“Bengaluru-based startup Dognosis is developing an early cancer detection system that combines trained dogs with AI. Patients provide breath samples through masks, which dogs then sniff for chemical signals associated with cancer. Sensors monitor the dogs' behavior, breathing, movements, and other responses, while AI analyzes those…”
Plain restatementAn Indian company is testing a breath-based cancer screening method in which trained dogs sniff mask-collected breath samples while wearable sensors record canine responses that machine learning models interpret. A published Phase 2 case-control study reported roughly 90 percent sensitivity including for stage I to II disease across seven cancer groups. A larger Phase 3 study is underway.
This claim is mostly accurate and checks out against a real peer-reviewed source. Dognosis is a genuine Bengaluru startup, and its Phase 2 results were published in the Journal of Clinical Oncology, a leading cancer journal, with the trial registered in India's clinical trials registry. The study reported about 91 percent sensitivity overall and about 90.6 percent for early-stage cancers across seven cancer groups, which matches the claim. Two important pieces of context are missing, though. The study was a case-control design in which roughly one in five participants already had biopsy-confirmed cancer, which is nothing like a real screening population where cancer is rare, so these numbers do not tell you how many false alarms the test would produce in general use. The research was also funded and co-authored by the company, and has not been independently replicated. The "more than 20 cancer types" detail comes from the company rather than the published paper, which reports seven cancer groups. The Phase 3 trial of about 10,000 people across 10 hospitals is real but so far rests on company statements, and the company itself describes the product as a pre-screening tool that flags people for further testing, not a cancer diagnosis.
[drifted from the evidence:] Bengaluru-based startup Dognosis is developing an [drifted from the evidence:] early cancer [drifted from the evidence:] detection system that combines trained dogs [drifted from the evidence:] with AI. Patients provide breath samples [drifted from the evidence:] through masks, which dogs then sniff for chemical signals associated with cancer. Sensors [drifted from the evidence:] monitor the dogs' behavior, breathing, movements, and other responses, [drifted from the evidence:] while AI analyzes those signals. A Phase 2 study reported [drifted from the evidence:] about 90% sensitivity for [drifted from the evidence:] early-stage cancers across seven [drifted from the evidence:] broad cancer groups [drifted from the evidence:] covering more than 20 cancer types. The company has begun a Phase 3 [drifted from the evidence:] trial involving around 10,000 people across 10 Indian hospitals to further validate the technology." (Source: The Economic Times)
An [added by the neutral restatement:] Indian company is testing a breath-based cancer [added by the neutral restatement:] screening method in which trained dogs [added by the neutral restatement:] sniff mask-collected breath samples [added by the neutral restatement:] while wearable sensors [added by the neutral restatement:] record canine responses [added by the neutral restatement:] that machine learning models interpret. A [added by the neutral restatement:] published Phase 2 [added by the neutral restatement:] case-control study reported [added by the neutral restatement:] roughly 90 [added by the neutral restatement:] percent sensitivity [added by the neutral restatement:] including for [added by the neutral restatement:] stage I to II disease across seven cancer groups. A [added by the neutral restatement:] larger Phase 3 [added by the neutral restatement:] study is underway.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- **The company and its location are correct.** Dognosis is a Bengaluru-based startup founded by Akash Kulgod and Itamar Bitan.
- **The mechanism description is accurate.** Patients breathe into masks sent to the company's facility, where dogs are exposed to the sample , and the process involves a patient wearing a custom-designed mask and breathing normally for around 10 minutes, capturing volatile organic compounds .
- **The sensor and AI layer is accurately described.** The claim's list of "behavior, breathing, movements, and other responses" matches the documented sensor suite, which also includes EEG-based neural activity and heart rate.
- **The 90 percent early-stage sensitivity figure is correct and comes from a real peer-reviewed paper.** The published figure is 90.6% for stage I to II. "About 90%" is a fair rendering.
- **"Seven broad cancer groups" is correct** and matches the published study.
- **The Phase 3 description matches reporting.** 10 hospitals, roughly 10,000 participants, started in April.
- **The Economic Times attribution is plausible.** The story originates with Bloomberg and was syndicated widely, including to Indian outlets. ---
- **Omitted qualifier (significant): the study design is case-control, not screening.** A 90 percent sensitivity and 91 percent specificity in a cohort that is roughly 19 percent cancer patients does not translate to useful real-world performance. In a population where cancer prevalence is well under 1 percent, a 9 percent false-positive rate means the large majority of positive results would be false alarms. The paper's own framing acknowledges this by calling for evaluation in true screening populations . A reader encountering "90% sensitivity for early-stage cancers" without this context will overestimate what the test currently does.
- **Omitted qualifier: the evidence is company-funded and not independently replicated.** One outlet flagged this directly, noting that the accuracy and detection figures are, for now, reported by the company and its researchers rather than independently verified .
- **Unverified specificity of "more than 20 cancer types."** This is a company-stated figure carried through the reporting. Digital Trends attributes it explicitly to the company: the startup claims that the system targets more than 20 cancer types . The published study reports seven groups, and one report lists the seven as breast, cervical, colorectal, lung, oral, ovarian, and prostate . I could not confirm from the primary paper that 20-plus distinct histological types were represented with meaningful per-type sample sizes. With only 283 cancer cases spread across seven groups, per-type numbers are necessarily small.
- **Framing drift in the wider coverage.** Several outlets convert paired sensitivity and specificity into a single "90% accuracy" figure, which is not the same statistic. The user's version avoids this error, but the surrounding media ecosystem does not. One headline reads "Detect Cancer From Breath With 90% Accuracy" .
- **"Detection system" versus screening triage.** The claim says "early cancer detection system," which is defensible, but the product is explicitly designed as a pre-screening flag, not a diagnostic. A positive result would direct someone toward further medical testing rather than provide a cancer diagnosis . ---
- **How much the AI and sensor layer actually contributes to the headline number.** The Bayesian fusion explicitly incorporates historical dog performance and participant-level covariates . The preprint describes assessing whether a Bayesian fusion integrating individual-dog historical performance, participant/sample covariates, and sensor-derived behavioral signals improves accuracy . I could not retrieve the incremental contribution attributable specifically to the sensor and EEG data versus the dogs' raw indications plus statistical modeling. The claim implies AI analysis of dog signals drives the result; this is not established from what I could verify.
- **Whether including participant covariates in the model inflates apparent accuracy.** If demographic or clinical covariates correlate with case status in a case-control design, the fusion model's performance may partly reflect those covariates rather than olfactory signal alone. I could not resolve this from abstract-level data.
- **Independent registry confirmation of the Phase 3 trial.** I found the Phase 2 CTRI number but did not locate a separate public registry entry for the 10,000-person Phase 3 study. The Phase 3 details rest on company statements to Bloomberg.
- **The "11th cancer type" generalization claim.** Reported as trained on the scent of 10 cancer types, they identified an 11th they had not encountered before, Kulgod said . This is attributed to the founder in an interview, not to published data I could verify.
- **A minor citation discrepancy.** The preprint text references two different CTRI numbers in different passages (CTRI/2024/10/075938 and CTRI/2024/03/061847). This is likely a drafting artifact but I could not confirm which applies to which component.
- **Regulatory status.** Coverage mentions a regulatory question around this category of product in India, but I did not verify the current regulatory pathway or approval status. ---
The core scientific claim traces to a real, peer-reviewed paper in a top-tier oncology journal, not to a press release. The study was an assessor-masked, multicenter case-control study across six hospitals in Karnataka, India, enrolling 3,275 participants (1,773 training, 1,502 testing), with the test cohort including 283 treatment-naive, biopsy-confirmed cancer cases spanning seven major cancer groups and 1,219 controls (healthy volunteers, non-oncologic chronic disease, or benign biopsy) . Breath was collected on cotton surgical masks, stored under cold-chain conditions, and evaluated by trained detection dogs, with individual dog indications integrated using a Bayesian fusion framework incorporating historical dog performance and participant-level covariates; the fusion system achieved 90.8% sensitivity (95% CI, 87.2 to 94.5) and 91.3% specificity (95% CI, 89.7 to 92.9), with an AUC of 0.962 . Critically for the user's specific claim: sensitivity for early-stage disease (stage I to II) was 90.6% and remained consistent across major cancer types . On the system description, sensors and AI interpret signals including the animals' movements, respiration, brain activity and body language to determine whether the sample carries signs of cancer . The company describes a proprietary canine brain-computer interface capturing 8-channel EEG at 1000Hz, with a smart harness integrating a respiration sensor and IMU capturing breathing patterns and movement data during detection sessions . On Phase 3: Dognosis started its Phase 3 trial in April with 10 hospitals and plans to recruit about 10,000 people over 12 months, testing asymptomatic people at higher risk of cancer and survivors at risk . ---
Complete reasoning
The reply receipt is formatted for pasting into the thread where the claim is circulating.
Compact share page: verify.trueseeker.com/s/f6a7976e6f7e/d2lxKWPUFa9EgFn4-S6gDSa
Ask this case
Answers come only from the case file above; nothing is added.
Is Dognosis a real company, and is the science behind it published?
Yes. Dognosis is a real Bengaluru-based startup, and its Phase 2 results were published in a peer-reviewed journal, the Journal of Clinical Oncology, with the trial registered in India's clinical trials registry.
Does the 90 percent sensitivity figure mean the test correctly catches 9 out of 10 cancers in general use?
Not necessarily. The 90.6 percent early-stage sensitivity comes from a case-control study where about one in five participants already had confirmed cancer, which is very different from a real screening population where cancer is rare, so the numbers do not show how the test would perform, including false alarm rates, in general use.
Does the test cover more than 20 cancer types, as the company claims?
The published study only reports results for seven broad cancer groups. The claim of more than 20 cancer types comes from the company itself and could not be confirmed from the peer-reviewed paper, which has just 283 cancer cases spread across those seven groups.
Has the Phase 2 study been independently confirmed?
No. The research was funded and co-authored by the company and has not been independently replicated.
Is this test meant to diagnose cancer?
No. The company describes it as a pre-screening tool that flags people for further testing, not a tool that provides a cancer diagnosis.