not give much detail to the specific training used in each paper. That is largely because many papers do not go in depth on training protocols. When methodology has been described, most studies report using clicker training and operant conditioning to reward the dog for a trained response to the odor in a lineup or scent wheel.
Sniffer dogs as a screening/diagnostic tool for COVID-19: A proof-of-concept study
Eskandari et al., 2021, in BMC Infectious Diseases10
This is the first paper I have found where the dogs weren’t just tested using axillary sweat samples or nasopharyngeal swabs. Six dogs were trained, three to alert to nasopharyngeal swabs from COVID-positive patients, and three to alert to masks or clothing worn by COVID-positive patients. All samples were verified to be positive or negative based on RT-PCR testing.
Like many of the studies in this area, positive samples were taken from ICU patients. This means that we cannot say whether or not these dogs were capable of detecting asymptomatic, mildly symptomatic, or presymptomatic patients.
The second group of dogs is the more interesting group in this study, since several other studies have already shown dogs to be highly capable of detecting COVID-positive cases in swabs.
When the three dogs were sniffing masks and clothing, their sensitivity was found to be as high as 86% and their specificity was 92.9%; identifying 43 out of 50 positive samples and 65 out of 70 negatives.
However, this paper did not give details on the exact lineup procedures. The lineup testing was done in a single-blind fashion rather than the standard double-blind fashion. No explanation is given for this choice. This paper also lacks details in the format of the test. Generally, these studies are done in a lineup fashion with three to seven samples; a variety of blanks and distractors mixed in with between zero and two positive samples. No mention is made of distractors, blanks, or even the number of samples presented to each dog.
While the authors broke down the performance of each individual dog for the swab portion of the study, they did not for the mask and clothing study. We therefore do not know how variable the dogs’ performance was for this setup.
While it’s exciting that this paper showed dogs were capable of detecting COVID-19 in masks and clothing, the lack of details in training methods for the dogs and lack of breakdown in performance leaves a lot of questions with this paper.
Scent dog identification of SARS-CoV-2 infections in different body fluids
Jendrny et al., 2021, in BMC Infectious Diseases11
This paper describes dogs that were trained on samples from 93 individuals who were healthy, suffering from a non-COVID respiratory disease, or suffering from asymptomatic to severe COVID-19 symptoms. This paper is an explicit follow-up to Jendrny et al., 2020,5 but designed to include milder cases of COVID-19. The researchers also tested if the dogs were able to generalize from inactivated samples to non-inactivated samples, and to see if the dogs could generalize to sweat and urine after being trained on saliva.
This paper clearly explains how the researchers used double-blind and controlled trial setups to test the dogs after eight days of training. The researchers reported that the dogs did successfully generalize from inactivated saliva to sweat, urine, and non-inactivated saliva. Dogs were most sensitive (95%) and specific (98%) with urine samples. Dogs were able to identify different COVID-19 disease phenotypes and phases of disease expression (sore throat, cough, cold, headache and aching limbs, fever, loss of smell and taste and/or severe pneumonia)
Furthermore, this paper tested the dogs against distractors of other respiratory diseases, which is especially useful given how broad the symptoms of COVID-19 can be. The dogs were tested on a total of 5,242 randomised sample presentations.
The authors point out that they had a sample prevalence of 18.5%. Before the current Omicron wave, that percentage would be much higher than what the dogs would be exposed to in the field. Careful training to build resilience and endurance for the dogs, or intermittent “gimme” rounds could help avoid frustration or extinction on the part of dogs in the future.
The promise of disease detection dogs in pandemic response: Lessons learned from COVID-19
Otto et al., 2021, in Cambridge University Press12
This paper does an excellent job of looking deeply into the nuts and bolts of creating a successful disease screening dog program around the world. Rather than examining if or how dogs can be trained to sniff out COVID-19, Otto and her co-authors dive into the economics and best practices needed in order to learn from the COVID-19 pandemic and respond more effectively to future pandemics.
This paper outlines 12 factors important to the success of detection dog training programs:
- Samples from various sources. Samples must represent people of a variety of ages, genders, races, lifestyles, and comorbidities. For example, if samples only represent healthy young men versus intubated overweight women with COVID-19, the dogs cannot be adequately trained for all circumstances.
- Confirmation of disease state in samples. It is imperative that the samples used to train the dogs are accurately identified as positive or negative. This is particularly tricky as the R-PCR tests that are gold standard for COVID-19 has a high rate of false negatives in presymptomatic cases.
- Negative samples include similar diseases. One of the most useful implementations of detection dogs, aside from screening large crowds, is to quickly screen people with similar diseases. Therefore, it is imperative that samples include both COVID-19 positive samples and samples that are positive for other diseases such as the common cold or flu, which have similar symptoms to COVID-19 in some cases.
- Positive samples from lifelike positive cases. Rather than training dogs on samples from intubated COVID-19 cases in the ICU, the dogs should be trained using samples from asymptomatic, presymptomatic, or mildly symptomatic cases.
- Adequate number of training samples. Otto and her co-authors suggest at least 100 positive samples for initial training plus additional novel samples for proofing, testing, and maintenance.
- Adequate sample storage and handling protocols. Samples must not be contaminated or contaminate other substances in how they are stored and handled.
- Training prevalence mimics real-life occurrences. While initial training numbers may be 50/50 between negative and positive samples, eventual training and testing must mimic real-life prevalence. For example, if 10% of people are infected, there should be nine negative samples for every positive sample.
- Dogs are introduced to blanks. If positive samples are always available in training, researchers cannot claim to know how dogs will respond when there are no positive samples to find.
- Training reflects operational usage. Initial odor imprinting will likely be done with boxes or wheels, but for a paper to claim that dogs can search people, training must be done with the dogs screening people.
- Alert is appropriate and well-trained. In order to safely use the alert in a public space, the dog should not bark, scratch, or dig as part of the alert. Furthermore, in order to act on an alert by a dog (for example, to remove a student from school for testing or deny entrance to a concert), the alert must be consistent, reproducible, and clearly defined.
- Ongoing training. The dogs should be regularly tested with double-blind testing and novel samples in an operational setting to ensure ongoing performance.
- Standardized training protocols. Training protocols should be defined and followed to ensure consistency and reproducibility.
This paper also details the in-depth understanding of accuracy, specificity, and sensitivity needed to utilize disease detection dogs. They also point out that past studies have found dramatic drop-offs in performance when switching from testing to real-world applications.13
A successful program also relies on an adequate supply of dogs, trainers, and handlers. This is especially difficult if the goal is rapid disease response such as during the emergent COVID-19 pandemic, when handling procedures are unclear, tested samples are difficult to come by, and it is unclear how severe or long-lasting the pandemic may be.
COVID-19 sniffer dog experimental training: Which protocol and which implications for reliable identification?
Angeletti et al., 2021, in Journal of Medical Virology14
In yet another study exploring the efficacy of detection dogs, Angeletti’s team trained three dogs using 20 COVID-positive samples and 15 COVID-negative patients. Like many earlier studies, this study suffers from a very small sample size and limitations in the utility for real-world COVID detection.
All COVID-positive patients presented with pneumonia and therefore the dogs were not trained or tested on mildly symptomatic, asymptomatic, or presymptomatic cases. The COVID-negative samples did not include flu patients, though they did test against patients with non-COVID pneumonia. The researchers did not test the dog’s ability to sniff people, clothing, or masks and did not test the dog’s ability to generalize to samples beyond the initial 35 used to train the dogs.
Like several other studies, this paper is a good initial step but fails to examine many important steps required to actually use detection dogs to screen crowds for COVID-positive individuals.
Highly sensitive scent-detection of COVID-19 patients in vivo by trained dogs
Vesga et al., 2021 in PLoS One15
This experiment aimed to test the dogs on screening live humans rather than swabs or other samples. The researchers compared the limit of RNA strands per milliliter that the dogs could detect in the lab as well as in a live setting. This paper also investigated whether a pit bull and a husky compared in performance to typical working breeds like Belgian Malinois.
The dogs were initially trained on samples from 12 patients admitted to three different hospitals. The dogs were trained using a positive sample rate of roughly 10%, locating the positive sample among saliva samples from healthy individuals. This initial phase of training did not include samples from mildly symptomatic, asymptomatic, or presymptomatic cases, nor did it include distractors with other respiratory illnesses.
For the in vivo portion of training, the dogs were trained comparing 100 hospitalized patients (40 of whom had COVID-19) with 300 health care workers (21 of whom had COVID-19) over the course of 56 days. The dogs were allowed to scent any part of the subject’s body, but generally initially scented hands.
After the team was satisfied with the sensitivity and specificity of the dogs in vivo, the researchers conducted an efficacy trial. The description of how this was done is unclear – potentially due to translation issues – but it appears that the dog teams were taken into the room of COVID patients in the ICU, and also screened healthcare workers in groups. This setup appears problematic (there are big context clues to differentiate between a lineup of workers and a single patient in a hospital room that could tip a dog off to higher prevalence), but later validation in testing suggests that the dogs successfully generalized to a more real-world setup.
The researchers had a final phase 4 study that they did not warn the trainers of: bringing the dogs to the Medellin subway station to examine passengers. Three dogs (one Malinois, one husky cross, and one pit bull) examined 550 passengers. The researchers asked for volunteers in the station to submit to sniffing by the dog and provide a saliva sample.
The researchers tracked efficacy over time to see how the dogs adjusted to the new environment. The dogs took roughly three hours of work to adjust to the new environment and perform with over 90% accuracy.
Out of the 550 people screened in two days, the dogs detected 17 COVID-positive subjects, 15 of whom were asymptomatic or presymptomatic.
The researchers also found that some cases scored as “false positives” (false alerts) were unrewarded, yet those nurses went on to develop COVID-19 symptoms within four to seven days. It’s possible that the dogs were therefore sensitive enough to detect COVID-19 in cases missed by rRT-PCR. However, the dogs also performed a variety of “false positives” in alerting to cell phones or other belongings from nurses who had recently worked with COVID-19 patients.
Further questions for the implementation of COVID-sniffing dogs
Aside from more papers examining the efficacy of COVID-sniffing dogs in vivo (such as in airports, subways, and stadium events), it would be exciting to see sociological papers examining the reaction of the public to such dogs. The global health impact of COVID-19 is massive and will be ongoing for years. However, the areas that may benefit most from COVID-sniffing dogs (poorer countries) may also have a variety of cultural responses to dogs in public places. Vesga found that the dogs occasionally refused to search a given person, and some people refused to be searched by dogs.
Further research is also needed to validate that the dogs can differentiate between various coronaviruses and COVID-19 in vivo. It is also unknown how dogs will respond to highly mutated variants like Omicron.
In countries with high rates of vaccine hesitancy, fatigue with the pandemic, or outright conspiracy theories around COVID-19, safety for the dogs from the public may also become a consideration.
Many papers point out that the risk to dogs of contracting COVID-19 from patients is exceptionally low, which is unquestionably good news.
- Rooney, N. J., Morant, S., & Guest, C. (2013). Investigation into the value of trained glycaemia alert dogs to clients with Type I diabetes. PLoS ONE, 8:8.
- Cornu, J. et al (2011). Olfactory detection of prostate cancer by dogs sniffing urine: A step forward in early diagnosis. European Urology, 59:2, 197-201.
- Guest, C. et al (2019). Trained dogs identify people with malaria parasites by their odour. The Lancet Infectious Diseases, 19:6, 578-580.
- Bomers, M. K. et al (2014). A detection dog to identify patients with Clostridium difficile infection during a hospital outbreak. Journal of Infection, 69:5, 456-461.
- Jendrny, P. et al (2020). Scent dog identification of samples from COVID-19 patients – a pilot study. BMC Infectious Diseases, 20:1.
- Grandjean, D. et al (2020). Can the detection dog alert on COVID-19 positive persons by sniffing axillary sweat samples? A proof-of-concept study. PLoS ONE, 15:12.
- Hag-Ali, M. et al (2021). The detection dogs test is more sensitive than real-time PCR in screening for SARS-COV-2. Communications Biology, 4:1.
- University of Oxford. (2021). Covid-19 data explorer. Our World in Data.
- Centers for Disease Control and Prevention. Interim guidelines for rapid antigen testing for SARS-CoV-2.
- Eskandari, E. et al (2021). Sniffer Dogs as a screening/diagnostic tool for covid-19: A proof of concept study. BMC Infectious Diseases, 21:1.
- Jendrny, P. et al (2021). Scent dog identification of SARS-COV-2 infections in different body fluids. BMC Infectious Diseases, 21:1.
- Otto, C. M. et al (2021). The promise of disease detection dogs in pandemic response: Lessons learned from covid-19. Disaster Medicine and Public Health Preparedness, 1–6.
- 13. Guest, C., Harris, R., Anjum, I., et al. (2020). A lesson in standardization – subtle aspects of the processing of samples can greatly affect dogs’ Frontiers in Veterinary Science; 7:525.
- Angeletti, S. et al(2021). Covid‐19 sniffer dog experimental training: Which protocol and which implications for reliable identification? Journal of Medical Virology, 93:10, 5924–5930.
- Vesga, O. et al (2021). Highly sensitive scent-detection of COVID-19 patients in vivo by trained dogs. PLoS ONE, 16(9).
Kayla Fratt is the founder of K9 Conservationists, a nonprofit that helps biologists gather data with the help of conservation detection dogs. She also runs Journey Dog Training, an online dog behavior help center. Kayla shares her life with two working border collies.