Edited excerpt from the discussion:

DR LUSKEY: This 55-year-old woman’s mother and sister both had breast cancer at approximately age 50. She had been undergoing routine mammography, and an abnormal mammogram ultimately led her to have a lumpectomy for a two-centimeter, Grade II, invasive ductal carcinoma. No lymphovascular invasion was seen, and sentinel node biopsy was negative.

The tumor was ER/PR-positive. HER2 was 2+ by immunohistochemistry but negative by FISH. She also has hypertension and diabetes mellitus and was moderately obese. She is single, works in a secretarial job and has friends through her church, although she always comes to the office unaccompanied.

We discussed her options for adjuvant therapy and opted to send her tumor tissue to Genomic Health for the Oncotype DX™ assay. The result showed that she fell into the low-risk group, with a recurrence score of 16. That corresponded with an average rate of distant recurrence at 10 years of 10 percent in the tamoxifen-treated patients in the validation study (Paik 2004a).

DR LOVE: If the result had been high, would she have been interested in chemotherapy?

DR LUSKEY: She would have been, yes. DR LOVE: What about hormonal therapy?

DR LUSKEY: We opted to treat her with hormonal therapy alone after postlumpectomy radiotherapy. As I’ve done with most postmenopausal patients, I presented an aromatase inhibitor as the first choice for adjuvant hormonal therapy and obtained a baseline bone mineral density test.

She has been on anastrozole for three months and is tolerating it well. Her baseline bone mineral density was excellent.

The real deliberation in her care was whether to pursue adjuvant chemotherapy or not. Before receiving the Oncotype DX assay results, this patient fell into a gray zone for me. Looking through the data and Adjuvant! Online sheets and bar graphs, it was difficult to come to an obvious decision about the right course to pursue. We used the Oncotype DX assay to help guide that decision for or against chemotherapy (1.1).

Whether that’s entirely right or wrong, that is the way we used it.

DR LOVE: Cliff, as you reflect on this common scenario, what are your thoughts about where Oncotype DX fits in and how you would go about selecting hormonal therapy in this situation?

DR HUDIS: This is a typical case of breast cancer. Postmenopausal women constitute the majority of women with breast cancer in America. Most of them have ER-positive tumors and negative nodes, so the truth is that this is a large subset of women. Until recently, we would have routinely said that a 55-year-old woman in reasonably good health with a two-centimeter tumor would be a candidate not only for hormone therapy but also for chemotherapy.

The Oncotype DX validation studies look at two questions. The first one is: Does the collection of genes that they were able to study in tissue prognosticate? And the answer appears to be yes. The second question is: Does it predict response to a specific chemotherapy regimen? And the answer also appears to be yes, although I would hasten to add that we could certainly use even more validation data.

I believe approaching this the way you did is exactly right. That is, rather than simply sending off the test to wait and see what happens, you say at the outset, “I’m ordering this test because I’m undecided about what to do, and the patient is undecided about what to do. Based on these results, we will make a more well-informed decision than the one we’ve already made.”

I think “the rub” in this case is that the patient is right on the cusp of the intermediate- and low-risk groups. It is clear from the validation study that a patient with a high score stands to gain substantially from a CMF- or MF-type chemotherapy regimen, given in addition to tamoxifen in that setting. What is a little less clear is where the cutoff is to say that there’s too little benefit for an individual patient to be treated. If you look at the confidence intervals for the 10-year point estimates of benefit — and that’s a slide that Dr Paik showed at San Antonio — there is a pretty wide confidence interval, especially for the intermediate-risk group (Paik 2004b).

For some patients, the upper margin, which is around five percent, is low enough for them to say they don’t want treatment. However, surveys of patients have found that even one and two percent marginal advantages over a five- or 10-year period are justification for treatment.

That brings us back to the patient. I think you have to go to this patient and say, “Our guess is that you would stand to receive a modest incremental improvement. That doesn’t mean you won’t get any improvement.” Then you and the patient together have to decide what to do. I find these cases easier if you come back with a recurrence score of five or six or 40. When you’re in the middle, it’s a little tough. I have some trepidation myself about withholding therapy that we might have called standard on the basis of a relatively novel test. If the patient agrees with that, that’s fine, but there is a lot of open communication that has to be well documented in these cases right now.

Having said all this, the other point to keep in mind, of course, is that if she is taking an aromatase inhibitor, then arguably she is receiving a slightly more effective therapy than tamoxifen, which, in fact, raises the bar for the chemotherapy benefit. This may change the whole equation — for this test, anyway.

On the issue of the aromatase inhibitor, I have to say — and this has become a bone of contention for some of us — that the various aromatase inhibitor versus tamoxifen trials essentially all used different definitions of disease-free survival and, to a large measure, have conflated prevention benefits with classic adjuvant therapy benefits.

When you are considering whether to give this patient chemotherapy in addition to hormone therapy, your reason for giving her chemotherapy is that you wanted to prevent distant metastatic disease and death. When we start to talk about the choice of adjuvant hormone therapy, especially with the aromatase inhibitors versus tamoxifen question, in most of the trials, nearly half of the putative benefit is for nonlethal events — either ipsilateral tumor recurrences or new primary disease, which are usually cured.

I’m not saying that it’s wrong to use those as a basis for choosing the AIs; I just want to highlight for everybody why, for example, ASCO’s position seems relatively conservative (Winer 2005). I don’t disagree with the choice of the AI, but I don’t have a particular dogma about it either.

DR LOVE: In general, what kind of chemotherapy would you have likely used if the patient’s recurrence score were high? What kind of hormonal therapy do you think you would have used?

DR HUDIS: Having said all of this about the AIs, I would still favor them because I believe they have a small toxicity advantage, and I think you can deal with the bone mineral density loss. I just don’t want it to sound like I have unbridled enthusiasm that is all one way. A role for tamoxifen probably exists, and I certainly don’t think that it’s wrong to start with tamoxifen and plan to switch at some point.

In terms of chemotherapy for this patient with an ER-positive, HER2-negative, lymph node-negative tumor, I would have been comfortable choosing CMF, especially given her hypertension, diabetes and obesity. For all those reasons, if I were to give chemotherapy, I would want to use a nonanthracycline- containing regimen. Frankly, that is what makes the Oncotype DX especially applicable in this case, because it is asking a CMF-type question.

Incidentally, SWOG-S8814 (Albain 2004) — which compared tamoxifen with or without CAF for node-positive, ER-positive breast cancer and suggests that the benefit of CAF largely accrues to the low-ER and intermediate- ER patients — is, as I understand, a candidate for testing and validation with the Oncotype DX. If that study shows what we would expect it to, and you can predict the CAF benefit with Oncotype DX, then we will start to have the kind of broad evidence we need to think of the test as a global chemotherapy predictor rather than a narrow CMF predictor.

DR LOVE: Joyce, specifically, how would you have managed this patient or one similar to her with a higher Oncotype DX recurrence score?

DR O’SHAUGHNESSY: Since she is 55, probably has only been postmenopausal for a few years and has a two-centimeter tumor, I’d be inclined to give her chemotherapy.

If you look at the patients in NSABP-B-20 who were under 60, there was a clear survival advantage with CMF (Fisher 2004). I’m going to have to be strongly talked out of chemotherapy, if you will.

I have started using the Oncotype DX assay, but I emphasize to patients that it is only one piece of the puzzle. With the original work from NSABP-B-14, Oncotype DX is only validated on approximately 650 blocks, and the same goes for B-20. They have constructed a nice curve, but if you think about how many patients are along every point there, it’s not a whole heck of a lot. So by and large, I have only been using the assay for select patients.

Oncotype DX is mainly driven by proliferation. That is what drives the high score, and that makes a lot of sense. More and more, we’re realizing chemotherapy helps those with high proliferative fractions. That’s an emerging theme. I am using it as one piece of the puzzle, but I need a detailed pathology analysis to get a real sense for that cancer and to make sure that it’s lining up with the Oncotype DX.

I would be inclined to use chemotherapy with this patient, and I must say I am impressed with CMF. There was more than a 30 percent absolute improvement in this “luminal B-type” subset in B-20 with that regimen. These patients would likely fall into the luminal B (highly proliferative) category as opposed to the indolent, strongly ER/PR-positive luminal A category. So at least in this subset, CMF looks incredibly effective and impressive, and I certainly think it is a reasonable option.

Generally speaking, I have not been adding in taxanes with this kind of patient. The other regimen that I tend to still use in this group — because I find it well tolerated — is FAC every three weeks for six cycles. That would probably be my most commonly used chemotherapy here.

I want to pick up on something Cliff said that I completely agree with. Eighteen was the “low” recurrence score, but my understanding is that the National Cancer Institute is going to conduct the PACCT trial, and they are planning to bring the upper limits of the low-risk group down to 15 or maybe even lower. They are then going to randomly assign those patients in the “intermediate” group to chemotherapy or no chemotherapy.

The upper limits of the confidence interval for that group were four or five percent, and I will often recommend chemotherapy for two or three percentage points. This is probably not a patient I would have ordered the Oncotype DX for, but in her particular case, with all the social considerations, I think that is reasonable. I have been using the test for smaller tumors, particularly if they are under 1.5 cm or with indolent-looking biology, to make sure that we’re not missing anything.

DR LOVE: Cliff, can you talk a little bit about the data that we have at different points in time with the AIs? Can you also comment on the recent data that have been presented from the BIG 1-98 (Thürlimann 2005) and Austrian-German trials (Jakesz 2004)?

DR HUDIS: My take-home conclusion is that at any point in time, the hazard rate for all counted events is probably lower for a patient who has been treated with an AI than for a patient who remains on tamoxifen. That doesn’t mean that the AIs are the right therapy but that the hazard rate for the events that have been counted seems to always drop.

In anastrozole versus tamoxifen in the ATAC trial (Howell 2005), a lower hazard rate for ipsilateral, contralateral and distant disease events exists for patients who received anastrozole up front.

That has led some to argue that you should use it up front, and they point out that although many of the recurrences are after five years, a peak hazard for recurrence occurs at around year two in these trials. If you “scrape that off” and lower it with anastrozole, you’re saving some patients who indeed may recur later, but better later than earlier or maybe never in that case.

The second study that we dealt with is CAN-NCIC-MA17 (Goss 2003). It’s unique because of all the trials that we have, it’s the only one that compared an AI to placebo. Of course, this came after four and a half to six years of tamoxifen. That study has a pretty clean result; it’s better to be on therapy than not. I think that is true in the subset of patients at the highest risk with node-positive disease.

A preplanned analysis shows not just a disease-free survival advantage but also an overall survival advantage. I have to say that makes some sense because in the node-positive setting, you would expect to have a greater proportion of the events that will influence survival. So it doesn’t surprise us, in a sense, that we see that advantage.

The third study was the IES trial, by Coombes et al, reported in the New England Journal (Coombes 2004). This trial sort of splits the difference because it gave patients two to three years of tamoxifen. This trial and the MA17 trial both suffer from a limit that you have to recognize, and the limit is called conditional probability.

That conditional probability limit means that patients don’t enter these trials until they have already done well for a period of time. These patients weren’t enrolled at day zero to take two years of tamoxifen and then switch or not. They were treated for two to three years, did fine and were then eligible to go on the study. That is one of those important caveats that make cross-study comparisons meaningless. They can’t be done, but in that trial, the switch reduced the rate of events.

Now, I’ve just described three different study designs, and in each case, going on an AI nudged the hazard function plot in the right direction. That is why I started off with my conclusion. I think that the AIs always do this, and that makes a fairly compelling argument for starting earlier with them, the caveat being that we don’t know about the rest of the general health issues.

BIG 1-98 is a complicated study and was actually reported only in part at St Gallen this year. The reason I mention that it’s complicated is — and I certainly didn’t appreciate this until recently — the study began as a two-arm trial. It was letrozole versus tamoxifen for five years. After about 1,800 patients were accrued, it was amended to be a four-arm study. Again, the same five years of each drug or two years of one drug and three years of the other in both sequences. The analysis that was presented was just for the up-front choice. The patients who were on the sequences were censored at the end of 24 months. So the median follow-up is actually longer than the median follow-up that can be reported, because they’re censoring out some of the patients.

This trial, consistent with the ATAC study, essentially shows that an advantage exists up front when one goes on the AI, using their definition. So I see these data as consistent. Continuing that theme, the ABCSG-ARNO studies (Jakesz 2004), which were combined and reported, essentially now show the same thing as the Coombes study except they used anastrozole rather than exemestane. But essentially, they found the same thing.

All five of these studies demonstrate that an AI lowers the hazard rate a little for breast cancer events when you lump together both distant and local events (1.2). The real question is: What is the best strategy? And I don’t think we have a clear answer for that right now.

DR LOVE: Can you comment on the cardiovascular mortality data that were presented from the BIG 1-98 trial and related data from the other studies?

DR HUDIS: It’s pretty hard to relate the data to the other studies because they have not gathered or reported in the same fashion. The take-home message was that there was seemingly an excess of cardiovascular events for the letrozole-treated patients in BIG 1-98.

The first-pass thinking for me is that you have to recognize that if you lower the number of first events that are breast cancer, more patients are at risk for other events. But then, of course, you ask: Is this something that is being seen across the studies, and is there something to explain it? In a related fashion, a doubling in the incidence of hypercholesterolemia existed also. In the US, they would have gotten statins pretty quickly for that, and in many of the European centers, they probably did not. The link between this temporally related rise in cholesterol and cardiovascular events certainly deserves to be explored.

ATAC has not clearly reported the same risk bump, and Tony Howell had a letter to the editor addressing this in The Lancet in the weeks after St Gallen. In the other trials, I’m not aware of data one way or the other, although I think they probably exist and I have not read them carefully enough to answer this. I think this issue speaks to the fact that we don’t know what the global effect of these drugs will be on the health of a large population, most of whom are never going to have a breast cancer event again. That’s the other thing that gets lost in all this.

It’s absolutely true that the numbers of breast cancer events are reduced. But if you take a cohort of typical aromatase inhibitor candidates, you’re dealing with an older population with lower-risk breast cancer by definition and a relatively low and steady rate of events.

Many of these patients are not destined to die of their breast cancer at all, and so the nonbreast cancer health implications of using these drugs have to be weighed.

What we don’t have for the AIs is a global score of health benefit. The data are presented to us in narrow compartments. For your breast cancer, it does this. For your bones, it does that. For your heart, it does something else maybe, and I don’t know how to integrate this yet, but I think that is the real challenge.

DR LOVE: Joyce, what’s your take on this and also the issue of differential effects of the three available AIs? People have talked about letrozole having more of an estrogenlowering effect. Do you think that this will get played out in terms of either efficacy or side effects and toxicity in the adjuvant setting? Are these cardiac data the beginning of that?

DR O’SHAUGHNESSY: Each of the three trials had a different definition for cardiac toxicity.

Rowan Chlebowski has been the breast chair of the Women’s Health Initiative (WHI), and of course cardiovascular events are a key primary endpoint (Anderson 2004). To tell you how laboriously this was assessed, every time patients come for a six-month follow-up, they are asked 33 questions about possible cardiovascular events. If any of the answers are “yes” — EKGs, trips to the ER, cholesterol levels, hypertension, glucose, et cetera — 20 pieces of clinical information are put together.

Then all the data are brought together, and two independent cardiology groups adjudicate. When you look at cardiovascular events in WHI, you know what you’re looking at. This was not done prospectively in any of these AI trials. So what we’re evaluating in those trials is difficult to sort out, and the definitions are different from one trial to another.

The numbers are also small, and cross-trial comparisons are difficult to sort out.

I’m increasingly convinced that there may be some cardioprotective effect of tamoxifen.

The Braithwaite meta-analysis (Braithwaite 2003) showed a 25 percent reduction in incidence and a 45 percent reduction in cardiovascular mortality in randomized breast cancer treatment trials of tamoxifen versus placebo. So there may be some protective effects of tamoxifen.

I believe that in the same way we watch the bones, the onus is on us with respect to cardiovascular disease. Patients are going to die of heart disease, and it is our job to make sure they’re having their homocysteine and cholesterol levels, et cetera, evaluated.

Rowan Chlebowski has calculated the net health benefit and analyzed the global health index for the AIs, much as they did in the WHI. He found that the aromatase inhibitors come out ahead of tamoxifen. It is difficult to catch up to the adverse events of tamoxifen with regard to endometrial cancer, DVTs and thromboembolic problems. So it appears that the AIs are safer.

Select publications