What Model in Clinical Design Should We Use to Examine New Cytostatic Compounds? by Elizabeth Eisenhauer MD

SLIDES & TRANSCRIPTS
Tuesday, February 15, 2000

What Model in Clinical Design Should We Use to Examine New Cytostatic Compounds?
Elizabeth Eisenhauer, MD

Slide 1:

DR. EISENHAUER: Thanks very much for the opportunity to address this series of issues which have already engendered some heat. I am not sure I am going to shed so much light but actually provoke you to think a little bit about what some of the options are in the development of agents which may not cause tumor regression in animals.

Slide 2:

At a certain point in the life cycle of any drug there is a sufficient buildup of laboratory experiments to warrant crossing the bridge into clinical evaluation, and one of the messages I want to convey is how we go about making decisions about clinical trial designs will be more and more dependent on the type of lab experiments that have been done and an increasing back and forth dialogue between the two banks of this river is important.

Slide 3:

We know as has been alluded to earlier that our approach to cytotoxic or cytocidal drug development as Langdon put it has been based on two key preclinical observations. The first is that there is a direct relationship between increasing dose and increasing anti-tumor effect and between increasing dose and toxic effects.

Slide 4:

Similarly in the lab tumor regression is seen with cytotoxics so that increasing doses will result in some, and I don't think we know in all, but in some animal models at highest doses.

Slide 5:

So we can expect regression if you extrapolate from this. So the bridge to the clinic for traditional cytotoxics has been built on these parallel dose-toxicity, dose-effect relationships and the observation of regression in established tumors such that for Phase I trials a surrogate for an effective dose is one which is toxic making toxicity the end point for dose selection and for Phase II trials a surrogate for efficacy is regression making response the end point, and these two principles have allowed the selection of doses and the selection of drugs which later on have proven to be effective as evidence by their outcome in survival in randomized trials.

Slide 6:

Where we are faced with some differences are agents that are specifically designed to target molecules in signaling pathways or in the vascular compartment in which three measurements are important -- what happens to target? what happens to tumor? what happens to toxicity in animal models?

Almost by definition you would expect there to be a parallel relationship between target inhibition and anti-tumor effect, perhaps with a plateau as you maximally inhibit target, and toxicity probably doesn't plateau with some of these agents because the mechanisms of toxicity may be different than the mechanisms of anti-tumor effect, and we see this, for example, with antisense molecules where there is a non-target class effect on things like platelets and fatigue and other parameters.

Furthermore, the relationship between toxicity and optimal dose may be shifted to the right so that doses which are optimal are, in fact, non-toxic as in the example, the hypothetical example Neal gave us on his last slide.

Slide 7:

So this presents problems in terms of Phase I design and the other observation is the one we focused on in the last session in the discussion, which is what if all you see is this, a dose-related effect on growth delay but no regression of established tumors?

Slide 8:

So the bridge to the clinic for these agents runs into problems. Dose toxicity and dose-effect relationships may not be parallel. There may not be regression, and for newer agents, therefore, the end point for Phase I trials is uncertain, how do we select dose and for Phase II trials the same debate rages.

Slide 9:

Let us look at what the alternatives are for Phase I dose selection? The goal of the Phase I trial has to remain the same. How do we pick a dose that we think will work? There are a number of alternative end points that have been suggested. I am not going to talk about surrogates for activity. I think that has been explored mostly in immunologic-based treatments where measures of T cell recognition of a specific antigen, for example, has been used to define doses.

Slide 10:

I will talk in more detail about the other two. The first of these that has received the biggest amount of attention has been the use of direct measures of target inhibition to assess what dose is optimal for a specific drug, and the huge advantage of this is it makes perfect sense. There are a number of challenges, however, in translating this perfectly sensible and rational end point into a real trial design.

The first challenge is that the relevant tissue, of course, is tumor, and we all know how hard it is to get sequential biopsies in patients, particularly those enrolled in Phase I trials. It is possible, but it is challenging.

The available tissue such as blood, or as was described this morning buccal mucosa, for example, actually may not be relevant and at the very least if we are going to say that this is what we need to do to find a dose in patients, we need to have shown in animal models that that same tissue, not tumor, but that same tissue, when the target effect is measured, shows a parallel impact as the tumor tissue does so that the final closing of the loop in the animal experiments with the prelamin, lamin measurements might be actually measuring what happens to murine buccal mucosa, for example, and to show that that moves in parallel with the same change in target tumor tissue.

The other question that I think Ed Korn will expand upon is sample size related to when you are looking at differences, not between cohorts, but are we seeing anything we want to avoid which is severe toxicity, the usual Phase I end point? But are we seeing a sufficient proportion of patients with the end point we want to achieve which might be 80 percent target inhibition? One to three patients per dose level is not going to be sufficient, and the final conundrum links into what we heard earlier this morning, assay measurements.

The standardization of the measurement of target inhibition is usually not achieved by the time you are doing Phase I trials. Secondly, there is some debate about whether we should be looking directly at that tyrosine-kinase inhibitor phosphorylation, for example, if that is what you are targeting or whether we should be looking downstream from that which is ultimately what we hope to perturb by the administration of an inhibitor of tyrosine kinase.

The final disadvantage, I guess you could say, of focusing all your energy on target inhibition is that some of these agents may have targets other than those we think are relevant. We have seen that historically with drugs such as doxorubicin, and as has been pointed out some of the inhibitors that are VEFG tyrosine kinase inhibitors, they may also have EGF receptor tyrosine kinase inhibitory activity which is more relevant.

Slide 11:

So there are big challenges with this. Where most trials end up in their design phase is to go back to these two parameters, assessment of blood levels and assessment of toxicity. Of course, these are simple and if we do achieve doses that are at least moderately toxic, we can feel reasonably comfortable we are not underdosing patients, which is a big concern. But there has to be some preclinical dose effect and blood level effect relationship of the drug in question to justify these as Phase I end points, and of course, the disadvantage is really where I started, that we may be choosing doses that are too high.

Slide 12:

So what is one approach? One approach that I have seen evolve for a number of these agents is as follows. The Phase I portion of the design is pretty traditional. There is rapid escalation with one to three patients per dose level to doses that, on a number of parameters, people think are likely to be active, that is, they are well above the minimal target blood concentration. They are starting to produce toxic effects.

Then you stop for a moment, and whether you call it a randomized Phase II or a randomized Phase IB, I guess it depends on how tumor specific you want to make it, but then there is an opportunity to say, let us look at that highest dose and one 50 percent or a log lower, depending upon the range you have examined, both of which produce blood levels that we think should be working and actually with 15 to 20 patients per arm explore the effect on target. Is there a dose relationship? If not, you might choose the lower non-toxic dose that gives the same target inhibition and proceed with that, and that can happen either in parallel with Phase II or other development or in sequence.

Slide 13:

However, the dose is arrived at beyond the dose selection all the other these drugs are actually exactly the same as any other therapeutic in this or other cancers, that is to demonstrate efficacy, and fundamentally there are two types of trials that demonstrate efficacy, those that screen so that you are rejecting or accepting drugs for further development and those which are definitive, and the screening trials traditionally are Phase II, and they use surrogate end point such as response.

Response in and of itself doesn't prove efficacy in my point of view. It is only when you select a drug on the basis of response and put it to a randomized test and see a survival difference do you see that you have a really effective drug that has made an incremental gain in treatment, and I think the same standards should apply to cytostatic agents in terms of this end point.

Slide 14:

So the options for development after Phase I can follow basically three paths, the traditional move to Phase II single agent studies to look for evidence of activity, moving right away into randomized trials, skipping the Phase II, often in a maintenance setting at maximal response, and we have seen examples of that or abandoning the idea of using a single agent at all and moving into combinations, Phase I and then randomized trials.

Slide 15:

What are the issues with single agent Phase II screening trials? I alluded to these earlier. If the agent doesn't cause regression in established tumors in animals, we feel -- we are not certain -- but we feel that response may be improbable. So, "Are there alternatives to response?" is the first question, and the second question is, "Do we really need to see that evidence of efficacy in non-randomized trials before we move into randomized studies, if the animal data are sufficiently compelling?"

Slide 16:

So the first question is, "Do we need to do Phase II?" I think not always, no. We don't always need to do Phase II, provided we have some basis for the rejection of this step and the only basis, again, is going to be lab data, but to mount a 1500-patient randomized trial certainly consumes resources. It certainly takes a lot longer to sort out whether your drug has made an impact, and if you have 10 angiogenesis inhibitors at the same time coming to that point it would be really nice to know before you did ten 1,500-patient trials which one was the most likely to succeed. I think there are substantial ethical considerations to exposing hundreds of patients to an agent for which you cannot honestly say that there is any clinical evidence that this is going to make a difference.

So there would be huge advantages to finding some screening endpoint in a non-randomized setting before you proceed to Phase III.

Slide 17:

So if there are alternatives to response which can be validated, it would be very useful to know.

What are some of the objective measures that people have suggested we use? Change in serum markers, and as many of you will know this is the approach that British Biotech took, not so much a fall in markers as a change in the slope of the rising curve for the marker. That has not really been validated as yet.

Time to progression has been mentioned. Time to progression or medians in time to progression with small patient numbers have big problems in terms of the confidence intervals around that observation. So there are some problems there. Measures of target inhibition I think it is important to clarify that actually showing that your target is inhibited is not a measure of clinical efficacy. It is a measure of whether you are inhibiting a target, and that is all.

The best example I can think of to make that point is in the aromatase inhibitor development. We knew if those drugs worked, serum estradiol levels should fall way down. Clearly they did and in a dose-related way but no one would have said that showing that you hit the target by measures of serum estradiol levels was enough to mount a randomized trial in metastatic or adjuvant disease. There needed to be some clinical correlate or clinical surrogate of benefit first.

Functional imaging, PET scanning, I think probably that will be discussed tomorrow. We do know in some tumor types that a change in tumor function is assessed by PET scanning, foretells when shrinkage will take place. What we don't know is if we see those changes in the absence of shrinkage whether it allows us to say the drug is going to show clinical benefit, and I am going to talk a little bit about progression rate separately.

The main point I want to make about all of these is that none of them have been validated, and by that I mean none of them have been used alone to select a drug which has later been shown to improve survival.

Slide 18:

Just a moment on progression rate. This came from two ideas within our own group. The first was a stopping rule that Benny Zhee described about a year ago when we looked at our Phase II database, we found that looking at not only the proportion of responses, but the proportion of patients progressing early on therapy, in other words in whom progression was the best response, that that allowed a little more refinement in our ability to appropriately reject drugs early, if they had high progression rates.

So beyond response progression rate contributed to correct decisions. The other observation hearkens back to something Langdon Miller was referring to, and that was you know a drug that produces a low response rate that has a survival impact how do you kind of put that together, and one way of putting it together is really to acknowledge that patients with partial response and stable disease may actually not be that different in their behavior to therapy. In some studies it appears that the benefit of therapy is actually due to the prevention of early progression rather than to tumor regression in a proportion of patients. An example within our own group again was a trial done some years ago in which best supportive care was compared to chemotherapy in non-small cell lung cancer. There was an overall survival benefit but only 20 percent or fewer patients actually showed a response. So the question was why should a change in 20 percent of the population of a partial response contribute to a shift in survival of the entire population

Slide 19:

and some hint of why was shown here when we looked at survival by best response seeing that patients who had stable disease and partial response really weren't substantially different from each other.

This has been shown in our ovarian cancer trials, in small cell lung cancer trials so that maybe the reason the population did better was there were fewer patients with progressive disease as their best response rather than a handful of patients with partial response, and I think it would be interesting to see similar data from the irinotecan study because there was a large proportion of patients with no change in that trial.

Slide 20:

So if we use this idea to select or reject new agents you kind of turn the stopping rule that we are used to on its head. Instead of setting a minimum response rate above which you choose a drug, you set a maximum progression rate above which you will reject the drug, and the progression rate of interest would vary according to the tumor type that you were interested in.

So for breast cancer we would be interested in a drug with a 30 percent response rate in untreated disease and a progression rate of less than 20 percent arbitrarily chosen, for non-small cell lung cancer a 20 percent response rate but a 30 percent progression rate. It would have to get below that limit to pass moving to the second stage and so on,

Slide 21:

and when we went back and looked at some of our data in these tumor types we saw that looking at response rate and looking at PD rate you did seem to see that, even though response rate was low, there was a cut point in the progression rate data suggesting that if we had looked at this alone we might have picked the same two active regimens, ignoring response altogether

Slide 22:

and for those of you who see glioma patients no drug we studied in untreated first line glioma showed sufficient responses, but you can see that a proportion of patients with early progression really did vary greatly.

We will never know whether those drugs would have made a difference in a randomized trial because they were rejected on the basis of response, but here is an idea of a way to screen which might be explored.

Slide 23:

Another point that has been raised, these are the growth characteristics of the animal tumors that are used to show whether a drug is interesting or not, the doubling time practically going straight up.

Human tumor, in this case breast cancer, has a much slower doubling time. So it is plausible that something that would only slow the growth of this might cause regression in that. So maybe we need not be so obsessed and worried that these drugs won't cause regression

Slide 24:

because non-cytotoxicity in vitro at a cellular level may not be equivalent to no change at a tumor mass level in vivo. The tumor mass is composed of different compartments of cells and halting the proliferation of one faction may actually cause or lead to involution over time.

The final point with respect to this is that inhibition of critical signaling pathways may in some tumors actually cause an apoptotic cellular death.

Slide 25:

The last two slides, this one, and Ed will talk more about this, when might we consider skipping Phase II to go back to my initial comments about Phase II? When the single agent data are compelling and there is no expectation of tumor regression as might be seen with metalloproteinase inhibitors. The Phase III trial is planned in a setting where no standard therapy exists such as in a post-adjuvant setting, and there is the appropriate early stopping in place.

For combinations, again, if it is foreseen that the place of this drug is going to be in combination with chemotherapy, expending a lot of time and resources doing single agent exploratory trials may not make a lot of sense. So moving right into a Phase I of combination and then a randomized trial of chemo plus or minus the new agent, again, might be logical,

Slide 26:

but finally, just to summarize novel non-cytotoxics really may require new approaches. We are not actually certain that we do. A lot of this is nervousness about this as opposed to genuine evidence that we need new approaches, but the message I want to leave you with is that none of the things I have talked about are really the ultimate test of efficacy which each of these agents will have to show.

We are not going to expect there to be a drug market on the basis of the change in progression rates. We might select a drug to do a randomized trial in on that basis, but ultimately we need to see changes in cure rate survival or quality of life to make a drug worth using at all.

Thank you.

(Applause.)