Czech trainers teach dogs to sniff out Covid: 95-percent success rate | Hacker News Reader

Czech trainers teach dogs to sniff out Covid: 95-percent success rate

france24.com

165 pointsrespinal73 comments

Where is the study? I couldn't find the link. "95-percent success rate" is meaningless without context. Specificity/sensitivity, prevalence and the gold standard used all matter. You, too, can detect coronavirus with 100% "success" (and much higher than 100% that depending how you calculate it).

bool carriesCoronavirus(human *subject) {return true;}

There doesn't seem to be a study. There seem to be no scientists or statically-trained people involved at any point. The article is the same everywhere I look (provided by AFP, French news agency) and only mentions "dog trainers working in their own time" who "rely on scant financial means".

On the other hand, if you tested everyone using this strategy and quarantined those who tested positive, you'd eliminate Covid in two months.

If adding a magic doggie in to deliver the news would help compliance, all the better!

> [by treating everybody as infected] you'd eliminate Covid in two months

Not to mention a ton of other illnesses. I do wonder if “the world’s governments pay everybody to stock up on food and stay home for two months” would actually be a huge win for humanity, and maybe even in the short-term it would work out cheaper than two years of 75%-lockdown :P

> the world's governments pay everybody to stock up on food and stay home for two months

What's that going to do other than to cause the price of "food" to go up and cause a bunch of people to die in the worst Black Friday reenactments?

Note that while it is possible to calculate a weighted price index of food, people do not eat the weighted average of all food sold.

It is fantasies of control by people that got us to the point where every day every where governments are burning away their countries' wealth and their citizens' futures.

Many of us in countries with low Covid infection rates have willingly isolated ourselves over the scientists and doctors told us that this was an effective measure. We haven’t seen it as “government control”, and instead as measures to ensure that the virus doesn’t spread and infect more people than the medical staff can handle.

> We haven’t seen it as “government control”,

The quotation marks are misleading because I never referred to government control. Instead, I am referring to the "fantasy of control."

However, now that you mentioned it, in these countries with low observed infection rates, are people free to gather in groups? Are they allowed to engage in usual activities? Are they allowed to travel freely?

If the answers to those and similar questions are not all no, then, clearly not everyone is willingly isolating themselves, and people who entertain fantasies of control or directing the power of government to limit their freedoms.

They have done so without ever paying attention to the costs of such limitations on the usual functioning of human beings.

I apologise for adding the quotation marks, and I’d added them since there’s no one else controlling.

As for the freedom of gathering - we already know that with just a few weeks of not being in contact we can protect ourselves, our families and each other. There’s no martial law or curfew imposed on us - just a lockdown or a Circuit Breaker that we’ve seen as inevitable.

I am always amazed to hear this argument about government control, you do realise that having a civilised society relies on 'limitation on the usual functioning of human beings' such as beating up the guy you don't like, urinating wherever you please, etc. Presumanbly you are happy to limit those natural human instincts.

But somehow spreading deadly disease is you essential liberty

> if you tested everyone using this strategy and quarantined those who tested positive, you'd eliminate Covid in two months.

You have too much confidence in that statement.

If false positive, false negative, and prevalence rates are all 5%, and let's say you subject 350,000,000,000 people to the dog test (ignoring the time it takes to do that and logistical difficulties with having enough dogs and handlers etc).

9.5% of society tests positive resulting 33,250,000 total positive results. Half of this number are true positives and half are false positives. That means, quarantining 16,625,000 uninfected people along with 16,625,000 infected people which results in the loss to society of their output and might also now get them infected depending on how they are quarantined.

Out of the 17,500,000 people who are infected, there will be 875,000 false negatives who will not be quarantined who will continue to spread the infection. In fact, 5% false negative rate sounds kind of optimistic. If false negative rate is 25%, 4,375,000 infected people will not be detected using this method.

You can play with the numbers using this calculator[1].

Second, let's say you put a dog at every company and every day everyone needs to be sniffed on the way in. If the false positive rate is 5%, the probability that you will not be quarantined due to a false positive during the year is 0.95*250 approx = 0.0003%.

[1]: https://www.covid2020.icu/false-positive-false-negative-simu...

It's also possible to add real tests and not taking dog sniffing as a source of truth, but more as an indicator.

"Real" tests are also subject to false positives and false negatives.

What you are suggesting is to ignore the possibility that dog sniffing may result in false negatives, and administer a "real" test to anyone who's identified by a dog to be positive. Keep in mind that under reasonable assumptions, half of those people are false positives. A portion of them will falsely test positive again. However, some of the true dog test positives (anywhere between 2% - 25% of them) will falsely test negative. Therefore, this strategy will result in more infected people not being quarantined.

But way more people with covid will be quarantined than without those tests.

My scenario was to give everyone the "always return positive" test, as suggested in the original comment, though it was ambiguous as stated. So there is no false negative rate, since there are no negatives. The magic doggy is just to encourage compliance.

You do accurately describe the issue with relying solely on tests with false negatives to control Covid.

> give everyone the "always return positive" test,

I had assumed your comment was sarcastic.

Note that if you believe prevalence is 95%, there is no reason to spend any resources on dogs, tests, or even vaccines.

Making all that effort only makes sense in a world where prevalence is assumed to be low and the effort is to keep it low.

So, let's say prevalence is 5%. Then, you method has a false positive rate of 95%. Are you OK with the implications of that?

Here in Finland authorities use them at least in the airport and it’s claimed that “testing has shown an accuracy level of nearly 100% even 5 days before actual symptoms appear” [1].

[1] https://unric.org/en/finland-first-in-europe-to-use-dogs-to-...

As an anecdote, our complete family was infected early December. I lost my smell and taste capacities for about 4 days. My wife and oldest son (13) still have very limited smell and taste.

But now, I still cannot smell my body odour when I sweat. Normally, it is really a strong odour. But not any more. As my wife and oldest son cannot smell, I cannot ask them and going next door and ask if I smell or not after doing a workout is not something I want to do right now :-D

I would not be surprised if dogs could smell a disruption in the "normal human body odour pattern".

This is just a unique sample point without statistical value but if you have the same experience, do not hesitate to contact me.

> I would not be surprised if dogs could smell a disruption in the "normal human body odour pattern".

But isn't COVID induced Ansomia makes us loose the sense of smell and not that we stop generating body odour?

I think the parent comment meant that the infected individual might not be addressing the smell because they could not perceive it, and it would therefore be greater than typical for them.

Yes, the three of us lost our sense of smell. Now, I am back and can smell again, but I cannot smell my body odour after a workout even so normally it smells really. This is at the moment the only thing I cannot smell.

The questions are: is it because the infection had for effect a change in my body odour? Or is it because I recovered all my sense at the exception of the capacity to smell my own body odour?

Nobody in the family can answer (too small or still suffering from anosmia) and I am not yet ready to ring the bell next door after a hard workout and ask the poor soul the smell!

Interesting, I am Czech and I only hear about it here on HN. But it is plausible, Czechs reeeeally like dogs and love to train them.

I wonder if an artificial nose could be constructed doing the same thing. Smell is a potential diagnostic tool.

For example, it seems that certain people can smell Parkinson's disease and that the biomarkers have actually been identified:

https://www.acs.org/content/acs/en/pressroom/presspacs/2019/...

We are not very good, as a civilization, in artificial smelling. Being humans, we develop machines that try to 'see' and 'hear'. A canine civilization would probably develop a smell-distinguishing AI first.

I'm also Czech, and I don't trust any of this. Methodology is suspect, and barely described. And as someone already pointed in the comments, this needs to be double blind, to avoid suggestions from the dog trainer. I'd also like to see control samples from people with other respiratory illnesses, etc.

I definitely concur that strict verification is necessary.

That is one of the reason why I would prefer an electronic nose. For all their problems, machines do not want to please their handlers.

I remember there being a project during the Iraq War that sought to duplicate the effectiveness of bomb sniffing dogs with some kind of chemical device.

Long story short: millions spent and it was concluded that dogs are still both the best and most cost effective method for sniffing explosives.

Too bad the dogs sometimes get injured and are put to sleep. I've read about rats trained to sniff landmines. Probably their training is more expensive?

It would be nice if reporters were required by law to define exactly the numbers behind statements like "95% success rate". Under penalty of being sent to live on an iceberg for violating it.

Why hold the reporters to that standard? They are just regurgitating a press briefing that came across their desk or outlining an interview with someone from the project.

I don’t where everyone gets this idealistic reporter archetype. Unless you’re a super famous investigative journalist with the clout to get funding to fuck about for months/years for a story you’re basically a glorified copywriter.

We can read press briefings ourselves. The only reason for reporters to exist is to dig at least a little bit deeper for truth. We've all heard about the economics of journalism, we just don't think it's acceptable.

Sorry, but I don't think you're paying enough to deserve someone to spend the time digging for deeper truth. That's the sad reality of modern journalism in a capitalist country.

I think the bigger problem is that it's too easy to copy after somebody has done the digging work, so journalists are not incentivized anymore.

I tried to pay for journalism, but usually I get lower quality material than what I can read in HN comments for free.

> They are just regurgitating a press briefing that came across their desk

I would argue that's one of the problems. It's easy to just regurgitate press briefing, but that's _not_ what their job should be. I'd argue they could at the very least impose some form standards (don't publish "x% success", ask for exact metrics before publishing the press briefing). Asking them to vet that the study methodology was sane is too much - but asking them to use specificity & sensitivity instead of "success percent" isn't too much to ask.

why do your think this the the work of a journalist? If this is important, then we should have some regulation saying how companies need to report experiments.

well, for starters, it's published on france24.com, that's a pretty good hint that it is the work of a journalist.

I believe one shouldn't regulate _all speech_ (e.g. company press releases, or scientific publications); but I think it's fair to regulate reporting, to some extent. And you can even have a somewhat wide definition of "reporting" (e.g. if you collect audience size metrics (e.g. in order to get revenue from advertising), and your audience is greater than X, then you are reporting).

It is the job of the journalist to at least be able to ask "what do you mean by 95% accuracy?" and to publish the answer given or the fact that no satisfactory answer is given because the reader is not in a position to be able to ask that question to the entity whose results are being publicized.

The number is the story.

So reporters are just PR agents? Just cc everyone on the press release and forward misinformation because they can't understand it? Just fill those column-inches? We've got a paper to put out, doesn't matter if it's true. Sell those ads!

The whole point of journalism is to filter information by relevance or value, validate it for accuracy, and present it to the audience.

"Regurgitating" press releases has sadly become the norm for many outlets, but it trivializes the whole existence of news media.

You have a misunderstanding about "validate it for accuracy". From the point of view of journalism, this validation only means that you're reporting that a statement was really made by someone, or that a fact occurred as the journalist saw it happening. It doesn't mean that the journalist will validate experiments, or investigate statements made by someone else.

I hope to never meet someone who's violated an iceberg.

Thanks, I almost inhaled my keyboard.

Don’t worry. Captain Edward Smith died.

> Vlachova said the Czechs would like to work together with the Finns or with French and German teams working on similar projects.

> Unlike their western peers, the Czech team works in its free time and relies on scant financial means provided by a local dog food maker.

This is the kind of cooperation where Europe could really shine. It's not big pharma or financial money abstraction, it's something people can relate to.

I have a deep skepticism about sniffer dogs. This article by the National Narcotic Detector Dog Association just about sums it up:

https://nndda.org/the-double-blind-attack/

>While the scientific community highly endorses double-blind testing, it can have disastrous effects on canine teams. As there is no way to determine whether the canine responded correctly at the time of the alert, the question is, should the handler reward the canine? This is a vital question due to the fact that detection canines are largely trained on a fixed ratio reward system where the canine is rewarded immediately for almost every correct response.

It’s easy to go beyond fixed reward to variable, which is also very powerful.

Why can't a third party immediately confirm (or deny) the dog's alert in the double blind test?

That's what I was wondering. The whole argument that double blind testing can't work because the dog needs positive reinforcement seems specious.

I'm imagining a room full of opaque cake domes with drugs under some. The dog identifies one, it is lifted to reveal the truth. No one knows how many domes hide drugs, the test proceeds until the dog/handler conclude it.

Similar approach taken in France[1] not that long ago. What seems interesting here is the fact that France uses Malinois dogs, which seems a bit odd. They are insanely trainable and smart(I am saying that as someone who owns a working line Malinois) but their noses are nowhere nearly as effective or sensitive as say terriers(which is what they have gone for in Czechia). I'm guessing it's a "work with what we have already" situation but still curious to know if there is something more to it.

[1] https://www.france24.com/en/video/20201103-researchers-train...

As cool as this development is, I can't help but think about how this could be grossly abused and used as another means of detaining or abusing certain populations state or federal gov might not like. "Pointing", or the provable technique of eliciting a "positive" detection response from a detection animal, has long been a huge point of contention for use of drug dogs during traffic stops - especially in California and Illinois. A huge proportion of these stops also involved the use of "pointing" drug dogs to lead to egregious 4th amendment violations.

https://www.chicagotribune.com/news/ct-xpm-2011-01-06-ct-met...

Cops need a reasonable suspicion to search your vehicle or belongings.

And an easy way of claiming they have one is having dogs point at things, usually on command. It's a legal loophole.

Other tricks include claiming you have a slurred speech, or that they smelled alcohol breath or marijuana.

Then they can search your stuff, find nothing and say it's a false positive.

Having dogs point at things is already (or should be) "search".

In many states, a dog indicating is used as the threshold of probable cause.

Huawei worked on AI that can recognize Uyghurs. Seems like a reasonable extension of that.

Of course, when a random PR declares they have trained a ML to detect Covid with 95% accuracy, the public answer is "Why are governments not doing this?" .

But whenever another random PR declares they have trained a dog to detect Covid with 95% accuracy, the public reply is skepticism.

For whatever reason, the public seems to have a “newer is better” bias, where new techniques and materials are assumed to be inherently better. This often creates confusion and surprise when people learn that sometimes newer techniques are cheaper or more convenient but not better in every dimension; such as the fact that well developed 35mm film beat out the alternatives until very recently.[0]

It’s not clear to me if this bias is a result of marketing/current society, or if it’s an innate part of what drives us to make new stuff as a species.

0 - https://youtu.be/rVpABCxiDaU

Apparently, the Miami Heat (NBA) are also using dogs to screen fans for coronavirus. https://www.espn.com/nba/story/_/id/30770833/heat-use-corona...

I wonder what dogs' ability to detect general sickness is. It would be good to have a control with other sicknesses.

Dogs, at least some breeds, can detect enormous amounts of "smell" information.

Only problem is how to make them tell people what they sense.

Maybe some kind of dog brain MRI would be able to get the info directly?

At least they can detect some forms of cancer:

https://en.wikipedia.org/wiki/Canine_cancer_detection

I have heard about dogs helping diabetic kids control their glucose level.

Eg https://abcnews.go.com/US/10-year-boy-diabetes-dog-monitors-...

If you have another illness, you shouldn't be on the streets either. Therefore, it doesn't reduce the usefulness of the test.

Assuming it's an infectious illness.

I guess nobody will know until we have methodologies that test the odor of people. We don't have. But at least we now have ML techniques that could be used to do that in the future.

I don't think they even know definitively that it is odor. The dog could be working off any number of nonverbal cueues.

That's right, but we need to start from something we know dogs are very good at.

Occam's razor applies here as well.

Depending on how accuracy is defined 95% might be a bad result, as for all imbalanced class problems

Dog catches Covid, sniffs and passes it to person of interest, smells on Covid said person.

Slightly bizarrely, I was wondering yesterday if this might be possible.

I suspected it probably would be, but logistics would be a problem - because you would need a lot of trained dogs+handlers to match the current need for tests.

just a question, because this virus came from bats and now infects minks and humans: why should dogs be exempt from spreading or getting infected with it?

They're not. Like with the original SARS-CoV, there are reported cases of dogs getting infected. Felines seem much more prone to developing illness though.

so why are dogs still allowed in public transit then :/

I'd be somewhat worried that the dogs get or spreads covid..

This makes the avenue for detecting covid the same avenue for getting it. Covid is transmitted through breathing which is just a less aggressive form of sniffing.

Thus the doggo would get covid then spread it to everyone he sniffs.

Lies, it's actually 94.643

Reply on news.ycombinator.com