Olfactory detection of human bladder cancer by dogs: proof of principle study

"We needed them to learn to recognise an odour signature for cancer
from among the hundreds present in urine, without recourse to the "pure"
source of the odour. This makes it very different from training dogs to
detect, for example, drugs or explosives."

Actually, detection of target odor is detection of target odor. We
do not know exactly what the K-9s are indicating on when they are taught
to find marijuana or cocaine or C-4 or gunpowder. The dogs don't talk, so
we have no way of knowing the exact signature scent in the whole whole
scent of any compound they are responding to.

Using pseudo scents to train a drug dog illustrates this. Anyone who
has used pseudo narcotic scents to train with knows that the pseudo scents
do not smell like the whole scent of the target odor. A dog trained only
on pseudo narcotics odors will respond by going to the full trained
responce when put on real drugs. The reverse, however, is not always
true.

The reason is that the pseudo scents are only a small make up of
signature scents that are exclusive to the target odor. Other signature
scents that are found in the target odor that are not exclusive to it
(found in other odors) are not present. The dog trained on only real
drugs, is looking for the signature scents found in the pseudo along with
other nonexclusive signature scents found in the target odor. Without the
other nonexclusive signature scents found in the real drug odor being
present with the exclusive signature scents, the dog does not recognise
the pseudo as the target odor he was trained to find because he is looking
for the whole scent picture, not just the signature scents exclusive to
the target odor. The result is that he will show interest in the pseudo
odor, but not go to trained final responce.

As far as teaching the dog to respond to cancerous tissue or urine,
the signature scents of the cancerous urine and tissue exclusive to the
urine and tissue of cancer patients has yet to be determined. I believe
the dogs percentage of accuracy could be greatly improved by training the
dogs specifically on the cancerous urine and tissue first until a reliable
trained final responce is obtained, and then working the dogs on lead and
using lineups of samples rather than letting the dogs wander and try and
locate sample on thier own. Any good detection dog team is just that, a
team. The handler guides the dog through the areas to be searched, and
interperates the dog's responces. The dog does the scent work. One
without the other is useless, reguardless of how good they are. Both the
dog and the handler have to be trained, as well as conducting regular
maintainance training with thier trainer once the initial training is done
to ensure reliability.

The anology of the dog being trained to find a bomb within a bomd is
a little misleading also. If a dog is trained to locate a specific type
of wood, say Redwood, and then trucks full of logs are examined to find
hidden Redwood (this has been done) is the dog not able to do it because
it is really "a tree within a tree"? That is absolutely not the case.

As long as the target odor used to train the dog is there, it would
not matter that you are inspecting mutiple samples of tissue or urine.
The only tissue or urine that would cause the dog to go to trained final
responce would be the samples with the trained target odor.

I also would like to see what the criteria was to select the dogs you
trained in the first place. I believe if the dogs were selected by the
types and amount of drive needed to do detection work, the percentage of
accuracy would have been dramatically higher as well. THE MOST IMPORTANT
PART of training detection dogs of any kind is having the proper K-9
candidates to train in the first place. A mediocre detection dog
candidate with all the best training in the world will still be a mediocre
detection dog.

Just my own ideas looking at it from a dog trainer's point of view!
Steve Phillips
www.thedogguy.net

Competing interests:
Other than being a detection dog trainer with an interest, I am not currently involved in any cancer detection program.

SD asked for permission 15.10.04

Willis and colleagues are to be commended for their fascinating, well
-designed and well-executed study. (1) Up until then there was only
sparse, anecdotal evidence suggesting that dogs might truly be able to
detect human malignancies. Their study compliments a recent study in which
dogs were trained to identify melanoma in tissue samples and were then
tested for their ability to detect melanoma lesions on human subjects. (2)

In both investigations, the dogs detected malignancies in supposed
controls. These studies bolster the idea that dogs may someday prove
useful in the early detection of cancer. Such experimental investigations
are of great interest to me, as I had encountered a patient who claimed
that her dog’s incessant sniffing led to her finding of a breast mass
that proved malignant. Due to the lack of substantiating literature to
back up her claim, most of her medical care team brushed this off as
coincidence, particularly since the lesion was internal rather than
cutaneous as in other anecdotal reports. Professor Hywel Williams' well-
known case of a melanoma being detected by the patient's dog (3) is now
supported by experimental evidence. Our unpublished case suggests that
dogs, in rare situations, may also detect non-cutaneous malignancies. Work
such as that by Willis at al should spur on further rigorous investigation
into this intriguing area.

1. Willis CM, Church SM, Guest CM, Cook WA, McCarthy N, Bransbury AJ,
Church MR, Church JC. Olfactory detection of human bladder cancer by dogs:
proof of principle study. BMJ. 2004; 329(7468):712-4.

2. Pickel D, Manucy GP, Walker DB, Hall SB, Walker JC. Evidence for
canine olfactory detection of melanoma. Applied Animal Behaviour Science.
2004; 89: 107-116.

3. Williams H, Pembroke A. Sniffer dogs in the melanoma clinic?
Lancet. 1989; 1:734.

Competing interests:
None declared

Dear Editor

Interesting as this subject of dogs detecting cancer by smell is, I
find this paper dissappointing. As it is such a novel approach, surely it
would have been worth compiling a more comprehensive study. Firstly, there
is virtually no information with regard to the patients in respect of
grade or stage of the bladder cancer, and at the very least a comparison
should have been made with urine cytology. It has been clearly
demonstrated that the higher the grade the greater the accuracy of urine
cytology. Secondly, it would be interesting to know the results of other
more experimental laboratory tests that have been suggested for the
detection of bladder cancer such as Immunocyt, BTAstat, NMP22, Lewis X and
urovision. Non of these provide the diagnostic accuracy that is required,
such that cystoscopy and upper tract imaging with IVU or ultrasound remain
the gold standard investigations for suspected bladder cancer. One has to
say that this study suggests that the canine olifactory sense is unlikely
to replace these. It would appear that in the haste for sensational news,
that appeals to a wider audience these omissions have been ignored.

Competing interests:
None declared

In response to comments by Michael Leahy, we would like to point out
that we were well aware of the need to consider lifestyle and
environmental factors associated with bladder cancer. As stated in our
Methods section, we collected comprehensive data on each participant,
including occupation, dietary intake, hobbies, medication, and, most
importantly, smoking habits. We specifically designed the training regimen
to take into account the association between smoking, bladder cancer and
the possible presence of tobacco related odours in the urine. During
training, the dogs were taught to ignore the urine of control individuals
who smoked, and were presented with positive bladder cancer urines from
non-smokers, as well as smokers. In the evaluation tests, all runs
containing a bladder cancer patient who smoked included at least two
controls who were smokers.

We did not measure metabolites from tobacco in the urine, but based
on self report there was no obvious tendency for the dogs to
preferentially select the urine from smokers. Specifically, 41% of
selected urines were from self described “current or recent smokers”
compared to 43% of non-selected urines. 37% of selected urines were from
individuals who reported smoking one or more cigarettes on the day before
their urine sample, compared to 33% who reported smoking no cigarettes on
this day.

Moreover, adjustment for smoking status has essentially no effect on
the observed association between presence of cancer and odds of selection
as indicated in the following table.

Table. Effect of adjustment for tobacco smoking status on the odds of selection

Model, Odds ratio*

______________________________________________________

Base (cancer) 4.1

+ Usual cigarettes per day 4.1

+ Cigarettes on day of urine sample 4.2

+ Time since stopping smoking in week** 4.2

+ Usually smoke (any amount) 4.1

+ Smoked on day of sample (any amount) 4.4

______________________________________________________

* Odds of selection among individuals with bladder cancer (transitional cell carcinoma) compared to among those without from a conditional logistic regression model.

** Zero for current smokers and 100 for non-smokers or smokers stopping more than 100 weeks before sample.

None of the bladder cancer patients whose urine was used during
testing was in a high risk occupation for the development of bladder
cancer. Furthermore, there was no over-representation in terms of dietary
intake (including tea, coffee or alcohol), or exposure to chemicals used
for domestic purposes, such as paints, solvents and pesticides, amongst
this group.

Competing interests:
Authors of the paper