Canine Olfaction Combined With Bayesian Modeling for Multicancer Detection From Breath Samples: A Phase II Study in...

Area of Science:

• Oncology
• Biomarkers
• Veterinary Medicine

Background:

• Low-cost, high-sensitivity cancer screening tests are crucial for population health, particularly in low- and middle-income countries.
• Existing screening methods face challenges with low cancer prevalence and accessibility.
• Breath analysis using canine olfaction offers a potential low-cost, non-invasive screening solution.

Purpose of the Study:

• To evaluate the analytical validity of a multicancer breath detection system utilizing trained dogs and Bayesian fusion modeling.
• To assess the accuracy of canine olfaction for detecting multiple cancer types from breath samples.

Main Methods:

• A multicenter, assessor-masked, case-control study involving 3,275 participants (1,773 training, 1,502 testing) across six hospitals in India.
• Breath samples were collected using cotton masks, stored cold, and analyzed by trained detection dogs.
• A Bayesian fusion framework integrated individual dog indications with historical performance and participant data.

Main Results:

• The canine olfaction system achieved 90.8% sensitivity and 91.3% specificity in the test cohort (n=1,502).
• The receiver operating characteristic AUC was 0.962, indicating high diagnostic accuracy.
• Sensitivity for early-stage disease (Stage I-II) was 90.6% and consistent across major cancer types.

Conclusions:

• Canine olfaction combined with Bayesian fusion demonstrated high analytical accuracy for multicancer detection from breath.
• The study establishes the analytical validity of this approach for cancer screening.
• Further prospective evaluation in true screening populations is warranted.