Influences of breath sample re-use on the accuracy of lung cancer detection dogs

Abstract

Evaluations of dogs as lung cancer detectors using breath samples have produced a variety of results, some quite promising. Breath samples are typically collected onto a substrate and stored in a sealed container when not in use, but volatile compounds dissipate when the substrate is exposed during training and evaluation sessions. Collection of appropriate samples for training and testing dogs requires significant resources and strict control of recruitment and sample collection processes. Therefore, some researchers re-use samples while training dogs. No systematic evaluation of the effect of sample re-use on dogs’ training performance has been conducted, so the influence of this potentially important training factor is not known. We trained seven dogs to indicate the presence of lung cancer positive breath samples using an automated apparatus. The samples were stored at −60 °C or −80 °C. Samples from 460 individuals who were classified as positive or negative for lung cancer were used for training samples. Individual samples were presented to dogs up to four times over a period of 2 years. As sample re-use increased, sensitivity declined (−6.65, p = < .001, 95% CI [−10.56, −2.76]), specificity increased (2.87, p = .036, 95% CI [.19, 5.55]), and the dogs’ bias shifted in the direction of a negative indication bias (−.094, p = < .001, 95% CI [−.149, −.39]). However, there were no significant changes in the measure associated with the detectability of the target (−0.30, p = .285, 95% CI [−.087, .26]). All observed changes in performance across sample re-use were small. Therefore, these findings suggest that sample re-use may be appropriate for training, but additional research is required to determine which factors underly changes in performance as breath samples are re-used.