 Thank you. I'd like to thank Terry and co-organizers for organizing this important meeting and for Dr. Sheer and Chang for doing such a beautiful job at setting the platform and basis for our discussions today. I would see my task over the next 25 minutes or so really to frame what's already been discussed and maybe provoke you in a way to really think deeply about what we can do here and what is the way forward. So if we had our wish list about the SJSTN five-year vision, what would that be? I mean in an ideal world we'd be able to understand the immunopathogenesis and this would probably provide a roadmap not just for SJSTN but other immune and inflammatory diseases. We'd like to have a solid prediction and prevention program that is actually not only published but actually effective in clinical practice and we'd like to be able to measure decreases in morbidity and mortality associated with not only the human mortality but the death of drugs and clearly having a robust global pharmacosurveillance and collaborative networks is something that really the time is right to build this. And I say last and not least because this has come up I think already in the discussion, education is extremely important and nothing really happens globally without appropriate education and all of us in the room that work in the ADR space know that not enough space is dedicated to ADRs in medical school, in undergraduate science programs and we deal with physicians that are in clinical practice that don't understand ADRs, they don't recognize ADRs and they just haven't had enough background in ADRs and SJSTN is no exception to that so we need to have better mechanisms for actually disseminating information to providers. So what are the unmet needs currently? And I'm going to go through these in some detail but we've already heard the problems of actually even though this is a very dramatic disease about defining the phenotype and immunophenotype but importantly being able to assign causality of a specific drug. We've heard that it's very important to be able to collect samples and DNA and biological samples on these patients but this is something that's still in its infancy and we need collaborative networks that are representative across ethnicities, pharmacogenomic studies, there are some challenges regarding those and I'll speak to that when Hung and Neil have really nicely outlined the immunopathogenesis of this disease but there's much more to be done again that can give us broad insights across hypersensitivity and then there's issues with regards to management, prediction and prevention and above and beyond all of that really to make any of this happen there needs to be huge, huge capacity building. So challenge number one, defining the population. I've already mentioned that education of providers is important because people are not reporting these diseases, they're not recognizing them potentially. Pharmacosavailants has a reporting bias and is incomplete and often has been driven by industry reporting of newer drugs and many of the older drugs in FDA and other databases may be underrepresented. There's problems with the way we code reactions and this is going to be talked about later today but clearly the coding systems in electronic medical records and records in general, they're not geared towards identifying diseases such as SJST and they lack sensitivity and they lack specificity and they're just overall pretty lousy. There's huge challenges in retrospective causality assessment and an enormous amount of infrastructure is needed when you're studying less common diseases to build these collaborative networks and this is a paper actually that was published by Neil's group now some time ago, really to crystallize there's this huge gap in terms of actually what's actually reported in terms of pharmacosavailants and what is actually occurring. So this was actually comparing cases actually surveying burn units across Canada and 14 I think of them actually responded so only 64% responded but there was still a huge gap identified between the number of TEN cases that could be identified from burn unit reporting over a five-year period versus what was in the equivalent of the FAIRS database in Canada, the CADR MP and so really it was only capturing at best 10% and if you actually looked at a hospital discharge database this would go down to actually only 4% realizing that the ICD-9 codes is so terrible that that was not going to be a good representation. So under reporting is a serious problem if we look in FAIRS more recently and I thank the FDA for providing this data but you can see I would only make a couple of points here I mean many of the culprits here the ones we're expecting to see but then you know under number six we see a set of minifins so what's going on there I mean who in the room believes that a set of minifins is the number six cause of toxic epidermal necrolizes does anyone really believe that. So what happens in these databases is they're very important for calling data but they're not sensitive or specific enough to look at causality and you can imagine a disease like SJSTN which we've already heard has profound systemic and constitutional features like fever patients are going to get prescribed a lot of acetaminophen in the early stages of that maybe even before the spots come out so it's this propathic effect that we're seeing with a set of minifin even though there may be a tiny number that are actually real cases and then if we actually look at formal causality algorithms which have actually been developed we can see a number of things stand out and this is actually a causality assessment called the Alden score which was developed by the EuroScar group that really hones in on some key features of SJSTN and why do we need a specific score for SJSTN well first of all if you look at the delay from the initial drug dosing it tends to be much shorter for SJSTN than other types of or many other types of immunologically mediated drug reactions like dress and this causality takes that into account so most of SJSTN will occur within a four to 28 day window and if it's significantly outside of that then you start to wonder and the drug is not a drug that is has been associated with SJSTN as outlined here in terms of actually the likelihood of the drug present then you really would start to think that this is not going to be a probable or likely reaction these types of things like D challenge, re-challenge they can often be hard to measure and they may be predicated on things like the half-life of a drug so if you have a very long half-life drug like nivarapine someone gets toxic epidermal necrolysis the D challenge may not be that useful in that case because the horse is out of the barn essentially. Just to make a couple of points about reporting and pharmacovigilance as well this is actually from the Canadian adverse drug reaction databases is actually a sample of the report that exists and I would just make the point that in some cases it can be very challenging I mean this is a patient that was reported to have toxic epidermal necrolysis but there are three suspect drugs and all of those Bactrim, Dylantin, phenobarbital all of those are drugs that actually could cause toxic epidermal necrolysis and in terms of the duration of therapy it's probably a little bit short for Bactrim but look I mean it's really within the window period that you can see so there are complex cases where from a date from a pharmacosurveillance alone and from a ADR database we won't be able to tell with precision drug causality. Here's another case where there's two suspect drugs Advil and clindamycin but you can see the Advil had been taken and perhaps off and on for almost a year whereas the clindamycin was more recently prescribed but they're both listed in as being suspect drugs and we would obviously have a very low suspicion for the involvement of ibuprofen here. So just to move forward in terms of the second challenge in terms of biological samples again along the lines of what I've previously said robust phenotyping is needed because we need to make if we're actually ascertaining drug causality in looking for immunopathogenetic and pharmacogenomic studies we need to actually have robust phenotyping and causality assessments so this could include for instance not only the clinical phenotype but an immunophenotype so some in vitro x vivo or in vivo measure that there actually is an association with that drug. There's also currently few examples of electronic health records paired with biological samples although that's growing and and Josh Denny from Vanderbilt is going to talk about that a bit later but we need to actually be prepared to help well-crafted sources because often so often you get just very valuable cases of material and if you're not it's very hard to get this material retrospectively or to look at things as to what happens with drug responses over time so paired samples time samples tissue specific samples those are all important and and the very nice data that when hung showed is that these these responses these T cell responses in SJST and are largely tissue specific and so the actual TCR repertoire the T cell receptor repertoire that you see in blister fluid is actually going to be very richly representative of the immunopathogenesis that in PBMC's 20 years down the track may not be so so getting the appropriate samples early is very important in terms of pharmacogenomic studies again we need robust phenotyping but but not just robust phenotyping appropriate reference and control populations and I think what this actually is is becoming less clear over time with population admixture what actually is the appropriate reference and control population and how do we define race and I think there's a lot of now discussion back and forth about that doing genetic pharmacogenomic studies there are other ways of defining race other than self-identified race and these are probably going to be more precise and more precise markers of actually defining the appropriate race match control populations and actually simply a a large reference population which can give founder effects especially when you're only dealing with a small number of patients with a given disease like SJSTN and pharmacogenomic studies also ideally should provide a a robust roadmap for translation as well as providing a platform for understanding immunopathogenesis and I show this slide this is an old slide with only two digit HLA typing but really just to give you a feel that many of the actual HLA HLA serotypes actually exist all over the world but there are there but the diseases that we see that are HLA restrict are going to be driven by the population prevalence of this so you've already seen with B-1502 and B-5801 that all appear in all and carbamazepine SJSTN or prevalent in Taiwan but but those HLA restricted diseases not so much in North America and Abacavir and the association with B-5701 would be an example of a Caucasian largely Caucasian disease and the other issue with pharmacogenomic studies that's very important and I think this came up in the initial question period is how many cases do we actually need to establish risk if we've got a disease like SJSTN that is that is uncommon but the the effects tend to be profound and the odds ratios tend to be high how many actual cases against a control population would be need to define an effect and this is actually nicely showing you from the studies that have already been published that these diseases that have high odds ratios like Abacavir allopurnal and carbamazepine that actually the numbers needed are reasonably small to identify an effect in fact with Abacavir there was actually 15 a total of 15 cases against a population of 200 controls that were actually needed to define with high odds ratios a very strong association with B-5701 and Abacavir hypersensitivity in terms of actually common common themes in pharmacogenomics and this is unpublished data from our group in terms of all of our this is actually all of our serious T cell mediated drug reactions across phenotypes and I guess what this really shows is that although there's over over 5000 HLA B alleles the same ones are coming up over and over and over again in association with these reactions so we're seeing and and if you look at this list if anyone does work in the infectious disease space you'll notice that a lot of these alleles that are associated with bad drug reactions are also associated with protective effects from serious infectious diseases which is fascinating and it's probably not by accident but we are seeing across our populations and across different phenotypes that these are the alleles that's the HLA B alleles that seem to be coming up over and over again across different drugs and clearly in a predominantly Caucasian European population so in terms of in terms of road maps I guess HLA B-5701 provided a powerful translational road map for discovery of a marker and translation of that into routine clinical test testing but then it didn't stop here after the clinical testing was adopted there was a rich science that continued to evolve that has given us profound insights into the mechanisms of drug hypersensitivity and so I would propose an HLA genetic SJSTN translational road map where it actually continues on a common pathway until the identification for instance of a risk allele an HLA allele or a risk factor but then there's a divergence that if it's actually a strong effect with a favorable number needed to test to prevent one case ratio that this actually could be translated into clinical practice with excellent safety ratio and cost effectiveness if there's 100% negative predictive value to prevent future cases of SJSTN but in parallel to this is really understanding more about this process more about the immunopathogenesis the structural biochemical and functional relationships that we've learned about vis-a-vis a vacavir in the altered peptide model but then being able to understand how drugs actually interact with HLA in the immune receptors and the types of drugs that are actually interacting with HLA in the immune receptors to create these reactions because there are common themes across these drugs they're often aromatic amine derivatives and then being able to develop preclinical predictive models that actually can inform drug development and design so it's like a parallel pathway at several steps along the way including preclinically we can actually we should be able to pull out risk early and actually define what the risk is and then work backwards to define what are the structural elements of those drugs that are actually that are actually driving these types of reactions what are the specific structural elements and this can be quite challenging because we often see quite subtle differences in drugs that actually define devastating effects I think in the 1960s there was a drug called ibuphenone that caused terrible hepatotoxicity and it's only one methyl group different from ibuprofen which we know actually quite uncommonly causes true hepatotoxicity and there are other examples where where subtle structural differences make huge differences in terms of the propensity of a drug to cause a serious adverse reaction so immunopathogenesis I've talked a lot about this but again we can get more insights from in vitro and in vivo studies and broader issues are at play here as Wenhang said this is also a fertile platform for defining therapeutic targets so for instance if we could target granulicin without severe infectious morbidity or at least protect patients infectious against infections while we're actually you know targeting some important target for SJST and this would obviously be paradigm shifting and then also predicting which includes pre-clinical prediction as I've already spoken to we know with reactions like a back of your hypersensitivity and and Stevens-Johnson syndrome toxic epidermal necrolysis that the immunity seems to be long lasting and this is not this is important because it's not like this for all T-cell mediated drug reactions dress and other other hypersensitivity reactions they seem to for for reasons we don't understand their immunity may not be as long lasting but this is an example of in a back of your patch test and in vivo test where a back of your is put on the skin becomes positive and patients with a back of your hypersensitivity 10 to 15 years after that initial reaction and you can see these are gamma interferon over night le spots again showing the same thing at several time points several years after the reaction we're still seeing robust immunological responses and with carbamazepine the same thing happens this is a patient that was nine years post carbamazepine TEN and again robust ex vivo early spot gamma interferon early spot responses greater than 17 years post SGN so the question is what is actually maintaining this long lasting immunity and I think that's sort of a question for future discover we you know we've got some theories as to why that might be in terms of cross reactive memory T cell responses with with chronic prevalent pathogens and this is a model that outlines that that so called heterologous immune model that would not be mutually exclusive to any of the other models that we've discussed such as hapten PI or altered peptide so I'm just gonna I'm gonna need to wrap up here but management I think will be discussed by other individuals but it is a huge problem because there's no randomized controlled evidence there's very little evidence to suggest even that that any treatment intervention steroids whatever you want to define actually has any incremental benefit over robust supportive care in terms of prediction and prevention I've talked about the translational roadmap and some of the hurdles on a on a population basis and the fact that there are common drug structures and common HLA associations the actual fact of whether something does get translated into clinical practice actually depends on a number of properties of the test the drug the drug toxicity and if we've got a drug like alapyrinol for instance where there might not be another alternative or the alternative likes a phoboxostat is much more expensive or causes hepatotoxicity in three to four percent that may have a different connotation than a drug where there's several other drugs with with better safety profiles that don't need genetic testing this is the so-called gap between a high association and what might work in clinical practice and it just basically identifies it just because a drug has a high odds ratio it won't necessarily you still may have to screen thousands and thousands of patients to actually prevent a single case and this becomes infeasible just like flu clocks is still on where there's a strong association between B5701 and drug induced liver disease but you'd need to screen almost 14,000 individuals to prevent one case this is something that there was already a discussion about it sees Kwan study from Hong Kong and again it just identifies effectiveness efficacy gaps that the one thing we don't want to do is shift practice away from from actually from from from one unsafe practice to another so if carbamazepine causes SJSTN and a screening test is available then that screening test ideally should be used and acted upon and has already been articulated what happened in this study was that that practitioners got the test back and maybe even before getting the test back they had already prescribed phenatone and so there was actually an increase in phenatone SJSTN which offset any decrease in carbamazepine SJSTN and you can see that represented in terms of prescriptions on this on this graph here and so we really would want to actually provide providers again with the education and confidence to to do that I think just to offset this with the abacavir example really the reason why this worked with abacavir why abacavir was adopted reasonably early and was effective is the fact that practitioners could reality test it worked so carbamazepine is a diffuse drug it's prescribed by a lot of different people TEN is rare it you'd have to you'd have to treat thousands and thousands of patients to be convinced that you are preventing SJSTN in your population a busy HIV practice within two weeks you would start to see that abacavir hypersensitivity is going away through screening so that's what this happened here in these observational studies you just saw it disappear so I just wanted to perhaps end and maybe we can move into the discussion period now to think about how are we going to achieve these objectives with SJSTN think of this from a SWAT type analysis thinking about strengths weaknesses opportunities and threats and you know really because you know for those of us in the in the room that have been in this field a long time sort of it's like wheels and roundabouts I think we'd really like to be committed to a strategy that would move things forward as a timeline a strategy what what can be done to really make something effectual in terms of translational SJSTN research you can see from the preceding talks that an enormous amount has been achieved but it's really a question of sort of having a vision and a timeline moving forward so I thought I'd just open it up for discussion Terry if that's okay well we might have some questions about your talk thank you very much Dr. Phillips I have an eighth question how come that there's so few mhc type in in european cases there's almost none is there there's no association or the work was not done most of the data is on asian patients right in terms of in terms of mhc association yeah class one association for for which drugs well any drug I mean this I think I think there's been a lot of work across different syndromes I think the associations the fact that anti-convulsants aromatic amine anti-convulsants were you know one commonly prescribed in asian to the allele that is actually the urisk allele is actually prevalent in asia 10 to 15 percent it was just a convergence of of those two factors but the same was seen in european populations with a bakavir hypersensitivity where that association you know 5701 one of the first associations between a drug in an HLA class one allele was defined in a predominantly european population so I mean it somewhat has to do with the facts that in early on you know for for drugs and for you know devastating diseases like sjstn it was recognized that this was more common in asian populations in southeast asian populations and that is a huge clue when you pair an immune reaction with a racial predilection that is a huge clue that there's going to be an HLA association and I think once that was recognized then more and more of these studies were being done and there have been you know other GWAS studies like the flu clocks is still in 5701 that clearly again that was a european story yes Steve so I think that's true not only for for drug related diseases but for other diseases that are clearly HLA associated for example celiac disease and dermatitis and pediformis associated with that with B8 you don't see B8 in Japanese you don't see the disease in Japanese the typical disease I enjoyed your talk very much many of the points most of the points I agree with I think the most challenging and we've talked about this through initiatives that that Terry has led is the penetration into the educational aspect the educational dimension of what needs to be done people just this is not in the arm and turn of people so the penetration certainly has to be at the medical schools and the medical centers and at the residency programs because it's much harder as you say when you have a diverse population using tegritol much harder to educate them when you know that the frequency is so infrequent so one of the challenges I think is really the penetration and this is true from in many areas of medicine but this in particular because this was not a part of the arm and tearing of most of the practitioners at least in this country for generations so it has to be some approach and maybe Terry wants to talk about it but there is an approach in terms of education of physicians in terms of of the use of genetics well maybe maybe I will comment on that if you don't if you don't mind and and and actually I'll might refer to mark in my in my response one wonders how many things one can stuff into the brains of medical students residents and physicians and a lot a lot they are they are indeed and and then how many of those that they'll actually retain over over time and and while it's it's wonderful to see Neil at your hospital I was amazed as well that the residents were saying you know this person should be tested for this for the salil there may be maybe automated ways that we can get at this within you know smaller medical care systems as as is being you know done within Taiwan if you can flag that somebody has this variant and and if you had a system for sort of universal prescribing and dispensing of drugs where this drug was not given out until there was a clearance that the patient didn't have that allele wouldn't that be great now that's not a simple thing to do I think if you put it on the board exams it would get into the medical school I think you're right mark do you want to comment on that yeah resistance is futile that that that educational approach has been there's plenty of evidence to demonstrate it won't work and we know this we need new ways to do it which you know involves leveraging electronic health records point of care just in time education I want to try and tie up a couple of different things and bring it back to a point that I think is really critical in terms of research gap that we haven't addressed at all so one of the reasons why this is challenging in countries like Canada and the United States is because the allele frequency is so low that there's really not a justifiable case for saying that you should be doing screening and you know the point that was made I think by Elizabeth was that you know you can't you don't know necessarily what somebody's ethnic background is just by either taking a history or or looking at them and so we don't know who in this room is carrying these risk alleles and so what that raises the question for me is is what is the role as we move into the world of sequencing as opposed to one-off testing can we in fact solve the problem that we currently haven't solved of using massively parallel sequencing to define hla types because if we did that it would really resolve a lot of issues as we move into more and more sequences I mean in our system and geisinger you know within two to three years we're going to have 250 000 exomes on people and if I can get hla information out of those exomes then I can know before anybody prescribes a drug who's at risk can set up the best practice alerts within the electronic health record to let people know wait a second we think we should do something different here we can educate them at the point of care related to that and we can then begin to answer some of the big questions that we can't answer now which is you know what is the real risk because if we can do what josh is going to present later which is electronically phenotype for adverse drug events and we have these large collections of data that we can then we can do the association studies way easier so but this is the rate limiting step right now is we can't apply massively parallel sequencing to define hla typing so I would identify that as a very high priority for a research agenda to solve that problem because I think we could move forward a lot faster and just to throw in a few comments and I wrote them down so I wouldn't forget so I have to read them but just about the educational aspect if we did have a system however fast it turned around etc that was in hospitals whether in patients or in academic clinics at least residents would be exposed to the the question and the relevance of testing or not but I think Elizabeth's right after you see a lot of negative tests even when we started talking about anti-convulsant toxicity in the 1990s neurologists would say oh I've treated thousands of patients and I've never seen that there's lots of places in the world I've never seen but I believe they exist and it just always bothered me that that was the attitude that is a challenge the negative impact of testing is quite strong that one you've perhaps labeled a person and we now know that penicillin allergy is not a benign diagnosis people end up with all kinds of drug resistance and well proven so we'd like to disprove that are we creating a new group of so-called penicillin allergic people by saying you have a certain hla gene on the other hand there is a safety aspect and I would question and maybe you mark like if you were going to get carbamazepine but there'd be a little bit of you inside you you just wouldn't mind being tested just yeah so I mean he's nodding so you know I think it's something that if you sort of know enough you would probably want to do it but it then comes down also to this increased complexity and cost and patients would say well can I get this so what drug could you give me where I don't have to be tested it'll be cheaper it'll be faster just give me that one then and what they're doing is wandering away from the known universe into the unknown universe saying no I'd rather be in an area which is basically where you get to with natural products is the world of the unknown you say I want to get closer and closer to the unknown more information is scary to me I don't want more information and there's a real aspect of that so just my bottom line for that what we had hoped for many years ago when we started doing genetic and vitro studies was that if you had the phillips list there of hla b genes and you could just be tested for that at birth and then we learned how to manage that information it would be done let's say and and you would know you say okay well look here's what we've learned or you follow those people or cohort of those people in a control group and and you see who's better or not but there may be ways of doing this at a population level that is not impossible single sniff in down the road from hla b for alleles like that it's very simple if you had a list of alleles you know you could very simply without typing hla identify the people who are very likely to have b5701 for example I don't know if there's one for 15 oh two who's at least 16 sniffs is my understanding yeah I mean I think that's that's a good I mean we've done um you know some work with that at Vanderbilt recently in terms of imputation and you know I think in particularly in European populations and increasingly more racially diverse populations it seems to be um it seems to be you know a reliable strategy I think I you know and I like the idea of upscaling sequencing approaches um I I think where we're going to be with this so I think you know the the sort of single alleles single disease associations are not are not going to be so common there's going to be diversity across races there's going to be complex associations and I I think really the future is still going to be and what ultimately is the immunopathogenesis of these reactions if we can actually identify the populations at risk through HLA realizing that there's this huge gap a large number that actually carry the risk allele will not get the disease what is actually explaining that and are there other are there other sequencing strategies like T cell receptor sequencing um that we need to be pursuing in in order to actually really define what the population of risk is across potentially different HLA alleles because you know with SJS TEN I think the the results across you know specific populations of being quite clean so far but if you look at nivarapine for instance nivarapine dress there's maybe about six alleles and a couple of protective alleles and it's really diffuse you know um and it looks like that's probably going to be the rule rather than the exception so I think we need to be ready with robust bioinformatic tools and uh and very sound and progressive science to be able to tackle this problem that was really interesting this idea that these the alleles that are involved in hypersensitivity drug hypersensitivity seen also to be involved in protection against viral infections and I'm wondering if you've given any thought to what you think the the mechanism could be that there's a relationship there if that relationship is real what would the mechanism be that an allele that's involved in drug hypersensitivity on the other hand is involved in protection against HIV? Yeah I mean well there's other differences with those those types of alleles in B5701 and there's been a lot of work for instance with B2705 and ankylosing spondylitis looking at early and restricted responses and there's evidence with those alleles that they do tend to sort of make they do have tend to have a restricted repertoire of peptide repertoire early on and tend to recognize more things as as foreign and that's sort of part of part of the control I mean it's a simplistic way of looking at it but certainly the angst spond story has has evolved in that direction and you know in terms of issues of immunodominates and what drives immunodominates with those alleles so I guess it's you know it's that's that's sort of one one theory the other theory could be that there are other sort of more prevalent pathogen effects that apply to both systems so that cross reactive memory T cell responses that help fight an infection also will be bad for a drug so you know for cross reactive for instance in theory cross reactive memory T cell responses to a herpes virus that may help fight HIV would not be good in the setting that of a development of a neoantigen with a drug online first and then you gentlemen so um I mean I think this is a really good you made a really good point that you know we're moving past sort of a simple single you know gene single phenotype and we need to get a little bit deeper than that when you were talking about sample sizes and the ability to use small sample sizes though you were really sort of focusing on that simpler case and I know it's just a matter of power calculations but can you talk a little bit to what types of samples and what types of designs would be needed to get at these slightly more complex questions yeah well I mean I think I think at the present I think it all starts with the it all starts with the with the phenotypes you know being both the clinical and the immunological phenotype and so the ability to actually obviously make sure the cases that you have or true cases obviously does narrow down on your sample size so you're not having to deal with a lot of noise outside of that but I think I think the the JAMA study that that Wenhang presented was a beautiful example of how you may not see you may you know the actual a reasonably important effect may not be seen until you actually hone in on you know on a different approach and in that in the case of phenitone it was very important to actually to have the GWAS information to actually identify the drug metabolizing genotype that actually then filtered down into into the HLA risk and I think so I think having approaches like that given that we're now in a in a state of paradigm shift where the so-called idiosyncratic dose independent reactions of the past are now dose dependent and and genetically predictable we have to sort of keep an open mind in terms of other effects that we might need to you know that that might need to be recognized so so there's several different layers of phenotype you know I guess and that last example is a good example where if you are actually hone in on the drug metabolizing population at risk and you're likely to find a lot more pearls buried in there in terms of the people that are actually at risk to get phenitone and it might not be restricted to B-1502 and southeast Asians and you know maybe Wenhang you can you can comment to that as well and because I think that's a very nice example in terms of the CYP2C9 star 3 group when you actually look at that group and you you see B-1502 in some but not all of them what do you actually what do you actually see before we publicly resolved there are many doctors in Taiwan they they did the prefer the B-1502 before prescribing phenitonein although the phenitonein the look not strong association but in in clinical many doctors because the information will be labeled in the packets of phenitonein so you saw that now many doctors they they they did prefer the genitalia of B-1502 for phenitonein but in reality is that the disease cannot prevent by only genitalia marker so if we want to have a successful formative applying to the clinic we need to know how is the transfer of the real link to to the drug specific association and and then also that there's a possibility of negative belief or the number need to be tested so there are several you should consider before we push our effort to push the doctor to to to do the genetic test before the prescription so phenitonein and after we we we found actually the between only part of the risk factor we are doing the dual study also find that if we comment to to earlier then the sensitivity so-called can protection of the phenitonein steaming joints you know they can as high as 60 percent although not 100 percent but I think it's good enough better than before we that's just a physical drug and take the risk to within patient device steaming joints you know I think there are several effort we need to to do it's not so simple it sounds easy but it's not so easy yes what I think I just wanted to go back to one of the issues that was raised by Mark and Neil and that is one of the concerns I have is how high the bar seems to be set for determining whether a test has clinical value and you know I toss the question back to mark if you found out that you were negative for any polymorphism whether it's cytochrome p452d6 or or around 1502 would that have value to you as an individual and if so how do we include the value that a negative test might have in the metric that's used to determine whether or not testing should be done or or more supported yeah this is a challenge because of course it really tests the paradigm of what we've traditionally done which is population-based assessments versus individual utility and you know the utility for me to know my 2d6 status and my hla status and my cip status would be quite high and I might be willing to pay out of pocket to determine that prior to utilizing medications the studies that have been done that have looked at willingness to pay indicates that people are probably willing to pay a couple hundred dollars for information around one specific risk and so that translates you know to I think what we heard from Neil about you know the cost at least for this specific test but if I had to pay out of pocket you know I would be much more likely to buy a 2d6 test because of its impact on a variety of medications as opposed to a 1502 which I'm highly unlikely to carry just on the basis of my race and ethnicity so but that's people have tried to move that into economic valuation and David may talk about this in his talk but how do we measure the personal utility of this as we do our economic analyses as opposed to always modeling things from the societal perspective of cost effectiveness but the implementation of that is very challenging now in the United States of course we're everybody wants their own thing we're highly selected for that in this country but ultimately that you know has also led to you know a per capita health care expenditures that are significantly higher than other parts of the country with significantly less value associated with them so we have to find some sort of a rational way to do this and that's why you know the appeal of sequencing is that your then your acquisition cost is really just an information retrieval once the sunk cost of the sequencing is done assuming it can be done at high reliability and the information can persist yes yes well we modeled the cost effectiveness for uh genotyping for 1502 in Singapore actually we learned is that the value was more in the negative predictive value and being able to tell people that you know you've got this disease you're going to probably need lifelong therapy and it's very little chance if you take carbamazepine that you'll have this reaction and that was really more the economic value there um I had a question for Elizabeth on um I was very interested that it's just a small number of alleles that predict um or that seem to be keep popping up and do you think that one strategy would might just be to have an HLA your HLA type as part of your electronic record I mean it's not that much information to store and keep or maybe I see Josh right nodding his head and I mean we store blood type right so um I mean just yeah I think as things evolve um you know I think that that might be one strategy I would hope that we would get to a level of sophistication where we can actually define what the gap is and and therefore provide information that's actually going to be the most clinically relevant possible because I think where we'll struggle with that approach is that clinicians and providers already can't handle a simple yes no um in in the medical record and there is not a hundred percent compliance even with very simple algorithms and maybe Josh will will talk to this a little bit with the predict models that have been set up at Vanderbilt um but I think if you put it it's just not an area that's out there in and widespread knowledge enough I think that uh that it would be difficult in its current shape and form uh to be to have widespread clinical uptake I think we've seen some of the challenges associated with with using HLA markers that way um you know I'd mention the back of your example where it was a reality test for physicians in in clinical practice and that's that's the primary reason that that took off if you start dealing with HLA alleles you know people can't keep the the nomenclature straight you know when we test medical students you know and they'll be saying you know I don't know if they've copied someone else's paper but they'll be saying it's beef you know 1702 associated you know though it'll be everything but the right allele that's that ends up being reported as being associated and um I think it's hard for people to keep this information straight because it's not part of their fundamental medical education at this point so I think we have to develop better strategies based on robust evidence and I I think although that would be kind of a a dream to have all of that information in there that would improve the safety of drugs I think it would be it would be a big ask currently to be able to get that translated in that in that shape and form I think that as the science moves along we'll be able to refine things um to a point that there'll be better informatics better reporting strategies better uptake strategies and better science to define what the actual risk is for for these reactions. Disclose your name and affiliation please. Thank you Juan Lertora from the NIH Clinical Center. This has been alluded to in the previous presentations but I wonder if you could expand a bit in terms of how the HLA B1502 positive individuals and the population that is tolerant to carbamazepine how is that population being leveraged in terms of further studies of uh uh tolerance and immune mechanisms that are protective relative to the susceptible population? Right well that's clearly a very important population and I think a population that um that you know that went on each one you have been avidly studying I mean it's not a population in in European and North American populations that we have a lot of access to we do have some access um and again a lot of the work that's actually going on relates to specific specific TCR responses and clonotypes the Wen Han has actually presented some of that very nice data where you could actually see in their initial paper a difference that fell out between positives and tolerance and then in their subsequent paper really quite striking results associated with a specific TCR clonotype in patients in in in blister fluid from patients with acute carbamazepine SJST and so that that may explain a large part of the gap but it's still there's probably still an underlying story associated with why we would actually see that and uh and and again so one area that we're actively pursuing is the is the um uh a different model that actually that the these types of responses to drugs actually uh you know that they can't that why do they occur so quickly you know the latency period for for SJSTN is pretty quick it's like from four to four days for three weeks pretty much you know we tend to see this on the on the first exposure um and we see memory responses that will last over several decades and so one model that we're actively pursuing is that there could be a cross-reactive response to a prevalent pathogen that's driving uh that's driving this and that clearly could also explain why there's differences in positive predictive value between drugs between alleles even between family members thank you and you don't have any idea what pathogen is right sorry you don't know which pathogen pathogen are talking about oh well i mean we've got we've got some important leads uh i mean the most that there's a large and rich literature on how human herpes viruses how the human immune system has has evolved um in conjunction with human herpes viruses which are clearly viruses that are laden my laden endowment within us and i think uh that we always think of these as being villains that cause you know that cause horrible diseases but in fact there's probably an evolutionary basis for the for the development of our immune system and why those group of viruses have have uh have evolved specifically to become human like because there is a within uh within the human suite of those viruses it is specific to humans in terms of the evolutionary stream so those are the best candidates but that doesn't mean that there can't be other other examples that would occur over time or that the actual sequence of infection that happens to us all which is very individual which is different between the developing and the developed world even in terms of when we actually get our RNA versus DNA viruses that that also couldn't explain some of the diversity and differences between populations yes you argue for personalized medicine but maybe we should look at the deep personalized approach which is uh why do we need a back of ear there are many good nucleoside analogs for hiv that that are very safe take for instance um flucloxacillin which shares this susceptibility gene we don't need that it's not available in the united states it's available in england right and uh someone could argue that that these drugs should be replaced and what we should focus our attention on is looking at uh an animal model or an in vitro system to screen for this type of susceptibility and then modifying the chemistry one of the problems with our anticonvulsants is that we're still using the anticonvulsants that were developed 70 80 years ago you know in fact phenobarbital I believe the FDA has declared phenobarbital ineffective right they've it's basically been marked as ineffective so perhaps that should be the approach which is a kind of a chemical one to look at how we can screen drugs flucloxacillin is is very remarkable because if you look at flucloxacillin and dicloxacillin you have to look at them a long time before you can tell them apart there's just one little molecule so f the fluorine and why is it so different or is it different does dicloxacillin cause the same injury that flucloxacillin does in susceptible people or could we shift to uh drugs that are safer so I mean I think there's a there's a little picture and then there's a big picture and the little picture might be thought of as the drugs that we have right now where we can approach screening but there may be several hurdles as you say to there being other drugs other approaches that that are make make those approaches difficult but then there's a big picture of how these same drugs have taught us so much about drug reactions and without these drugs the paradigm even the paradigm shifts that have occurred over the last five years would not have been realized so we are continuing to learn an enormous amount about these drugs that are causing problems like abacavir, carbamazepine, phenyto, and alapyrinol that will drive forward preventive and prediction approaches that can then inform drug development and design but without that we won't actually have an approach of how we're going to move forward with preclinical screening and strategies to actually inform drug development because as you as you've identified the the differences are subtle I mean I gave a couple of examples there's a lot of differences between drugs where subtle differences make the difference between a dangerous drug and a safe and tolerable drug so I mean it's uh but but our science to understand that is still evolving so we need to we need to put resources into actually uh moving that along uh with consistency. Yes, Dr. Pimomar. Just just to answer your question so you're quite right they are quite similar to ecoxilin and flucoxilin there are some cases of hepatotoxicity with ecoxilin but nobody's actually studied it because there's not enough cases to be able to do HLA association studies and to some extent you know it could be replaced in England in the UK but it's only historical really and unfortunately when you have a historical setup and so on and the drug is widely available everybody knows how to use it it's sometimes difficult to change clinical practice because of that but the frequency of that is one in five to ten thousand individuals it was just the we's of work together with the SAAC and daily myself and so on and we were able to collect that many samples to be able to get that uh that that uh um kind of analysis done with the GWAS just to the um also in terms of all drugs versus new drugs if you take the epilepsy example I don't think there's any new anti-epileptic drug which is any better than the other drugs in terms of efficacy and most of my add-on therapies um they're also even though they may not cause the hypersensitivity reactions that you mentioned that carb may has been does they have other problems which are even more difficult sometimes to actually uh deal with so unfortunately whatever you do there's an equal and opposite reaction to whatever you do so it's always important to be able to look at the whole picture yes over there some threads of ideas together um and maybe move back into the precision medicine mode as opposed to the depersonalized medicine mode so it was mentioned you know exome sequencing whole genome sequencing all these things are you know going to be relatively routine how do we do all this but um and does it make sense from various points of view but you know um if we think about just HLA a few tens of thousands of base pairs and I were going to choose one region in the genome that I really wanted to know something about myself it would be HLA and not just for Steven Johnson syndrome but for all of the other drug reactions we talked about all of the autoimmune disorders that are associated with HLA many other diseases for which HLA pops up on genome-wide association study scans so while about so easy to sequence there are uh longer read sequencing platforms that are coming around not so hard to sequence a few tens of thousands of base pairs and so when we think about what is the value you take clinical utility of these tens of thousands of base pairs not just for Steven Johnson syndrome but across all of the diseases I mentioned I'll bet you it turns into a very cost-effective exercise I just have three things I wanted to say first about the medical school education you know the p450s are really hard for people to get a hold of and there's just too much stuck to memorize so Dave Blockart's p450 card is a pretty simple way to get people to understand that I don't know if everybody has seen that but it took kind of years of understanding the different p450s and pathways of metabolism in which drugs go through them but something like that could eventually be used for for this also I wanted to say oh my name is Sally Yasuda and I'm from the FDA and we are very concerned with pushing people to the wrong drug for example with carbamazepine labeled for 1502 we don't want people to go to another drug that we don't know anything about and that is very concerning for us so I just wanted to reiterate our interest in that and the third thing I wanted to say is that um for a recent drug that we approved we imposed a post marketing requirement to um do a genotype test for serious skin reactions and I don't know if Mike wants to say any more about that but that's something that we are very interested in pursuing and one way we thought we could get at it and one thing I might ask could you just identify yourself we haven't been asking you great thank you and and other speakers who we didn't have a chance to introduce if you would please identify yourself it'd be very helpful could you tell us more about that drug that was just approved that you and why what the what the rationale was in terms of of asking for these post marketing studies particularly for skin reactions proved and we knew that it had serious skin reactions and um we were involved in the carbamazepine labeling of course and we've been very interested in in this area so we just wanted to have more information as we go forward so I think the study is 20 years to 20 years long you know in case there's a very small signal but we just wanted an opportunity to to pick up anything that we could so with these with these types of studies where patients have acute reactions um or maybe there's some mechanism of identifying patients I think what's really important um and what can speed things along considerably as if the right samples are collected at the time the patient actually has the reaction so if we can get a system in place not just for for pharmacosurveillance but actually for biobanking getting cells collected in cryopreserved um and and and DNA uh plasma etc at the time the patient has the reaction then we can actually you know with with the resources and and tools that we've got available currently we could probably get a pretty good idea of what the HLA association is or what the mechanism of the reaction is that's our hope and it's not just for Steven's Johnson sorry Mark but for Dress and other serious skin reactions and um we are asking for genotype collection sample collection and a limitation of this as you said is this is going to rely on um post marketing surveillance and reporting of adverse drug reactions to the company and okay the last the last question please we we have to stop I'm sorry one of the issues that was raised when you mentioned back of your as an example where differences of prevalence have an impact on the utility of the test very clearly between different demographic populations is the experimental challenge where not only is prevalence an effect on utility but also there's a sliding scale on the risk effect size of certain alleles in different subpopulations as was mentioned with 1502 in Europe for example so in a experimental design case control study where you pull samples and you look for a difference between the control group and the exposed group one of the problems is that you may miss out on a subset which is actually has a different kind of risk effect or threshold than another group which doesn't and so it goes back to the experimental design question of discovery how will you discover a risk market let's say an HLA risk market if the effect size is actually not always constant because you have a mixed group of patients who have different genetic interactive combinatorial effects or even non-genetic effects right and so what actually is your thinking as you talked about how many patients you need to make discovery with this challenge I mean I think that's that's sort of where you have more distal approach is where you actually have patients identified that have a well-defined phenotype and you've got drug tolerant populations but but I guess in an ideal world I would see things moving much more approximately and that would that would include having having only a few cases but being able to apply the the cellular techniques and sequencing to those few cases to be able to say with with reasonable certainty that this is I mean if you actually identify if you've got an allele that has a prevalence of like 20 in the population or something from that study then you're you're not going to be that far ahead but if you've got an allele if you happen to strike gold in that setting and it's very early in the development of the drug only few patients have had this you know terrible reaction and even only one or two cases if you happen to get you know if you if the same HLA allele was with confirmed cellular studies and then you may have an answer because the chance of getting that in two consecutive samples or two or three or four or five would actually be you know like highly improbable so I mean I think the the moving backwards to be able to you know I think with with sort of some machine learning bioinformatic approaches that we could potentially move things backward to early clinical development to start shaping what the real risk is a bit earlier before we have to get into population based studies because when you think about it I mean no one really wants to be in a position where you're actually doing a case control study because that means that 15 to 20 30 people would have had to have had a devastating illness to do that study of and ideally we would be sort of moving forward with more more intelligent approaches to be able to define these at a point where there's only very few cases so I think it's a different you know and this is obviously specific for SJST and if you've got a different phenotype of disease or a signal you know has been you know identified in a in the preclinical in in the pre-marketing phase rather sort of like for a back of ear where it was a prevalent syndrome so I mean it was identified there were a lot of cases it was there was a robust clinical safety program that made it still a usable drug but I think that I think the issue is when you know if we're going to try and identify there's two different questions I guess if we're going to try and identify HLA associations to drugs that are currently on the market where there could be an ongoing safety signal versus drugs that are currently in development and I would argue for drugs that are in development that the rational approach is to try and define things as early as possible with the with the science that's moving forward. So we are going to have a working groups at three working groups at 4 p.m. so we have to stop here thank you very much for all the speakers. And thank you Ricardo and just I didn't answer the question where the bathrooms are so if you go out out here to the straight out and then take the first left and they are on your left there we will start right promptly at 11 30 the four speakers for that session need to have their slides up and we'll see you at 11 30