Residual cancer burden after neoadjuvant therapy and long-term survival outcomes in breast cancer
Prof William Symmans - MD Anderson Cancer Center, Houston, Texas
Just with respect to disclosures, we did receive a patent for the mathematical formula for the residual cancer burden about eleven years ago and from the very beginning the residual cancer burden has been a freely and publically available website tool and will continue to be so. So I just wanted to clarify that.
This is really about organising the workflow in pathology to standardise the way we evaluate response after several months of neoadjuvant treatment. All of these data relate to post-chemotherapy response. The basic principle is that we estimate the area that still contains residual disease, map that area to the slides that we’ll be looking at under the microscope and create an image so that we can really reconstruct it and then be able to determine what area still contains actual cancer under the microscope. Then we combine that with the fraction of that area that still contains cancer cells that are invasive as well as the number of positive lymph nodes and the size of the largest metastasis. So there’s no special testing here, this is just organising what we would otherwise report anyway in pathology but doing it in a quantitative and standardised manner.
The website, you enter these six variables and hit ‘calculate’ and you get a residual cancer burden score where zero would be a pathologic complete response, i.e. no cancer left, and any increasing score above zero means there’s some cancer left. That is then categorised into a class, there are three classes, one, two and three – minimal, moderate and extensive disease. The website calculator page gets visited quite frequently, it’s just breached 16,000 visits per month. So it’s being used out there. We don’t know how many patients that actually affects because it could be visited multiple times for a patient experience but it is being used. The website has instructional videos and educational materials and protocols and illustrations and the diagrams I showed on the previous slide to really be there as a resource for pathologists who are using this or training people how to use it.
The purpose of this study was a multi-institutional pooled analysis. It was really led by ourselves and Laura Esserman who leads the ISPY consortium which we participate very actively together and Laura is in the room here if there are any questions and the discussion. So we identified sites and convened sites that we knew had an experience with this. Two of those sites, the I-SPY as a prospective trial, that’s the only one that’s really using it prospectively, collecting it prospectively in their trial, and the Cancer Research UK ARTEMIS randomised trial did a central review of RCB and so there are two trials embedded within this consortium of 5,160 patient responses.
So the headline result here, the most interesting result to my opinion, is that when you look at the residual cancer burden index scores we’re showing on the x-axis and you look at a function of survival, or in this case it’s the risk, on the y-axis, it’s plotted on a log scale, you can see a log linear relationship here. Now the implication of this result is that you can take an individual patient score of how much residual cancer burden they have and calibrate that to an accurate estimate of their risk over time. The scale on the y-axis, as I said, is on a log scale, it’s the relative risk of an event relative to patients who had a pathological complete response who had no disease left.
So the classes of residual cancer burden are shown in this plot by the vertical cut-offs and the classes are indicated at the top of the graph. These classes stratify risk in terms of event free survival and distant relapse free survival almost identical results. So we pretty much just carried on with event free survival as proof of principle for the further analyses which are to look at the subtypes of disease.
When we look in triple negative breast cancer, which this audience knows is the most naturally aggressive form of breast cancer, we see the classes clearly and strongly separating future prognostic risk. But I will point out that the format of the tables we’re showing in the subtypes have the frequency in the white rows, the five year event free survival estimate and confidence interval in the grey rows and the ten year estimates in the black rows. So one of the points here is that about half the patients get a pCR or an RCB1, in this case it’s 55% in the series. But we did find in this large meta-analysis that there was a lower survival probability in the RCB1 compared to the pCR in triple negative breast cancer.
In hormone receptor negative HER2 positive breast cancer, the patients who received HER2 targeted therapy as part of their neoadjuvant treatment, we see a couple of striking findings. First of all nearly 70% of these patients had a pathologic complete response, just really illustrating how effective these current treatments are. The 20% of patients who had an RCB2 in grey or an RCB3 in red have a significantly worse survival than the rest. So a small residual quintile of patients still at fairly substantial risk.
When we look at hormone receptor positive HER2 positive breast cancer, again we see the RCB classes are stratifying the risk. The interesting point to note here is that the RCB1, the yellow curve, really track with complete response for approximately five years and then there are some later events that we observed in this population.
In hormone receptor positive HER2 negative breast cancer, where there still remains a bit of residual confusion about whether or not chemotherapy can help patients with this type of disease, we see that the extent of residual disease is strongly prognostic. So there’s clearly an effect on prognosis from the chemotherapy but the most prognostic aspects of the distribution of response are when there is the most residual disease, the RCB3 in red and the RCB2 in grey, and it’s a long-term risk that’s still continuing beyond ten years.
Returning to this continuous relationship with the log of risk, we can see in each of the subtypes in the order that I’ve presented them, from left to right triple negative, hormone receptor negative HER2 positive, hormone receptor positive HER2 positive and on the farthest right hormone receptor positive HER2 negative, we see that that strong relationship with risk is very clear in each and every subtype of disease. Look how perfectly linear that risk curve, that log linear risk curve, is for the triple negative at left, and very tight confidence intervals around those risk estimates.
On the farthest right, the hormone receptor positive HER2 negative, this is the group where you get a slight sway to the right on the curve rather than a true diagonal linear effect. We’re interpreting that as the influence of adjuvant endocrine treatment but that will obviously deserve further study.
So in multivariate analysis with pre-treatment clinical and pathologic parameters we learned the following: that in each of the subtypes residual cancer burden index was independently significant, that a small minority of patients who present with T4 extensive disease in the breast still remain at significant risk independently. They’re a small group but it’s a dangerous group. And that in hormone receptor positive HER2 negative, pink on the farthest right, that the clinical nodal status before treatment and the histologic grade of the tumour continue to add independent prognostic information. In the triple negative breast cancer, blue on the farthest left, you can see that compared to the T2 presentation the T3, the larger tumour size, still adds some additional independent risk to the multivariate model.
So, to conclude, the prognostic association of residual cancer burden, both the index score and the class, was generalizable across a very large multi-centre pooled analysis. I didn’t show these data but they’re in the presentation – it adds prognostic information to the stage categories. Residual cancer burden was prognostic in each phenotypic subtype of disease, it’s independent of pre-treatment clinical and pathologic information. We also just want to highlight in the hormone receptor positive HER2 negative some of these pre-treatment clinical characteristics remain prognostic.
But the most important conclusion, in my mind, for the here and now is that there is a strong potential to calibrate an individual’s residual cancer burden index score to her residual prognostic risk. There is a generally linear relationship between RCB index value and the log of risk and that it is, as we have demonstrated, entirely feasible to have phenotype specific calibration risk curves for use in the communication of risk and the interpretation of clinical trials. Thank you.