PRODUCT UPDATES · JUN 19, 2026 · 3 MIN READ
A cross-event adjudication conflict resolves one way, and attribution is accepted on its own
A cross-event adjudication conflict now resolves once per CTCAE term, the server refuses a differing decision, and attribution is accepted separately.
Burna AI Team
Week 25 of 2026 (15 to 21 June) turned a cross-event adjudication conflict into a single decision. When two blind raters record the same CTCAE term on separate rows and grade it differently, the adjudicator now resolves every row of that term at once, and the server refuses any later resolution that differs. Accepting an adverse event's grade and accepting which drug caused it became separate reviews, and protocol documents opened inside the trial view.
The work serves coordinators and adjudicators running multi-rater blind grading, and anyone who has to reconstruct a grading decision months after it was signed.
What is a cross-event adjudication conflict?
A cross-event adjudication conflict is a disagreement between blind raters over the same CTCAE term recorded on two or more separate event rows in one case.
In blind grading, two raters grade the same encounter without seeing each other's work, and an adjudicator resolves the terms where they disagree. Each rater can add an adverse event by hand, and every addition creates its own event row. When both raters add Fever and grade it differently, the disagreement spans two rows.
Conflict detection keys on the normalized CTCAE term rather than on the row. Every instance of a term in a case surfaces to the adjudicator as one conflict.
How the adjudicator resolves a cross-event conflict
- One decision per term. The adjudicator decides once for the group, and the decision is stamped on every row in it. Each row keeps its own audit trail.
- Rater A, rater B or a custom grade. When neither rater's grade fits the record, the adjudicator can author a grade with a written rationale. The source of the final grade (rater A, rater B or custom) is stored in the electronic signature metadata.
- Locked on the server. Before a decision is saved, the server checks every other row in the case with the same term and rejects a decision that differs from one already recorded, however the request arrives.
- Rationale and note kept apart. The adjudicator's note is stored separately from the grade rationale, and the signed payload is rebuilt from both, so the signature still verifies.
Judgments about the same event are known to drift. In the placebo arms of two phase III trials, when the same adverse event was reported for the same patient at several visits, its attribution category changed at least once 36% and 31% of the time (Hillman et al., JCO 2010). The standard a trial needs is one recorded resolution per term, with its source and its signature. How Burna AI handles disagreements between raters covers the adjudicator workflow end to end, and how agreement is measured covers the evidence side.
A second blind rater can join while grading is under way
The assignment window now stays open while a case is in progress, so an admin can bring on a second independent rater after the first has started. Work already done is kept, and the blind holds. When an assignment cannot go through, the admin sees why: the case is past its assignment window, the rater is already assigned, or the case already has its full set of raters.
Organization admins run blind grading for their own organization
Organization admins and owners can now run blind grading for their own organization: assign raters, oversee cases and open any case in it. The organization match is checked in the query itself. An admin from one organization is refused another organization's case, and platform administrators keep cross-organization oversight. Regression tests cover all three paths.
Signing out now clears the data cached in the browser tab, so the next person to sign in on a shared workstation sees none of the previous user's cached data.
Separate attribution acceptance: the grade and the cause are two reviews
Accepting an adverse event's CTCAE grade and accepting which drug caused it are different clinical judgments, and a clinician may settle one while still working on the other. They now track separately. Accepting a grade updates the grade; accepting attribution is recorded drug by drug.
| Attribution status | What it means |
|---|---|
| Accepted | Every suspect drug's attribution has been accepted |
| Partial | Some suspect drugs have been accepted |
| Pending review | None has been accepted yet |
This sits on top of per-suspect attribution with the working shown, where each suspect drug is scored on its own.
Review decisions that survive reloads and navigation
An accepted adverse event now keeps its status, including Adjudicated and Manual override, when the workflow reloads in the background. Returning to an encounter visited earlier restores the full workflow state every time, even when nothing in the underlying data has changed.
Protocol document preview in the trial view
The protocols table in a trial gained a Documents column. Coordinators preview up to two documents inline and open the rest from a dialog, each through a secure link that expires.
Deleting a protocol document asks for confirmation first. It then marks the extraction drafts built from that document as stale and removes its passages from the search index, so a deleted document does not resurface in a review queue. Every deletion is checked against the signed-in user's organization.
Since then: In July 2026 the web provider app these features shipped in was retired. Blind grading, adjudication and trial protocols now run in app.burna.ai.
AI suggests, clinicians decide. Human-in-the-loop, always.
To walk through blind grading and adjudication for your trials, book a 15-minute conversation.