Impact of individual training parameters and manner of taking breath odor samples on the reliability of canines as cancer screeners

The aim of this study was to assess the progress in operant conditioning of dogs during 3 training phases up to a level where the dogs performed a minimum of 40 correct indications without false alerts and hesitations in at least 100 trials in each phase, and to compare the training results with the working phase. Breath samples from 57, 45, and 118 patients with breast cancer, melanoma, and lung cancer, respectively, were used as patterns, with samples from 305 healthy volunteers as controls. Six dogs in 2 age-groups (20 months vs. 6 months old) were used for the training. In phase I, the dogs were trained to indicate by sitting or lying down before the breath samples taken from cancer patients (pattern odor) coupled with food odor and placed among 4 blank samples (a lineup of 5 samples). In phase II, a pattern sample without food odor was placed among blank samples. In phase III as well as in the working phase, a pattern sample was placed among control samples. Significant individual differences were found in dogs’ performance during the training. The percentage of trials with prompts to achieve self-dependent work decreased in older dogs in consecutive training phases, but increased in phase II in the younger dogs. The increasing level of challenge in consecutive training phases resulted in a higher percentage of false alerts and misses, and in younger dogs, decreased motivation for sniffing the odor samples was observed in training phase III. Nonsignificant and negative correlations between the training phases and working phase in the percentage of correct indications and false alerts show that it is difficult to predict the performance of fully trained dogs on the basis of the results achieved during the training. Fully trained dogs discriminated pattern breath samples (cancer) from healthy controls with 79% sensitivity and 78% specificity, or with 50% sensitivity at the probability of correct response by chance of 50% and 20%, respectively. The sensitivity decreased to 68% and 37%, respectively, when each sample was tested only once, although being better than by chance (P < 0.05). The percentage of false alerts and misses is difficult to reduce during the training; thus, the percentage of correct indications tended to decrease in the course of training phases II and III. Odor samples taken in hospitals may contain components of “hospital odors,” which may be a confounding factor for dogs’ indications. Using trained dogs for cancer screening on the basis of detection of odor markers still requires further study.