It was the purpose of this study to evaluate the capability of a classically conditioned domestic dog to accurately discriminate urine and breath samples from lung cancer patients of all tumor stages from healthy controls in a strongly standardized and controlled setting of a prospective trial.
The study population as defined by the in- and exclusion criteria reflected the population in which the development of cancer is more likely and would therefore benefit most from lung cancer screening. Based on the data of the German S3-guideline for lung cancer, males are more frequently affected by lung cancer than females , hence 149 male and 97 female patients participated in this study. According to the Robert Koch Institute, the mean age of disease onset was 66 years in males and 64 years in females in 2016 . The incidence increases starting at the age of 35–39 years and peaks at the age of 80–84 years. The mean age of the study population was 65.6 years corresponding to the mean age of patients published by the Robert Koch Institute, however, an extension of the inclusion age up to and including 85 years would have defined the affected age group more precisely.
Some authors describe that the amount of dissolved substrates or VOCs in exhaled air depends on the lung volume . Therefore, FEV1 values of cancer patients were determined. We could not confirm that the ability of dogs to correctly detect cancer samples was compromised when evaluating samples from patients with a limited FEV1. In this study, the detection rate in breath samples from patients with normal FEV1 was only slightly better than that of breath samples from patients with severe obstruction (88.9% vs. 83.3%). However, this statement is impaired by the low number of patients with a medium or severe obstruction. Mass spectroscopic analyses are warranted to get further insights into the influence of a low flow perhaps resulting in a lower concentration of VOCs.
Sniffer dogs can distinguish VOCs from cancer patients and set up during the study
The race of the conditioned dog as well as factors like age, diet, motivation or health status could influence the suitability of a dog. To get more objective results it might be useful to have the same samples evaluated by several dogs, as it was already tried in other studies [9, 13, 17, 18]. Another factor influencing detection results had been demonstrated by Biehl et al., who described different outcomes depending on previous experiences of the dog, e.g. police dog vs. household dog vs. tumor sniffing dog with a better outcome for experienced dogs. However, due to the low number of dogs included in detection dog studies, it is difficult to prove an influence of the individual dog on the result statistically [13, 17].
Dogs have a pronounced olfactory memory. Therefore, repeatedly presented patient samples during the conditioning and/or the study phase may result in a dog recognizing an already known sample and thus being incorrectly conditioned to the detection of already presented samples rather than to the olfactory detection of cancer-specific VOCs. In order to exclude this possibility, all presented patient samples had been presented only once during the study phase. This procedure reduced bias to a minimum due to residence or food within the same patient. Control of any lifestyle factor was not intended, as the aim of this study was to develop a screening method for lung cancer that could be used everywhere and for everyone, independent from the lifestyle. While the development of an electronic nose is the ultimate goal of our study, the sniffer dog’s identification of lung cancer, as presented here, was only the first step towards the electronic nose and will be followed by identification of a VOD signature within the same samples that had been analyzed here. However, we employed highly stringent criteria to avoid any bias due to smells derived from preparation of test tubes, sample preparation and acquisition. The carefully controlled preparation of our test tubes is described in the Material and Methods section; the same study group member preparing the test tubes for breath samples was also assisting patients during sample acquisition. Furthermore, all samples were taken in the same room of the municipal hospital Darmstadt. Similarly, during the engagement of the sniffer dog, actions were taken to rule out any cross contamination of samples or smells. For example, the setting for the sniffer dog was always prepared by the same person, which was different from the person accompanying the dog; funnels applied to the test tubes facilitating the dog sniffing at each sample were used only once; position of positive samples among the negative samples was randomly assigned and blinded to the person leading the dog.
In the present study, encouraging results of the olfactory detection of lung cancer from urine and breath samples of patients could be shown. Experimental studies using trained dogs to identify breath odor markers of human cancer have been analyzed and compared with the authors’ own results. The mean sensitivity reported in all previously published studies dealing with the olfactory detection of lung cancer was 78%, the mean specificity was 71.5% [9,10,11, 14, 15].
Our results concerning sensitivity for the urine samples with a lower bound of 85.4% were better than the reference value of 78%. Breath samples alone might be inferior regarding sensitivity – but only if specificity is above 98%. Specificity was substantially better than the reference value of 71.5%, both for breath and urine samples. Combining breath and urine samples is likely to lead to higher sensitivity as in 40 of 41 cases the cancer sample was detected, this will come at the cost some decrease in specificity, however, it is unlikely that the combined specificity will be below 85%.
A limitation of the design is that was impossible to obtain precise estimates for sensitivity and specificity. In order to obtain not only bounds but precise values for sensitivity and specificity, some kind of replication of the sniffing experiment with the same dog would be necessary.
It could be shown that the dog was generally better in detecting lung cancer from urine and breath samples than bronchoscopy. In this study, a sensitivity of only 56.1% could be achieved by bronchoscopy vs a sensitivity of at least 84.5% in urine and of at least 73.7% in breath samples by olfactory detection. With a sensitivity of 100%, CT is the diagnostic gold standard but due to the associated radiation exposure, its use should be limited to a population at risk. The use is also limited by the high rate of false positive results . Here, the combined analysis of urine and breath samples yielding an identification of 40 out of 41 samples corresponding to a detections rate of 97.6% is a promising approach, which might fill this diagnostic gap.
Thus, the olfactory detection of lung cancer is a promising alternative: The collection of breath and/or urine samples is non-invasive and there is no risk for the patient. Therefore, a precise specification of a risk population is not necessary, since even with a sensitivity and specificity < 100% there would be a benefit for the patient. Due to the simplicity of the sample collection, this test would be feasible at many locations; it is less expensive than a CT examination, requiring a relevant radiation exposure and a contrast medium with their associated risks and can be repeated at any time. Another major advantage of sniffer dogs is that they could start patient screening right after their training without further delay for patients at risk, while the development of an electronic nose requires a yet to determine profound understanding of the VOCs indicating lung cancer with comparable sensitivity and specificity.
Disadvantages and limitations of this method - general aspects: Lung cancer is among the most common cancer entities, therefore a high number of sufficiently trained dogs would be required, which seems to be unrealistic. Training of dogs is very individual and time consuming and needs to be performed by a number of professional dog trainers taking race, age, previous experience, etc. into account. Furthermore, the sniffer dogs would have to maintain their high level of training over a long time period and - even in case this could be feasible - their performance still depends on their daily status and motivation as living individuals. Considering these aspects only, this seems to be hardly feasible in clinical practices of family doctors. In addition, factors that could disturb the olfactory detection by sniffer dogs needs to be identified.
There is still another point, which had already been discussed in the literature and is of importance: It could be assumed that, apart from the smell of smoke or the change in VOCs, e.g. due to a urinary tract infection, other circumstances such as drugs or hospital odor could influence the dogs’ accuracy . Finally, variables like room temperature, climate conditions or humidity might also affect the sensitivity and specificity of sniffer dogs during the patient screening process.
Since it is not known which VOCs or combination of VOCs indicates cancer, it is hardly possible to determine processes in the body or environmental factors, which might have an influence on the expression or detection of these VOCs.
A study on chronic obstructive pulmonary disease (COPD) patients identified drugs as possibly disturbing factors but the inflammation itself was excluded . In addition, the specific scent of hospitals was identified as potentially disturbing in other studies [12, 20].
In a study by McCulloch et al. household dogs were trained to accurately distinguish breath samples of lung and breast cancer patients from those of controls. A correlation between current tobacco consumption and the sensitivity of the olfactory detection could be demonstrated . In contrast, in a study of Ehmann et al.  lung cancer was identified with an overall sensitivity of 71% and a specificity of 93% but the detection was independent from the presence of tobacco smoke and food odors. However, if tobacco consumption should result in a poorer sensitivity and specificity of olfactory detection of lung cancer, this method would be not suitable for screening.
In conclusion, lung cancer screening by sniffer dogs seems to be affected by too many variables to become highly reliable. Therefore, the aim is to identify electronically the VOCs that indicate cancer.
Application of an electronic nose in cancer detection
Using an electronic nose for diagnosing cancer had been used for a variety of different cancers with lung cancer being the most intensively studied. Bladini C et al. provided an extensive overview on the variety of cancers studied as well as the current technically available systems discussing their advantages and disadvantages . The authors identified 37 articles dealing with commercial and non-commercial electronic nose systems published until January 2020. These studies used various statistical approaches, e.g., principle component analysis, support vector machine and logistic regression analysis, to identify “the lung cancer breathprint” being able to distinguish between healthy controls and lung cancer in different stages. Among these Li et al.  used an array of 14 different sensors in combination with an in depth data pre-processing and, among others, support vector machine processing for classification, reaching a sensitivity of 91.6% and a specificity of 91.7%. The authors concluded that these studies displayed satisfying results with different technologies, which could be used in clinical practice with desirable technological development .
In contrast, identification of a breathprint of lung cancer patients using currently available electronic nose systems was the approach presented in this study: Identification of VOC biomarkers by analysis of exhaled breath in lung cancer and non-cancer samples.
Although 24 VOCs have been identified to date as potential lung cancer markers, e.g. aldehydes, 2-butanone and 1-butanol , there is little consensus among studies dealing with identification of lung cancer-specific VOCs. In addition, only few studies link VOC levels in patients’ breath with approaches that employ sniffer dogs, as we will do following this study. This approach is to serve as positive reinforcement.
For a successful olfactory detection of lung cancer it is crucial to understand which compounds are tumor-specific, in order to use such VOCs as true tumor markers. As already mentioned, previous studies comparing the dog’s detection rates of synthetic samples vs cancer samples achieved varying results [24,25,26]. Lung cancer cells seem to produce VOCs, which differ from normal cells, or a larger amount of VOCs compared to normal cells. Sponring et al. demonstrated that certain compounds like 2-ethyl-1-hexanol and 2-methylpentane can be cancer cell derived and thus may be indicative of the presence of a tumor . Another difference between normal cells and cancer cells is the increased metabolism of substances like several types of aldehydes and butyl acetate by cancer cells resulting in a lower concentration of those VOCs . Currently available literature suggests that a combination of several VOCs may indicate the presence of cancer better than a single cancer-specific VOC, . Buszewski et al.  stated, that the signature odor of cancer that dogs use for differentiation between samples may be related to specific qualitative or quantitative olfactory impressions produced by a mixture of VOCs.
The identification of cancer-specific VOCs will therefore be a challenging objective of further studies in order to develop a non-canine screening method for lung cancer that provides an identical or higher sensitivity and specificity as achieved by the canine nose. Among these studies Li W et al., outlined a research protocol for construction of a model for early prediction of lung cancer based on exhaled biomarkers using gas chromatography-mass spectrometry . As follow-up analysis of the study presented here, we will use the second sample of each study participant to identify potential target molecules for cancer detection by mass spectrometry similar to the outline published by Li W et al. .