How does Burna AI measure agreement between AI suggestions and clinicians?

With blind, adjudicated comparison. Raters grade the same case without seeing each other's grades, disagreements route to an adjudicator, and agreement is computed as Cohen's kappa alongside an agreement matrix and the rate at which adjudication changed the AI's grade. Every signed grade also records what the clinician accepted and what they overrode, so agreement is measured in daily work too. For context, human reviewers identifying CTCAE toxicities in the same notes agreed at a published kappa of 0.59 to 0.68 (Fairchild et al., 2020). Ongoing internal testing demonstrates strong agreement with expert clinicians. The external measure is a concordance study Burna AI is running as a Mayo Clinic Platform_Accelerate participant, across three campuses: 1,600 cases drawn from a de-identified dataset of 3 million patients, against an adjudicated two-clinician reference panel, reading out in December 2026. The study design is at burna.ai/validation.