The use of dogs as a medical detection tool is one of the most impressive applications of their olfactory capabilities, and one of the most active research areas in canine biomedical detection.

The principle relies on the dog's ability to discriminate profiles of volatile organic compounds (VOCs) produced by the human body in different pathological contexts.

These compounds can be present in exhaled breath, sweat, urine, or saliva.

Scientific literature documents promising results, but with varying levels of evidence depending on the indications and the methodological quality of the protocols. This article reviews three particularly studied areas: cancer, epilepsy, and Covid-19.

Biological Basis: Canine Olfaction and Volatile Organic Compounds

The dog's olfactory acuity is based on several well-documented anatomical and physiological characteristics: a more developed olfactory epithelium than in humans, a nasal architecture that promotes contact between inspired air and olfactory receptors, as well as a very large number of functional receptors.

Sniffing produces an airflow dynamic distinct from ordinary breathing, which improves olfactory sampling.

These specificities explain why dogs are used to detect complex odor signatures in many contexts, including medical ones (Jendrny et al., BMC Infectious Diseases, 2021, PMID 34412582).

Diseases do not produce a unique odor in the simple sense, but rather VOC profiles resulting from metabolic, inflammatory, or infectious changes.

In cancer, these profiles appear to be partly linked to alterations in tumor cell metabolism.

In infectious diseases such as Covid-19, they reflect the interaction between the pathogen and host metabolism.

In epilepsy, available data support the existence of an olfactory signal associated with seizures, and suggest the existence of a pre-ictal signal in some patients, but this point is still undergoing scientific consolidation.

To select dogs suitable for medical detection, recent work highlights motivation to work, ability to learn by reward, concentration, general health, and olfactory acuity.

Important traits vary depending on the role: biodetection dogs require a strong tendency to search by smell and tolerance to distraction, while medical assistance dogs require more attachment to the human partner and ease in various environments (Bistre Dabbah et al., Journal of Veterinary Behavior, 2024).

Cancer Detection: A Solid Body of Work, Acknowledged Methodological Limitations

Canine cancer detection is the most historically documented field.

One of the first famous reports dates back to 1989, with a letter published in The Lancet describing a dog's persistent attention to a skin lesion that turned out to be a malignant melanoma.

Since then, numerous studies have explored different types of cancer and various biological matrices: exhaled breath, sweat, urine, saliva, tumor tissue, and blood.

The systematic review by Bauër et al., published in Integrative Cancer Therapies in 2022, provides a reference synthesis (PMID 36541180).

It included studies on the detection of cancer and infectious diseases in humans using dogs or rats.

Of the 226 dogs that participated in the included studies, 68% of the work focused on cancer detection, with the lung, prostate, and breast as the most studied locations.

The authors conclude that the results are promising but that the heterogeneity of the protocols remains a major obstacle to clinical interpretation.

A prospective double-blind study published in Scientific Reports in 2024 evaluated a bio-hybrid platform combining trained dogs and artificial intelligence for the simultaneous detection of four cancers (lung, breast, prostate, colorectal) from exhaled breath samples in 1,386 participants. The overall sensitivity reported was 93.9% (95% CI: 90.3-96.2%) and specificity was 94.3% (95% CI: 92.7-95.5%) (Half et al., Scientific Reports, 2024, PMID 39548246).

These results are remarkable, but not sufficient on their own to justify widespread clinical integration without independent replication and robust standardization.

In France, the KDOG program of the Institut Curie is a research and innovation program aiming to develop a reproducible, reliable, and ethical methodology for canine breast cancer detection.

This is a serious scientific endeavor, but not yet a clinically deployable large-scale validation.

Epilepsy Seizure Detection: Two Distinct Capabilities Not to Be Confused

In epilepsy, scientific literature carefully distinguishes two realities: seizure response dogs, which intervene during or after a seizure, and seizure alert dogs, which are supposed to signal the imminent onset of a seizure before its clinical manifestations. The level of evidence is not the same for these two functions.

Response dogs are trained to act during or immediately after a seizure:

• push an alarm button,
• fetch help,
• position themselves to protect the person.

Their effectiveness relies on a behavioral response trainable to an observable event, and the level of evidence for this function is more solid.

The question of anticipatory alert dogs is more complex.

The review by Lacombe et al. published in Epilepsy & Behavior in 2018 concludes that available data remain limited, very heterogeneous, and often based on owner-reported observations rather than robust experimental protocols (PMID 30502330). It supports the interest of the subject without firmly validating anticipatory alert capabilities for all dogs or all patients.

A decisive advance was made with the study by Catala et al. published in Scientific Reports in 2019 (PMID 30923326), which experimentally showed that trained dogs could discriminate an epileptic seizure odor common to different patients, with unexpected generalization between different types of seizures.

This result clearly demonstrates the existence of an ictal olfactory signal.

A subsequent study conducted in an epilepsy monitoring unit, published in Epilepsy & Behavior in 2021, clarified the nature and temporality of this signal (Maa et al., 2021). Trained dogs discriminated ictal and interictal samples with a probability of 93.7%.

The detection of the signal preceded the clinico-electrical event with a probability of 82.2%, with a reported average delay of about 68.2 minutes before the seizure.

Psychogenic non-epileptic seizures were not associated with this signal.

These results are important but remain exploratory on limited sample sizes.

Covid-19 Detection: Documented Performance in Screening Conditions

The Covid-19 pandemic generated a rapid and voluminous series of works on canine detection of SARS-CoV-2, now constituting some of the most well-documented data in this field.

A prospective cohort study conducted in two community screening centers included 335 ambulatory adults, 143 symptomatic and 192 asymptomatic (Grandjean et al., PLOS ONE, 2022, PMID 35648737).

The overall sensitivity of canine detection compared to nasopharyngeal RT-PCR was 97% (95% CI: 92-99), reaching 100% in asymptomatic individuals.

Specificity was 91% (95% CI: 87-95). Canine sensitivity was statistically superior to that of the nasopharyngeal antigen test (84%, p = 0.006), although specificity was lower.

A Spanish study trained 6 dogs for in vivo detection on 848 human subjects in three populations of different prevalence (Vesga et al., PLOS ONE, 2021, PMID 34587181). Sensitivity was 95.9% (95% CI: 93.6-97.4), specificity was 95.1% (95% CI: 94.4-95.8), and negative predictive value was 99.5% (95% CI: 99.2-99.7).

A pilot study documented a sensitivity of 98% and a specificity of 92% in the testing phase, with 96% sensitivity and 100% specificity during a hospital deployment of one dog on 153 patients (Maurer et al., Open Forum Infectious Diseases, 2022, PMID 35818366).

The systematic review by Meller et al. published in Annals of Epidemiology in 2023 (PMID 37209927) synthesizes all of this literature and concludes that canine SARS-CoV-2 detection shows a generally high level of performance and potential usefulness as a rapid screening tool, with strong dependence on the quality of protocols and standardization of training conditions.

A separate pilot study also showed that dogs trained on acute Covid-19 samples also identified samples from long Covid patients with high sensitivity, suggesting the persistence of an olfactory signal beyond the acute phase (Twele et al., Frontiers in Medicine, 2022, PMC9245071).

Cross-Cutting Limitations to All These Applications

Several structural limitations are common to this entire field of research.

Inter-individual variability among dogs is documented: detection performance varies depending on the animals, their motivation, their state of fatigue, the work context, and the characteristics of the samples.

Lassitude, hunger, or environmental distractions can affect reliability in real-world conditions, which analytical instruments do not share.

The standardization of training and validation protocols remains insufficient internationally.

The case/control ratio, the type of samples, the nature of the blind or double-blind protocol, and performance criteria vary considerably from one study to another, making direct comparisons difficult.

The precise molecular mechanisms remain under elucidation.

Identifying the VOCs responsible for the detected signal is a prerequisite for understanding false positives and false negatives, and for the eventual development of artificial sensors.

The issue of the welfare of medical detection dogs is rarely addressed in the literature, but studies indicate that repeated exposure to intensive work sessions can be a source of behavioral stress if recovery and enrichment protocols are not adapted.

Conclusion

Medical detection dogs constitute a serious research area, with sometimes remarkable but still unevenly robust results depending on the indications.

For cancer, experimental performance is often high, but standardization and replication remain essential.

For epilepsy, the existence of an ictal olfactory signal is well supported by available data, while pre-ictal detection is suggested by several experimental works without yet being consolidated.

For Covid-19, the data are among the strongest in this field and show that trained dogs can achieve high performance under certain conditions.

The real challenge in all three cases is no longer whether dogs can detect something, but under what conditions this detection becomes truly reliable, reproducible, and transferable.

Sources 

1. Bauër P. et al., Remote Medical Scent Detection of Cancer and Infectious Diseases With Dogs and Rats: A Systematic Review, Integrative Cancer Therapies, 2022, PMID 36541180
2. Half E. et al., Non-invasive multiple cancer screening using trained detection canines and artificial intelligence: a prospective double-blind study, Scientific Reports, 2024, PMID 39548246
3. Jendrny P. et al., Canine olfactory detection and its relevance to medical detection, BMC Infectious Diseases, 2021, PMID 34412582
4. Catala A. et al., Dogs demonstrate the existence of an epileptic seizure odour in humans, Scientific Reports, 2019, PMID 30923326
5. Maa E. et al., Canine detection of volatile organic compounds unique to human epileptic seizure, Epilepsy & Behavior, 2021
6. Lacombe M. et al., Dog alerting and/or responding to epileptic seizures: a scoping review, Epilepsy & Behavior, 2018, PMID 30502330
7. Grandjean D. et al., Diagnostic accuracy of non-invasive detection of SARS-CoV-2 infection by canine olfaction, PLOS ONE, 2022, PMID 35648737
8. Vesga O. et al., Highly sensitive scent-detection of COVID-19 patients in vivo by trained dogs, PLOS ONE, 2021, PMID 34587181
9. Maurer M. et al., Detection of SARS-CoV-2 by canine olfaction: a pilot study, Open Forum Infectious Diseases, 2022, PMID 35818366
10. Meller S. et al., Canine olfactory detection of SARS-CoV-2-infected humans: a systematic review, Annals of Epidemiology, 2023, PMID 37209927
11. Twele F. et al., Detection of post-COVID-19 patients using medical scent detection dogs: a pilot study, Frontiers in Medicine, 2022, PMC9245071
12. Bistre Dabbah S. et al., Survey on the importance of different traits for medical detection dogs, Journal of Veterinary Behavior, 2024