It is difficult to prove that it is impossible to train dogs to detect hypoglycemia. All that can be done to disprove this claim is to show systematic efforts to train dogs have failed, and that no-one has been able to demonstrate that it is possible.
In 2013, such an attempt was tried:
Note: This was just a letter, and probably did not pass a full peer-review.
They took three dogs that had already been trained by an organisation, and whose owners and trainers believed were capable of detecting hypoglycemic skin swabs.
Trained dogs were
largely unable to identify skin swabs obtained
from hypoglycemic T1D subjects. [...] To our knowledge, this is
the first controlled study to address
whether dogs can detect a hypoglycemic
scent, though there are anecdotal and case
reports suggesting that dogs can respond to
hypoglycemia (2–4). Our results addressed
only whether there is a detectable hypoglycemia
scent on the skin. In future studies, it
may be helpful to include behavioral elements,
such as studies in the presence of
human companions. It might also be helpful
to obtain swabs from the usual human
companions of the dogs. We found that
trained dogs were unable to correctly identify
skin swabs obtained during hypoglycemia
in subjects with T1D. Further studies
are needed to address the role of other factors
that the animals might use, such as
behavioral cues.
Given the lack of evidence, so far, it seems that trained dogs are not to be trusted. This provisional position might change if people can find other cues that trigger the dogs, or other ways of training dogs.
Given that blood glucose meters, while arguably not as cute as a dog, are fairly cheap and fairly accurate, it seems dog trainers will have a large hurdle to train dogs to be sufficiently sensitive and specific to warrant their use.