What problem does Burna AI solve?

Adverse event attribution in clinical research is inconsistent, and the inconsistency reaches the drug label. In the placebo arms of two phase III trials, 47% and 50% of adverse events were recorded as related to a study drug the patients never received, and when the same event recurred in the same patient its attribution changed at least once in 31% to 36% of cases (Hillman et al., J Clin Oncol 2010). Two reviewers identifying CTCAE toxicities in the same 100 notes agreed at a weighted kappa of 0.59 and an unweighted kappa of 0.68 (Fairchild et al., Int J Radiat Oncol Biol Phys 2020). Burna AI prepares each grade with its source sentence, the CTCAE criterion it meets and per-drug WHO-UMC and Kramer attribution, so every investigator reviews the same kind of evidence and makes the call. Grading one event by hand commonly takes 15 to 20 minutes; with the evidence prepared, the investigator's time goes to review and signature.