Martine Moossdorff

In this thesis has provided some answers that may help us avoid comparing apples and oranges in breast cancer research.

The first chapter showed that we actually are comparing apples and oranges, even when we focus on local and regional endpoints: many different endpoints are used, definitions are often not provided and if the endpoint is defined, the definition of the same endpoint may vary between studies.

The second chapter showed that worldwide experts in the field of breast cancer indeed disagreed at first about definitions of local and regional endpoints, but finally reached consensus on what we should classify as local event, second primary breast cancer, regional event, and distant event in breast cancer studies. Some issues were subject to debate as the expert panel considered that the available evidence was insufficient, namely whether contralateral lymph node recurrences are distant or regional events, whether we should distinguish between “true recurrences” from “ipsilateral second primary breast cancer” when a recurrence in the ipsilateral breast occurs, and whether irresectable recurrences should be considered distant regardless of location.

The third chapter builds forward on the question regarding contralateral lymph nodes. Whether a contralateral lymph node recurrence (CLNR) should be classified as distant or regional depends on its prognostic impact. This systematic review showed that prognosis of published cases of CLNR was more similar to regional events than distant events, namely 82.6% overall survival and 65.2% disease-free survival after a median follow up of 50 months. It also suggested that CLNR alone without any other recurrence has inferior prognosis compared to CLNR and simultaneous ipsilateral breast tumor recurrence (IBTR).

Because this review was limited by a small number of patients and a high risk of publication bias and therefore at risk overestimation of prognosis, Chapter 4 explores prognosis of CLNR in a larger population. It includes data from two national cancer registries and three individual hospitals. In contrast to the systematic review, this population had OS of only 30.2% after 5 years. This was worse compared to prognosis after ipsilateral lymph node recurrences (5-year OS 57.4%) but better compared to distant metastasis (5-year OS 10.1%). The study was limited by the fact that the CLNR population was diagnosed earlier (20% before 1995), potentially leading to underestimation of prognosis due to suboptimal treatment of both the initial cancer and the CLNR. Furthermore, information on both the presence of IBTR, and influence of detection method (repeat SN versus clinically evident CLNR) was lacking, and no conclusions could be drawn about those situations. Despite that, this chapter suggests that prognosis after CLNR is not as good as after ILNR, but considerably better than after distant metastasis. Therefore, all CLNRs are not necessarily similar to ILNR, but curative treatment may be suitable for individual patients.

Chapter 5 builds further on classification of lymph nodes, but focuses on infraclavicular lymph nodes at initial diagnosis. If affected they are classified as pN3a according to TNM, similar to presence of >10 affected axillary lymph nodes. This chapter shows that prognosis of patients staged as pN3a based on infraclavicular nodes is better than prognosis of patients staged pN3a based on >10 axillary lymph nodes, and suggests that the next TNM classification should not classify them in the same category. In conclusion, the first part of this thesis shows that in breast cancer research, many different endpoints are used and there is a need for more consistent definition. Regarding some issues with classification, evidence was unavailable, and Chapters 3-5 provide some answers to these questions.

The second part of this thesis focused on individual risk and the timing of local and regional recurrence. A first step towards more individual risk assessment is using the characteristics of tumor biology that we routinely measure in breast cancer patients: hormone and Her2 receptor status. We can divide tumors into subtypes with different biologic behavior and different response to therapy.

Chapter 6 studied the risk of local recurrence after mastectomy in these different subtypes. This study showed that these subtypes are important: their absolute risk of LR varies (triple negative tumors were at the highest risk), and also different risk factors are important in different subtypes. Furthermore, other studies have suggested that different subtypes may respond differently to treatment such as radiation therapy. This means that the decision which patients need local treatment such as radiation therapy should be tailored to subtype, and research investigating local treatment should describe the results separately for different subtypes of breast cancer so we can do better in the future.

Chapter 7 and 8 focus on the timing of local and regional recurrences in different subtypes of breast cancer. We often express prognosis in terms of 5-year risks, for instance “the 5-year risk of local recurrence is 3%”. But if a woman has finished treatment and has been breast cancer free for three years, is her risk of LR in the next 2 years still 3%? As an answer to this question, Chapter 7 firstly revealed that the risk of local recurrence in the first 5 years after diagnosis was already quite low at 3% directly after treatment. Secondly, after 3 event-free years, the risk of LR in the next two years was only 1%. A similar pattern was seen in Chapter 8 for regional recurrence (RR). The different subtypes showed different patterns of recurrence for both LR and RR: the subtypes with the highest risks (triple negative and Her2+ breast cancer) showed the fastest decline. This information can be reassuring to individual patients who have remained event free for a number of years. It also suggests that follow-up beyond 3 years may have low yield (although recurrences do happen). This is particularly important for breast cancer research, to estimate whether continued follow up will change the message of the study. This may lead to acceptance of earlier results, although longer follow-up may be necessary for other outcomes and for ER+ breast cancer. In individual patients, this information may be used as a starting point to tailor follow-up to individual needs, although there are many reasons for prolonged follow-up besides detecting local and regional recurrences.

In future breast cancer research, these findings may have a positive influence on the quality and fast availability of reliable results. Using clear and consistent definitions throughout breast cancer research will facilitate reliable comparison of results. Tailoring follow-up to subtypes of breast cancer is a first step towards reacting to the biologic behavior of the tumor, instead of a one-size-fits-all approach. The low absolute yield (as a result of the low number of events) may lead to evaluating (preliminary) outcomes after 3 years instead of 5 or 10 (at least for local endpoints). This may speed up certain studies although it will certainly not be possible for every trial.

Particularly the low risk of recurrence will be a challenge for future breast cancer research, as this will make it difficult to obtain statistically significant results. The focus on statistical significance of the results and the lack of power due to low number of events, however, sometimes distracts our attention from the actual size of the benefit. Lack of power because not the expected 5% but only 1% developed a recurrence, does not mean that the study is of low quality; it means that both treatments were really good. Furthermore, a difference between two interventions (even if statistically significant) may be so small, that we do not consider it clinically relevant. A very striking example was the ACOSOG Z11 study, randomizing women with cT1-2N0 breast cancer and 1-2 positive sentinel nodes after breast conserving therapy to either watchful waiting or axillary lymph node dissection. The study was closed early because of slow accrual and consequently, it was underpowered and no significant difference was seen between the treatment arms. If we look more closely at the actual risks, the absolute 5-year risk of regional recurrence was 0.9% without ALND and 0.5% with ALND after a median follow-up of 6.3 years. This difference was not significant, but had it been, it would not be clinically relevant and certainly not justify exposing all patients to the potential morbidity of ALND. Additionally, other meaningful outcomes such as OS and DFS were also not significant. The authors concluded that although prognosis is inferior in women having 1-2 positive sentinel nodes (compared to women without affected sentinel nodes), the axillary lymph node dissection did not improve this prognosis. Despite this, the results have not been implemented in our standard of care in The Netherlands.

When facing the challenge of low event rates, underpowered studies and non-significant results, we should be more flexible than to dismiss a study simply because few events prevented statistical significance. We should focus more on actual results and find new ways to reliably compare treatments.

This study brings us to a second challenge (or opportunity) for future breast cancer research. As Monica Morrow commented on the results of this ACOSOG Z11 so clearly: “Bigger surgery doesn’t overcome bad biology”. But how do we overcome bad biology? Can we recognize it? Can we target it to treat the cancer? First steps can be taken if we take breast cancer subtype into account, for instance based on receptor status (such as in Part 2 of this thesis). This information is already available for all breast cancer patients diagnosed today. Larger steps have already been taken by genetic profiling of individual tumors, and these tests are even commercially available at this time. Even bigger steps are being taken by studying the tumor even more closely, and find out what is actually happening on a molecular level within the tumor before, during, and after treatment. Through dedicated research, we are slowly learning what these tests mean for prognosis and for predicting which patients benefit from which treatment.

I believe this is the future of breast cancer treatment, but these studies will face the challenge of long follow-up and low event rates. Carefully choosing endpoints, ensuring clear endpoint definitions, balancing the expected yield of continued follow-up and reliable results, and dealing with low event rates in a way that benefits patients most are the keys towards better, faster, and stronger results for breast cancer patients.