 So, this was the talk that I was assigned, and I'm going to try to do this and get us right back on schedule. So, pharmacogenetics is one of the things that I do, and there are two ways in which people think about pharmacogenetics. In general, one is this idea of rare, serious adverse drug reactions that everybody thinks they come out of the blue and therefore must have a genetic basis. And the other problem is that clinical response to virtually any drug that we ever use is characterized by striking variability and efficacy. These are data on from two sites in the Pharmacogenetics Research Network, looking at response to Simvastatin and response to an antihypertensive, and you can see that most people are at the average, at or around the average. And there's always a little bit of, there's always some people who are way away from average. So the question is why are some people up here and some people down here? Most people are in the middle, and I'll come back to that over and over again. So the notion of studying pharmacogenetics, understanding the genetic basis of variability in response to drugs, is what the Pharmacogenetics Research Network is all about. So the Pharmacogenetics Research Network has been in existence since 2000. This is the current makeup of the network. There are 14 clinical sites, and there are a couple of other sites that do network-wide kinds of things, and I'll tell you about that in the next slide. This is the structure. These are the 14 sites around the outside, and I'm not going to tell you what they are. You can read them if you want, or you can look on the web later. But they cover a range of disciplines, neuropsychiatry, implementation, cardiovascular oncology, endocrine and inflammation. The network is also a consistent working groups and has taken on for itself the idea of organizing national and international consortia to study specific drug responses. So this is the International Warfarin Pharmacogenetics Consortium, and C-PIC I'll talk about a little bit, and I'll say a word about PharmGKB in a second. In the last cycle of the network, one of the things that Mark Retain and Kathy Giacomini spearheaded was an alliance with the Genomic Medicine Working Group at Yokohama, led at the time by Yusuke Nakamura, who's now at University of Chicago with Mark, and the notion was that we would bring to the Japanese group proposals for genome-wide association studies for drug response phenotypes, and they would undertake for themselves at their own cost the genotyping that was involved. There are about, there are over 20 projects that have been done. It's a little slow to get the data out the other end because there's replication involved and that sort of thing. They're very much involved in scientific partners, and that's been a really great asset to the network. So in the current iteration of the network, the idea was to create network-wide resources that would be accessible to network investigators for a variety of uses, and that's what the network looks like now. There's a statistical core, there's a next-generation sequencing core, there's ontology cores, and there's an EMR core that we're involved with or that we direct. PharmGKB was originally a site for the network, but has since sort of spun off on its own. It's the Pharmacogenetics Knowledge Base, and their goal in life is to accrue data on the relationship between drugs and genes and variability and drug responses, and this is a screenshot from their front page, and they've been enablers for some of the international consortia work that we've undertaken. So I'm fond of this slide. It appeared in the New Yorker in 2000 when the famous press conference occurred in the White House announcing that the first human genome draft had been completed, and I'm fond of saying, when I show this, that we all look to this vision. Don't think it's going to be on paper, and it's interesting that she's handing her prescription not to a physician, but to a pharmacist. So it reinforces this idea of implementation science, at least, or delivery of genomic information as a team sport. So what I'm going to talk about is here. I'm going to talk about our Predict program, which is our toe in the water for implementation, and serves as a model for some of the other things that are going on in PGRN, and then this alliance between PGRN and the Emerge Network that you heard about yesterday, and I'll tell you a little bit more about in terms of a pharmacogenomics implementation project. So Predict was an idea that grew out of a mandate from our leadership to people like me and the leads in informatics to develop a system that would deliver pharmacogenomic variant data within the electronic medical record system at Vanderbilt. We started planning in the fall of 2009 and started implementing in the fall of 2010. This is the front of one of our brochures. So the notion is you select populations who are at risk for receiving a drug that has some actionable pharmacogenomic story. There are, and the way we define that is a drug that has something in its FDA label, basically. And I'll tell you how we've gone about identifying people at risk in a second. The genotype, not for the drug that you think they're going to get, but on a multiplex platform that assays many genotypes for many drug responses. And then you archive the data. You pull out the genotypes that you think are important, display them in the electronic medical record, develop decision support tools, track what happens when those decision support tools fire. I call that the easy stuff. Right, it is the easy stuff. So while we were planning, the FDA did us a favor. And that is that they changed the label for clopidogrel, one of the most widely used drugs in medicine at the time. And they changed the label to reflect the fact that clopidogrel is a pro-drug, requires bioactivation through multiple pathways, primarily CYP-2C-19. There is a common variant in CYP-2C-19 with a minor allele frequency of somewhere around 20%, about 2% of the population are homozygots. I'll show you data on that in a second. And the variant is a loss of function variant. It's a clear loss of function variant. So consider alternate treatments or treatment strategies in patients identified as CYP-2C-19 per metabolizes. That set the cardiology world sort of a flame. They hated this idea because their sense was that there hadn't been a randomized clinical trial showing that there should be alternate strategies. The FDA stands was, well, we know enough about the biology to say that this drug doesn't get bioactivated. So it's not going to work in treating people who are homozygote poor metabolizers with this drug. It's like treating them with placebo. And that's the tension that we all feel in pharmacogenetic implementation in a nutshell. So we have been implementing since September 2010. These are numbers that are about a month old now. And this is what happens when you look at 12,521 subjects. They're 2.7% that are homozygots, 18.9% that are heterozygots, 66% have no common variant, and 12% have something else that we don't know what to do with. And we don't deliver any advice. The only ones we deliver advice about are these two here. This is easy if you think it's only star two, but it gets a little complicated. And this is important because this is sort of where the sort of rubber meets the road, the blood and guts of these kinds of things. There are other variants that we have to take into account. And so we call people who are hypometabolizers. These are 334, mostly star two, star two. But there are other variants. This is star two, star two. So there are other variants that contribute to this, what we turn the homozygot per metabolizer. So we have decision support. This is what the decision support looks like. I'm fond of showing this and saying, if you see a typo on this, don't tell me about it because it's been through the pharmacy committee. It's been through the genetics committees. It's been through the informatics committees. It's been through the legal committee. And it's been through the pharmacy and therapeutics committee, which is the last gatekeeper to the way we use drugs at our place, just like everywhere else. And if there's a typo, I just don't want to know about it. This is we're not the only ones that are doing this, as I'll tell you about in a moment. The University of Maryland and other places are doing it. They have developed their own decision support. And this is what theirs looks like. So very, very similar. Ours is prettier, but it says the same thing. So this is the slide that I wasn't sure I wanted to show because it is a little bit early for this. But Josh Peterson, one of the people at our place, has been looking a lot at outcomes. And this is what happens when you take 7 and 1 half thousand, predict patients, and ask the question, what happens when they get clopidogrel? Now, only 1,600 of them got clopidogrel. And the reason for that will become clear in a moment. But bear with me for a second. So you 1,620 get a stent. And then we looked at what happens at 90 days. So if you're a normal metabolizer, the chances are that you end up on clopidogrel, a cheap, effective drug. If you're a poor metabolizer, that's the 2.7%, the chances are about 50-50. And we think this is a hard number to come up with because it's probably more like 70-30 that you're on an alternate therapy. The reasons that it's not 100% are there are lots of reasons why that is. If somebody wants to know what they are, I can tell you about them. It has to do with contraindications to drugs. And it has to do with physicians not listening to advice and those kinds of things. And what's very, very cool for me is that there's a gene dose effect. So if you're an intermediate metabolizer, doctors are smart enough to know that the evidence is a little mushier. And so some of them convert to something else and some of them don't. So there's a very nice gene dose effect. So that's our first look at outcomes. Dan, for the intermediate, do you have any signal related to the recommendations that some have made about Dublin clopidogrel dose? We're looking at that right now. I mean, that's a great question. Supposedly, it doesn't work. Supposedly. Well, quadrupling might work. That's the sort of, the 300 instead of 75. Yeah. You know, I don't want to go there. So this is the way that our electronic medical record looks right now. I'm fond of showing this slide because blacked out all the identifiers, except this one, just to make sure you know that I do see patients and this is one of mine. And I wouldn't say that it's the prettiest way of displaying it. But we display now five drug gene, what we call drug gene pairs, propitogel warfarin, simvastatin, thymopurines, and tecolimus. And this is the medicines that this person is on. Actually, the list goes down further. What's very cool is this particular patient has a variant in CYP-2C9, which would make you predict that they would need less warfarin. In fact, they take four milligrams on one day and three milligrams a day the other day. So that is an unusually low dose and that's probably why. So how did we get from one drug to five drugs? I'll tell you about that in a second. But this is what happens when you start to do multiplex testing. So you now, it's the same population of patients, but now we can look at the, remember we do the genotyping on this uniform platform and we pull out the genetic variants as the P&T committee says to us to pull them out. And only 11.8% of people have no variants. And I think, you know, this is pretty obvious. You know, the more you look at, the more likely it is that somebody's gonna have a variant. So at the end of the day, this is the argument for multiplexing the testing because if you did it one test at a time, you're always gonna have, the test won't benefit most people. If you do it 50 tests at a time, I tell people you're abnormal for something, you just don't know what it is. And so this is the way around that. So that's what we do. So another group that is at high risk came out of this study. So we used our informatics capabilities to find a cohort of about 50,000 people who make their medical home advantageable. They get all their care there as near as we can tell. So how many over the course of five years received one of 58 drugs that include pharmacogenomic information in there? The answer was a bit of a surprise. 65% got one medicine like that within five years. So we, and when I say we, I don't mean me, I mean the guys who do the work, Josh Denny mainly, developed a predictive algorithm that said, is this patient, does this patient have characteristics that make them likely to get clopidogrel, synvastatin or warfarin? Those are the three big ones for us in the cardiovascular space at least in the next five years. And so if you see a patient in your internal medicine cardiology, nephrology or diabetes clinic, the first thing you, if the first thing you see is this, then you click on that, you get this pop-up and the pop-up will tell you this person has a calculated risk score of 68%, which means that they're a good candidate for getting this testing preemptively before they get exposed to any of these drugs. So this is why we have 7,500 patients, not all of whom have been exposed to the drug. Cause some of them come into the program through this preemptive way. Others come into the program through an indication way. So patient, people who are gonna get a, who are going to the cath lab, they're tested. People who are scheduled for elective knee replacement who are gonna end up on warfarin, they get tested. People who come into the children's hospital with acute leukemia, they get tested because they all have indication specifics. Plus there's this preemptive thing. So this is the way it looks right now. I wouldn't swear to these numbers here, but around 6,000 of the patients come in through this prognostic testing and around 6,000 come in through this indication-based testing. This is where the number comes from. And what's interesting is that there are four drugs here. I haven't listed tackling this because that's just been added. And we would fire clinical decision support in 22% of the people on clopidogrel, 25% of the people on simvastatin, 100% on warfarin new starts. Because we always give advice on what the dose should be, but only for the new starts. And then around 3% for thiopurines right now. This is complicated, but it sort of gives you a sense of what the flow looks like. Start down here, if you're the provider, there are two ways in which you get the testing. One is that you order a drug and the system says, oh, this person has been genetically tested. Make sure that you adjust the dose appropriately. That's one way the provider hears about it. The other way is that the provider sees that predict little stickum and orders the testing prognostically. In either case, the genotyping is on this Illumina platform. Most of the variants are sequestered. Bad word, I know, but there it is. And the actionable variants are delivered to the electronic health records. So that's about six or eight actionable variants right now. The other reason to show this is that we do engage patients and there's a way that patients can actually see their own results. And the other part is up here, a lot of the work that has to go on before anything happens at all is an evidence review, review by P&T and other committees, and then development of the clinical decision support. St. Jude does it very much the same way. I'm fond of saying the two places that have the most advanced preemptive programs like this in the country are both in the great state of Tennessee. Or the state of Tennessee, anyway. Um. So what Mary Relling would tell you is that they looked at 4,000 patients admitted to St. Jude and about half of them will get exposed to at least one of a list that they have of 33 high risk medications, including a lot who get codeine or tramadol and those are patients who are expected to have high risk diplotypes and will need dose adjustment. And they have clinical decision support of the type that we have. The mechanisms are a little bit different, but the idea is the same. This is a TPMT genotype recommended before a drug is being given. Here is a person who has a CYP-2D6 ultra rapid metabolizer who's about to get codeine. So you have to be very careful with those. Those are the kinds of decision support that they've developed. So Mary has also driven the development of CPIC, the Clinical Pharmacogenomics Implementation Consortium in collaboration with the PGRN and PharmGKB. And it's now extended way beyond that. You heard from, has she gone? Mission Health? Where's she gone? Mission Health person has gone. She's not on the, she's confabbing. She's confabbing. So you heard that she's part of this as well. So lots of institutions, lots of countries, and we have observers as well. So that's CPIC. I'll come back to what CPIC does, but Mary, Teri mentioned it yesterday as well and showed some of the publications. So one of the things that distinguishes the Pharmacogenetics Research Networks is that there are 14 sites, and each of them is funded to do their own pharmacogenetics stuff. And to creating networks, creating intersite synergies has been a challenge. And one of the ways in which we decided a couple of years ago when the last cycle started was to create something called the Translational Pharmacogenomics Project that would bring together sites that were interested in implementation. And the overarching goal is to implement the CPIC guidelines into diverse real-world settings. So that's what the implementation part is. I'm gonna talk a lot about implementation, but I just wanna make sure that you remember that pharmacogenetics needs a lot of discovery as well. It's not just about implementation. How am I doing for time? So there are four aims, which I'll go through in a second. There are six sites in the original incarnation, and then there are two other sites that are about to be added, including Mark's site at the University of Chicago. And they interface with CPIC and with PharmGKB to develop educational tools, a toolbox of implementation solutions, and some annotation that it turns out to be important for implementation. So CPIC, you heard about yesterday, lots and lots of guidelines. I stole these slides from Alan Schildinger who directs the TPP, so they're all mine. So these are the guidelines that have been published, and there are many others that are in progress, including a bunch of updates to these guidelines, which are scheduled every two to three years. These are the ones that are coming and updates are coming. The aim, too, is implementation. So most sites are looking at CYP2C19 and Clopidigral. It says all sites, but I'll say most sites because we've added these two new sites. There are other implementations that are going on at various sites, and these are listed here. And the two models, one is the targeted, you sort of see a patient who's going to the cath lab, so you do CYP2C19 genotyping on them. That's at the University of Maryland, and Mayo has projects in both spaces. And then we're doing this preemptive genotyping along with St. Jude, Ohio State, University of Florida, and Mayo's doing some of that as well, and Mark's program is sort of a hybrid, but mostly like this, mostly the preemptive program. These are some of the counts. You've already seen some of this. These are a little older, but University of Florida has some data on CYP2C19. St. Jude has some data on TPMT, just to show you that many places are doing this. And then we have a very, very long list of implementation metrics that we're looking at. So the lessons that we're being learned, I think I will read you this slide, just to say that it's a whole lot more complicated than we would have thought, although anybody who thinks about it for a while would have realized how complicated it's gonna be. You have to have institutional support. You have to have the leadership of the institution say this is what we wanna do. Like at Mission Health, like at Vanderbilt, like at St. Jude, the people at the top of the pyramid say this is what we're going to be known for and this is where we're gonna invest. The decision support is obviously important. The problem of education that sort of goes in a recurrent loop, you have to do it over and over and over again. And then as you're implementing, you have to watch what you're doing to see what's working and what's not. Aim three is to develop formats to report results to prescribers. As this business of star two, what does that really mean and how do you translate a star two genetic result to something that is actionable or not? That's what this is all about. And then aim four is dissemination. So we've written a lot of papers, including this paper that sort of summarizes what we have done, we think. So that leads me to the Emerge-PGRN partnership. You heard something about Emerge yesterday. Emerge has great strengths in electronic phenotyping, in capturing large populations in privacy science and in other things that I've probably forgotten about. The Pharmacogenetics Research Network has this capability in developing drug gene guidelines in the CPIC and developing a platform which I'm gonna tell you about in a second. I'm thinking about CLIA, although we all think about CLIA. So there's this interaction between the two networks. The interaction starts with this platform. So, and this platform was not designed to be an implementation platform, but it's not a bad way of doing this, I think. So when we developed this, we have this next generation sequencing capability within PGRA, and Debbie Nickerson and others said, well, we should develop an abbreviated list of important pharmacogenes that everybody's gonna be interested in. Her vision was that everybody's gonna genotype all their patients on this platform. So there'll be thousands and tens of thousands of people. I don't think that's gonna happen, but it is being implemented in Emerge. So each of the genes was nominated by one of the PGRN sites. We had lots of balloting. The rules were every site got to have at least two of their favorite genes on the list, but we all had to agree that these were important pharmacogenes. So CYP2D6, for example, probably was nominated 14 times because it's an important pharmacogenes. And these are mainly what they are, metabolism transporters and targets. Targets are things that individual sites study. There's a, and this is on a nimble gene capture in the next generation. Sequencing platform is currently high-seq being transformed, transported to my-seq and perhaps ion torrent. This is what happens when you interrogate individual subjects and look at how many single nucleotide variants they have. They have around 1,300, around a couple of dozen that are novel and around, again, a couple of dozen that are unique. And the problems are that everything works. I'm on top of it. Everything works except the two genes that we think are the most important ones. And that's CYP2D6, and we're spending a lot of time trying to get CYP2D6 right. Mark knows all about that stuff, Mark. Mark and Mark know all about this stuff. It's problematic because there's dozens of variants and the phenotype that you're interested in or the genotype you're interested in are people who have loss of function variants on both alleles. And then there's a second problem that you're interested in, people who have gene duplications. There's a pseudo gene right next door and there's fusions between the pseudo gene and the real gene, so it becomes pretty problematic to ask. And we think there are solutions to that, but it won't be just a plain old next generation run. It'll probably be PCR-based selection of targets and then next generation sequence and the way it looks. We haven't attacked HLA much, but there's some real HLA variants that are of interest and then I won't talk about this. So the EmergePGX project grew out of a conversation with Terry and I had about trying to figure out how to use this platform in an implementation space within the electronic medical record. So the notion is develop a list of actionable variants and you can do that through Emerge, through CPIC, through whatever you want at an institutional level. Identify the target patients in a nutshell. Most people are using some kind of variation on the Josh Denney algorithm. Come back to that in a second. Re-sequence on the platform. Identify the actionable variants. They're gonna be six or eight or 10 of them right now. And those, you stick in the electronic medical record, you deploy the decision support, just like we've done within Predictor. Then you track outcomes of some sort and Mark Williams has been working with Josh, Peter Sutter on the outcome side of it, but the sorts of outcomes that I've shown you are the kinds of outcomes we're gonna focus on. And then aim three is to take all the rest of them and put them in a repository for somebody to do something with. So this idea I keep on coming back to of a genotype, phenotype repository. This is gonna be about 9,000 subjects, so not enough to do very much with, but a start to thinking about how this should be done. We're planning to recruit, this is the important part of this slide, around 8,000 subjects, but if the Air Force wants to contribute another 5,000, we're happy to take them as well, since they're... 8,000, 9,000. Okay, right. So the initial target drugs are no surprise to most people. These are the nine sites within Emerge, the top seven are adult sites mostly, and the top, the bottom three are children's sites mostly. So the children's sites are focused a little bit on different drugs and a little bit different algorithms, but basically the adult sites are mostly focused on clopidogal, warfarin, simbostatin, some of the five hearings, the things that you've heard about already. Subject selection, I've just outlined the ones that are using some kind of preemptive selection strategy. Gail Jarvik has also come up with this other strategy where she's gonna start to look at patients who have unusual drug responses or unusual genotypes and then try to figure out whether she can get them incorporated into this dataset to look at some rare things, and this is the last slide. Okay, so where we are right this instant is we've collected about 3,000 of the total 9,500, so 9,500 is close to 7,800. The first 300 samples are in process right now. We want to be able to deliver by the end of the year at least 100 samples from each site. Some sites are able to just go back into the biobank and grab 1,000 samples because they're consented. So other sites have to go and consent people prospectively to do them. So the accruals are gonna be different at different sites. We're working on the second part, which is pretty standard and involves the EMR staff at each place. And then the third thing is to think about what this repository would look like and we're starting to think about that. We're having an Emerge meeting next week and one of the things that we've decided to do is to create use cases for what, you'll hear about this next week Rex, we'll create use cases for what this repository might look like. So one use case is we're really interested in knowing how many instants from OSES there are among people with cupidical. So we're gonna collect specific information that will enable those use cases. And another thing to point out is that among the 84 genes are five that are on the ACMG list. They're actually eight, sorry, six, they're actually eight. Two of them are drug metabolism genes. So we sort of, I didn't count those because we're gonna be counting those as part of the implementation of the pharmacogenetics part. But these are gonna be the ones where there may be incidental findings that we're particularly interested in tracking down and we'll have the electronic record so we'll be able to sort of see what the frequency is and what the phenotypes look like along the lines of what was told you about before. So that's a brief overview of where we are with implementation within eMERGE and within PGRA. Great, thank you, Dan. Questions for Dan? Oh, yes, please. In general, how are you viewing and handling informed consent for this being, I noticed on your order sheet it did make it very clear this is genetic testing and you had some little boxes for opting out and those sort of things. So how are you thinking about consent in this context? So the, well, let me talk about what we've been, I think that's sort of the extreme case and that's that the predict project itself which looks at a very restricted number of pharmacogenes is viewed as a quality improvement initiative and so we actually tell the patients about it and we hand them a brochure but there's not an IRB consent form that goes with it. They can say, oh, we don't want to do it and you, of course, wouldn't do it. This, of course, is quite different. Dan, could you step by the mic so that you could? This, of course, is quite different because we're doing sequencing and we're going to find other steps. So there's a formal consent process that we have gone through and it's IRB approved and we have a consent form that says this is what's going to happen. Yeah, I just, I'm asked because I'm interested in trying to sort of, if you will, crank things down a little bit and my view is I've often felt that the routine pharmacogenetic testing, that is the interrogation for the known variants, is the most opposite thing to Huntington's disease testing that there is in genetics and that this is just trivial and I really think we can waste a lot of time with a lot of angst about risks and benefits and informed consents and risk to relatives. So who cares? Just do it. We agree with you and we're doing it, we're doing it exactly the way you would want us to do it. So we don't get informed consent to get creatinine to adjust a dose of a renal excreted drug. And we sort of look at the pharmacogenomic variants the same way but obviously what we're doing within this particular project includes next generation sequencing of genes that are on your list. So there there's a consent. That's different. But I think it's great that you guys have this more graded response and thinking about it instead of having the deep tendon reflex of oh my God this is genetic testing. Well the problem then becomes if you want to report what you're going to do then you have to go to the IRB and say now it's not quality improvement only, it's we're looking at a data set. Other comments? Yes, Adam. Just curious, I'm noticing the data you're going through from the predict study that you had clinical data decision support fire 24 or 24,000 patients and then you had 6,000 progressed to actually ordering. Have you done any assessments as to why why some of those aren't going through? Or is it music? One of the things we're struggling with right now is that EMR thing that says predict test up in the corner. Not you know every time people see it they look at it and they say you know it doesn't really require consent but it still requires a discussion with the patient about what's going to happen and so in a very very busy clinical environment we think in a very busy clinical environment we think that that's probably the major reason that people don't want to do it. I know that there are patients for example that I see because I'm in a referral environment where the algorithm will say fire and yet I will see that, I see that patient once a year and they get all their care at some other hospital and so it's not useful for us because we really want to follow what happens to these patients so I don't do it in those kinds of environments all kinds of reasons and we're looking into it but basically the capture rate is small. Great, any other comments? If to come. Just to mention in response to the earlier question that there's a site specific variation on the return of incidental findings in the context of this project and some sites are returning the incidental findings and others are not. That's part of the informed consent which varies by site. Great, thank you. So our next presenter will be Terry Minoglio who's gonna talk about James.