 Thanks very much. So all of these sessions were titled, State of the Science in Gaps. And when I saw that Bob and I had economics, I said, well, that's a real narrow field to try and cover for State of the Science in Gaps. I reached out for some guidance from staff and he said, no, we want you guys to decide what you want to do. I decided to talk more about what I consider to be potentially a State of the Science that could be applied that hasn't been used too much, although Ignite is an example where I think it has been applied to some degree relatively effectively. And that's specifically the use of economic modeling to help us study design and implementation. So for the talk, I'm going to briefly describe economic modeling, its strengths and weaknesses, discuss some potential applications for study design and implementation, and present some successful applications of modeling in genomics. So what is economic modeling? And obviously I should have taken out my parenthetical statement there because that was a roender to add a definition, which I did, but then I didn't take it out. So just ignore what that means. This is part of my amusing part that Bob wanted me to do. So it's just my deficit here. So in economics, a model is a theoretical construct that represents an economic process or processes by a set of variables, a set of logical and or quantitative relationships between the variables. And the economic model is a simplified framework designed to illustrate complex processes often but not always using mathematical techniques. So that sounds pretty intimidating. But in the long run, it's highly useful. And I think we're moving into a time where it can be applied more generally. Now there's some strengths and weaknesses. The good, on the strength side, you don't need to have a complete data set to do an economic model. The model can really help you decide, if you have a whole bunch of different potential data elements that you could collect, the model can let you know which ones are the most important to collect and which ones really don't make as much difference. So rather than trying to collect everything, you can focus on just trying to collect the most important bits. And you can run it from a different perspective. So you can help use it to answer questions from different stakeholder perspectives. Now the weaknesses are it's very sensitive to assumptions. There's the old joke of what's the difference between a statistician and an economist. Statisticians actually require data. So you can build a model assuming anything you want to assume, but if the assumptions don't really reflect some degree of reality, you can go wrong very quickly. If you try and do rigorous modeling by the book, it's extremely complex, it's resource intensive, and it requires significant expertise and experience. Probably there are fewer people that can do this than there are genetic counselors, which we heard earlier is a constrained resource. And if you really try and reflect a real world, the models can become unmanageable. And I'll demonstrate an example of that the first time out. We tried to do this. References are at the end. That was meant to be included. So most of this is work that I've published with colleagues at some point or another. I want to present four examples here. The healthcare system perspective using universal Lynch syndrome screening, a hypothetical analysis to facilitate future decision-making using IL-28B testing to inform the use of protease inhibitor and hepatitis C viral genotypes 2 and 3, a patient perspective, which is pharmacogenomic testing to inform warfare and dosing, and then generic approaches to modeling, which is actually something we've been doing through an IGNITE supplement with the University of Florida, looking at HLAB 1502 and carbamazepine, and applications that we might use for Lynch syndrome. Now, I mentioned that trying to represent a real world example can be complex. This is one limb of a decision tree that we created when we were first starting to look at universal Lynch syndrome screening, looking at tumors. The actual full model that we worked on took up the entire door of my modeler at about this font size. So pruning we recognized early on became extremely important to be able to get this down, and it took us a little while to do it, but when we did, and we ran this model from the perspective of our healthcare system, we believed the EGAP working group report that said there was sufficient evidence to do tumor-based screening, but it was silent on which way to do it, how best to do it, and we knew that there were lots of different ways that you could do it. And so you could do it with just immunohistochemistry straight to sequencing. You could do immunohistochemistry with BRAF testing, but no methylation testing. You could do methylation testing with no BRAF, and you could do immunohistochemistry with BRAF and methylation. I'm not going to go through the rationales for this, but the bottom line is that if you, oops, sorry, if you look at, from the perspective of the healthcare system, how much am I going to need to spend to detect a case of Lynch syndrome? You can tell that there's a very significant difference between this one and this one. Now this was back in, you know, maybe about 10 years ago, so we could argue that maybe this number is going to be different now, but the reality is, is that from an implementation perspective, we didn't lose any sensitivity, really, if you look here, but we saved significant amount of money, and so from the healthcare system that had made the decision that we wanted to implement this, this told us how best to do the implementation. We subsequently used this model when questions arose as to whether or not we should implement an age cutoff, where we could demonstrate the impact of using an age cutoff on sensitivity and how that would impact the cost of screening that ultimately led the healthcare system to say, okay, we're not going to impose an age cutoff. IL-28B and protease inhibitors in hepatitis C. This was routinely used with hepatitis C viral genotype one, which is the most treatment-resistant viral genotype, and there were a number of economic analyses that supported cost effectiveness, but viral genotypes two and three are much more responsive to therapy, so the question was whether to use protease inhibitors as part of this therapy, standard therapy is dual therapy and did not include protease inhibitors. However, even with genotypes two and three, this patient genotype in IL-28B protects response to treatment across all of those genotypes, but there was very little evidence in hepatitis C viral genotypes two and three, so the questions were could IL-28B genotyping be used to select candidates for use of triple therapy for these more responsive genotypes, and how much improvement in sustained viral response is needed to cross a threshold of cost effectiveness, and SVR is the intermediate outcome that has a strong chain of evidence to the clinically relevant endpoints, and in fact what made this model very nice was that we already had a good economic model towards end-stage liver disease that we could plug our output into, and the answer is you didn't have to get a lot of improvement in sustained viral response to really be cost effective, and so what we argued was that if you did the IL-28B testing, did triple therapy for those with a resistant genotype, that all you needed was a 2% improvement in sustained viral response across this cost effectiveness threshold, and that if you treated all patients, you would need an improvement of 11%. Now, we published this and then the new antivirals came out and made this all irrelevant, so that's one of the disadvantages of fighting the last war, I guess, of what we did there. We haven't talked much about warfarin dosing and pharmacogenomics, but I wanted to present an important perspective from economic modeling. So this, we used perspective trial data from Intermountain Healthcare, and we then used a policy model to assess the cost effectiveness of doing preemptive genotyping to inform warfarin dose. Now, in the, it was a randomized controlled trial, relatively small numbers of about 200. What we found was that the outcomes in the testing versus no testing arm were essentially equivalent. However, the prospective trial data showed that tested patients required 2 to 3 fewer INRs. Now, if you look at it from a cost perspective, the 2 or 3 fewer INRs essentially offset the cost of doing the pharmacogenomic test, but if you think about this from a patient-centered perspective, it would strongly favor testing because every time that patient had to come in for an INR, they had to take time off work, they had to come into the clinic, they had to sit in the waiting room, they had to have the blood test, they got a callback. If you could reduce that by 2 to 3 fewer times, I think you could argue that the patient would be highly interested in doing that. And so the perspective from which you can do the modeling is very useful. Now, I mentioned that modeling is really complex. And so we said, well, is there a way that we could perhaps do this more simply, where you could create a model that people could just plug in data elements and get an answer that would be good enough? Maybe not perfect, but do it from their perspective. And so as a supplement to the University of Florida's IGNITE grant, we looked at modeling cost-effectiveness analysis for preemptive genetic testing for a pharmacogenomics adverse event, which is HLA B-1502 and carbamazepine, which if you have this particular HLA type, you are at increased risk for a severe ketanase adverse event, such as skin reactions, such as the Stevens-Johnson syndrome or TENS. This is, we actually included international partners to the GM6 meeting, which was our international meeting. We had some folks that were talking about using this in their area because of the increased prevalence of this particular genotype. And so we took a published model from Thailand that had looked at this cost-effectiveness. And we took the elements there and created a generic model where you just had to plug in certain of the data elements like the prevalence of HLA B-1502 in your population, the cost of drugs, the cost of alternative drugs, the cost of the episode of care. And then we used data from Singapore and Malaysia to compare the results of customized models, which each of those groups built, again using the Thai model, against our generic model. And what we're finding is that the generic model is delivering results that are probably good enough. They're within a range of acceptability in terms of the right answer. We have a manuscript on that that's in preparation. We're further taking some of the work that we did in Lynch syndrome at Intermountain. And we're now looking at this as a broader implementation project. So we're using the business case model. We will populate that model with local inputs from several different healthcare systems across the country and are going to measure the impact of the model on decision making for these systems that are in various stages of implementation of Lynch syndrome screening. And so this is part of a R01 proposal that is going to be resubmitted to the NIH's dissemination and implementation study section. So in conclusion, defining perspective is critically important. You need to know from what perspective are you really looking at this. So is it from the payer perspective? Is it from a healthcare system perspective? Is it from a patient perspective? But we can use these tools pragmatically to rationalize decision making both on the implementation side but also to ask questions about what is going to be the most important things that we need to be sure and measure as we're looking at outcomes. I will tell you that doing this type of modeling, which most academic economists look at as you're taking my baby and abusing it, is really tough to publish because it just is, they say, well, you can't use the endpoint of cost per case detected. That's not robust. Well, maybe not, but a quality doesn't mean anything to my healthcare system. They don't know what to do with that. So it's tough to publish that, but we managed to find a few journals that were willing to take a crack at it. And I believe that we're just scratching the surface regarding this application in genomic medicine. And I think Ignite is to be congratulated for really embedding ideas about thinking of the economics of genomic medicine as part of this. And I think that having more modeling opportunities going forward could be good. Shameless plug. I did co-author this book on economic evaluation in genomic medicine. It just goes to show what you can accomplish if you have unlimited ego and a feeling that you, because the only economics I've ever had was economics 101 in 1974. But I still thought, oh, yeah, I could contribute to this. So why not? And then here are the references of the studies that have been published. Thank you. Thank you, Mark. And Todd Scar was on the agenda to speak about the Ignite highlights and opportunities in this area, but unfortunately he's six, so Ann Holmes is going to present that for the Ignite network. Thank you, Ann.