[MUSIC] I've showed you this cartoon before. And driven by this vision of being able to deliver genetic information to the healthcare system, that is important for drug response, but deliver the information before the drug is prescribed, is the challenge that we face. And that at Vanderbilt, we created in a program called Predict, whose mantra is to deliver the right dose, and the right drug the first time. So the barriers to this kind of implementation project are considerable. The major barrier right now is the level of evidence that is needed to implement. So, we've talked a lot about rare variance, we've talked a lot about variance with modest effect sizes, and there are examples of variance with relatively robust effect sizes in the drug response field. And so, that was the place that we could start with the delivery of genomic information. There are other variants that may have large effects on people's health, but they're either rare, or the evidence around the fact that they have an impact is modest. So, doing this kind of project required review of a lot of published, and sometimes unpublished data, around the relationship between genetic variation and response. There are other issues in the personalized medicine implementation space that are important, and I'll mention here, that impacted our Predict project to a somewhat lesser degree. There are issues around regulation of genetic testing, that's something that the FDA has become involved in, and the FDA is quite interested in using these approaches, but in the context of appropriate evidence. For some of the tests, not the tests that we were using, but for some of the tests, there is an intellectual property issue. Somebody may have a patent around it, or may refuse to allow other laboratories to implement testing, and that's a subject of active debate in the courts right now, up to, and including, the Supreme Court. Our project was done as a pilot project, within the Vanderbilt environment, supported by the medical center. But one of the goals is to create an evidence base that will enable a discussion around reimbursement, because at some level, we think this information is going to be useful. There are issues around privacy. People are very concerned about the use of genetic information in their electronic medical records. And issues around privacy, interestingly, if we ask people, and we did ask people in focus groups, what about the use of genetic information to guide drug treatment? And the response is not, don't do that, the response is, I thought we were doing that part already. That's something that people are very receptive to. The idea of using genetic information to predict disease or predict susceptibility to certain kinds of complications of disease makes people a little bit more leery. And then, the whole idea of engaging broad populations across the medical center, and across the country, and across the world in these efforts to relate genetic variation and other personal variation to health care outcomes is one that encounters barriers at cultural levels, at socioeconomic levels, and so we want to embrace as many patients as we can in these kinds of efforts. So, again, it is not possible to conceive of a system in which you would deliver preemptive genetic information into an electronic medical record to be used at a time when a drug is prescribed, for example, in the absence of an electronic medical record system. It just cannot be done in a paper records system, I really believe that. I can't conceive of a way in which we could do that. This is a screenshot of an actual electronic medical record of a patient at Vanderbilt. All the identifiers have been blacked out for privacy reasons, but you can see that there's a list of physicians, those are all blacked out, a list of diagnoses, a list of drugs on the bottom right, and then above the list of drugs and below the list of drugs allergies and drug sensitivities, is drug genome interactions. That's a list of genetic variants that have been discovered in this particular patient for drugs that they may or may not be taking. The Predict program currently implements five drug gene pairs, based on, again, evidence in the literature, and based on a platform that we're using for genetic testing. So the idea is, somebody is identified as a potential participant in the Predict program. We have elaborate ways of identifying patients, which I'll talk about in a moment, and once they're identified, they're not tested for the drug that they may get in the future. They're tested for variants that are relevant to that drug, and many others, so we develop an archive of data, and I'll show you why that's important in a moment. So as we were planning the program, the FDA relabeled the drug Clopidogrel, and actually added a black box warning, which is their most extreme form of warning, saying that CYP2D2C19 variants are important in determining response. And look at the last bullet point, it says, consider alternative treatment, or treatment strategies, in patients identified as CYP2C19 poor metabolizers. It doesn't say do the testing, and it doesn't say exactly what you should be doing, and that's one of the problems in personalized medicine. All they're saying is, there are data that the poor metabolizers have decreased response. Now that's incontrovertible. The question is, do alternate strategies improve that poor response? And do you actually go out and genotype absolutely everybody? So one of the parts of our Predict program was to identify patients who were at high risk for getting Clopidogrel or other drugs that are in our program within the next several weeks or months. Patients who are going to have a cardiac catheterization are at high risk for getting Clopidogrel, so that was one target population. Another population that we chased, was patients in the internal medicine clinic, who have a number of conditions that make it likely that they're going to get drugs like Clopidigrel, not over the next couple of weeks, but over the next couple of years. So a middle aged obese patient with hypertension and diabetes, that's the kind of patient who should have this kind of testing, we think, in order to use that information later. And of course, this is all a pilot program designed to asses how to do this at a first, it's what I call a baby steps. So these are the data for the first 13,500 patients studied in the Predict program and simplified, so there are 13,500 subjects, 361 of them have two copies of a loss of function variant in CYP2C19. The variant that is the commonest is CYP2C19*2, but there are also star three, star four, star six, and star eight, and they're also included in our assays. One comment that I need to make, is that the genetic information has to be as close to 100% correct as we can possibly make it. The analogy I like to make around this kind of information is, rotating on the wards, I will have a resident come to me and say, Mrs. Smith is in renal failure, because her serum creatinine is eight today. And I will say to myself, well, Mrs. Smith's creatinine was one yesterday, and people don't go from one to eight in a day. And besides, her electrocardiogram looks normal. And she feels fine. And so obviously, that's a lab mistake, do it again. And everybody, any clinician would recognize that. The problem with genetic information is that we don't have a context. So the genetic information has to be correct the first time, if we're going to act on it. That's a generic problem across laboratory medicine, of course. So out of the 13,000 patients, we have 2.7% who are homozygous, 19% who are heterozygous. The evidence there is probably to do something, but the evidence not quite as strong, and then 88% who have no variant. So, one way to look at these data is to say, well, if you’re going to genotype everybody who starts Clopidogrel treatment, you're going to do it in 100 patients, to find two or three who are homozygous. And that's one of the arguments against the use of personalizing therapy, based on CYP2C19. In order to get outcome data, you have to study thousands and thousands, probably hundreds of thousands of patients, in order to figure out what's going on with that 2.7%. The paradox of large numbers in personalized medicine I alluded to in the last module. Now I told you we do five drug gene pairs. So these are the data for five drug gene pairs and what's interesting here is, that when you look across all five, there's only 11% of patients that have no variants. So the way to look at this is to say almost everybody has a variant that could affect their response to one of these five drugs. This is a very simple experiment. And so you don't know what the variant is and you don't actually know whether that patient's going to get that drug. But it highlights the virtue of multiplex testing, because remember, the cost of this testing is the same whether you do one variant, or five variant, or in fact 200 variants, if you have 200 variants that are important. So the virtue of the preemptive testing is that you do all the testing at once, and then you're able to act on it for an individual when the time comes. And we think this is a model for delivering genomic information into the course of health care, and it will depend on the strength of evidence, it'll depend on the allele frequency, and it'll depend on the ability of the physician, or the health care system, the health care provider, to actually interpret the data. So what happens in our system, is that if a patient with genetic information in the chart receives a prescription for Clopidogrel, the system will look at if the genetic data are variant, then the system will deliver a pop-up. And this is what the pop-up looks like. I'm not going to read it all, it just says in red letters in the red box at the top genetic testing has been done, and you may want to change the dose, and it offers alternative treatments. And the alternative treatments may change as a function of time, as a function of the evidence that we accumulate, or it offers the option of staying the same. And one of the things that we're looking very hard at now, is how often this fires, how often do physicians pay attention, how often do other health care providers pay attention, and what are the long-term outcomes in terms of preventing in stent thrombosis, preventing recurrent myocardial infarction, because that's the name of the game. And it's too early to say much about that, but this is the dataset that we need to accumulate in order to answer that kind of question. I've been very Vanderbilt-centric, and I'm very proud of the efforts that we make at our center, but there are other centers that do very similar kinds of things. This is point of care decision support for Clopidogrel at the University of Maryland. And this is a warning that pops up at St. Jude's Children's Hospital in patients who are receiving, or about to receive, codeine and have the CYP2D6 ultra rapid metabolizer status. So this kind of coupling between genomic information obtained preemptively, and the electronic medical record system warning capabilities is being built at a number of places. And one of the things that we've come to realize is that there's a very important evidence review. That needs to be done for each one of these drug chain pairs. And, so, one of the tasks that the Pharmaco Genetics Research Network, another network sponsored by NIH, and the Pharmaco Genetics knowledge base at Stanford have taken on together with investigators within the PGRN, is the Clinical Pharmacogenetics Implementation Consortium and basically the idea of a consortium, which publishes its results, and those are widely available now, is to ask the question, if you have these data, what should you do with them? It does not pose the question, when should you get them? That's a different question. But if you have the data available to you, how should you act on them? And the idea is on the left in general, you assign a genotype, and then a combination of genotypes called a diplotype, then you make a guess, not a guess, you make a prediction of what the phenotype, the drug response phenotype or the drug metabolism phenotype will be, and then a recommendation based on that phenotype. So an example is a patient is discovered to have CYP2D6*4*4, star four is one of the most loss of function alleles. The prediction, therefore, is the patient is a poor metabolizer and the prediction therefore is to avoid codeine, because that drug will not be biotransformed, and will therefore produce less than optimal anelgesia. So that's the idea behind CPIC, and it requires evidentiary review which is one of the major steps, and so this is an important step in implementing pharmacalgenomics. I want to close with a picture, and the picture is the one on the left. When we were implementing our Predict program, we started with Clopidogrel, We were looking very hard for a patient who has the CYP2C19*2*2 genotype, and this is her. She's seated next to her doctor, the doctor who actually put the stent in her coronary artery. And I love this picture because it epitomizes, for me, personalized medicine. It's not about large data sets, it's not about statistics, it's not about GWAS, it's about individual physicians taking care of individual patients. Now that patient has other diseases, that patient takes other drugs, that patient shouldn't take certain drugs because they're beyond a certain age. That patient has values of her own that dictate whether she wants aggressive, or not aggressive treatment, for certain kinds of diseases. That's all part of the personalized medicine equation. And then her physician can turn to her and say, well, for you, in addition to all those other considerations, it's important to treat you with these medicines because of your genetic makeups. The genetics is part of a personalized medicine relationship. And Sir William Osler, the great Canadian physician, said it well when he said, the good physician treats the disease; the great physician treats the patient who has the disease. That, for me, is the epitome of personalizing medicine. Now there are lots of people who are interested in the idea of personalized medicine, I just want to close by giving you a sense of the kinds of resources there are around the web, mainly, to access further information. One of the things people are very nervous about is using genetic information to discriminate in terms of healthcare delivery. There's a law that was passed by the Republican and Democratic House of Representatives and Senate, signed by President Bush, in the late 2000s, in 2008, called the Genetic Information Non-Discrimination Act, or GINA. And GINA prevents discrimination based on genetic test information in terms of health care. There are other issues around GINA that make it not the perfect law for example, you can still be denied long-term health insurance, which of course could be an issue for some genetic conditions, and it specifically excludes federal employees, strangely. The FDA has become very, very interested, and has issued a large and relatively thoughtful White Paper on personalized medicine, and how they're going to try to regulate personalized medicine, and how they're going to try to help move the field forward in a rational, evidence based way. It's called Paving the Way for Personalized Medicine, freely available on the Web. And then there are many other sources, I've already alluded to the idea that the National Human Genome Research Institute has many programs, and they are posted on their website, there is something called the Personalize Medical Coalition, which has a course called Personalized Medicine 101 on their website, there are many places you can read about that. And then every disease that I've talked about, the rare diseases where there's a big genetic story, has very active web presences, and very educational materials around their particular site. I will highlight the Cystic Fibrosis Foundation, because when I talked about the idea of developing Ivacaftor for this one particular rare variant in the Cystic Fibrosis Transport Regulator, the Cystic Fibrosis disease gene, it took accruing patients from many, many, many physicians, many sites, to develop enough numbers to show that the drug works, and that was really enabled by the Cystic Fibrosis Foundation working with physicians who take care of the patients, to accrue the right number of patients. So they've been very active partners, in not only public advocacy, but in actually developing new therapies. And I'll close with this slide which I've shown before, again, emphasizing the idea that we now have the opportunity, by looking at large data sets, to identify patient characteristics, and that includes genetics, which I've spent a lot of time talking about, and other kinds of characteristics that drive individualization of therapy. So the future is very bright. There are lots of problems that need to be overcome, and the important thing about what I've talked about in this module, is that we're attacking those problems. We're trying to develop methods, trying to learn how best to do this in a way that helps patients, helps physicians, and actually controls costs as opposed to driving them up. [MUSIC] [APPLAUSE]