Summary of findings
Given the extraordinary claims made about dogs detecting cancer on the basis of odour,1-4
our aim was to design and conduct a simple, yet stringent, experiment to establish whether dogs have this capability. We achieved the successful detection of urine samples from patients with bladder cancer 41% of the time (rather than the 14% expected by chance alone), providing convincing evidence that dogs do, indeed, have this ability. Multivariate analysis suggests that the dogs' capacity to recognise an odour signature characteristic of bladder cancer is independent of other chemical aspects of the urine detectable by urinalysis, such as the presence of blood. Although this multivariate model does not fully allow for the lack of independence in the data, because each dog did the same set of tests, it is, nevertheless, able to assess possible confounding of a specific cancer signature with other features of bladder cancer urine. Exactly what the chemical composition of the cancer odour signature is we can only speculate at present. Evidence from gas chromatography and mass spectroscopy studies indicates that elevated levels of formaldehyde, alkanes, and benzene derivatives occur with some cancers,5-8
but other volatile molecules are probably produced as well.
Rationale for training approach
When we embarked on this project we had no relevant peer reviewed publications to refer to. The trainers on the team were experienced at teaching dogs to scent-match, but this was not the task being demanded of the dogs here. 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. At the beginning of the study we considered using surplus tumour material obtained during surgery. We dismissed this, however, largely because the tissue could not be chemically fixed without irrevocably altering the smell, and the use of unfixed tissue had serious health and safety implications for the dog trainers.
Having decided that we would concentrate on urine as the source of tumour derived volatile organic compounds, we then had to consider whether to use each participant's urine sample separately or whether to pool those of the cancer patients and, separately, those of the controls. Although pooling might have led to a greater concentration of the desired odour signature, we foresaw some important disadvantages and pitfalls. Firstly, we had no idea whether certain foods, drinks, or drugs, for example, may obscure, interfere with, or even mimic, the odour of tumour related compounds. Only by taking detailed histories from each participant, and introducing each sample separately, could we gradually eliminate these possibilities. Secondly, pooling specimens would lead to many fewer samples being available for the dogs to smell. The very real possibility then existed that dogs would merely scent-match with known samples, rather than learn to pick out the distinctive odour signature common to the cancer urines. Lastly, and perhaps most importantly, we were concerned that “rogue” control specimens from people with undiagnosed cancer elsewhere in the body may be inadvertently added to pooled samples. We did, in fact, have an occasion during training in which all dogs unequivocally indicated as positive a sample from a participant recruited as a control on the basis of negative cystoscopy and ultrasonography. The consultant responsible for the patient was sufficiently concerned to bring forward further tests, and a transitional cell carcinoma of the right kidney was discovered.
We next had to consider the physical state of the urine when presented to the dog. We felt that air dried samples would have greater applicability in a clinical setting, by virtue of easy handling, transport, and storage. However, the overnight drying process may result in the loss of volatile organic compounds important to the overall odour signature. We therefore opted to train one cohort of dogs on wet samples and another on dried samples. When tested, the dogs trained on liquid urine performed significantly better, suggesting that the more volatile molecules are of importance in the cancer odour signature. However, the small sample sizes, together with other potentially confounding variables between the two groups of dogs, such as breed, age, and environmental conditions during testing, limit confidence in this observation. Further work to determine the optimum physical state for the urine will therefore be needed.
What is already known on this topic
Canine olfactory detection of cancer has been anecdotally reported but has not, until now, been the subject of scientific scrutiny
What this study adds
Dogs can be trained to distinguish patients with bladder cancer on the basis of urine odour more successfully than would be expected by chance alone
This study provides a benchmark against which future studies can be compared
Lastly, we gave careful consideration to the selection of patients and controls. During training, we exposed the dogs to urine from patients presenting with a broad range of transitional cell carcinomas, in terms of grade and stage, as we felt this would increase their likelihood of recognising the common factor or factors. We took particular care to train the dogs with control samples containing elements likely to be present in urine from patients with bladder cancer but also commonly occurring in other non-malignant pathologies. In this way, we could teach the dogs to ignore non-cancer specific odours. This led to the inclusion of urine samples, during both training and evaluation, from a variety of patients, such as people with diabetes to control for glucose, those with chronic cystitis to deal with the influence of leucocytes and protein, and healthy menstruating women to control for blood. Given the prevalence of benign prostatic hyperplasia in the age group of men most likely to have bladder cancer, we also included this condition in both training and testing.
Our approach to training was vindicated by the results achieved when the dogs were formally evaluated. Despite the fact that we had not used dogs with proved scenting abilities, and despite the inclusion of age matched diseased controls, we achieved a statistically significant success rate. We learnt a great deal during the study, and we are confident that improvements in the success rate can be achieved by modifications to the training regimen. In particular, we need to work on suitable reward mechanisms when the trainers are blinded to the samples, so as not to confuse the dogs. Also, for this approach to cancer detection to have more clinical relevance, we would need to teach the dogs to respond to more than one positive sample at a time, and to have a signal for “no positive sample present.”
In summary, our study provides the first piece of experimental evidence to show that dogs can detect cancer by olfactory means more successfully than would be expected by chance alone. The results we achieved should provide a benchmark against which future studies can be compared, and we hope that our approach to training may assist others engaged in similar work.