I presented this morning a study, we called it the PD-L1 AI Blueprint Project. The PD-L1 Blueprint Project started many, many, many years ago. At that time there was one sPD-L1 immunohistochemistry assay connected to one specific drug. We had four or five different assays, four or five different drugs. So that was very problematic for the pathologist to deal with four or five separate assays. So the Blueprint project was to compare the performance of the different assays and the conclusion from the Blueprint project was that three of the four studied assays performed very similarly and could be interchangeable. That was a huge relief for the pathology departments because they didn’t need to deal with four different assays.

So the current study is… While the first Blueprint project was only manual, of course, IC assessment, we have now applied an AI platform to the same specimens, compared the agreement between 24 expert pathologists using manual assessment versus an AI-performed assessment. It showed up that at least the AI was not inferior to the manual, it was rather superior, slightly better for three of the assays, clearly better for the fourth assay, compared to manual assessment.

So that is the conclusion of the study, that the AI PD-L1 assay was at least not inferior, rather slightly better.  The 24 pathologists participating in the study, they are all top experts in the field. We are not talking community pathologists – in that case the AI might perform even much better but that has to be studied in a different way.

What is the clinical implication of this?

The clinical implication of this is it needs to be applied to a clinical situation where we have a clinical trial, where we have outcome data. We want to be sure that not only can the AI reduce the variability but it should also give a better prediction of outcome to immunotherapy than the manual assessment. That part of it we haven’t studied in this particular presentation so that will be the next step.

Is there anything else you would like to add?

AI might be more precise, it might be faster and might be… I don’t discuss cost here but could be also cost effective. So all these things. We need to apply AI assessment in clinical trials and hopefully very soon in clinical practice.