Read the Full Video Transcript

Ruth Etzioni: Short answer, in case I don't get to my three minutes. What can models see? Just nothing yet, but maybe a little more by the end of the afternoon. Modeling is a way of defining "what if" scenarios and then projecting quantitatively the outcomes corresponding to each one. There are two scenarios, relabeling versus not, that we want to understand what the implications of each are going to be relative to one another. A principle model can help us to compare those outcomes. And I've got outcomes of harm and benefit, but we need good data, precise and sometimes detailed specifications. That's what I'm going to focus on and usually some assumptions.

So what do we need to specify a model? There are three big buckets here. One is, first of all, when you say Grade Group 1, how was it ascertained? Was there a single biopsy? Were there two biopsies? A lot of upgrading. Dan talked about [inaudible 00:01:01]. A lot of the upgrading happens early on. Maybe if we have accurate biopsy, we have a confirmatory biopsy and we then call someone Grade Group 1 if they pass both. That is a different animal than if we say Grade Group 1 on the basis of one biopsy. Is there an MR involved? Is there imaging as well? And so what actually led to the calling of the Grade Group 1? And we can abstract it and say maybe we just want to talk about the accuracy of classification. How accurate do we think we can be? If we can be this accurate, maybe we can start talking about relabeling. That's just abstracting this question of what led to the definition of Grade Group 1. Because if you just say Grade Group 1, that's not fully specified.

The second one is, how's it going to be managed? Dan talked a little bit about this. In both settings we have to talk about what are we doing now? What are we going to do, and that's also heterogeneous, right? And then what might we do? And that's left out of the question. And then there's a question about how will patients comply. All of those three together are going to drive the outcomes. I just want to show you that, well, first of all, I want to say that modeling requires one other thing. It requires an underlying consensus about the natural history of the progression of Grade Group 1. And I'll say that the active surveillance studies are really helpful with this, but we have to go beyond the empirical data because the inclusion criteria and the monitoring frequencies and technologies in each of the studies lead empirically to different frequencies of progression in those studies. And those are the empirical frequencies of progression that are very influenced by the surveillance protocol.

So we have to kind of get to the underlying risk of progression. Now, the reason why I've been thinking a lot about this is because we are working with Ingrid and Scott and with a very accomplished statistician in the Netherlands who's really done fantastic work that I think will help us to get to this question of the natural history. So is the question specified? Not yet. And so if we can properly specify the question by filling up those buckets, then I think that we can get to modeling actually being helpful and helping us to understand what the tipping point is in terms of harm and benefit.

Speaker 2: Thank you.