Which drug caused it, and how you can tell.
Grading tells you how severe the event was. Attribution tells you whether your agent caused it. Burna scores every suspect drug with WHO-UMC and Kramer, carries comorbidities and concomitant medications into the reasoning, and shows the investigator the working before anything is signed.
Whether your agent caused it is the harder question.
Give a new drug to a thousand people and some of them will have a bad night with another cause entirely. Separating those two cases is most of the work, and it's the part a grade on its own leaves open.
“It's a question that has been there forever since we started clinical trials. And it's always very polemic, because when you have 10 doctors around the table, you have 10 different points of view.”
Is it the backbone, or is it your agent.
Modern trials stack agents, so the attribution question stops being yes or no. Burna scores each suspect drug separately across 42 oncology regimen profiles and 26 organ system classes, so your investigator sees which agent the evidence points at, agent by agent.
WHO-UMC, applied to one event
Pembrolizumab
Carboplatin
Timing fits the known onset window
Yes, week six
Yes, same cycle
The reaction is documented for this agent
Yes, immune colitis is labelled
Rarely reported
It improved when the drug was withheld
Yes, within four days
Drug continued throughout
Another cause explains it better
None found in the record
None found in the record
Re-exposure reproduced it
Not attempted
Not attempted
What your investigator is shown
Probable
Unlikely
Figure 1. One event, two suspect agents, each scored on its own. Illustrative rendering; the investigator signs the result.
“You see so many combination trials, and especially with sophisticated agents, you want to be more objective on which part of the treatment is actually contributing to the AE. Is it the backbone, or is it my investigational agent? This is what pharma and biotech care about.”
A first-in-human agent brings no history with it.
For a first-in-human agent the adverse event record is still to be written, so Burna requires the protocol. From it, the engine runs pharmacokinetic modelling and scores the known agents in the regimen, which turns the question into elimination: what the established drugs explain, and what is left. Attribution for a novel agent lands as possible or probable more often than conclusive, and it says so.
Attribution is a clinical judgment and it stays one.
Burna proposes a score with the full reasoning attached: the algorithm, the per-drug probability, the comorbidities and concomitant medications it weighed, and the lines in the record it drew on. The investigator accepts it, changes it, or rejects it. The audit trail records what was suggested and what was decided, under 21 CFR Part 11.
“It's often a black box, because the clinician doesn't necessarily display any rationale whatsoever if it's not in their clinic note.”
Scored the same way every time, so results compare across sites and studies.
Because every attribution is scored against the same algorithms and recorded in the same structure, the results are comparable across sites and across studies. That is the precondition for the questions that come next: which comorbidities and concomitant medications predict who is at risk, and which early events predict adherence or outcome later.