Samuel Broder – The Human Genome: Sequence and Assembly Was the Easy Part

my great pleasure things with this our birthday brother my brother is cheap many clothes as a cry distinguished career went just to highlight towards through things it was appointed by President Reagan the National the founding scientists punks leadership the official channels litters hidden scientific this was my colleague in local for under full discussion corporations noticeable measure discourse medical informatics was inspired by my discussion about how is the sciences could involve clinical aspects any last things to mention is loud here’s Daisy nature’s are these researcher with the energy rom and I just I mystical policies thank you very much Soren it’s a wonderful pleasure to see you again and it’s a really great pleasure for me to to present to the class here and I hope that we can have a lively discussion and I hope we can have an enjoyable hour or so I’m going to talk about my own personal views on the human genome and they may not coincide with what more canonical representation of that that field might say so I’m going to present the viewpoint of someone who is actually primarily a doctor md person trained to initially to take care of patients and in some manner means healed the sick or prevent unnecessary deaths or premature deaths and that will be my perspective and that may not be the perspective that others would bring to this particular topic so the title for my talk and I had a little bit of difficulty coming up with it because in part it kind of hard to condense it I he classroom topic into a short sentence but at any rate I thought I would call it the human genome sequencing and assembling the three billion letters of code was the easy part and there hopefully some ironing that because at the time largely due to sore nin his colleagues we were able to sequence and assemble the human genome using whole genome shotgun sequencing and assembly a feat which was thought to be for practical purposes technically impossible so there is an irony in that title but that was the easy part is my point so this is sauron alluded to this way solera in the summer of nineteen ninety eight this was the entire company as a practical matter there were four strange or not strange looking guys assembled in the shell of a building on the Rockville Pike in rockville maryland about 25 miles from the White House and this was it this was the official picture and you can see Craig you can see me in the background hoping that no one would ask me any questions that would reveal my total ignorance of the topic I show this slide because this was the slide taken at the White House when bill clinton was the president in which we assembled in the PDF did the Treaty of Utrecht or treaty of paris or whatever comparable concord you wish the name in which the nih human genome project organizers met and we sang kumbaya and had a wonderful time at the White House and this was the celerra team that went into the white house at the time I’m showing this partially because it has a nostalgic practical value I’m also showing it because historically it is exceedingly rare to see Craig in a tie and it’s even a more rare to see ham Smith the Nobel laureate who helped us with many of the library’s involved in the genome project where art I in fact that’s the only time I’ve seen him wear a tie in my entire recognition of him and the guy a strange looking guy on the left is here I want to introduce later on I’m going to talk about the

concept of stage migration and I want you to remember the guy on the left lower left there as you look at it because it will illustrate a point probably more effectively than any other metaphor so this was the Human Genome Project announcement the solara part of the equation this was a wonderful lead science article and basically it was published in February of 2001 and and an entire new era was declared the human genome at least a reference genome had been essentially sequence and assembled in large degree and could become a major element for future research it was a very important event no doubt about it but one of the questions that comes up is well you have the whole genome how come you haven’t cured cancer or how come unit you see you fill in the blanks how come you don’t know how to prevent diabetes totally and how come you don’t how come you haven’t cured aids whatever the topic is and you after all you have the genome it is people use different ways of characterizing it and but they all have an element of this is a revolutionary paradigm-shifting event which it is but then there is a logical disconnect with you if it is such an important event how come we haven’t cured how come we’re not here sitting talking about something that has nothing to do with human biology and human diseases because all answers are in and I’m going to try to address that in part so you can see a series of articles that were written some written immediately before the genome was finally assembled but when there was a certain inevitability about the project and in it we made all sorts of predictions here’s one article that I found at random where we made comments and the question that comes for us today is how many of these predictions of the of what the human genome would mean have actually come true in the approximate nine or ten year period since that event occurred so here’s a quotation that I took from a another paper and and we can judge for ourselves or you can judge for yourselves well what the with how how prophetic these statements were so this was a paper that Craig and I wrote well let’s be candid I wrote it and Craig Craig made useful comments but this could be a good thing for Craig if you disagree with the topic which is which is important so our DNA sequence and his variation provide a special record of human evolution and the migration of populations this was written before the genome was actually completed so we will learn how this sequence varies among populations and individuals including the role of such variation in the pathogenesis of important illnesses and responses to pharmaceuticals we will localize and annotate every human gene and the regulatory elements that control the timing issue site specificity and the extent of gene expression and for any given physiologic process we will have a new paradigm for addressing its evolution its development its function and its mechanism this will revitalize medicine by identifying important new targets for prevention diagnosis and therapy so if this were a multiple choice do they give multiple choice exams at Brown do they give to false you keep do they give exams okay if you had an exam it had to answer this true or false and only true or false what would the answer be I would say it’s more true than not and therefore true and if you had to rank it by percentage accomplishment i would say significant percentage of what this says insofar as it says it has in fact had been accomplished in many ways and yet there is an enormous gap between what we need to do and what we have done yet and that’s where the some of the issues come up that I hope to cover today the reason we have a human genome is because Craig in part we have a human genome because evolution but we really have a human genome sequence and assemble the reference for discussion is because Craig and the visionaries he was able to recruit of which Soren was certainly one were able to do what I think was from my naive way of looking at it technically an exceptionally difficult if not impossible task at the time and probably even in retrospect whole genome shotgun sequencing assembly was at the outer limits of Technology and yet it was done so I mean Norma’s visionaries and craig Venter certainly was one of those I want to cover one special application that is in that arena I’m from celera genomics in the celerra corporation these are some discoveries that have been made since the genome was a sequence and assembled and in my opinion in consequence of its assembly so I’m giving this as the prelude as a foundation for where we’ve come and

where we might be going but for those who have heard me speak before and yet are brave enough to hear me repeat the exercise you will see that I always try to set things up this is obviously a setup in part I’m going to talk to you about positive things but then I’m also going to talk to you about why we haven’t done or so I’m setting up the equation here that yes we have done a lot so I’m going to describe one element and the idea is now that they’re there is a fully sequence human genome to deal with one can either use whole genomes whole genome Association studies of which you no doubt have had many examples in this class or you can use a slight variation which is the understanding functional genomics which is essentially trying to do an exon by exxon ree sequencing the genome and then focusing only on the coding regions and then looking to see if there are variations and associations you can find in diseases and by both techniques we’ve chosen the latter but by both techniques you can make discoveries you can make important discoveries no question there is no question to all discoveries that I think would have been difficult or impossible to contemplate prior to the genome i’m using this is an example so here is an example of a gene the kindness and like proteins the gene that encodes candies in like 13-6 which was not thought to be associated with heart disease and there’s a DNA polymorphism a snip in this kinesin like protein qaeda sins essentially like robotic proteins if you will that walk along tracks and carry cargo around and and there is a DNA polymorphism a single nucleotide polymorphism that is associated with heart disease and more significantly in my view is associated in many ways two responses to statins but not by necessarily the classical canonical lipid pathways that we’ve heard about in so many over so many decades so provides a totally new way of looking at the problem and perhaps offers new ways of understand the pathophysiology or even the new targets for treatment and this is a consequence of Human Genome here’s just about one example of this so this is a study in which one examined what’s called the prove it timmy 22 trial this was a study in which individuals who had acute coronary syndrome which were the purposes of our discussion here I’m going to simplify is basically either a heart attack or a cousin of a heart attack I’m simplifying but that’s acute coronary syndrome so basically this was a famous study that randomized people to an ordinary dose of a drug called pravastatin a kind of a statin versus an intensive dose of another statin called atorvastatin and what happened is that study clearly showed that intensive atorvastatin worked better than the ordinary common dose of pravastatin that was before the genome came in we analyzed that study using this gift six carrier approach and we found that individuals who carried one allele of the variant were under we’re discussing here the kif 67 19 are invariant so and we actually looked and we observed that all of the benefits all the clinical benefits we’re assigned to individuals who carried this variant allele we’re using clinical endpoints that is the highest standard you can do at a clinical trial and the clinical endpoints were a recurrent event death or something related to that another heart attack or need for East and restenosis repair or something and basically with these clinical endpoints you could see the curves pull apart in fact you can see that they’re actually starting to pull apart at three months by this lady’s clinical endpoints however if you look at people that are non carriers you don’t have a demonstrable difference so what you have here is a situation where a certain genetic endowment carried the whole population and I’m sure you’re familiar with that notion that it is possible for a particular benefit all to be too confined to a certain population and then depending on the size and the effectiveness that that population essentially can carry what you think is a total effect for the total population okay so that’s what we think is happening here and what we observed was that we could not ascribe the difference in benefit to a difference in the response to what everybody thought would happen which would be a difference in the degree of lipids or the degree of c-reactive protein which is a kind of

special if effectively a marker of inflammation so there were differences in that high dose atorvastatin did give a more significant drop but there were no differences in the between the two carrier states if you see what I’m saying it isn’t that we’re accidentally tracking some unusual response to the lipid profile we’re tracking something that is different we think it is related to a process called vulnerable plaque we think is related to a tendency to predict who is going to get vulnerable plaque rupture and the effect of pleiotropic effects of statins which have many effects including against inflammation and against certain processes that go on in the endothelium and we think that’s what we’re tracking we’re not tracking mere cholesterol drops or improvements along that line which is a kind of reversal of the normal cannon that people thought we’re not I’m not saying that cholesterol or lipid profiles are immaterial to this process but in this particular experiment where the we are finding that a drug atorvastatin high dose 80 milligrams a day seems to be working in acute coronary syndrome by an unanticipated mechanism and I don’t believe and that mechanism one assigns who’s going to derive all the benefit from statins in this context high dose tens it also provides a whole wealth of opportunities to discuss brand new ways of targeting these diseases so this is just a schematic why should something that normally carries cargo that’s what kind Neeson’s tend to do I would venture to say without too much fear of contradiction that virtually everyone in this room knows more about kinda sins than I do but I can certainly tell you this and eason’s we’re not big on the top ten list of causes are related molecular mechanisms heart disease before these discoveries and now we have a whole new method of dealing they carry cargo they’re actually need to watch a picture because they actually look like a science fiction robot with little I’m not making this up that walk along tracks that seem to have headless dimensions but like arms it will pick up something and walk on the track it’s really eerie to think that that’s going inside our bodies even now as I speak in as you’re listening very very very eerie but nevertheless some DNA polymorphism in this probably cargo-carrying region of this protein seems to be related to heart disease and seems to be related to it seems to be related to a mechanisms of how statins work in certain situations that please I’m not saying that high-dose atorvastatin doesn’t do something through lipid profiles but what I am saying is this doing something beyond mirror canonical lipid biology and that would have been very difficult or impossible so I’m going to switch gears now so I gave you my bona fides yes the human genome has yielded a number of things I’ve give you a concrete example the kinds of things and there’s lots of other things you can talk about the kinds of things that it made possible and but I’m now going to move into another area which is now the really dominant theme that I want to cover and hopefully some of you in this room are going to solve some of the problems that I’m going to identify and the problems are well it can be reduced to the notion of well why is it take so long to do anything how many of you are in a pre-med course or may intend possibly to become a physician someday one a bit more of you will do it but at least one I guarantee I guarantee you’re going to hear the following form relatives and friends as you go to medical school or in your early training why can’t we do more about this disease or how come we’re not doing there what supra and they’ll hold you personally accountable by the way just in case you I’m just preparing you no matter what your rent friends and relatives are going to come to you on the party and they’ll hold you personally responsible for the lack of progress okay so I have two observations from a doctor’s perspective that being me statistics as a discipline is not intuitive and it is hard actually there’s gonna be three observations but I’m gonna call us and then conditional probability is even harder it is really hard it may seem easy because no doubt many of you are facile and you can handle the issues but conditional probability under the right circumstances and the right presentation and in the right meal you can stump full professors of mathematics as Sauron no doubt has shown you i will show you and then there there’s this quotation it’s obligatory to have some quotation from Henry the fifth or from Hamlet or something or some Greek thing and I’m choosing to do this in these talks so basically there’s a lot more in medicine and statistics and biology than you might think and in particular many assumptions that take the view of being almost axiomatic not worthy of further discussion so

fundamental that they can’t be analyzed may be incorrect so that leads to some issues so statistics I wanna show this slide because none of you in this room is old enough to have actually seen or even heard of this kind of advertising I however do not fall into that category so I can remember that within my lifetime that basically cigarettes were advertised not only in the kind of way that we give you a warning smoke at 0 you know the way they do now the proforma smoke cigarettes and you will die so that’s it now please buy our brand okay that didn’t used to exist in fact the opposite was true the opposite was true there was a time when cigarettes were promoted as a health benefit or if not a health benefit one step below health benefit I’m making none of this up okay so what were the tools that were used might this is really cool they used first they used a symbol in there was a narrower believing that people respected Authority and there were certain figures governmental figures in particular that people believed in there were people that worship franklin roosevelt if you said a bad word about him you were going to get beaten up and people felt that way about government officials or other officials at high rank and there was a belief in authority figures which was capitalized by the cigarette companies at the time and at the time the highest authority figures that people accepted beyond almost anything were physicians and nurses so what images were used to advertise cigarettes class physicians and nurses here’s one example so they used doctors here my favorite was doctors would be in regalia they’d be looking over microscopes and they’d be smoking so what you see here is a doctor enjoying Lucky Strikes smiling the clear implication is he likes this brand and this is the reason I’m showing the slide other than just to take you down memory lane look at the statistic they got they gave you they give you five figures twenty thousand six hundred and seventy-nine they didn’t say twenty thousand seven hundred they give you twenty thousand six hundred and seventy-nine physicians say Lucky’s are less irritating and then you can’t read it but they actually had a can affirm that did the accounting and whatever and this was a form of statistical inference of the day which we find amusing and a source of merriment and Wonder but I’m not so sure that comparable errors if you will or fallacies could not be repeated in another context obviously not this blatant and not this with ludicrous but it is possible to come up with other various and so I’m using this as the prototype for people i love this your throat protection against the initiation against cough irritation against cough so there’s actually a website that has a lot of these in there and they’re really it’s very interesting but statistics were used in this kind of unusual way to to identify and induce individuals to take up a habit that obviously was an extreme health hazard so then i present this now soren if you have any questions i’m not going to go into the problem here because i’ve learned that it will divert the entire talk but let me just show you this for those of you who believe that you can master any problem if Sauron hasn’t already done so he should present you the what’s call the Monty Hall problem you heard of this no doubt it’s a game show you make a few little assumptions but basically in brief and I’m taking a certain number liberties the guest host will offer you the prize or gives either goats or nothing be is behind the doors and invariably he will never oh he will the Monty Hall omniscient actor will never actually or at least I never have seen him open the door if you win it correctly first time so you pick if three Doors you pick a door which has the prize it may be a car they will never open the door that you’ve identified say you won that’s it now go home what they will do is part of the game is you pick a door and then they will say he will open another door the other door being opened presumably at random is always empty it will not it would be a stupid game if he opened the door just showed you so and then you are asked at that point and assuming there’s randomness to this process you or then asked should you switch or should you should you stand with your original selection now I’ll just tell you what the answer is but please if you don’t agree with me will let it go the correct answer in the terms that I’ve just given you is that this actually is a specialized form of conditional

probability and no the decision is not random the decision whether to switch or not at the point and under the conditions I’ve just given you is a form of conditional probability and you should switch now at any one time to get philosophically you could say what does it mean at one time that you have a higher probability of succeeding by one strategy another in one case well okay but if you in a play of game 100 times if you did opted the strategy as I’ve just developed it and you switch you cut and you had a bet going with somebody you’d make more money you all find this just understandable nobody is outraged and wants to tell me that it’s can’t be that they’re all either yards have to be equal you guys are good because there are full professors of mathematics who get this wrong who write letters objecting that this is the wrong answer that you’re an idiot schools are failing and usually people come up with a truly violent response when they to this side but anyway the reason I’m doing this mainly to illustrate the notion that conditional probability is hard you guys can get it but a lot of people can it’s hard and many of the problems that we really deal with in medicine are modified forms of what I call a conditional probability saw in may not but to me they’re kind of conditional probability in that you have to understand certain premises and then one premise actually conditions another so I’m going to now describe the issues that we faced with certain types of clinical trials not all but certain types bear in mind in my definition at least genome-wide associations and functional genomics in effect are an example of case control studies right very specialized example does anybody disagree with that their case control it you could not in practical terms in my opinion designed as an exploratory study a randomized trial where you do you’re doing genome-wide associations on randomized patient populations and then I personally don’t think that that would be under some conditions it might be practical almost all genome-wide Association studies that i’m aware effectively our case controls you know what i mean by a case-control i don’t need to define that right okay so what we’re going to talk about applies mainly to case control studies of which you know what associations are part but their ply for much knowledge but and then some of these principles conceivably could also enter an even two randomized trials prospective randomized trials so not all but some of these so I’m going to go over certain topics that I find especially burdensome in applying the knowledge any scientific knowledge yeah some of you probably know much of this already you know so I’ll go fast but there are several types of bias and this is a short list that can enter and their fundamental and they’re really important to know and they’re not mutually exclusive so one of the biggest bias features that you have is selection bias which means you have differential access to a study population or you have an unrepresented nature of samples and this can include biases that relate to self selection and intuitively completely non-obvious things for example how many of you would find it intuitive quickly without doing a lot of thinking if I were to say that if you have as your controls people who are at a job site and working routinely at a job site fully working active and you were using some kind of endpoint survival mortality susceptibility to Z and you were taking as the control working individuals that that would introduce the selection bias I’m going to intuitively find that how many of you would find that if I said that that if you did that you’re actually going to buy a 20-percent artifact and survival just on that fact alone probably would be more how do you all find that intuitively obvious you do that’s pretty good because a lot of people in science don’t and they will use as a control for a certain population workers there is a defined well-established well validated well understood well characterized thing called the healthy worker effect in which it’s a definition it’s a kind of selection bias if you are actively working and you continuously work and you’re showing up for work every day you’ve already been eliminated from having certain illnesses right because the sick people got selected out so at any rate if you’re having a a case control study you have to know that fact and you’ll be able to correct for but you don’t know that fact and you’d be surprised how many studies are actually implemented to don’t know that fact you can run into issues you can run into issues related to many issues with parallel to that there’s something called the diagnostic or work up by us that is if you’re developing a new diagnostic test and somehow even in a subtle way that you’re not aware someone is using that diagnostic test to make a decision I’m studying whether a certain diagnosis or diagnostic tests and to

pick up appendicitis and I use that tests in some way that’s not obvious as I’m performing it and I use that to make a decision in real time as to whether I’m going to go in and do an exploratory laparotomy on people and I don’t realize what I’m doing I may actually influence the test and in effect what you do it is complicated way which you don’t have to get into you inflate the specificity your test in your diminishes you inflate the sensitivity and you diminish the specificity of your test you artificially inflate the sensitivity and that you don’t realize it okay so uh because you’re making you’re triggering the decision to go ahead and therefore you’re picking up more cases like your specificity you pay it to enormous penalty and specificity when you do that and a test must have both you follow a mean by sensitivity specificity if I wanted to say that that one waited one way to detect whether such and such a person test had a certain disease if I Ted if I tested if I had a test that was positive in everybody in the population that would be a highly sensitive test right but it would be specificity would be garbage right hey taking all comers is it gives you a hundred percent specificity you don’t miss anything right okay but it’s a useless test okay all right so that gives you these kind of issues there could also be unrecognized differential electrician or participant drop out or do the deaths or the things where you may not be you may not be tracking the records correctly there’s information or miss classification bias you can have errors in the measurement of a key factor outcome you could have subject variation observer variation you have a deficiency of the technology you use you can have recall bias this is a big problem when you are asking individuals to come in and remember whether something happened you can have and then that’s a cause of something you get a recall bias you can have reporting bias there are many different issues related to that if you are doing a study whose subject is a certain condition Association and if it relies on people calling in to tell you that there are themselves response to experience something you’re going to get a unusual kind of statistical by us on that right because the people that call in may not be the same thing as the population that you’re trying to they may have features about them that induce them to call in so that’s an issue there confounding bias is unfair comparison or distortion by some other factor there can be phenotype suppression by unaccounted or unrecognized introduction of a therapy if you’re trying to study whether something is can related to hypertension or not whether something causes hypertension or not but you don’t recognize that something in your population is being administered to people to block their hypertension and they don’t have that phenotype and you’re not aware of that your records aren’t able to record that or capture that or there’s an irregular usage pattern you make you can get error in that in that dimension as well my favorite when people arguing with me on this point is if you didn’t if you had if you had a a very everybody was on a diet that had extremely low phenylalanine in it everybody was on such a diet then you would not have picked up the gene for phenylketonuria you would not recognize it as an illness probably right or it would be you see the point I’m making if you in biology they call that the notion of an ox atrophic mutant right where you is essentially you correct the problem then and you in effect would provide the conditioned media that can actually correct the problem so so basically I want to talk about stage migration this is another thing particularly in cancer now remember I asked you earlier on to remember the picture with the guy that was in the extreme left so now I’m going to illustrate the problem if the guy that was on the extreme left the guy that was sort of poking into the camera trying to get himself in the picture had been asked to leave the picture before the picture was taken and the remaining three people g Myers and ham Smith and craig Venter were remaining you would see by that exiting you would see the average IQ would go up right if you understand that feature then you understand the fundamentals of stage migration now I will illustrate stage migration my experience when I give these talks is that people always not always but frequently say this is so obvious why are you bothering me with this and I’m bothering you with this in part because this is a mistake that is made over and over and over again and almost has the same pattern someone forgets this concept writes a paper about it and then there’s a series of comments and letters to the editor pointing out that

the person forgot this notion so maybe that era is over but I don’t think so what up dude you know what i mean by the concept of stage migration it is called the Will Rogers effect but I will not repeat why because it’s politically incorrect for me to give you all the different reasons but will rogers was one of the comedic American figures who first described this phenomenon he was not a physician but he essentially captured what this is what it really basically boils down to is let’s say you have I’m going to make it simple because cancer usually is not divided this way but let’s say you have local disease and advanced disease just two categories usually that’s not only two categories principle local disease in advanced disease as defined by a certain technology of the time supposing you then come up with a brand new technology an imaging technology that is able to see tiny tiny bits of tumors that have actually spread thereby rendering the patient advanced in it’s no longer at the local site anymore its advanced that’s what i mean by advanced metastatic it’s gone to another site but you’re now using a new technique and it shows certain people that previously would have been fought as having local disease only what will that do to the apparent survival features of the people who previously were it would have been categorized local disease what would that do to the new category of local disease those new people classify the new way what is it due to their life expectancy it improves it what does it do now that you are reclassifying those people and you’re putting them into the more advanced group what does it do to that group speak it pardon precisely right you now have a phenomenon in which you have actually improved both groups right but accomplish nothing you can’t cure cancer by fiddling with numbers what you are doing at least the cancer cell is unimpressed although people have tried to cure cancer that way it never works the cancer does not care if you are making statistical biases or not it will keep growing and keep injuring people so what we actually have is a phenomenon in which I new technology by redefining something in a restaging people if you want to look at it that way now creates an artifact if you’re not careful of effectively seeming to improve people that I have local disease only purely local and also because the new population of advanced disease but they’re really better than they’re really better than the in the old-style advanced they come into the new category and they dilute out the previous badness of the disease so both compartments now got better right but nobody got better to me sword may disagree this is a kind of specialized regression to the mean kind of thing you’re just you’re just deluding everything out you’re not changing anything you’re changing the mix but you so you’re you’re creating more of an average between the two without realizing it that’s my way of looking at and not as a statistician but that’s so you do not change the median survival of the population as a whole and you will not by having that maneuver do anything to the public health statistics of so how many people die per unit hundred thousand in an age adjusted way you won’t do a thing death rate will be the same and then people say well how can it be I had a neighbor that had lung cancer and they found that it spread to X Y or Z and they live 10 years well yes but it didn’t wasn’t the same thing and there’s a related come phenomenon called we have to discuss this called lead-time bias which is related in a broad way to what we’re talking about today which is if you develop a brand new technology and it could detect a cancer even a primary cancer much much earlier than a prior technology from the of your detection you will think that the person has lived a long time so for example if you say that people with lung cancer may live only two or three years let’s just make up that statistic if you have some new way that diagnosis and way way way earlier than you ever did before you may say those people ha they’re living for 10 years well not really you’ve just set the clock back eight years you have done a thing you’ve just discovered it in earlier phase so these are concepts that I think you have to keep in mind and what do they end up doing they end up making it much more difficult to come up with us evidence-based conclusion and they force you sometimes into the domain of having to do randomized prospective trials which are now and have been the gold standard but randomized trials important

though they are are slow and you can come up with any metaphor you want but it’s in my way of looking at it almost like I’ve invented the internet i’m going to send an email from san francisco to providence doesn’t quite go the speed of light because of all the changes and servers and everything but it was pretty fast so let’s say a few milliseconds a second whatever it takes assuming the servers are all working well okay then my technology involves that i have to get on a train and ship the message by hand to boston and then i have to put on the steamship and take it to europe and the dolt of a destination from my message now that may be unfair but those are the timescales not that off that the process of doing randomized trials yes there are improvements or technical improvements there’s this and that but if it takes five or ten years to understand an answer it’s going to take five or ten years and you can’t shorten that part of the equation with randomized trials so that’s their the gold standard but they’re very difficult in the processes so they do certainly apply many of these concepts certainly do apply to genome-wide associations everybody recognizes you can have false positive results and selective reporting these issues so type 1 errors are particularly relevant to the conduct of genome-wide associations when you’re searching hundreds of thousands of genetic variants you always have issues that could be by you can find false positive results you can do your bond for any corrections or whatever but still you can get into issues and because at the end of the day a genome-wide Association study is a case-control study and all of the biases that I told you and more that could route that can perturb a case-control study can confer turbo genome-wide Association study so those are certain issues that we have to deal with then there’s a different and evolving knowledge now much of what I’m going to talk about today is related I’m going to focus on prostate cancer but some of these in a variable way will apply to other cancers so let me focus on prostate cancer first let us define our terms there’s a practicable none of the people in this room who carry two X chromosomes need fear it those of you who carry only one x and by implication a y may get it at some point in your life it’s a male disease prostate cancer occurs there has been a huge effort to develop early screening technologies to catch prostate cancer when it’s early and in that capacity potentially prevent deaths from prostate cancer so how many of you like the notion that catching a tumor early is a good thing do you not trust me would I ask you trick questions so yes and no here’s where my definition of conditional probability comes in Sauron will not probably like this way but it’s might remember I’m a doctor noticed a decision I can I i can say i do anything I want really statisticians will clean up my mess it’s a kind of conditional probability in the sense that it all depends on what you believe and what exists as the predicate for the screening so let’s walk through this I’m focusing on prostate cancer the caveat that I give you some of what I’m going to say may or may not apply to other cancers on the other hand they might and they will therefore allow you a framework for asking relevant questions which will perhaps allow a more intelligent debate the notion that catching tumors early is a good thing is very well established and it’s it’s true in some senses of the term it is true you would like to catch a cancer early but certain implications of that philosophy if left unexamined lead to frustration and endless debates what we have to this day is an enormous debate about whether PSA screening prostate specific antigen screening is going to translate as a public health benefit to men okay now when I go to the doctor once a year I get a PSA to check it you the men in this room are too young but someday probably by the time you hit 50 the somebody’s going to start asking you about esa and i’m not in any way saying you shouldn’t get a PSA test but what we have as a screening test you know what i mean by a screening test you know the difference in the screening test and a

diagnostic test although the distinction is sometimes blurred a screening test is performed on a perfectly well individual without any clinical provocation other than certain demographics maybe age or something you don’t if you’re getting a test because somebody’s sick or is complaining of something and then you’re starting to work up that person as part of the test is not a screening test okay it’s a diagnostic so a screening test is something that I use the term you is done unprovoked you come in you get it that’s it what has happened is that there has been this endless debate about whether PSA screening really has been able to reduce the death rate from prostate cancer in this country or else other places where it’s been tried and the answer is there’s a lot there’s more heat than light being generated on this epic and many people quite fervently feel that PSA testing has not had a drop caused the drop of the death rate others feel that it it has benefits and I’m not going to get into that debate with you I’m going to let you define your own definition but let me tell you some of the parameters that make this much more complicated and make the notion that it yes it’s a good thing to catch cancers early and the end in the reason the point i’m raising here it depends it all depends it is in my definition a form of conditional probability all right so supposing we have a cancer let’s focus on prostate cancer for a moment because it’s these things actually may have some relevance and we’re talking about cancer can be microscopic localized to an organ regional spread and metastatic spread okay cancer progression on the ordinate on the abscissa is time all right so let’s do the next slide let’s l’examen tumor a and tumor be first tumor a tumor a remains undetectable and without morbidity during the patient’s lifetime okay and that by the way is my personal definition of something that’s indolent it will not bother you during your expected lifetime this is not to theoretically project that if you have lived if you’re expected life expectancy is 80 if you had made it 150 would have showed up that’s not we’re talking about we’re talking about you’re living a normal life and before a competing cause of death supervenes this thing causes a problem okay do we all agree to that definition okay so this is the tumor remains undetectable and without morbidda during the patient’s lifetime in other words yes you see a tumor there it’s little nests of tumor it hasn’t invaded invasions quite critical it’s in sight do so to speak is there and you can find lots of those in postmortem examinations of men when you look at their prostates little nests of tumor they’re not going anywhere they don’t appear do anything would screening be very useful in this category no could it cause harm yes not just economic harm because you might need to do an invasion or some kind of quasi invasive procedure or we’ll talk about this in a minute there may be certain operative procedures that follow tumor be tumor be grows until it becomes detectable but never really causes symptoms or leads to death in other words it’s there the PSA might be bumping up is this likely to be benefited by screening is it possible that it could cause harm if you had a screening program okay because this category of tumor might be the very one where somebody says okay I need to do a radical prostatectomy or I need to use early radiation on you or I need to do the robotic in surgery that people are doing now what are the complications of that well some men will get incontinent someone will get impotent yes they will be bleeding another effects maybe those are small maybe there pendant and the surgeon but it isn’t entirely fair to say that what’s the harm of making an early diagnosis so this is a reality that occurs and if your disease is for dominating in a population under screening if this category of patient will be a high percentage of what you find what will it look like on your cancer statistics will you have an effect on the cancer incidence yes you will you will have a

surge in incidents at the time the technology is introduced will you have an effect on death due to the disease you’ll have an effect on death I would vote no I would vote that you won’t pick this up on a population-based thing because these are not tumors that would have declared themselves necessarily but for the screening they’re not going to kill you right as a cause of death does everybody go with me on this is I’m not going on to the next slide until I have an absolute unanimous consensus it may look that way it will certainly look like you have an epidemic of the disease you will certainly get noticed that people saying my god we’re having all this prostate cancer showing up right but it you won’t see a difference in the death rate the death rate is not subject except for recording errors or diagnostic errors which are not realistic here because death rates are recorded by death certificates and in with in large measure every death in the United States has recorded in the national center for health statistics and there’s pretty good integrity on that and so if somebody writes down the death certificate died from prostate cancer the odds are they died of prostate cancer and that kind of even if they’re having a systematic error if you have a doctor somewhere who’s crazed and won’t write that down that’s going to be a systematic error probably is not going to give you an artifact unusual sense so death rate is really a rock-solid statistic that’s eight what’s called age-adjusted death rate you understand what even by age adjusted I’ll take your silence as a no okay aids adjustment is a statistical tool that needs to be done whenever crude statistics are introduced and what you do is you essentially for want of a better term you normalize or standardized it’s analogous but not literally the same thing is grading on a curve do they grade on a curve at Brown it must be some tough curve okay it would be analogous to having a pool of institutions and grading on a curve but not correcting for how many brown students there were okay so let me go with that if you don’t know that a population has changed in age and a tumor is related to age which it is but you don’t know the population is aged if you have certain raw figures come in you’re going to say bo have an epidemic of cancer well really it’s not that at all it said you didn’t realize the population it shifted if you see what I’m saying so what this is is for want of a better term a way of normalizing to a given standard population usually you normalize to a two thousand year population it’s called age adjustment if somebody in this probably won’t happen but in the odd event that somebody does a presentation and you pick up the notation that they’re not age adjusting their statistics you can stand up if you’re in a bad mood and ask her if the age-adjusted and if they didn’t they’ve committed a serious thing it’s analogous but not homologous to failing to report something in constant dollars if you want to know if something was more expensive in 1910 versus 2009 are you with me it would be hard for you to know if something is more expensive or not unless it was reported in constant dollars against some standard in brief that’s that’s not correct literally but it’s spiritually correct so you need to have age adjusted statistics so basically we have not seen a dramatic drop in the age-adjusted prostate cancer rate in this kind three we see but you know same number roughly there’s been massive screening program so I think those are maybe there’s been a little trend down but it really certainly not dramatic and that has introduced issues for people and they’re also controversies when randomized trials have been done it’s been difficult to show a survival advantage I’m not saying there aren’t some studies that might but there’s always complications in this field at any rate my main point is that you need to segment in your minds whether I can’t what how a profile of a cancer looks within these parameters this is worked by esserman and her co-workers in san francisco and i thought was unusually clear and that’s why I’m heritage it okay so the point here is both tumors a and B represent low-risk indolent or indignant lesions of epithelial origin their tumors that are indolent they can be there they’re probably not going to contribute to death and therefore screening is not likely to do you a lot of good if a lot of the patients in your tumor pool are in this category it seemed like it wouldn’t necessarily

be an issue to do early test PSA screening everyone accepted and understood that this means you know your conditional probability of getting to maury could possibly be quite high you know it’s and it seemed like the problem comes when your Institute shooting a policy that previously existed for maybe diagnostic cuts right so that you know you’re applying the same policies to what’s really just what I’m really not I think that is a brilliant and utopian point of view it is true what you said is one hundred percent correct and insightful but I think that in the practical reality of something it is very difficult to ask a patient to come in for an annual test get it giving what you’ve described his essentially statistical informed consent I’m going to explain to you I’m doing a test and these are some issues that we’re going to talk about these are the limitations the test it is very difficult then when you get a positive result not to act on it so for the one person who is declared to become a physician you’re going to encounter another phenomenon many of you are going to be doctors you just don’t know it yet so you’re going to encounter another thing somewhere along your career I promise you you’re going to be in the following situation you will either have ordered a test or will be in the room when you ask somebody who ordered as a test why did you order the test if the test apparently had no relationship to what you’re worrying about on the wards at the time and the test is way off the categorization of what you thought you’re working up and dreadfully comes back positive ok I promise you’re going to be in that situation and basically what happens is people somebody is in the on the wards for a pneumonia and somebody gets a test that has nothing to do with pneumonia nothing due to the patient and it comes back positive and now what now you’re going to have to work it up it doesn’t matter how it got there and that’s a real phenomenon for example we now are seeing imaging companies that have these little bands that come around they’ll do a total body imaging for you they’ll image you from head to toe and and they’ll find early lesions guess what they’ll find something that you didn’t know you had a live poma of the spine a little thing here or there whatever it is and guess what you’re going to have to work it up now you probably cannot ignore it probably for many reasons including human nature but there may be other reasons as well so it is very difficult for me to envision a paradigm of screening public health screening in which you in effect screen the patient but give a number of statistical caveats that would be true if this were Plato’s world and both patient and doctor were philosopher kings and you were then able to have a learner discussion about that but I will tell you that when you are a patient no one is a philosopher king I don’t care if you’re a Nobel Prize winner if a doctor says you might have cancer then the then all the stuff that Woody Allen used to have in his movies all is true you hear the word you might have cancer that is a life-altering experience it is not as something you could say well I can statistically worry about this and I’m a bayesian and whatever doesn’t work that way so please you disagree with my interpretation Arden know there’s somebody be and etc and they’re not even against to answer they given an age adjusted risk of the chance of having a child with an employer which trisomy 21 yes indeed I like that analogy it’s a good counter but but what I would want to just rebut or give you my reply post is I would venture to say that on a statistical basis what you’ve described as a lot more sensitivity specificity than PSA screening even with the caveats and so I would not having studied the matter i would say that yes you do have those corrections demographic corrections but even with those corrections that the inherent validity of what you’re offering is much stronger because it is it is this mix but i could be wrong so it it may well be that you’re doing that but this this mix is really what it determines the public health value of whether you’re going to essentially say everyone over the age of 50 every male over the age of 50 is

going to get a PSA that’s everybody and that introduces special issues yes you may you may be able to find examples where it works I I’m not sure that it could be applied as quite so easily in prostate cancer screening but we can have a discussion it’s certainly open now to Mercy is destined to become metastatic that is essentially the surrogate definition of fatal if there are some exceptions of what I’m going to tell you and yes certain primary tumors depending on their position and so on can be lethal all right but usually when you’re talking about a cancer death usually you’re talking about a cancer that has become metastatic it has gone from its primary site that’s typically what we’re talking abut it’s typically so to Mercy is destined to become metastatic and fatal but can still be detected while while still curable that is surgically removable well this category be benefited by PSA screening in prostate cancer yes this is it this is the Holy Grail this is what you want you want this particular group if this would everybody were like this this would be the group then you would have screening would be would be relevant okay now tumor D is destined to become that aesthetic but it grows so quickly that by the time it’s detected it may no longer be curable is this likely to be benefited by screening that is that a phone call that is my timer okay so it’s a very interesting point of view it is not the classical point of view for springing though because the way this has been set up is that the tumor by the time you detected by your whatever technology it has already crossed some Rubicon within the patient and it is immutable what you’re going to do if you see what I’m saying so yes you you could create scenarios where you do want to know you could argue that it is always better to know a person needs to know their fate you could argue that there may be certain additional benefits but at least from the bed for showing if you’re arguing the benefit must be reflected in the drop and death rate then probably this wouldn’t work let me give you a tumor where we do not have debate cervical cancer I doubt if you’re going to find very many articles in literature that argue that cervical cancer screening as we do it now which is either pap smears or DNA capture whatever you want cervical cancer is an HPV associated cervical cancer human papillomavirus associate yeah I got even find a lot of articles saying I you know the it’s just not effective love of law you won’t see very many of those because there isn’t any debate on that practically speaking we know and that’s reflected in the population statistics and you can make all sorts of correlations screen women despite the potentials for artifacts that I’ve described for us women virtually always have a lower death rate than cervical cancer you can see in the national population statistics that cervical cancer is dropping no doubt related to those kinds of interventions so there’s a classical example where you just don’t have the debate you’re catching the cancer early you snip it out or do whatever you wish to do and it go that’s it the woman will not die for disease you can always get into a philosophical question about whether you’re intervening on certain number of women who won’t benefit they wouldn’t their tumor wouldn’t gone anywhere either but there is no question that you’re getting a benefit and the kind of intervention you’re doing is relatively cost-effective another thing which is not quite as dramatic and that quite a solid but still pretty solid in most people’s views is colonoscopy to remove polyps yes you’re going to remove a lot more polyps than will become malignant but most people sort of no longer consider colonoscopy that invasive and hey you feel good with the propofol and whatever and you get the screening and and the polyp is removed and the person will not have a polyp that as a potential to become malignant and it’s in a relatively accessible position there’s usually not much in the way of untoward action so I’m not saying that that’s quite at the same gold standard level but it’s still pretty high but for prostate cancer and by the way we are seeing colon cancer statistics drop for many reasons in this country so it all fitting together in some manner means other screening approaches even hemoccult might be doing something but in prostate cancer it’s this may or any tumor it’s really this mix right and if you don’t know the mix you can’t be sure what your screening is going to do so there have been suggestions well by the way I want to introduce one other concept for you to think about you know

what the concept of an interval tumor is so supposing I have a screening schedule and I’m screening somebody in this case I’m going to include breast cancer for a moment to make a point I’m going to do PSA screening and I’m going to do mammography okay let’s say I do mammography some people like every year let’s say though I would do every two years okay if I pick up a tumor on the scheduled annual or biannual date is that likely to be more or less lethal than a tumor that comes in to the woman discovers it during the interval and reports it to the doctor it not having been seen in the previous test assuming in this discussion that we’re not talk about technical errors the radiologists didn’t miss it or anything no technical errors is is it is now that’s what I mean by an interval the thing popped up clinically between normal screening times and the patient is now coming in between screening times and by definition if a patient comes in with a problem it’s no longer screening whatever your definition even in my opinion if the patient comes in says doc I’m just not feeling well if you do test at that point I do not believe it’s screen well hell you could have intuition you can have other things but there it’s not screening as I would as some doctors might but with me they’d argue with me if you’re wrong better they’ve been out with me maybe I’m wrong but you see that which is the tumor which is likely to be the serious more serious situation pardon exactly now you now know a fact that many doctors don’t know so intervals tumors typically classically are more to be feared unless there’s some error why because they have a certain growth velocity that they’ve declared they weren’t there the last time you checked so statistically they may be some serious growing going on there may be a serious velocity going on here and those more or less are not good tumors whereas if you’re picking up something in a regular scheduled well then the statistics might favor that it’s you know growing in a more mannerly pace okay so if you’re going to screen in my opinion and many other people believe this is the group you’ve got to have in the proportionality of that group is going to determine the effort effectiveness of your screening so while I’m deliberately being agnostic with you about whether PSA screening is worth it or not I am introducing that just because you’re catching small tumors and my god has got to be good to catch a small tumor is not always correct I have had very high ranking people in the oncology community high grade authorities tell me that I was an idiot which I said may be true but nevertheless small tumors do not always kill you so that is the issue the the point is that we can’t be trapped in our own discussion from first principles you can’t just assume that small tumor it has to be susceptible to a benefit by being removed early it may be true but it may not be true you may be you may be causing a diagnosis of cancer and what and what is that called by the way that’s called overdiagnosis that’s another kind of statistical error over diagnosis which in my opinion is defined by the term of identifying a tumor that otherwise would have been benign and not cause problems for that patient during his or her lifetime and if you diagnose a cancer in that context and say it is requires an evaluation that otherwise wouldn’t have required then you could be open to something related to to a statistical artifact called overdiagnosis now why is this important this isn’t important to make a point or two get into debates it’s important because many times we have public health policies and we’re disappointed when they don’t work okay and so we need to think about it this way it also is a warning about how there is not necessarily linear D in making discoveries from the genome making something that could be a biomarker or whatever and so on so this is causing people to rethink this is an article I urge you to look at it from estimate at all in JAMA just came out 2009 vol302 and basically you’re talking about additional costs you’re talking about a number of issues to the healthcare system but it could also be some of what I’m talking about though not all applies to breast cancer i will say that mammography particularly in certain age groups clearly as a benefit it has been shown to reduce deaths that does not mean however that some of these issues do not apply to it so this is another issue probability of death from breast cancer compared with death from other causes at a time interval from breast cancer diagnosis so this is another way of looking at the problem even if a

woman has breast cancer has breast cancer if she is caught very early her lymph nodes are negative it’s very very early it does not mean she will not die of breast cancer and that the entire dosis was in vain it certainly will she made to have a died of breast cancer it’s certainly an issue and it was good to do the operation in this context but other causes of death shown in yellow can still predominate this is if she already has early breast cancer she may face a higher death rate from a competing cause of death heart disease stroke certain other conditions certain other tumors so this is a reason to individualize in my opinion we need to treat the entire woman in this context human beings are not simply the sum of their organ parts and it even if someone has early cancer it still might be important from a statistical point of view to assess what their risk profile is for the diseases that they face not just the cancer or the problem that’s in front of a particular doctor and this is a reminder of that so another slide to show drug discovery and development this is a slide put out by the industry Pharma I’m merely copying it but I think it’s probably correct and this shows you the timelines that it takes to develop a drug and you want to add these up it could take about 15 years to develop a drug from the identification of a compound yes sometimes things go faster and I’ll even show you some examples where it can go faster but my only point here is that it takes 15 years if the genome gives you a clue there’s no guarantee that it’ll give you a drug I’ve yet to see a genome-wide Association study that shows this is a link to a disease l by the way the genome-wide Association study has identified a new drug that will work in this context at the same time it doesn’t work that way it’s an invitation to begin research important invitation but it’s still invitation so you still have these timelines so this is another answer that somebody is going to come up to you at a cocktail party and say why is it take so long to develop drugs and you will say to them why you asking me and they will that will I promise you they will not stop so this is the phenomenon hurry up and wait the human genome we have genome-wide associations but then you have to your back to an old paradigm your back to the paradigm of how do I develop new Diagnostics how to do the clinical trials those are not moving any faster yet this is one exception to this now this is the death rate due to hiv/aids age-adjusted death rate in the population do you notice anything unusual about this curve I love this curve I find myself looking at it longingly this shows something that’s basically unprecedented it shows that the death rate of a disease a major disease this was the major killer of young people ages up to 40404 in the year 95 96 it shows that that but this is a total population the death rate has dramatically fallen without an effect on incidents the epidemic did not stop there was not an effective vaccine there was not a sudden cessation of transmission this is unprecedented as far as I know for a lethal chronic virus to have his death rate precipitously fall absent some dramatic paradigm shift such as a vaccine or some incredibly effective prevention so this happened now there’s another issue here if you had been a planner for years 87 through and including 95 or let’s make it 94 from simplicity you might have concluded if you were bayesian that further addition of money into this field this is antiretroviral therapy would have been futile right if the curve is going up you’re having all this research done year after year is going up maybe it’s you know call it throw a towel in right okay so anyway I’m introducing this because I really love this curve you don’t get to see curves like this too often i’m not saying you never get to see them but you don’t just see curves like this too often the death rate in adults has dropped maternal-fetal transmission in this country least has drops very significantly it is possible in children to restore growth velocity and even to reverse certain forms of with certain antiviral drugs so it’s a very interesting and gratifying example and this again is to show you in this case I’m not showing the total population 25 to 44 year old you will see that about 94 95 AIDS was the number one cause of death in this age group you’ll also see as an internal control that for practical purposes there isn’t some statistical weirdness going on you’re not you’re not the other causes

of death on a cancer heart disease suicide homicide chronic liver disease and so on in this age you did not suddenly have a disappear whatever it’s selective before you even know what happened in this year 95 what happened is highly active combinatorial antiretroviral drugs finally came together were implemented and within a year the death rate drop started drop I’m just showing you this because this is my labs first work with azt which is one of the components there’s the structure we showed that a CT actually had activity in in laboratory setting in in submitted this paper in june of 85 police if you would remember that date jun of 85 the drug itself we did it we had the clinical trial the phase 1 clinical trial reported by march of 86 than the drug was approved in march of 87 so that’s a real outlier for what i showed you for the 15 years so under some confluence of events you actually can move pretty fast the other thing i would say is this was before the genome and this is before we even had pcr we had no molecular diagnostics so this was really an ancient time we communicated by phone fax machines were still relatively novel i’m not making this up you know basically if you want to do a paper you had to sit down and a Selectric and God forbid somebody made you make suggested changes because it meant yeah I mean there were no power points out if I were doing this talk i would do glass slides or kotah Chrome’s or whatever okay so why did this work why did this what is this is this a positive example what happened here well for one thing there are a lot of factors I won’t cover them all but one key factor was the development and acceptance of reliable surrogate markers viral RNA by rt-pcr glimmer a chain reaction became a surrogate marker became accepted it being possible to develop new drugs or drug combinations and see very quickly if you’re making progress or not and therefore if you weren’t you could abandon a regimen these select as they say euthanize the drug and go on to other drugs but you also could approve the drug quickly if it had surrogate markers related to viral RNA you could begin a kind of conditional approval by the Food and Drug Administration what they call accelerator pool or subpart H approval which would mean yes the drug can be marketed you’ll have to come in with your clinical endpoints later but in the meantime people can reliably depend on an experimental drug and it can be marketed and released and available so i think this changed everything in both adult and pediatric when we were on our pediatric wards we collaborate with philip is 0 and we had results in children that were literally unbelievable even to colleagues we watched children have reversal of what was being called cerebral atrophy who watched increases in IQ we watched increases in growth velocity and we had colleagues who literally and specifically told me they did not believe the results because such results were impossible so there are ways in which you can get around the paradigm so now I want to talk about very briefly viral load testing surrogate marker was able to accelerate everything along this pathway but in particular new drug development you had a very rapid cycle you did not have to wait for clinical endpoints the first randomized trial with AZT used death as an endpoint placebo-controlled trial I guarantee you don’t want to do that more than once very very difficult to do all sorts of ethical issues many different issues but I believe to this day however that had a placebo-controlled trial not been done it would have been possible to receive a consensus that AZT was working every one of the artifacts that I was telling you about would have been invoked a placebo-controlled trial does not end all of that but it markedly if it’s properly conducted it markedly reduces many of the kinds of artifacts and selection biases that we talked about okay let me talk about surrogate markers and what I mean so that those of you in this room with your formidable intellects will solve this problem because I cannot the problem is this we have different definitions of surrogate markers is it an endpoint measured in lieu of some other so-called true clinical endpoint that’s one definition and observed variable that relates in some way to the variable of primary interest which cannot be observed directly that’s another definition but this is my favorite and this is I think the operational definition a surrogate marker is a response variable for which a test of the null hypothesis of a relationship to the treatment groups under conspiracy is also a valid test of the corresponding null hypothesis based on the true almost always clinical endpoint so in other words a surrogate marker can be used and it will give it will give you the same answer rejecting the null hypothesis you know what i mean by that right that’s clinical trial is said to be positive in favor arm be versus a or the other way around if it

rejects the null hypothesis that is the two cannot be assumed to be the same it rejects that with with whatever confidence intervals and whatever p value choose to select so if a if a and usually the p-values point 05 but if a surrogate marker can can do obey that principle and conform to that then it’s a surrogate marker in my view this is the so-called Prentice definition not everybody accepts that but I think it’s the most useful three attributes of an ideal surrogate marker surrogate marker must be prognostic for the disease or disease progression the circuit should be affected by treatment if you’re talking about treatment research and the effective treatment of the circuit should mediate the effects of the treatment on the true endpoint that’s what we need that’s what I hope the computational biology will identify for us novel hints of what is a true surrogate marker issues to consider a prognostic marker or factor is not necessarily a surrogate marker sedative ly obvious something can be prognostic but it isn’t a surrogate for how drug works okay a valid validating the mechanism does not always imply a good surrogate status that is effects on the expression of a putative marker may not necessarily translate to effects on survival disease progression or related technical end points and a positive result in an early trial on a putative circuit market is not out i’m going to imply that one has a truly validated surrogate so those are all pretty obvious and then i leave you with this this is a slide that I like so this is a slide which I label think outside the box and don’t let first impressions fool you so basically you have these settlers marveling at the stupidity of the people protecting their homes about to like that burning arrows at them what an idiotic thing to do and then they realize they’re doomed so I think that I think that’s an important lesson tooth 22 for all of you and I hope that I hope that this has been a good dialogue in conversation thank you in question so do you think that farmers were I think in the center is chaotic movie which the ionic compound distributed contains everything get by the corners of the next level strategy other than maybe randomly they are going to just something is your critical what’s going on inside I’ll try going I think this gets back at the tools that are available for pharma and I’m not sure pharma or other sources currently have all the tools they need to develop drugs for exceptionally difficult diseases I think we can expect farmer to do a good job probably coming up with new generations or drugs to treat high blood pressure things along those lines I think in some cases Pharma can act pretty quickly I think they did act pretty quickly in hiv-1 and particularly they were able to act with great speed and effectiveness once it was clear that they had a reliable surrogate marker I can’t emphasize that enough now I would love to tell you in an act of self glorification that well we developed a ziti and several other drugs and actually they went to NDA status new drug application status they were approved without any molecular diagnostics without anything really to speak up it can be done but it’s warning professional stunt drivers do not try this at home you can’t have an industry based on that you have to have for farmers they have to have some reasonable road to approval for farmers everything depends on whether a drug will be approved by the relevant health ministry in the u.s. it’s the FDA they can’t afford to be interested in the product that has no or a pathway that has no likelihood of going to an FDA approval right they don’t earn any money unless they have a drug that can be shipped in interstate commerce and promoted that requires FDA approval and so they needed and they got with HIV because it was the original preposition or proposition a preposition end proposition or AIDS was that farmer would not respond that the problem was to intractable and they would not make a commitment one or

two companies of very few companies were an exception Burroughs Wellcome was one exception and I work with Burroughs Wellcome ezg but by and large once it became clear that you had a surrogate marker a reliable surrogate marker that would be a pathway to approval it made it possible to really have lots of companies come in it made it possible to examine combinations and instantly add or subtract they made it possible to detect resistance and develop profiles of drug resistance including sequence-based methodologies for identifying drug resistance in the reverse transcriptase of limerick gene the prettiest gene and basically it made it a totally different developmental program for them it did not have to use death or advanced disease as an endpoint and made possible to develop lots of things quickly and that I think led to the drop in death rate we don’t have that in many diseases however we actually have to do clinical endpoints and that almost by definition means very large clinical trials with rare exceptions with some exceptions perhaps in certain fields but by and large you need wat large clinical trials which are expensive and and and risky and if drug fails you don’t know why it failed necessarily and if it succeeds you don’t necessarily know where how to come in with an improvement so I think those are real issues farmers should write you an enormous unconditional grant and should allow you to apply new tools of computational biology computational genomics to begin a search for reliable surrogate markers in the totality of what that means now it doesn’t necessarily mean a blood marker but something a test the profile the diagnostic something that allows you to predict what the null hypothesis test is going to yield without having to wait for the clinical endpoint and once you have that you can drop an enormous amount of time off the development program you can drop costs and so on so the answer is farmer has a significant challenge and I don’t know I don’t know precisely the pathways that one would need to get them to adopt other paradigms also you’re going to have to have the health ministries you can have to have FDA on board on this they don’t like surrogate markers they will use them under some conditions but they don’t like them because they have been burned in all fairness to them they don’t even like pure studies based only only on hemoglobin a1c for diabetes which is pretty good surrogate marker for button for glucose control without integrating other things including overall health effects and so on so I think that those are real issues and I think in cancer in particular yes we sort of have surrogate markers but sometimes we think we have a surrogate marker some people think tumor shrinkage can be used as a surrogate marker and it can no question in some conditions but there are times when tumor shrinkage doesn’t translate the overall survival and then you have a problem so those are issues sorted response you try to offer I say they’re both red and Freddy Johnson spectacular acd we met other things come straight out of the lab spectacular success whether that is arrested at marginal differences or manufacture older boys over studied wasteful small incremental literal improvements so deferred you know new things like exactly mirrors and stuff they’re going to be invaluable but then they’re going to be a whole rationing push the next as well right now no they are making great contribution the point is if they have a strategy other than surrender between kings of buying companies for which the small gamble you discover this room well let’s be honest some farmers will do that as a strategy they some farmers have enormous capital reserves and their strategy will be can be I will allow a smaller biotech company to get to a certain point I will look to see that they have gotten through some hurdles they have survived all the obstacles that are there there they’ve won a kind of a lottery and at that point we will buy them we will buy the whole company or at least buy the product from them and they have the resources to do that I’m not criticizing that but it does require a kind of change in the classical role its association give your self contracting rd so this seemed to me that a lot of the problems that you’re describing really boil down to gaps and education alright so my problems of the surrogate markers being misinterpreted by assistant

studies studies and the interpretation and use of early screening test it really what that means me it sounds like it boils down to the statisticians that understand like sources of biases correlations and these things don’t necessarily understand the clinical woman medical issues that go into identifying those event perhaps some of the positions that are at all in these studies on had an understanding of the statistical implications of how to study is set up I mean it seems like a lot of the confusion that you’re talking about his talk and has to be resulted model for education I agree and I’m also a stage in my career where I’m replicating what used to irritate me to no end when attendings of other elder statesmen would say things to me along the lines of you guys don’t even know anything about what it was like back in the day and that invariably irritated me to an extreme degree and so that is what I am saying now you don’t know is like back in the day I think that I think that what you’re saying is one hundred percent correct I think some of it is an education issue but I think that that would be a simplification yes education is the ultimate answer but I like the Monty Hall problem because it is a warning and I’m big on warnings the Monty Hall problem is a warning that even exceptionally transcendentally brilliant people can miss a problem can misunderstand the scope of the problem and can misunderstand the format of the problem can miss that the problem essentially must be solved as a element of conditional probability and believe that it is independent probability and I think that that is how much greater educated we’ve had math professors mrs. we’ve had math professors not only miss it but engage in highly vitriolic denunciations of those who got it correct right am I making this up how much smarter can politician you can you find the politician right I mean that this is if they gave out Nobel Prize for mathematics so I think the issue is I wish it was that simple i do believe it’s education but i think that yes you’ll be able to educate but i think the larger lesson is and i don’t mean this in a cliche ridden way that may come out that way I think it’s good to examine the first principles of where conclusion is going what are the assumptions you are making if you say that screening is going to work what are the assumptions that must come true there must be true then and that’s why I broke that curve up by lower Esserman and so on without saying it it was that group see you had to you had to say that if you were going to do group if you’re going to speeding and it was going to work from public health basis it had to be see the tumors had to predominantly bc right in that and if it and if they were too many in a or b or d your screening would fail it would be worse than failing it would actually cause a certain amount of harm because you’re going to once you go down the pathway you can’t do these things halfway I know I know you you offered the notion of what I call it a statistical informed consent and you were supported by learning learned colleague behind you and I think that’s true I don’t think that in a typical public health setting it will apply in my opinion I could be wrong but if you have a new screening test for prostate cancer or a new test for breast cancer or something yes you can give the conditions in the caveats but that isn’t what believe me in my opinion many patients will end up saying doctor will you just please sit down just tell me bottom-line do I get the test or not I don’t want you I don’t want to know all this stuff so the doctor then we’ll have to then be in a position of evaluating it but the doctor is going to require professional societies and and various organizations they’re going to have to take the leadership role on it and sometimes they drop the ball so that’s why it’s going to be up to you guys to you know who had meetings and when you hear presentations and when you see public policy being made it’s not infallible some dogmas are self-perpetuating because there is too much invested in them if yeah I think peer review is good if truly peer review by disinterested but not uninterested peers and you know make sure that the person doesn’t have her some vendetta we would not have done very well by some

peer reviewers when we submitted the Human Genome Project there would have been some peer reviewers to a moral certainty who would have said reject this paper the one in science there’s no question they would have said reject and they would have found all sorts of great reasons why they were they weren’t registered or anything it was just the science of it so I guess there is that element but I think education will certainly go a long way statistical literacy will go a long way but I don’t I think each group is going to lb other problems in our area well people are still gonna be human they’re still going to be a mistake see all the people who come up with these things walk them over them and patience when it comes down to it I want to have every test possible so the two selfish actors kind of screw up the whole economic situation there’s sciences well I would agree with you except I wouldn’t necessarily use word selfish I would say that they these are people who have certain views and it’s in the case of screening it’s just two incredibly beautiful and self reinforcing to take the view that if you catch something early it can’t hurt and almost assuredly will do good and that concept just goes against the grain now in some tumors that may one hundred percent be true i gave you some examples I think at breast cancer it’s probably true screening does more much better than thank you for coming but I think for some tumors prostate the jury is still out I’m sorry there was a question back there did you still have your heart oh yeah okay I wanted to ask you about what the process was for developing a ZB what do you go through traditional process of screening many many compounds beginning down and that was not really a method you see you in a lot about the one who actually had one that worked out the army so you were the lucky lab that you’re the one you know 10,000 compounds that can work out well we had an arm we had a belief system that was actually opposed to the reality of the time and I actually I would have to say it’s a very complicated story but but we did not believe here are the four things in were true at the time many people believe retroviruses pathogenic retroviruses didn’t exist in human beings at that time it exists so my father tried to feed something that doesn’t exist to even when it was acknowledged that summit reverses existed they thought they were like htlv-1 they worked at very rare conditions very rare cancers like subacute t-cell leukemia or neurological signals so eats them to do with that three even if all those other things might be acknowledged to be true it was a belief that vector viruses were inherently untreatable so you so basically other words it’s okay it’s a retrovirus but it’s inherently people because they have entirely and extremely high error rate and also their degree in the genome so apparently you can’t possibly treat it and then in addition there’s a whole group of people that believe that next scenes were the only way to go and it was just a point it trying to broke therapy that a vaccine would be right right around the corner and solve the problem that was 25 years ago if you can believe the euro that people believe that a vaccine EP simple sir save your lab didn’t proper into the assumption that HIV or the no symptoms causing hug by HIV were caused by a virus and stuff like that you’re sorry so you’re glad your operator different assumptions that we currently have an operator for HIV so your slang well I don’t know what you mean we accepted that paycheck okay so you hit 70 that hiv-1 what’s the cause of 80 and that the retrovirus was the cause of AIDS we accepted orange with this moving novel we accepted the data okay data were overwhelming so then we said the only error the only area where we were we had a strong disagreement was the view that it was inherently untreatable well we did believe that it was inherently approval we believe that certain drugs if they inhibited viral replication would allow a plateau or restoration of the immune system so you had a precocious view of HIV I guess well I’ll be out you know I’d like to say that I had on stanyan foresight and gazebos it was able to understand the universe in totality and could predict things without having to see empirical data and I wish that that were true I think we had a belief system we were prepared to test it and fortune smile so it could have gone the other way but I believe at the time there was an inherent flaw of the thinking of the time even for the vaccine advocates and that is at that point in time depending on what statistic you have 30 or 40 million people had already been infected

but we’re about to become affected and therefore if you believe in a vaccine in a classical primary prevention vaccine you’d be leaving that number entirely to their fate in other words even a vaccine that was a hundred percent effective in preventing new infections would leave dekhi would leave tens of million people to die and my position at the time was that that was ethically and scientifically unacceptable and so we were going to take a chance we were going to do we believe they were data to support the notion that if you blocked retro viral replication you would undo the viruses capacity to cause damage so easy he mimics like a okay a CT is a nucleic sight of you or of a Sanger sequencing reaction we put in a dye deoxy keep going you’re on my track that’s perfect right yeah what does it do well just stops the replication yeah but how does it do it well it lacks the oxygen ROH group I thing to add another nucleotide really good it’s a chain terminated it’s called the chain terminator it gets incorporated that’s the old way be you guys certainly wouldn’t remember this but there was a time when you actually had to when you sequence something you would actually I put a didactic aside and look at a gel and look where it stopped okay and it would actually stop the chain and at that point you knew that that was the incorporation right son remembers you do not you know animated you know automated DNA sequencers and stuff with electro Frederick glory flowable things and and and and fluorescent emitter diagnostic signaler England you know that stuff but there was a time when people actually had to hold up with a momma to an x-ray and do that well that’s what’s being done to the virus we think we think with the virus actually causes a it causes a chain termination of virus and forces the virus that commits suicide it’s one operating its it the virus once that that once that stop codon once that stop nucleotide is inserted as the virus attempts to go from RNA to DNA or make the second stand of DNA it stops it’s it it’s gone see so you chose that specific chemical because you knew it had that effect so you didn’t like try out pull library get try libraries of compounds because we’re prepared for the notion that you could have unexpected activity but once we recognize that AZT and others other die deoxynucleotides could do that and critically that they would be anabolic lee phosphorylated that is the right phosphorus groups would be put on by the cellular enzymes not the viral enzymes the cellular enzymes once we realized we had to do experiments to prove that point once we realized that it was intuitively obvious to us that the drug or and could do so at a ratio that was much more damaging to the virus than normal cells a couple orders of magnitude three orders of egg do we believe very strongly that that would be clinically relevant we believe that that was the hypothesis we were testing you you need to have an intuit in my opinion it is okay it is required that you have some intuitive I process when you frame a hypothesis I don’t think you within broad limits unless you’re saying something insane that or violates the laws of thermodynamics or something you don’t necessarily have to defend your hypothesis beyond a certain point it’s the proof of the Kapaa Pappas’s that you have to be Riley objective about there you have to stand aside and say look it’s going to come out the way it comes out there’s nothing I can’t want it to work and that will make a difference so we came up with a hypothesis that I drugs like AZT would cause clinical improvement for the reasons i just gave if you block replication the virus can’t hurt you unless it’s replicating and so if you block replication our theory was at a minimum you would keep the patient in the state they were in but hopefully and this actually happened you would also allow restoration of immune system and that has been confirmed over and over again by the way this is an issue with the AIDS denialists of the hiv-1 denialists I know if you’ve been following the New York Times but there was a whole country that made it a policy belief that hiv-1 was not the cause of AIDS and went further than that essentially refused to support any antiretroviral drugs and the consequence of that was not a scientific debate it was not a minor anecdotal discussion or some sort of hyper sophisticated point the consequence of that was several hundred thousand people died unnecessarily because in the decade that there was that denial the people weren’t getting or there were barriers to their axis and now the relevant government for the relevant country has completely reversed course and basically basically said boy we’re sorry and even in resource-poor countries when you introduce these drugs you actually get a drop of the death rate it’s not just

something for resource-rich countries and if you’re interested in this topic you should come to hear my to talk i’m about to give the last question before you you just pass people talk about humidification parties there are so many steps in so many towards the w creepy color knowledge or step out again system here another lunch of Gustav is how is the company likes a lot of the promise should we can create others is it for Siri or you see major progress I think genome-wide associations are incredibly important as our functional genomics but they have to be viewed totally objectively for example there are more than one way to understand genetic interactions with the environment you don’t always need a genome-wide Association study to make an important advance you have to accept information and data wherever it comes from okay this number one number two you have to recognize the genome-wide associations have the problems that we talked about they can give you wonderful things but they can also give you problems so you need replication number three we need to understand that the genome-wide associations are really the start then at the end even when they’re well replicated so yes they’re important and they teach you stuff but what do i do if i have a genome-wide associate and I find that there’s something located in intron ten of a gene but I don’t understand what that intron is doing there I don’t what do I what do I do with that see where I’m going on that and some of the associations while absolutely real may suppose I have an odds ratio of 1.5 does and doesn’t the confidence level doesn’t cross one well that could be very critical or it could be that i’m still in a long road before i can apply that medically that’s what I’m telling you I’m giving you my doctor’s perspective for me yes I believe in basic science I love the beauty of it I think it’s important I don’t think everything needs to be defined in terms of a medical application as the justification for doing it so I’m not saying any of that but I am saying that if you make a major commitment in a biomedical field that one has to be highly sensitive to the productivity of that field related to the public health and that has to be by some eventually buy some quantifiable measure okay reduction in incidents but should in death rate alleviation of some definable suffering or pain syndrome or something loss of productivity or something related to that you it has to have that element of a practical application and we need to keep that as the metric I think genome-wide associations are well on the certainly on the right track but I don’t think they could be the only answer and I think we need other tools otherwise you’re going to have a genome-wide Association we’re going to go back to what I’m saying ok now you rush to the starting line you see where I’m going on that you now are beginning 10 or 15 year journey I can’t wait for 10 to 15 years more i’m down to measuring things by the month ok