I’ve written before about COVID-19 scent-detection dogs. I get lots of questions about them, and there are now several groups working in this area. There’s been a mix of information to date, ranging from encouraging to some pretty bad preliminary studies released on pre-print websites and other places. A dog’s nose is a wonderful thing (except when my dog sticks his in places I don’t want it to go), and dogs have been shown to be able to detect a wide range of different scents with great sensitivity.

The first question is: Will dogs be able to detect people with COVID-19?

If the answer is yes, then the bigger question is, will it be a practical way to detect people with COVID-19?

We may get more answers now that dogs are being used in a Finnish airport to sniff out COVID-19.  Ten dogs have been trained to detect people with COVID-19 based on smelling wipes collected from individuals. News reports include claims of close to 100% accuracy… I’d love to see good data on that, as I suspect it’s not 100% effective in the field. However, even if the dogs are moderately effective, they could be a useful tool when combined with other measures (e.g. rapid confirmatory testing of people that dogs flag as potentially infected).

My big questions at this point is, how effective is it really?

  • We need to consider both sensitivity (how good dogs are at detecting infected people) and specificity (how good they are at only detecting infected people).
  • For a screening test, we want a test that is highly sensitive, meaning it detects most infected people, even if it has some false positives (i.e. people who are mistakenly identified as positive but aren’t actually infected). That works if the false positive rate isn’t massive and if there is a convenient way to follow up to confirm who’s really positive. If we have a quick follow up test of another kind, the initial false positives are a bit of a hassle but not a big deal and easy enough to weed out, so we could tolerate some loss of specificity.
  • False negatives on the other hand (i.e. people who are infected but go undetected by the test) are a bigger concern.
  • So, knowing the sensitivity and specificity of these COVID019 detection dogs in a field situation (where there are lots of people of different types, with different stages of infection and with different smells) is key. Hopefully that’s being studied well.

Another question I have is, what’s the management plan for dogs that stick their noses in wipes from people with COVID-19?

  • Dogs have limited susceptibility to SARS-CoV-2, but limited and zero aren’t the same.
  • Will the dogs be screened in case they get infected in the process?
  • And (an oddball question perhaps) if a dog gets infected, does it lose the ability to detect infection in people? would the dog then smell the scent associated with the virus all the time?

There will be more to come, I assume.

I’ve had countless questions about the potential for scent detection dogs to be useful for COVID-19 surveillance. It’s an interesting idea, but it’s dependent on COVID-19-infected people producing some volatile compound detectable by the dogs than uninfected people do not.  (The virus itself is not likely to have a detectable odour.)

A recent pre-print paper on bioRxiv provides a rambling description of a pilot study about the potential for dogs to sniff out COVID-19 in people based on their armpit sweat.  This one can probably be filed under the “not likely to ever be published in a journal, but an interesting story” category.  (It may also be another paper in the “let’s get something online first, who cares about the depth and editing” category. Sixty-one authors is a bit extreme too, but I digress.) As I’ve said before, pre-prints can be useful, but there has to be at least a modicum of effort…

Anyway, they collected armpit sweat samples from people with COVID-19 and patients without signs of COVID.

They used 18 dogs that had been trained for explosive detection, search and rescue and colon cancer detection, but their table also lists an arson detection dog.  Further, they say that “We did not decide to work with drug detection dogs as there is always a possibility that COVID-19 positive or negative people use prohibited substances that would let catabolites be excreted by the axillary sweat.” And yet, their table lists a drug detection dog.

After training, the dogs were tested to see if they could differentiate sweat samples from COVID-19 patients in scent boxes, from the samples from the non-infected patients.

  • Three of the 18 dogs flunked out of COVID-19 detection school as they were “unable to adapt to an olfactive search on a line of sample”.
  • Eight others were removed because they were “late in their testing period due to this necessary basic “retraining”. I’m not really sure what that means. I guess they weren’t completely kicked out of school but had to repeat the year, and no one wanted to wait to rush out the pre-print.

The authors say that left 8 dogs whose results they used for the analysis; however, my math says that 18-3-8=7, not 8.

Numbers aside, results were interesting, as the remaining dogs seemed to have fairly high detection rates (84-100%).

It’s hard to say what this means and whether it’s relevant. I guess it means that there’s potential for some dogs, but you have to find the right dogs, and train them.  The authors make a fair point that they were looking at effective dogs, not whether it worked in the whole dog population.

Is this of any use?

It’s hard to say. They tested samples from people with overt COVID-19 disease, and that’s likely not very relevant. If someone has signs of COVID-19, we want to treat them like they have COVID-19 at least until they are definitely tested, regardless of what the dog thinks. Where this would be more useful is for detecting asymptomatically infected people, in which case the dog might provide an early warning of infection. The downside is that it requires a dog and a person to be in close proximity to the patient being screened. Dogs are low (but not no) risk for picking up SARS-CoV-2, but there’s also reasonable concern about handlers. The practical nature of this is also questionable. The study mentions that armpit sweat is “easy and safe to collect”, but if the “test” relies on people taking samples of their own armpit sweat, it’s no longer a quick, easy detection method. It’s more appealing if the dog can, for example, be used to screen people coming into a building or workplace.

The authors have a few excuses for some of the false negative results, which seem to be a bit of a stretch, or indicate potential issues applying this to a real world situation (e.g. distraction by a “too zealous television team”). Some other discussion points are a bit hard to follow. They attributed false positives to two male dogs and a young woman that “had been sampled during her fertile period”… they’d better sort that problem out since that’s a reasonably common subset of the population.  The paper’s discussion devolves considerably from there: “In Shakespeare’s day, a woman in her fertile period used to hold a peeled apple under her arm until the fruit became saturated with her armpit scent ; then she presented this “love apple” to her lover to inhale in order to provoke his sexual excitation. Pheromones are defined as substances produced by on animal which conveys information to other individuals by olfactory means. And in such a situation androsterone and molecules like benzoate derivates, excreted in axillary sweat, enhance sexual attraction by men toward women near the end of follicular phase of the menstrual cycle when fertility is at highest.”

Overall, it’s an interesting pilot study that shows more study might be warranted. In particular, it’s important to figure out whether dogs can be good detectors of COVID-19 infected people in situations like busy, high risk places in the community.  Criticisms aside, it would be really cool to have a dog hanging out at the entrance of busy places, picking out infected people, to help curb further spread of SARS-CoV-2 in the community.

A colleague asked me about scent detection dogs the other day. My response was that I hadn’t heard much after all the initial buzz, which might suggest things weren’t going well. However, as opposed to the horrible pre-print about COVID-19-sniffing dogs I wrote about previously, a paper in BMC Infectious Diseases (Jendry et al. 2020) provides some more robust and interesting information. It’s a pilot study, so it’s small, preliminary and underpowered, but it shows potential. Whether that’s “potential for dogs to be able to detect SARS-CoV-2 under certain circumstances” or “potential for dogs to be an effective detection tool” isn’t clear, but that’s the big question.

Here’s a breakdown of the study and some commentary:

The researchers collected saliva samples and respiratory secretions from hospitalized COVID-19 patients, and healthy people who were PCR-negative for the SARS-CoV-2 virus.

  • This may not be ideal, depending on the goal. My vision is using these dogs in the community to rapidly detect infectious people in high risk situations (e.g entrance to transit stations, public buildings, schools). In that case, people who are hospitalized with severe COVID-19 are likely not the best test population. A dog isn’t going to replace a PCR machine in the hospital.  It’s simply not practical in most cases to collect a sample from a patient, take it to a dog as a quick screening test, and then submit the sample for definitive testing.  We want dogs that can detect a mild case in the community, long before the patient needs to be hospitalized.
  • They didn’t test samples for other human coronaviruses, like those that cause the common cold. It’s a potential limitation, but I don’t think it’s a big deal in this case.
  • They also don’t explain where they got their negative samples. A clear description of the study populations is critical and it’s somewhat lacking here.  We want to be sure the dogs were detecting SARS-CoV-2 and not something else unique to the positive sample population, like a smell associated with being from a hospital.

Because of the potential susceptibility of dogs to the SARS-CoV-2 virus, samples were inactivated prior to exposing them to the dog.

  • That’s a reasonable step, but raises more issues of practicality and how the dogs could ultimately be used (e.g. can the dogs only be used to screen specimens collected from high risk patients, or can they be used to detect infection in someone walking by).

Eight dogs were trained using standard methods. They had a 2-week habituation process for the training system, then had 5 days of training until their rate of detection was greater than what would be expected by chance alone. They then started the study

  • The sample size was small, but reasonable for a proof-of-principle study.

The ability of dogs to detect positive samples increased over time. There was some variation between dogs, but all of them were pretty good. The overall sensitivity (percentage of positive samples that the dogs correctly identified as positive) was 83%, ranging from 70-95%. The specificity (percentage of negative samples that the dogs correctly identified as negative) was 96%, ranging from 92-99%.

  • For a screening test, we’d actually want the reverse, that is to say higher sensitivity at the expense of specificity. That would mean the dogs would catch most of the positives. Lower specificity is okay initially if the screening test (i.e. the dog sniffing) can be followed up with a more specific test, and if the implications of an initial false positive aren’t high. If a dog calling a person positive results in that person being sent home to self-isolate for 14 days, then a high false positive rate is a problem. If it just results in the person being pulled aside to have a swab collected for a lab test, that’s not as big of a deal (perhaps a bit of a hassle but maybe not a deal breaker).
  • A low sensitivity and high specificity means you run into fewer hassles with false positives, but the test will miss more positive people. The fact that 17% of prime samples from people hospitalized with active COVID-19 were called negative is a concern in terms of the dogs being able to detect the virus in less severely affected people and from less voluminous and close samples (e.g. detection directly in someone walking by).

I’d file this in the “interesting but preliminary” folder. Anything that can help identify infectious people is useful. If dogs can do it, that’s great, but they also have to be able to do it from a distance, because a handler and a dog getting very close to large numbers of people might cause more problems than they fix.

In my perfect world:

  • A SARS-CoV-2-sniffing dog would be parked at the entrance of schools, office buildings, transit stations, etc.
  • The dog would be able to detect infected people from a short distance away (i.e. without direct contact).
  • The dog would signal its handler when it detected a positive person.
  • That person would then (discretely) be pulled aside for testing, which would (in my perfect world) be done quickly, right there (there is lots of work being done to develop a more rapid test like this that can be done on the spot, but we don’t have one yet).
  • If positive, the person would be told right away and sent home to self-isolate. If negative, the person would be good to go (though maybe wondering why they smell like a coronavirus).

As I’ve said, it’s an interesting and useful preliminary study that shows potential. The key is to follow up preliminary studies with more detailed, rigorous work, which unfortunately often doesn’t get done. Nonetheless, I suspect media headline writers will jump on this and over-interpret the results. It’s also another example of the remarkable things a dog’s nose can do, but the potential practical applications (if any) are still very much up in the air.  I’ll be a bit surprised if this ever becomes a common/useful tool, but I’d love to be wrong about that.