Is There Life Beyond Histochemical Staging? by Daniel Sargent, PhD

SLIDES & TRANSCRIPTS
Tuesday, February 15, 2000

Is There Life Beyond Histochemical Staging?
Daniel Sargent, PhD

Slide 1:

DR. COMPTON: Dan Sargent will now talk to us about statistical issues. Don't go away, Bob. You forgot to mention that in the CALGB studies there are also a panel of histopathologic factors that are being evaluated.

DR. SARGENT: I went with the high technology, and it looks like that might be the way to go today given our success with the slides.

I would like to thank Carolyn for the invitation and thank all the speakers who have already gone before who really have provided a good introduction.

Dr. Hamilton suggested that we need a crystal ball or a fortuneteller, and there is nothing that I would like better than to be a fortuneteller that could sift through all of the data and come up with what are the, indeed, important prognostic and predictive factors.

What I am going to talk about today is why as a statistician I cannot do that right now and what we could do together so that is possible in the future.

Slide 2:

The challenge as we have already heard is that the number of possible prognostic and predictive factors is absolutely exploding, and I am going to talk in very general terms about prognostic and predictive factors. These could be factors that predict response to therapy, predict outcome, predict whether a patient will experience toxicity or not, so a very general setting for prognostic and predictive factors, and one of the premises that I am going to propose is that we are not meeting the challenge that is being posed to us right now.

Slide 3:

My first proposition is that we don't do prognostic studies very well, and they do not add significantly to our knowledge.

Slide 4:

How can I justify this statement? We could ask how many thousands of papers have appeared on new possible prognostic and predictive factors in colon cancer. I don't know, but it is many thousand, and contrast that to the answer how many new markers have been added to clinical practice in colon cancer B- very, very few. I think simply looking at the answers to those questions tells us that we could be doing a better job.

Slide 5:

What are some of the reasons for this? A lot of studies are completed on small series and convenient series. These are the series that were in my file cabinet, the blocks were; they were in the freezer down the hall. These were the 40 or 50 I could come up with pretty easily. What this leads to is we have very poorly defined patient populations that have different baseline characteristics. Some are Stage I. Some are Stage III. Some have received treatment. Some have not.

We have primaries and recurrent cancers lumped together, and we have varying quality of follow-up on these sort of studies.

Slide 6:

When we talk about analysis Carolyn has already mentioned that we really suffer from non-standardization of the assay techniques that are used, different staining protocols, different methods of interpretation and unless we reach some degree of standardization, and I think the CAP conference went a long way towards achieving that goal, we cannot make much progress.

We suffer from missing data in our analyses. The technology is not to a point where we can do all these assays on all the tumors. We have to exclude some tissue because we don't have both the normal and the tumor. So we have to work on technology based as was indicated by Dr. Warren in the last talk. Frequently we do not do multivariate analysis. So we end up with all these studies that show univariate predictive power, but when we pull together into a big multivariate analysis they don't turn out, and any time you do a study with 40 or 50 patients you are in a very low-power situation and even if it shows no prognostic importance that is probably of limited benefit because you probably had very limited power in the first place.

Slide 7:

As an example we have already heard today about p53 and colon cancer. McCloud and Murray in a review article in 1998 reviewed 20 studies on p53 in colon cancer, and they cited the vast variability in assay techniques, in the staining protocols, in the scoring protocols.

Therefore these studies have had vastly different results, some saying p53 mutations are beneficial, some saying that they are harmful, some saying that there is no effect whatsoever, and as a result we are left with no consensus.

Slide 8:

What are some ideas for solution? First of all, as Carolyn mentioned, we need to work toward standardization of our assay methods, at least publication of the assay methods so that we can understand what methods were used in each particular study.

Second, we should obtain patients when we can from well-defined patient populations such as clinical trials where we know what the baseline characteristics are, and we have good follow-up available. I think we are moving in this direction. Then the final premise I would offer would be that after initial or exploratory hypothesis-generating sort of work really the valuable studies are the large confirmatory studies that allow for multivariate analysis.

The reporting on my institution's experience here are 50 cases and this is the tenth published report on this that really serves to add more confusion than clarity in the literature.

Slide 9:

All right, my second proposition is that we are not taking full advantage of the current opportunities that we have.

Slide 10:

What can I say to back this up? The rates of block submission to cooperative group trials are, I would say unacceptably low. The best groups I know run about 80 to 85 percent in the block submission rate. If we ran a clinical trial, and we lost our outcome data on 20 percent of our patients, that would be unacceptable. However, we live with this every day in our tissue studies, and we should put the same priority on obtaining the tissue as we do on obtaining follow-up, and that alone would really benefit our tissue studies.

Secondly, the groups in the cancer centers have not developed or are just beginning to develop systems to collect blood and fresh tissue and for many of the new prognostic factors that are being studied we need either blood or fresh tissue, and I think that the groups in particular are running far behind in this area from what a lot of the cancer centers are doing.

Finally, many of the prognostic factor studies are limited to one group or one cancer center or one protocol. We just heard a good example in CALGB that they are going away from that by using tissue from two different protocols and combining them to do the same analysis.

That is what we need to do. We need to move even farther than that along the CALGB model of running intergroup tissue studies just like we run intergroup treatment studies.

Slide 11:

So the result of this is, again, small series, biased series. I will show you a couple of examples of that and again we don't have the proper data.

Slide 12:

This is some data that was kindly provided by Tom Pajak and Dave Grenion from an RTOG study. RTOG8610 was actually a prostate cancer study of 456 patients where they looked at six different tumor markers, and the goal was to get approximately 150 patients on each, test 150 patients with each marker,

Slide 13:

but unfortunately due to assay failures, missing data on one attribute or another, only 70 patients, they were able to obtain all six markers on only 15 percent of these patients. So as anyone who knows anything about multivariate analysis can tell you, to try to test six markers along with all the clinical and pathologic staging that already exists on 70 patients you are not going to get very far.

Slide 14:

A second example of selection bias is data on p53 expression from the same trial. They were able to get p53 determinations on 129 of their 456 patients which is really a not great place to start out from.

Slide 15:

They looked whether the patients with the p53 value and without a p53 value were balanced with respect to their Gleason score, which is an important prognostic factor, and they came out pretty well balanced on their Gleason score.

Slide 16:

They came out very well balanced with respect to what treatment they received,

Slide 17:

but then when they compared the patients that had a p53 determination with those that did not have a p53 determination they found a significant difference in survival between patients that they were able to test and patients that they were not, and I would suggest this has nothing to do with whether p53 is prognostic. What this indicates is they had a biased sample, and I would suggest that a plot like this should be included in every prognostic factor study to indicate whether the sample that you have is representative of the sample in your trial.

Slide 18:

Now, as much as it pains me to say this, statistics is not the solution. A number of different methods have been proposed including neural networks, recursive partitioning, the standard Cox regression model. I have done a bit of work in this area, and actually a lot of other people have as well, and a number of large comparative studies have indicated the new statistical techniques, while they have some advantages, they are not the solutions to the problems. We cannot just throw all the data into a black box and hope that the neural network will sort it all out.

If it were that simple we would have done it a long time ago. Unfortunately, it isn't.

Slide 19:

I don't want to be too bleak. There is considerable progress being made by a number of different parties. At the NCI, on the cancer diagnosis program web page, they have a very nice list of sites of different networks where people can go to get tissue. It doesn't have links to the tissue. Rather it says, "Call this person, go here, go there, they have this sort of tissue."

The cooperative human tissue network through the NCI is a source to obtain prospective tissue. However, this is not geared towards follow-up and therefore for a lot of the prognostic factor studies this is not particularly helpful.

Breast cancer is quite a bit ahead of the game compared to GI in this area. The cooperative breast cancer tissue resources which is also available on the NCI web page has a very nice searchable database where you can go in and say, I want patients that are age 50 to 59 that are ER positive, that received chemotherapy and were treated between 1985 and 1990, and it will give you a list of here are the 452 patients that meet those criteria.

We don't have anything close to that in GI cancer, and that is a very valuable resource. In addition, breast cancer has the intergroup breast correlative science review committee which is an active committee as opposed to the -- I sit on the GI committee. So I can criticize it. We are not terribly active on our GI committee. The breast committee meets regularly. They review and prioritize proposals. That is a model we could follow. In lung cancer SWOG has a new master tissue banking protocol that I think is an excellent idea. They basically have a protocol in lung cancer that says that any patient that goes onto a SWOG lung cancer protocol should have tissue submitted for undisclosed future purposes, and the consent form is generic in that sense. It does not specify that we will look at this target, that target and that target. It rather says, "We just want your tissue. Please give it to us."

We will see how successful that is or not. I cannot say. I will put up a slide in a moment that talks about what my dreams are at night.

Slide 20:

We are making some progress in colon cancer. ECOG and North Central have both done some studies that have combined tissue from a number of different studies and done overlapping analysis on tumor suppressor genes, immunohistochemical parameters. We just heard about the CALGB exercise doing these tests in a really grand scale 3500 patients. That is the kind of thing that we need to go to.

Slide 21:

So what is my dream and we may never get to this point, but in order to have a crystal ball, in order to go in there and really extract information that would be helpful first would be tissue collection on a grand scale to form a national, if not a tissue bank, at least a tissue database so that we could -- I am not suggesting that everyone give up their tissue but rather at least tell people in a central resource what tissue you have available so that it would be available to investigators who are interested.

We need better collaboration between the groups, between the cancer centers and between the groups in the cancer centers because a lot of these, in particular collection of fresh tissue and blood can be handled perhaps better in the cancer centers than in the groups, and ultimately what would be the best scenario would be to collect tissue on all patients not simply those who enter the therapeutic trials.

I know that a number of cancer centers are doing this already. Every patient who walks in the door we get tissue, and we get blood on them, and that is the kind of thing that we really need in order to move along.

We can simplify our lives quite a bit. A lot of times we only need very simple follow-up data. Is the patient alive or dead, and figure that out each year and that will go a long way, and of course, what we need here would be a generic sort of consent that would allow us to do all this.

Slide 22:

Final slide is with respect to how can we perhaps work towards achieving some of these. First of all we already have in place the GI correlative science review committee which could become more active to coordinate requests across groups and across studies and work to prioritize questions with input from all the relevant parties including surgeons, pathologists, oncologists and whoever else is interested, work towards more expanded national IRB or regional IRBs to deal with these consent issues which are very problematic, I understand and then work towards better pathology banking, if not banking then tracking systems so that we have information as to what tissues are available.

Thank you.

(Applause.)