Key considerations for the experimental training and evaluation of cancer odour detection dogs: lessons learnt from a double-blind, controlled trial of prostate cancer detection

Abstract

BackgroundCancer detection using sniffer dogs is a potential technology for clinical use and research. Our study sought to determine whether dogs could be trained to discriminate the odour of urine from men with prostate cancer from controls, using rigorous testing procedures and well-defined samples from a major research hospital.MethodsWe attempted to train ten dogs by initially rewarding them for finding and indicating individual prostate cancer urine samples (Stage 1). If dogs were successful in Stage 1, we then attempted to train them to discriminate prostate cancer samples from controls (Stage 2). The number of samples used to train each dog varied depending on their individual progress. Overall, 50 unique prostate cancer and 67 controls were collected and used during training. Dogs that passed Stage 2 were tested for their ability to discriminate 15 (Test 1) or 16 (Tests 2 and 3) unfamiliar prostate cancer samples from 45 (Test 1) or 48 (Tests 2 and 3) unfamiliar controls under double-blind conditions.ResultsThree dogs reached training Stage 2 and two of these learnt to discriminate potentially familiar prostate cancer samples from controls. However, during double-blind tests using new samples the two dogs did not indicate prostate cancer samples more frequently than expected by chance (Dog A sensitivity 0.13, specificity 0.71, Dog B sensitivity 0.25, specificity 0.75). The other dogs did not progress past Stage 1 as they did not have optimal temperaments for the sensitive odour discrimination training.ConclusionsAlthough two dogs appeared to have learnt to select prostate cancer samples during training, they did not generalise on a prostate cancer odour during robust double-blind tests involving new samples. Our study illustrates that these rigorous tests are vital to avoid drawing misleading conclusions about the abilities of dogs to indicate certain odours. Dogs may memorise the individual odours of large numbers of training samples rather than generalise on a common odour. The results do not exclude the possibility that dogs could be trained to detect prostate cancer. We recommend that canine olfactory memory is carefully considered in all future studies and rigorous double-blind methods used to avoid confounding effects.

Invasive melanomain vivocan be distinguished from basal cell carcinoma, benign naevi and healthy skin by canine olfaction: a proof-of-principle study of differential volatile organic compound emission

“…Our experience closely mirrors that of Elliker et al . when attempting to train dogs to detect prostate cancer, and serves to highlight the danger of citing canine accuracy rates based on training samples.…”

Integrating exhaled breath diagnostics by disease-sniffing dogs with instrumental laboratory analysis

“…Furthermore, each dog needs to be individually trained for his job (group training of dogs is out of the question); this can take months; and refresher training may be important. Finally, dogs must be continually evaluated for performance using known (prior) and new challenge samples to assure stability of results (Moser and McCulloch 2010, Johnen et al 2013, Elliker et al 2014, Hall et al 2015, Polgar et al 2016). …”

Loss of a Therapy Animal

“…Once again this potential began as anecdotal stories reported by pet lovers. Tests of dogs' ability to detect prostate cancer and breast cancer have shown mixed results and are continuing (Elliker et al, 2014;Gordon et al, 2008;McCulloch et al, 2006;Smith, 2014). A check by her physician showed early breast cancer (Smith, 2014).…”