Scent dog identification of SARS-CoV-2 infections in different body fluids

When non-inactivated saliva samples were presented to the dogs after training with inactivated saliva samples, dogs were able to discriminate between samples of infected (RT-PCR positive), non-infected (RT-PCR negative) individuals and distractor samples (RT-PCR negative but respiratory symptoms) with a diagnostic sensitivity of 84% (95% CI: 62.5–94.44%) and specificity of 95% (95% CI: 93.4–96%). During the following detection sessions, when the device was equipped with non-inactivated samples with the same body fluid (saliva, sweat or urine), the corresponding values for diagnostic sensitivity and specificity for saliva samples were 82% (95% CI: 64.29–95.24%) and 96% (95% CI: 94.95–98.9%), for sweat samples 91% (95% CI: 71.43–100%) and 94% (95% CI: 90.91–97.78%), and for urine samples 95% (95% CI: 66.67–100%) and 98% (95% CI: 94.87–100%) respectively (Table 1, Fig. 1). Disease prevalence was about 18% on average.

During the presentation of 5308 randomised sample presentations, the overall success rate was 92% with 723 correct indications of positive, 4140 correct rejections of negative or distractors, 214 incorrect indications of negative and incorrect rejections of 231 positive sample presentations (Table 2).

From 93 subjects in total, 46 were tested SARS-CoV-2 positive and 47 were tested SARS-CoV-2 negative via RT-PCR of nasopharyngeal swabs. The RT-PCR results of the sample material (saliva, sweat, urine) from participants with a diagnosed SARS-CoV-2 infection via nasopharyngeal swab and RT-PCR were only positive in twelve cases. The time interval between RT-PCR of the nasopharyngeal swab sample and RT-PCR of the sample material from the same individual ranged from 2 to 5 months, prior to which the sample material was stored and frozen at − 80 °C. Nasopharyngeal swabs from each dog, as well as from the outside of the membranes taken after each day of testing were all negative.

Rapid, affordable and accurate identification of SARS-CoV-2 infected individuals remains pivotal not only for limiting the spread of the current pandemic, but also for providing a tool to limit the impact on public health and the economy. Data from the current scent dog detection study confirm our former pilot study (sensitivity 84% versus 83% and specifity 95% versus 96%, respectively). In the current study, dogs were after only 8 days of training not only able to immediately transfer their scent detection abilities from inactivated to non-inactivated saliva samples, but also to sweat and urine, with urine having the highest sensitivity of 95% and specificity of 98%. These results suggest a general, non-cell specific, robust VOC-pattern generation in SARS-CoV-2 infected individuals and provide further evidence that detection dogs could provide a reliable screening method providing immediate results.

In the former pilot study from our group [22], only BPL-inactivated samples of COVID-19 patients and controls were used. The first step in the current trial was therefore to evaluate if dogs can transfer scent recognition to non-inactivated saliva samples, even when trained only with inactivated samples. The inactivation process with BPL did not impair the SARS-CoV-2-associated scent of the samples, as dogs were able to discriminate with a similar accuracy between inactivated and non-inactivated saliva samples from SARS-CoV-2 infected individuals and controls. This has implications for the training of dogs, as the health and safety measures other groups had to follow when using non-inactivated samples can be overcome by using BPL-inactivation. Data from the current study indicate that dogs can familiarise to a training device and be safely trained within little more than a week by using inactivated saliva samples from SARS-CoV-2 positive individuals and controls and become reliable SARS-CoV-2 detection dogs for untreated samples. Furthermore, the safety of working with the TADD-glasses was also confirmed by negative PCR results of the samples attained (canine nasopharynx and outer TADD-glass-membrane).

In a second step, untreated saliva, sweat and urine samples were presented to the dogs separately to evaluate if they can transfer scent recognition from saliva to other untreated body fluids. The detection rate for this experiment was also high, especially considering the dogs having not been trained with sweat or urine samples before. In order to eliminate the risk of recognizing an individual odour from a specific subject, samples used were different for each session.

The sample material of the individuals with positive SARS-CoV-2 status (nasopharyngeal swab tested positive via RT-PCR) was predominantly RT-PCR negative which could mean that dogs are able to detect the changes in metabolism of non-infectious secretions of SARS-CoV-2 infected individuals. This could explain some of the anecdotal reports from the scent detection work at Helsinki airport that dogs were able to detect asymptomatic SARS-CoV-2 infected individuals prior of them shedding virus. On the other hand, it is also possible that viral RNA has already degraded due to the storage process and is therefore no longer detectable via RT-PCR.

The fact that dogs were able to discriminate successfully between positive, negative samples and distractors represents evidence of a successful discrimination process, whereas the detection ability across three bodyfluids from 93 different individuals indicates a successful generalisation process. The study involves repeated measures since the same samples could be detected more than once in the same session. In any case, in the double-blind study, per detection cycle, dogs were confronted with samples they did not scent before and all negative and positive samples came from new and different patients.

Comparable to the current study, the prevalence in our pilot study was 18.5%. Furthermore, sensitivity and specificity were reproducible which was one of our goals. The high prevalence is due to the fact that always only one positive sample was presented next to several negative samples. It is important to note that prevalence is subject to dynamic processes and can impact predictive values of any screening method of a pandemic disease. Since the prevalence in our test paradigm is higher than in the current pandemic situation, with a growing number of people getting vaccinated, the real positive predictive values would be lower when sensitivity and specificity of dogs remain unchanged. In any case, a lower prevalence should not impact the performance of the dogs themselves, especially in the testing setting that we conducted, being rewarded with food for correct decisions. Level of frustration not finding a positive sample might increase when the prevalence falls below a certain threshold. Certainly, this ‘threshold of frustration’ depends on study design and mainly on personal traits of the dogs and training experience. However, the empty runs (presentation of only negative samples but no target scent) we used in training did not lead to excessive frustration in any of the dogs.

Several research groups that also trained SARS-CoV-2 sniffer dogs achieved good results, which support this work and consolidate the reliability of the canines’ olfactory sense for medical purposes. Grandjean et al. (2020) trained six dogs in 1 to 2 weeks using sweat samples and achieved success rates between 76 to 100% [20]. In addition to their work, where only sweat samples from hospitalised patients were used, the current study suggests that also asymptomatic SARS-CoV-2 infected individuals can be detected by the dogs. Our dogs were able to identify different COVID-19 disease phenotypes and phases of disease expression (sore throat, cough, cold, headache and aching limbs, fever, loss of smell and taste and/or severe pneumonia). Other scent dog detection studies, conducted by Vesga et al. (2020) and Eskandari et al. (2021) achieved promising results (Vesga et al.: 95.5% average sensitivity and 99.6% specificity; Eskandari et al.: 86% sensitivity and 93% specificity, respectively) and also planned real-life experiments [21, 27]. These studies support the evidence of canines offering a reliable screening method for SARS-CoV-2 infections. Future studies are important to address some remaining limitations such as the low number of distractor samples with specified pathogens (differentiation to other lung diseases or pathogens such as infections with other seasonal respiratory viruses, like influenza viruses, rhinoviruses, respiratory syncytial virus, human metapneumovirus, adenovirus, and coronaviruses other than SARS-CoV-2). This was however not within the scope of the current study. The laboratory identification of the specific VOC pattern is still in its infancy, but some current studies showed SARS-CoV-2 specific biomarkers in breath samples detectable by gas chromatography-ion mobility spectrometry [28, 29], which also support our hypothesis. Scent dogs should be considered an addition to the gold standard RT-PCR, for rapid testing in situations where great numbers of people from different origins come together. The accuracies may be increased by extending the training phase and selecting individual dogs with better scent detection accuracy. Dogs could also be trained to work directly on humans, but several factors need to be considered. People can be afraid of dogs, have strong allergies or be simply uncomfortable within the proximity of a dog. In addition, some infected individuals may feel stigmatized by being positively indicated by a dog. Therefore, the authors suggest a test scenario under real conditions as follows: Individuals to be tested should line up and swipe a cotton swab over the crook of their arm or neck. In the next step they present it to the dog through an opening in a partition wall which seperates the person to be tested from dogs as well as other individuals.

To date, there are very few reports of SARS-CoV-2 infections in dogs. Some studies confirm a limited susceptibility of the dog to this virus [30]. According to current data, SARS-CoV-2 could be detected in dogs via RT-PCR but seroconversion and mild clinical signs were also reported. However, serology in the canine population shows a very low prevalence [31]. Experimentally infected dogs neither shed nor spread the virus indicating no evidence regarding dog-to-human or dog-to-dog transmission [31]. Overall, these facts imply a low infection risk for working dogs.

As with any testing scenario, human and in this case dog daily performance could vary. This also applies to the most accurate diagnostic performance of the gold standard RT-PCR that can only be achieved under ideal conditions, which does not always reflects the real life situation. Peer reviewed and preliminary systematic reviews indicate PCR sensitivities ranging from 71 to 100% implying false negative results ranging up to 29% under real-life conditions [32, 33].

In order to generate rapid test results, a large number of over-the-counter rapid antigen tests are currently used. Test results are generated within about 15 min. According to the manufacturers, the tests approved in Germany have diagnostic sensitivities between 91 and 98% and specificities between 98 and 100% [34]. However, the diagnostic accuracy under real-life conditions is estimated to be much lower (pre-prints [35, 36]). A systematic review by Dinnes et al. (2021) evaluated the diagnostic accuracy of point-of-care antigen and molecular-based tests for SARS-CoV-2 infections and found sensitivities between 34.1 and 88.1% as well as an average specificity of about 99.6% [37] The Paul Ehrlich Institute (Langen, Germany) specified minimum criteria for approved rapid antigen test for SARS-CoV-2 infections. They require a diagnostic sensitivity of above 80% and specificity above 97% [38]. The scent dog method would meet these criteria. The purpose of validation of our screening method as a diagnostic test is out of the scope of the current study. The deployment of dogs as a real scenario SARS-CoV-2 screening method is just being implemented in some public facilities in different countries [39]. First reports are promising, however, further studies have to be implemented in order to validate dogs’ scent recognition capabilities as diagnostic tool for detection of SARS-CoV-2 infections.

Detection dogs were able to transfer the conditioned scent of BPL-inactivated saliva samples to non-inactivated saliva, urine and sweat samples, with a sensitivity > 80% and specifity > 94%. All three fluids were equally suited for SARS-CoV-2 detection by dogs and could be used for disease specific VOC-pattern recognition. Detection dogs may provide a reliable screening method for SARS-CoV-2 infections in various settings to generate immediate results that can be verified by the gold standard (RT-PCR). Further work, especially under real-life conditions in settings where many individuals have to be screened is needed to fully evaluate the potential of the dog detection method.