Seven dogs just outperformed most cancer screening programs in a Phase II trial

Doges Editorial · 2026-05-07 · 6 min read

A Journal of Clinical Oncology study across six Indian hospitals trained dogs to detect multiple cancer types from breath samples, reaching 90.8% sensitivity and 91.3% specificity — numbers that would transform access to early diagnosis in low-resource settings.

Imagine a cancer screening program that costs almost nothing to administer at the point of care, requires no blood draw, no imaging equipment, no specialized laboratory, and can be deployed in a community clinic with a folding table and a trained dog. That is the scenario implied by a Phase II study published in April 2026 in the Journal of Clinical Oncology, and while the study's authors are careful not to declare victory before the regulatory finish line — Phase II is not Phase III, and a research finding is not a clinical tool — the numbers they produced are striking enough that the careful language is doing real work to contain what they actually found.

What the study actually did

Seven dogs were trained to detect cancer from breath samples. The trial enrolled 1,502 participants at six hospitals across India: 283 individuals with confirmed cancer diagnoses and 1,219 healthy controls. The cancers represented were deliberately broad — head and neck, breast, lung and thoracic, gynecological, gastrointestinal, and genitourinary. Researchers applied Bayesian modeling alongside the canine olfaction results to sharpen the diagnostic signal and reduce noise. The dogs worked from breath samples collected passively using standardized collection kits, with no direct patient-animal contact required at the detection stage. The setup was designed with scalability in mind from the beginning.

The results: sensitivity of 90.8%, specificity of 91.3%, and an area under the receiver operating characteristic curve of 0.96. For context, a theoretically perfect diagnostic test would score 100% on both sensitivity and specificity; most widely used cancer screening tools, particularly when applied across multiple cancer types simultaneously, sit considerably lower. Early-stage cancers in the study yielded approximately 91% sensitivity — essentially the same as the overall figure — which is where the clinical value is most concentrated. Early detection is precisely when intervention outcomes improve most dramatically, and this test appears not to lose its accuracy when the cancers are most important to find.

Why India was the right place to run this

The choice of India as the trial location was not arbitrary. India has approximately one percent cancer screening participation across its population — one of the lowest rates among major economies — due to a combination of cost barriers, infrastructure gaps, and access challenges in rural and semi-urban areas where conventional screening tools cannot easily reach. Standard screening approaches require sustained investment in equipment, reagents, and trained clinical staff. A canine olfaction program requires trained dogs, handlers, and a breath sample collection protocol. The cost differential between those two infrastructure requirements is enormous, and the geographic flexibility of the canine-based approach is correspondingly greater.

Lead researcher Akash Kulgod framed the specific challenge the study was trying to address. Dogs have been known to detect cancer since at least the early 2000s — the fundamental phenomenon is well-established. "We've known for over two decades that dogs can detect cancer. The challenge has always been building a system around canine olfaction that is reproducible, scalable," Kulgod said. This study represents the most rigorous attempt yet to do exactly that: not merely demonstrate that dogs can smell cancer, but construct a clinical workflow that could be replicated across multiple sites, different dogs, and diverse patient populations without the results degrading across the variability that real-world deployment introduces.

We've known for over two decades that dogs can detect cancer. The challenge has always been building a system around canine olfaction that is reproducible, scalable.

— Akash Kulgod, lead researcher

What Claire Guest has been building toward

Medical Detection Dogs, the UK-based charity at the center of canine cancer research for fifteen years, responded to the JCO publication with the particular satisfaction of an organization that has spent a long time being right before the evidence fully caught up to the position. CEO Claire Guest, whose own life was saved when her dog indicated concern about a lump that turned out to be early-stage breast cancer, has been a consistent and sometimes lonely advocate for this research in periods when the clinical community was skeptical. Her response to the Phase II results was direct and compressed into a single observation that did not feel the need for elaboration.

The results are staggering yet not surprising to us.

— Claire Guest, CEO of Medical Detection Dogs

The road to clinical deployment

Phase II establishes that an approach works under controlled conditions at meaningful scale. Phase III — a fully randomized, prospective, multi-center trial designed to satisfy regulatory requirements — would be required before any health authority would consider the method for clinical approval. That represents a significant additional distance, both in time and in cost. But an AUC of 0.96 across 1,502 participants and six cancer types makes the case for Phase III funding considerably easier to argue than it would be with more modest numbers. Funders, institutions, and health ministries who are presented with this performance profile are looking at a different risk-benefit calculation than they were before this study existed.

The broader implication of the Kulgod study is not that dogs will replace MRI machines or liquid biopsy panels in well-resourced hospitals. It is that cancer detection does not have to wait for full clinical infrastructure to exist before it can be meaningfully offered. A program that can train and deploy dogs, standardize breath collection, and operate across community health settings could deliver early screening coverage to populations that currently have almost none of it. Ninety percent sensitivity in a setting where only one percent of the population is currently receiving any screening at all is a very different thing from ninety percent sensitivity in a hospital that already has CT scanners. The public health math changes profoundly depending on what baseline you are comparing the intervention against. The Kulgod study also advances something that has been building in the field for a long time: the scientific case that the discovery function dogs perform — moving through an environment and attending to it with their noses — is not a curiosity but a capability with genuine clinical utility. What has taken decades is building the methodology to translate that capability into something a health system can replicate and trust. This study does not complete that infrastructure. It advances it further than it has been advanced before, with a dataset large enough and a methodology rigorous enough that the skeptics who have dismissed canine olfaction as interesting but unrepeatable will need to engage with the numbers rather than the category. Seven dogs. Fifteen hundred participants. An AUC of 0.96. The baseline for this conversation has shifted.