It was not that long ago that I would attend conferences and other industry events and be inundated with questions about SEND. Conversations would often start with “Does my study need SEND?” or even “What is a SEND?!”. There would be a sense of déjà vu as I’d often seem to have the same conversation repeatedly. But I loved it. Ahh, fond memories. However, today the conversation around SEND seems to have progressed on from that. SEND has become just another accepted part of the workflow for regulatory toxicology. Yes, there were times some of us wondered if this day would ever come, but here we are.
This was never more apparent than at this year’s Society of Toxicology (SOT) meeting. Where previously there would have been all manner of talks introducing SEND and discussions about the challenges of converting collected data into SEND-compliant datasets, such discussions were patently absent to me. So, if we were looking to learn how to do SEND, then we would have been sorely disappointed. However, in their place, the SEND-related presentations in the schedule were all focused on Virtual Control Groups (VCGs). So, if we were looking for ways to get greater value from our SEND datasets, then we found some serious inspiration. Evidently, the proliferation of standardized nonclinical data has paved the way for us to explore new ways to reuse and repurpose existing study data. To be clear, we are not talking about inventing new synthetic data based on historical controls, VCGs are reusing control subjects from past studies for a current study in order to reduce the number of subjects needed to run the study.
While these talks did not focus on SEND itself, they each acknowledged its role in allowing for such techniques to be pioneered. Personally, I went into the event focused on HOW we could use SEND to create VCGs and was pleasantly surprised by the more passionate discussion around SHOULD the industry use VCGs at all and if yes in what situations should they be applied? The scientific validity was hotly (yet respectfully) debated. The general consensus, though far from unanimous, seemed to be that we should proceed in a cautious and limited manner, always applying an impartial scientific rigour to ensure that the conclusions reached when using a VCG are no different to those formed when using a concurrent control group. To achieve this, there seemed to be general agreement to start with non-GLP studies and to ensure that there are always some concurrent control subjects in the study along with the virtual subjects. As someone whose background is in data structures and comes from the technology side of the industry rather than the scientific side, it was fascinating to observe and participate in this discussion.
‘til next time
Marc
