Good afternoon. I want to welcome all of you to this wonderful webinar put on by the office of medical policy. My name is Marcia Halloran and I would like to introduce our guidance today is enrichment strategies for clinical trials. It will be presented today by Deputy center director for clinical science actor Bob Temple.
Good afternoon everybody. It's a pleasure to discuss one of my favorite subjects, which is enrichment. And I will just start off with the slides. As probably everyone realizes, there is broad recognition and concern that the cost of clinical trials is growing, and we're worried about our ability to get the information we need about the safety and effectiveness of treatments. Both when it comes to their effectiveness as new drugs and also when it comes to comparing one drug with another. Is all very much on people's minds. People are thinking of a variety of ways to make tiles more efficient. They're thinking about adaptive designs, collecting less information, only what is really critical. Better targeted monitoring procedures which are very costly. Even carrying out trials in healthcare environments where much of the data that you need as part of the trial is actually already collected. But today, I want to talk about something different, which is what I think is a major contributor towards efficiency and that is the use of a variety of methods that improve study power. Specifically, the likelihood of showing a drug affect if there is one, by choosing ready Haitians for the trials. As everyone knows, we don't do trials in a random sampling of patients. We try to make sure that they actually have the disease we are stating so we have criteria that they have a relatively stable disease, was stable measurement so we use lead-in periods and they don't respond to well to placebos so we use lead-in periods that they have diseases of some defined severity and don't have conditions that will obscure -- these efforts are -- uses them and have used them for years. There are other steps not as regularly used they can be added to these to increase the likelihood of the drug effect be detected if indeed there is one. In December 2012, we published -- FDA published a draft guidance called enrichment strategies for clinical trials to support approval of human drugs and biologic products. That is what I'm going to talk about. Enrichment is the prospective use of any patient. Or a stick. Demographic, pathophysiologic, historical genetic and probably others to selectively study the detection of the drug effect is more likely that it would be in an unselected population. This occurs to a degree in every trial even though it's not always explicit, and it really increases study power in three major ways. The first is by decreasing heterogeneity or noise, choosing an appropriate population. For example, people who definitely have the disease. Second is finding a population that have outcome events. The events will measure the effect of the drug. That is high risk patients or sometimes patients with relative severe disease. Prognostic enrichment. And third is to identify population if you can that is capable or more capable of responding to the treatment. This is predictive enrichment. The increased study power, increased size or -- effect size or decrease of noise facilitates proof of principle. That is, you definitely can show more readily that there is a clinical effect in some population. But depending on the specific enrichment mechanism used, it can sometimes leave concern about the generalizability of the results. Who does this result apply to? Can I find those people who respond? And similarly there are questions about how the drug will work in other populations. This raises the additional question, which always comes up when there is enrichment or at least certain kinds of her enrichment. How much data do we need before after approval in the non- selected group? If you have done predictive enrichment to find people with a particular genetic defect, how much do
you know -- how much do you need to know about people who don't have that particular defect?
As I said, the nervousness that people sometimes have about generalizability is present and is unavoidable. Sometimes you can use selection criterion. In people who have a particular genetic defect, you can define those. Some of the impaired designs that I will describe later which is doing studies and people who respond to an open screening people -- period, there isn't any way to identify the response a population besides giving them the drug. You do however know there is an effect and you do know there's an effect in somebody. Which we think -- and in some cases the remedy to reassure nervousness about this is to use these designs early, establishing beyond any doubt that there is a drug affect in somebody and then you can work on who it works on. The other is don't make the enrichment study the only study, at least not usually or sometimes you will. The third is to be very aware of what has been done to enrich the population and don't hide it or overstate the results. But at the same time although I recognize the nervousness people have about this, it's more more recognize that the selected population, however you actually selected, is in fact the one where treatment makes the most sense. After all, results in an unselected population, the way we always used to develop drugs may be driven by a subset of the population. You never know. Searching for who the responders are, which is very much on people's minds, is highly worthwhile and improves the likelihood of showing an effect and the risk/benefit effect for the treatment of patients. The guidance is primarily focused on studies intended to show effectiveness but it's pertinent do safety studies as well. I will mention that briefly. In the studies of oral hypoglycemic stew rule out a cardiovascular risk and increased risk of 1.8 or 1.3, we recognize that to have any chance of succeeding in such a study and a reasonable sample size, you need to include high risk patients to have a chance. More and more of the studies are including people who are older, would've had their diabetes for a longer time, who may have had a history of some cardiovascular risk and to succeed in this safety study you have to do that. Another interesting possibility is that if a drug class has an effect that bothers a lot of people, in an important way, one way to find out whether -- one efficient way to find whether a new drug lacks that affect his to do a study in people who have had the effect with the previous drug. This was actually done in people who had cost on an AC inhibitor to show that an angiotensin respect to a blocker did not have that effect. They picked people who unquestionably cost, 90% of them did on the AC inhibitor. And found a modest study to show there was a difference. This is an important aspect of comparative safety and bears attention.
I have mentioned the three basic flavors of enrichment. Let me go through them in somewhat more depth. The first form is to decrease heterogeneity or noise. And this is in many ways virtually universal. You define entry criteria carefully to be sure patients have the disease beings Duddy. You all know that this isn't actually always done. As well as it should be. And sometimes people snake into the trial who don't have a disease. That's a disaster. It's of interest, not easy, to find people who are likely to comply with a therapy. If they don't take it, they will not benefit from it. My favorite example is the early VA hypertension studies in the 70s. Where during a placebo -- they were giving a placebo contain riboflavin. When you shine -- a would slight or florescent light on your end that contains riboflavin, a close. The people who entered the placebo part of the opening session, if their urine didn't glow they were randomized. And as we know those trials showed a dramatic effect of lowering blood pressure. You tried various ways to choose people who won't drop out. You try to eliminate placebo responders in a lead-in period sometimes. You can eliminate people who give consistent treadmill results. And people whose blood pressure is very unstable from time to time. You try to leave people out there who have an illness likely to lead to an early death any kind to include people who are already taking a drug that has the same pharmacologic effect as the drug your testing.
Most of the time, people don't worry about generalizability when these approaches are used. Apart from that, enrichment strategies fall into two distinct types. The first is choosing people at high risk. That is, people who are likely to have the events you are trying to treat or prevent with the drug. That is generally called prognostic enrichment. It obviously has the the size implications. Or events means you can have a smaller study if the effect is the same. But it also has therapeutic implications. And a population where you get a 50% change in event rate and that is from 10% to 5%, the defect is considerably larger and more important in a less sick population where you reduce the risk from 1% to 0.5% and that could matter if the drug has some degree of toxicity. That can be important. And the other kind of enrichment is a predictive enrichment which is choosing people more likely to respond to be treatment. There's is a wide variety of ways that we now try to do that. Historically, this is sometimes based on pathophysiologic findings but increasingly, we are finding pretty iMac, cell surface markers or genomic bases for figuring out who the people who are going to respond. There are also a variety of things you don't necessarily understand but you can look for. Generally called empiric approaches. You can look for an early response on a surrogate endpoint. You may not know why that person has the response but they had a. Tumor response or radiographic measure or something like that. And then important subcategory of this kind of trial is they randomized withdrawal study where you take people who seem to be responding to the drug and you randomly take it away. I will talk about that more later.
Turning now to prognostic enrichment. Trying to find people who are at high risk of having the disease. There is increasing information about what puts people at risk. Genomic and pathophysiologic measures, but we have known about these before. If you're doing a trial to study the ability to prevent adverse cardiology outcomes the best predictor of the likelihood of having a heart attack is having have one fairly recently. Many trials that we are trying to show the effects of lipid lowering drugs or things like that use people -- a role people who have had a recent MRI. It's very helpful. Similarly a history of angina or transient ischemic attack or -- all make it more likely that you will have cardiovascular events. The level of your cholesterol and blood pressure, presence of diabetes. Turns out elevated C-reactive protein increases -- shows that you have an increased risk of having a heart attack that was at the Jupiter study. Family history, gender, race, age, all those things can play part towards predicting the likelihood of events. And certain individual measures that can be used all. Echocardiogram findings can tell you something about the likelihood, vascular injury can predict the likelihood of having a heart attack. Tumor histology or tumor genetics can say how likely recurrences. These are all being used. In one way or another, one always tries to find people at high risk so that intervention will have a chance to prevent. We see this a lot in oncology and cardiovascular. There are genomic predictors of the likelihood of rest or ovarian cancer. If you're doing ambitious trials there are ways to define who is at high risk. If you're studying lipid lowering drugs, the first successful studies, for as study to people with a recent history of heart attack as well as very high LDL. Those are always. Studies of antiplatelet drugs have a larger effect when you take people who have had a recent angioplasty because a are at high risk of having another check. People have been looking at these for a long time, and one of the things that is on everybody's mind is getting into Alzheimer's disease. There's a lot of discouragement of being successful if you wait until people have over disease. A lot of interest in whether you can some early way identify the people who are at high risk and study them.
When these methods are used there's also a question about the effects in the benefit risk relationship in the lower risk patients. Probably resolvable fully by only more study. Once again, you have unquestionably shown the drug works in some population which in my view puts everybody way ahead. Just a few examples of how this sort of thing has been used. In oncology, it was known from studies at him -- treatment of breast cancer that tamoxifen prevents contralateral breast tumors would -- when used as an adjuvant therapy. Obviously people are very high risk of having breast cancer. Great interest in studying people with more generally high risk. Notches people who are ready have one cancer. To show that tamoxifen could present -- prevent rest cancer you had to have a population with them -- enough breast cancer to be able to measure something. And it was toxic and also you didn't want to give it to people who were not at risk. It was studied in people who were in various ways using the Gail model defined as being at high risk of getting breast cancer and the trial was successful and it was been marketed with access to a Gail model calculator to be sure that you are the kind of high-risk person the drug had been shown to be effective in. Not that the size was necessary larger in that population, but the absolute effect was larger and you wouldn't be exposing relatively low risk people to toxicity.
And interesting selection method not used yet to my best understanding and I'm not quite sure why, is the following. Damico reported some years ago that in men with localized prostate cancer following radical prostatectomy if you look that was called PSA velocity, that is increase in PSA of 2 nanograms per milliliter you would predict the likelihood of prosthetic cancer mortality almost 100% over a 10 year period. As well as the likelihood of recurrence. I will show you the curb see presented. Essentially no deaths from prostate cancer. There were from other causes. In the population. These are the recurrence rates, the topline there is people with a PSA velocity greater than two and you can see that by five years there at nearly 50% recurrence and the other three groups are closer to 30. If you know look at mortality, in the sample, there were really no deaths due to prostate cancer in the population without the PSA velocity we saw 10% or 15% and the people did have high velocity. That's interesting and they keep waiting for someone to choose a population for regimen therapy that way.
There's been recognition for some time that the genomics of a tumor can predict the likelihood of recurrence in breast cancer. And Sam and coworkers looked at five different gene expression profile brooches that they apply to a 285 patient sample. For the five methods, I will show you one of them on the next like. A striking ability to predict who is going to get metastatic disease recurrence. And I will show you one of them, a 70 gene profile. Indications for patient selection are obvious. If people don't have recurrent tumors are metastatic tumors you will not be able to show that your drug improves that. There is been a relatively recent approval of something called Mono print which is an in vitro test based on gene expression profile that seems to be getting at the same thing. But the differences are striking. The people with a good -- on the left is probability is relapse, survival and on the right is overall survival. And you can see the people with a good 70 gene profile do very what -- much better than the people with a poor profile. It sounds like a good way to select people for your adjuvant therapy trope -- trial. Cardiovascular disease. For outcome studies, patients at high risk. These are generally called secondary prevention or post a Mike where they might've had a stroke are very high cholesterol, very severe heart failure and it makes just a huge difference. The study showed a survival affecting her failure was the consensus of the of enalapril. Done in New York heart Association. It is class for patients and involved only 253 patients. They had a survival benefit in 203 patients. That's because the untreated mortality was 40% in two months and treatment effect showed a 40% reduction. Later so these -- often couldn't show mortality effect. The combined mortality and hospitalization. The fact that these people were very high risk made a big difference in the trial. Similarly, the first lipid outcome trial that was successful, the forest study of Simba section -- simvastatin. 4440 for patients showed a nice mortality effect. Later trials have trouble showing mortality effect. Had to be much larger and usually had choose composite endpoints. Having a lot of events really helps you if the drug works.
The Jupiter study of [ Indiscernible ], people whose cholesterol was not nominally elevated. People who were 130 milligrams% or lower -- less. But in order to be sure that they would have some evidence to prevent, these people also had to have an elevated C-reactive protein which is a predictor for having a heart attack. Almost 18,000 people. LDL was low and C-reactive protein had to be above two. And had to be about four. These are basically not very sick people. And they were randomized -- the endpoint was the first event defined fairly broadly. And what you can see is they had presumably enough events they had a striking effect on their primary endpoint on a lie my of which I have a reasonable number. The rate of primary endpoints was fairly low. In this population. 1.36 per 100 patient years. On the untreated population but I'm sure that was higher than it would've been if they had taken an unselected population.
There's tremendous interest on high-risk patients. Talk important trying to prevent -- apart from the cardiovascular risk that we know but pretty will, genetic predictors of risk or Alzheimer's disease and cancers or there may be kinds of goals in this was an area of current interest. That was prognostic enrichment. Predictive enrichment, which of course is what is lighting everybody -- everybody's fire. Particular genetic or other characteristic that enables them to respond. Identifying people who respond to the treatment and then start doing them enormously enhances the power of the study and it has clear implications for how to use the drug. It's especially critical when responders are only a small fraction of all the people with the condition. The people of the Reagan or cystic fibrosis. If you study an unselected population you don't have a prayer. But if you study only the people with the right lesion, then you do. These kinds of selections can be based on understanding of the disease and tumor receptors they can be responsive biomarker for me give you some examples. Hypertension can be high read and or low rating, much larger effect -- angiotensin two blockers are beta blockers. Doesn't matter that much, you can treat people and see how they do but if that was important, if there was an outcome that wasn't so easy to observe, that can be very important. We study antibiotics and bacterial infections are we hope will be sensitive to the antibacterial and we certainly look at patients who do have a sense of organism even if that's not the only examination we make. A well established genetically determined difference could be the basis for a pathophysiological he selected population. Many tumor genetic or sue first markers are related to will understood effects on enzymes or gross Millis. Herceptin was an early example. ER positive breast tumors are responsive to antiestrogen treatment and there are a growing number of these. Even if pathophysiology isn't quite clear, their empirical ways to identify people who are likely to respond. And it's of interest that they don't always work that they are always a good idea. The cast study, cardiac arrhythmia expression trial was a very well-designed study and it was carried out in people who to get into the trial had to have a 70% reduction of BP be used. This is a trial to see who if people after a heart attack had a high rated premature ventricular beats and known to be an increased risk of dying suddenly. These very effective antiarrhythmic could decrease the likelihood of mortality in this population by preventing the French regular premature beats. They only put in people into the trials who had what was thought to be the relevant response, namely suppression of renter killer be. Unfortunately, the drugs were wonderful in preventing premature beats but they were lethal and they doubled mortality. That doesn't mean it wasn't a clever idea. Long time ago it tell us of topical my trade patches were carried out only of people who had a blood pressure or engine response to sublingual nitroglycerin. Why study and one who can respond to sublingual? Of interest is John Oates Ray Woosley and Dan Ronan who developed a lot of the antiarrhythmic in the 70s and 80s did so by screening and open population and then randomizing the responders. Often for a dose response study. I will talk more about randomized withdrawal studies. These are enriched with probable responders. You can look to see if the person has responded to request to see if he is likely to respond to a similar member of the class. That may not be therapeutically exciting but it can help develop the drug. As an example of that, you whether you can see that or not, I don't know. These are five studies of [ Indiscernible ] all involving several doses. The one in the middle was quite a small study. Specifically done people who buy his book -- history had responded to direct's. The effect size at 5 milligrams is 37, and the placebo so about 42 millimeters of mercury. Huge response. Far bigger than the response of many of the other unselected ones. I have long suggested that people do their dose response studies in no responders to a drug. To my best knowledge no one has done that but I am still hoping. And you can also select likely responders on the basis of some kind of -- biomarker that is randomized -- [ Indiscernible ] over a in week period and this was suggested by her seven. You could randomize patients whose tumors shrank. If they're stable over a certain amount of time those a B Gibb able to randomize as opposed to people whose tumors grew. I've always thought that if you are looking at the effect of cholesterol-lowering drug you might want to select patients who had a very large effect. I don't know that I've seen anybody do that but you could. You can also only put patients into a trial who had a CRP response if you think that's a reasonable predictor. Lots of possibilities. And you gain a lot. This is also taken from a paper I Rich Simon. This shows the sample size ratios as a function of the prevalence of market positive patients. Suppose you assume -- look at the bottom there, the prevalence of the marker positive patients is 25%. They have the right genetic or Provia Mike lesion. And the response and the Markert negative people is zero. So we have examples of that. In that case if it 25% marker positive and no response in the non- positive patients, the sample size to show an effect if you don't pick only view potential responders is 16 times as large. That makes in many cases that makes a study simply undoable.
As was noted in that figure, if the response in the -- is 50%, the game is considerably smaller although there is on and the samples rises only 2.5 times larger in the unselected population. There's also an enhanced risk/benefit ratio from doing this. [ Indiscernible ] is cardio toxic. Studies in patients with metastatic cancer as well as adjuvant studies were all conducted in patients with positive tumors. That made it easier to show an effect, that is true and the effect was considerably larger it would have been on unselected population but it also enhance the risk benefit -- by removing people who cannot benefit to get cardiotoxicity. It has two major advantages. One of the questions that always arises is if a trial is done entirely in the marker positive group, and efficient thing to do, it gives no information about the admitted patients. As a possible they have some affect? The enrichment guide -- really quite repeatedly says of a lot. Unless there is really no chance of an effect in the Markert negative patients, it's a very good idea to include some Markert negative patients, because first of all that might have some response and second, as you look at the response according to the gradation of the marker, you might learn something about what the best marker cut off his. If you stay positive means five and really people with for can respond, you would like to know that because you might want to treat them. None of that's tops you from making the primary endpoint in this trial that might include more people. The affected mere grids strata. You can get the advantage of enrichment without losing the information. A fair amount of time describing what you would generally call them. Approaches. That is where you necessarily understand why works in the group but you do an early look and it appears to work. I mentioned the approach to antiarrhythmic -- the cast, beta blocker heart failure studies were also the site. But was screen for here was tolerability. Or failure, Kuwait in some cases tolerate a beta blocker and other cases couldn't so the trials were only done in people who could tolerate it. Most of the trials. That's interesting and it could be controversial. Does mean that the effect to see a map population them what overstates the effectiveness in and unselected population and on the other hand it allows you to do the study. I mentioned history of response to a treatment class. Sometimes results in earlier studies can be informative. And largest bonds in the black subset of VA population into early studies. That was convincing enough to us to allow them to do a definitive study entirely in blacks which showed a reduction in the trial with about 1000 people. We calculated that Duddy in people -- in whites would have needed over 20,000 patients which would never have occurred. Another thing that is possible to adapt the study, after an interim look, there will be some correction for this. You could include more of the population that seems to respond better. Men respond much better than women or more severely old people respond much better. You could start adjusting the recommendation to include four of these people. That probably doesn't mean any correction but if you can't everybody in the trial. And then I will talk but the randomized withdrawal studies. There is a wide variety of pathophysiologic and genetic characteristics that are of interest. You could where only some people make the active metabolite as is the casing clopidogrel. Only people make the active metabolite or have a subset or stratum of people who make it. The effect could be larger. You have to do the study to see. Might make it possible.
You could study only people whose tumor takes up a drug. Going back to when I was in NIH, if we're treating someone was that thyroid cancer we would do the 31 test to see whether the tumor took it up and how much given the dose. You can still do that. You can look at effects on tumor metabolism, and of course there's tremendous interest these days in looking for genetic markers and predictors response. And recent cystic fibrosis and hepatitis C drugs make use of that kind of enrichment. That is plainly the wave of the future many ways. As I said, the predictive enrichment using genetic or Provia my characteristics predicting response have been principally in the oncology setting but we are seeing more. The cystic fibrosis drug is useful in people with a particular cystic fibrosis mutation that is only 4% of all cystic fibrosis patients. Unselected location would have very little chance of this. To genomic we directed drugs for hepatitis see. The use of this is clearly spreading. I want to talk briefly about the randomized withdrawal design. This was actually proposed in 1975 by a Belgian name -- who thought it was a more ethical way to do trials because at that time trials would often run six months and they thought it was cruel and him additional punishment to keep it within small on a placebo for that one. What he proposed was that people would additionally have an open treatment with the test drug and then be people who appeared to respond would be randomized to continuation of the test drug or to placebo and you could actually randomized more than one drug. More than one dose. The endpoint can be the time to failure. In which case people don't have to be -- feel miserable for very long. Or could be a conventional measure like him and he -- you would have. Intended to be enriched with people who are doing well in treatment. The screening period is defined who the people who respond will are. They are enriched in that sense. Another attractive -- they often need no. Meant. Especially Portman and rare diseases where finding patients can be very difficult. And a lot of places of drugs being developed, there may be a fair number of people on an open label therapy. This can be a way to get the trial date you need to take people on open label therapy and if the psychos away promptly you can do a randomized withdrawals Duddy if you can convince people to get into it. All antidepressants are not tested for maintenance of fax using this kind of design. People who responded reasonably well for three months or more a randomized to continue therapy or ending of therapy and the endpoint is recurrent depression. Interestingly although ordinary depression trials and acute depression fill out a rate of 50% and have for decades, these randomized withdrawals maintenance studies essentially never fail. They'll throw a 35 to 50% reduction in recurrent depression. This design was actually used to show that nifedipine was effective. That was the first use it was approved for. ASIC design was this. They get open nifedipine and it seemed to be doing well and they are observed for a couple of weeks and then put onto four weeks of continued nifedipine or four weeks and placebo. It was a very small study, only 20 people, and the early withdrawal -- the endpoint was early withdrawal because of current -- recurrent angina and it was five to zero or 620 depending on -- that was the basis for the drug already known and it's ethical because you can stop as soon as the failure criterion is met. Has a lot of attraction pain medication. Were keeping people on placebo for 12 weeks because you do want to know whether the drug continues to work, keeping people him put placebo for the on can be very difficult. There are other things that one can do, not commonly done. An interesting question always is whether some new drug works and people who don't respond to the previous therapy. Very attractive thing to know. Doing studies and nonresponders where you randomized to the new drug and the failed drug is a particularly useful comparative effectiveness study. I'm only aware of for the studies and all of history. They were the basis for approval of clozapine and captopril both of which call psychosis and wouldn't have been approved but for this that he showing an advantage. And calcium channel blocker the causes torsade de pointes and there's [ Indiscernible ]. You can do the same kind of thing randomizing patients who don't tolerate one drug to the drug they don't tolerate. If it's ethical to do that. And to a new drug, this was used in a very effective study to show that losartan does not cause cough even in people who who all -- [ Indiscernible ] very good study. There's a long section in the guidance on what to watch out for when considering predictive enrichment designs particularly any properties of of specific sites. I can only touch on highlights here. There's a lot of emphasis on knowing the performance characteristic criteria. You have to have a cut off point. Is it to plus or three plus or what are you going to do what the what you would like is getting data that shows that. By including a fair number of people and looking at the sensitivity and specificity of each level. That's what you'd like. Sometimes people already have a pretty good idea of what the cut off point is and they really want to get on with it. And that's not okay. We hope people will get more precise about these things. The guidance also talks about when to develop the classifier -- ideally early studies would enter a broad range but could a phase three study with broad inclusion criteria explore the impact of various thresholds in an analysis correcting for multiplicity using various thresholds? It probably could. Be sure that you have accounted for the multiplicity. There's always a question of who to include in the study that is only the enrichment -- where everybody would analyze those as the marker with the primary endpoint. A couple of designs to consider. If there's absolutely no possible effect in the marker negative group, you can screen all patients, find whether the marker is positive or negative and randomize only the marker positive patients drug or placebo. It clearly supports the effect for the enriched population. Overstates the effect for an unselected population. No information people blow the marker cut off. And it is suitable when there's little or no chance it is marker negative and the labeling would have to say who you did. If you use this more efficient design, this is certain information you would really like, but you know show a drug works in a disease with no treatment, it is highly likely we are going to make the drug available a look at the other people later. You don't want to lose a drug like that.
You test everybody for the marker, and you see whether its positive or negative. You do a stratified study in which both groups positive and negative get randomized to drug and placebo. You don't have to make the numbers the same. It could be all people marker positive and a quarter, a third of people who are marker negative. We don't have a position on that. And sometimes you can't tell at the time of randomization whether people are marker positive or negative because it takes -- whatever it is. In the case you have to randomize everybody. It would be perfectly acceptable to make the primary endpoint the endpoint in the marker positive people. Cell, that's the end of my slides. I did want to mention a couple of other things. And that is as I said before, there is always some concern about what about the people who aren't deselected population? If you're taking steps to reduce heterogeneity nobody is especially worried about the population. It wasn't suitable for getting into the trial. If you have gone through prognostic enrichment, you have to look through a wide variety of labels on drugs I have favorable effects on heart failure or lipid lowering drugs. But we tend to do is put the exact nature of the study and who the drug was study did into section 14. The clinical trial section. And then it's variable but how broad the claimant's. Some findings in a severely old population you might feel perfectly comfortable applying to the whole population smothers you might not be so sure. There some determination about that. And predictive enrichment you generally get a claim from the population that you have studied. That as I said, there is always some [ Indiscernible ] and if you've used an impaired design there isn't any way to identify the population. All you can do is describe the whole thing and let people work it out. Sometimes by starting the drug and seeing how it goes and so on. That is the end and now we have some time for questions. The first question, I will read it several years a. Would FDA consider there's one study approval in the enrichment of well-controlled studies are suggestive of the ability for phenotypic markers to predict efficacy in the -- population. Juveniles -- absolute answer without sing it I think if you have one definitive study in and a rich population and to higher ones major lean toward -- the marker was a good predictive marker. With the previous data as compared with evidence. You could actually described the case as a case. Where a single definitive study, but it was really definitive I have to say, and a black population led to approval but that was probably because without we were halfway there with the previous VA studies in the mixed population. I don't think there's any question that the other prior studies could be considered confirmatory evidence whether they would in a particular case depends on the strength. Next question. Oncology companies predicting response based on biomarker like teen expression or mutation in one type of cancer where the biomarker has not been validated in other forms of cancer. We have seen enough examples were particular marker was a predictive marker in one tumor and not in another. So I think you have to provide data to show that the marker is predictive in the tumor of interest. I wouldn't want to speak for people who might find something so predictive that it deems always to work and they would buy it in a more general way. But usually, the success of the predictive marker is specific to a particular tumor, sometimes even the stage of the tumor. Enrichment design small sample size -- smaller sample size be sufficient for safety assessment? FDA typically require certain number patient population for safety? We have written what we expect for safety in a broad and way and that -- and the magic number there something like 1500 patients. It really does depend on what you are talking about. If you treat a disease that has no treatment and enrichment design allowed you to do that, you can see approvals in oncology with at most a couple of hundred patients and we are prepared to do that. When you have done something very important. If it is for a widespread disease were a lot of people would be exposed for a very long time, we might expect additional safety that it. That's got to be case-by-case. Depends on improvement for what was available. Are there examples of enriched studies being used on medical devices? I can't answer that definitively. I think there's no doubt at all that some devices are studied in people with relatively advanced disease where you think that getting the device makes it very big difference in the likelihood of outcome. I'm afraid I don't know enough to answer that question. Sorry.
The guidance on enrichment and presence mentored randomize withdrawal studies any mention potential use potential use in pediatric studies. And tutoring children on placebo would be challenging. Has the agency had sponsors in recent years submit phase three randomized -- and pediatrics? In fact, they're regularly used in pediatric hypertension studies. Nobody likes to do pediatric hypertension studies including a placebo group. What they do is dose response. They randomize people to several doses of the drug. That works fine if there is a slope. At the high doses better than the Lotus you have learned that the drug works. Sometimes it turns out that all of the doses have the same effect because you didn't pick the right. In that case what we asked people to do is they randomized withdrawal study. To see the blood pressure goes up with a randomize from whatever does they were on to placebo. As soon as it goes up there out of the study so you don't expose them to a long placebo. It has been used in that setting. That focuses on the fact that depending on the endpoint, sometimes randomized withdrawal study, you can find out whether someone is not being treated will very quickly. The pain comes back or whatever and you can have a withdrawal criteria.
Go back, there something about Crohn's disease that I can't see. It went further. You're interested in the -- the drug had been shown to be safe. Could you do a randomized -- whether a randomized withdrawal study works depends on the mechanism of the drug. If it has a prompt affect and reduces something rapidly and randomized withdrawal study works fine. If the drug has a very prolonged effect because it modifies the fundamental disease, you may not really know how to do a randomized withdrawal study. You may have to wait six months to see anything. Not that you couldn't, but I'm not sure how attractive that would be for people. It depends on the mechanism of the drug. How early phase one should predictive or prognostic studies before form? I don't think that for the most -- I think we're talking about remote controlled trials. It may be very sensible to do an early study in a highly enriched population to get some evidence that the drug really works. That is what the antiarrhythmic studies I was describing the job notes on those people day. These were not large studies. They took the population, they screen the population for the ability to respond. You would have to call these phase two studies and then they randomize the people responded into these trials. Very efficient, very efficient and a very large. It made a lot of sense. And after that once the drug works, you might very well want to finish to see how works in most populations. Phase two but very useful in phase three for the definitive study as well. Please comment on random continuation study in high risk cancers such lung cancer. Patients who do not progress early on. This is to some extent what Rich Simon proposed. That is you screen everybody I'm a people respond you leave them on therapy, that is easy and if they progress, you take them off therapy. That is easy. If they sit there and don't respond, he fell you could randomize people who did not progress. How many people have done that? I'm not sure I know of any examples. That was thought to be reasonable. You do want to have an effect, you don't want to keep each -- treating people with potentially toxic drug if it doesn't get everything. That was a Rich Simon proposal. If the results from to well-controlled eddies are strongly suggestive of the ability for phenotypic marker to predict efficacy in a subset population what a confirmatory study in the population be required for approval? Perfectly good question. I don't think we have an absolutely definitive answer. Ordinarily, when you go scratching through the data that was already collected with no prior hypothesis, we are a little worried about whether multiplicity will allow you to reach a conclusion. What the suggestion here is you've got to studies and they are both showing -- let's say they really seem to show that this marker predicts effectiveness. Very well in the subset. Could we grant that without a confirmatory study? My answer is that is not out of the question. If it was strong enough, you would want to know about how sensible the mechanism is, how plausible it is, and you would clearly look at how strong they were. I certainly wouldn't rule it out. A very important factor is to be study when overall? no zero for substance to retrieve a failed study, that's not going to happen. I am assuming that the to well-controlled studies actually showed an effect of the drug. And the effects seem to be particularly present him in the subset. We sometimes do pay attention to those things especially if this study [ Indiscernible ]. I think it's possible. And that helps refine who should get the drug. It is very valuable information. What FDA consider a multi cancer confirmatory trial with a common enrichment based on a common genetic aberration? That's a good question. I'm not really prepared to answer that. I have to talk to the oncologist. As a general rule, more and more one is thinking of cancers is different. The not so much all the same. And then as I said we have seen phenotypic and even genotypic markers be predictive in lung cancer and not predictive in another cancer. I'm assuming here that one is talking about looking at each cancer separately and gaining strength from each one. I think that needs more attention and discussion. I can't answer that.
Can you give an example of publish Web investigational treatment was initiated after completing a set number of doses of the example, standard of care was sick sex old of therapy and you start the investigation treatment after three cycles. I'm not sure I understand the question fully, but I don't think there is any necessary problem if you started PII -- people instead are caring gave them every month or two and randomized to a new test to the addition of a treatment plus another treatment, I'm not sure I know of any cases where that's how you do it they don't see any inherent problem with that. And I guess the idea is you would presumably to enrich you would take people who maybe who had responded to a standard of care. Certainly, there are trials were people failing on care for some defined period of the people you put into the trial. That is an acceptable's to be design. And that probably lowers the overall rate of response but it means the overall rate of response in the control group was probably very low. I presume in that case you would continue the -- everybody in the standard of care that it already failed and add the new drug. That could be an acceptable design. The the details.
Last question. If an imaging agent such as [ Indiscernible ] is used to select patients for a therapeutic trial how with the labeling of the image and therapeutic agents be connected if at all? We have a basic conception that if a diagnostic test is a necessary part of the treatment approach with a drug, that needs to get through CDRH and be approved for that use. Our guidance does say almost always -- if you have done something spectacular in vast numbers of people's lives look like you could, and the imaging device is a yet, we say we would probably approve it while they work on the imaging. Some imaging localities are not CDRH approved. Used in private laboratories and we can imagine something that was important enough that we would improve on that basis. Usually it is supposed to be approved.
Thank you. We want to remind everybody out there. We send your comments to the docket. And if you're looking for this -- with this particular guidance and Dr. Tumbles talk. Good your website and and you can have that. Thank you so much for coming today. [Event Concluded]
Back to Webinar page: FDA Webinar: Webinar Draft GFI On Enrichment Strategies For Clinical Trials To Support Approval Of Human Drugs And Biological Products – March 25, 2013