May 21, 2021

But they’re all good dogs

Q: Did you see dogs are better than lateral flow tests for detecting Covid?

A: No

Q: The Guardian says: Dogs can better detect Covid in humans than lateral flow tests, finds study. With 97% accuracy!!

A: That’s the detection rate in people with Covid. The detection rate of Covid in people without Covid is 9%, which is a lot higher than you’d like.

Q: Where did they find the people?

A: It was people turning up to a testing centre in Paris, though since 109 out of 335 had Covid, it can’t really have been representative. The test positivity rate in France as a whole is only 4.5% at the moment and peaked at 15.9%

Q: Is 335 people enough?

A: Potentially, though the study initially planned to get 2000 people

Q: The story says lateral flow tests correctly identify on average 72% of people infected with the virus who have symptoms, and 58% who do not. That sounds really bad. Why does anyone use them?

A:  There’s a lot of variability between tests: some of them are better.  Also, they have much, much lower false positive rates than the dogs — around half a percent.  Since there’s a tradeoff between conservative (giving false negatives) and being sensitive (giving false positives), you can’t just compare the sensitivity of two tests that have a ten-fold difference in false positive rate.

Q: Still dogs would be quicker, and you could just use the real test in people the dogs picked out

A: That has potential, but dogs don’t scale all that well. You need to train and test each dog; they can’t just be mass-produced, boxed up, and mailed around the country.  And dogs aren’t that much quicker — this isn’t walk-past sniffing like the beagles looking for smuggled food at Auckland airport; you need to stick a pad in your armpit for a couple of minutes.

Q: How much of the spin is from the research paper and how much is coming from the newspaper.

A: The researchers are reported in the French source: “Ces résultats confirment scientifiquement la capacité des chiens à détecter une signature olfactive de la Covid-19“, souligne l’AP-HP, précisant que cette étude, pas encore publiée dans une revue médicale, est “la première de ce type réalisée au niveau international“.

Q: J’ai pas de clue que that means

A:  “These results scientifically confirm the ability of dogs to detect an olfactory signature of Covid-19 “, emphasise [the hospital], specifying that this study, not yet published in a medical journal, is ” the first of this type carried out in international level “.

Q: So it’s not just the Guardian

A: No.

avatar

Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »

Comments