HomeEventsPhD Defence Marissa van Maaren

PhD Defence Marissa van Maaren

Local Management of early stage breast cancer and clinical risk prediction of survival 

Marissa van Maaren is a PhD student in the Health Technology and Services Research (HTSR) her supervisors are prof.dr. S. Siesling and prof.dr. P.M.P. Poortmans from the Faculty Behavioural, Management and Social Sience (BMS) 

Breast cancer is the second most commonly diagnosed cancer worldwide. In the Netherlands, breast cancer incidence is rising, while mortality rates are declining. Still, 1 in 27 women diagnosed with breast cancer will eventually die due to the disease. Taking into account the fact that over 14,000 women in the Netherlands are diagnosed with breast cancer each year, breast cancer remains an important health problem.

As local surgery is currently the most important component of curative breast cancer treatment, a large part of this thesis focused on trends in breast surgery and survival following breast-conserving therapy and mastectomy. Since breast-conserving therapy includes the use of postoperative radiation therapy, this thesis additionally addressed the effect of timing of radiation therapy on survival outcomes. Furthermore, attention was paid to the use of proper methodologies and the value of observational research on comparative treatment effects as a complement to randomised controlled trials.

The second part of this thesis concentrated on clinical risk prediction of survival in breast cancer. Here, the focus was laid on the validation of an existing prediction model that is frequently used in clinical practice, on the impact of breast cancer subtypes on recurrence rates, and on the use of conditional survival, i.e. taking into account the number of years a patient survived. 

PART I: Trends and survival in breast surgery and timing of postoperative radiation therapy

Chapter 2 describes the trends in the use of breast-conserving surgery in the Netherlands before and after publication of landmark trials that put this type of surgery on the map. It was shown that the incidence of breast-conserving surgery has increased over time, but that large variation existed between the nine Dutch regions. Throughout the period 1989 to 2002, the mean percentage of breast-conserving surgery in the Netherlands was 50.6%, varying from 39.1% to 71.7% among the nine regions. From 2003 to 2015, the average percentage of breast-conserving surgeries increased to 67.4%, fluctuating from 58.5% to 75.5% among the nine Dutch regions. The variation was larger before publication of the landmark trials, and attenuated after the publication of these trials, even though the variation was still considerable. Explanatory variables for the variation, besides geographical region, were age, T and N stage, sublocalisation of the tumour within the breast, tumour grade, histological tumour type, multifocality, hormonal receptor status, HER2 status, axillary lymph node dissection, adjuvant systemic therapy, and targeted therapy. Remaining variation can potentially be explained by unmeasured factors such as patient’ and doctors’ preferences and the degree of shared decision-making, and are subject of further research.

Subsequently, Chapter 3 showed that breast-conserving therapy led to at least equal 10-year overall, distant metastasis-free and relative survival rates in T1-2N0-1 stage breast cancer in the Netherlands, when adjusting for a large number of potential confounding variables, compared to mastectomy. After stratifying these results for tumour and nodal stage, the significant benefit of breast-conserving therapy as compared to mastectomy was only present in T1N0 stage breast cancer.

Chapter 4 showed similar outcomes in T1-2N2 stage breast cancer. As N2 staged patients generally have the indication for radiation therapy, breast-conserving therapy was compared with mastectomy with postoperative radiation therapy in this population. Both types of local treatment were shown to be equal in terms of 10-year overall, distant metastasis-free and relative survival.

To further explore treatment effects in the Dutch breast cancer population, in Chapter 5, a linkage with the Statistics Netherlands was performed to obtain information on cause of death. In addition, treatment effects were estimated in the following subgroups: age (<40, 40–49, 50–65, 66–75, >75 years), hormonal receptor status (negative, positive), HER2 status (negative, positive), adjuvant systemic therapy administered (none, chemotherapy, endocrine therapy, both chemotherapy and endocrine therapy) and comorbidity at time of diagnosis (yes/no). This study showed that breast-conserving therapy led to equal or better overall and breast cancer-specific survival rates as compared to mastectomy in all prognostic subgroups, after adjustment for confounding.

In Chapter 3, Chapter 4 and Chapter 5, treatment effects were studied in the Dutch population using data from the Netherlands Cancer Registry. These observational types of researches may suffer from residual, unmeasured confounding, confounding by severity – meaning that the severity of disease is associated with the administered treatment – and hospital-related variation. All these types of biases may affect the estimated outcomes and consequently lead to incorrect interpretations of treatment effects. This emphasises the importance of adequately designed studies using proper methods. There are many statistical methods developed that may deal with these types of biases. In Chapter 6, four different methods – multivariable Cox regression, propensity trimming, hierarchical modelling and instrumental variable analysis – were applied on the research question: “what is the effect of breast-conserving therapy and mastectomy on 10-year distant metastasis-free survival?” This study aimed to show the difficulties in selecting the right method and to inform others to carefully handle results from observational research. Results of propensity trimming (excluding patients with the highest probability of being treated with either one or the other treatment) and hierarchical modelling (using hospital of surgery as hierarchical level, thereby correcting for hospital-related variation) did not differ from those of conventional multivariable Cox regression: they all showed a similar significant benefit of breast-conserving therapy as compared to mastectomy on 10-year distant metastasis-free survival. Results of instrumental variable analyses, using hospital of surgery as instrumental variable, showed no significant difference between the two types of treatment. However, the assumptions of this statistical method were not considered to be met. This study concluded that, although assumptions of all methods need to be very carefully thought of, results from observational studies may be a valuable complement to randomised controlled trials, as long as they are interpreted cautiously.

Chapter 7 further emphasised the importance of observational research as compared to randomised controlled trials. Here, the strengths and weaknesses of both study types were discussed. In general, randomised controlled trials are close to provide perfectly unbiased treatment comparison estimates, due to the random allocation of patients to treatment groups. Nevertheless, translating overall treatment effects of a trial to an individual patient’s response is very difficult. Especially when taking into account the fact that randomised controlled trials are restricted to predefined limits in inclusion. Using population-based registries, several specific prognostic subgroups – such as older patients – can be studied which are not covered in the trial population. Observational studies can certainly not replace randomised controlled trials, but they give complementary information on the unselected real-life population where randomised controlled trials are not performed.

As breast-conserving surgery is typically followed by postoperative radiation therapy, one can imagine that the timing of radiation therapy following surgery may be important for survival outcomes. For this reason, the effect of different time intervals between breast-conserving surgery and postoperative radiation therapy on 10-year disease-free survival was investigated in Chapter 8. Here, two populations were studied. In population 1, patients receiving chemotherapy before radiation therapy were excluded, as the use of chemotherapy in between is largely related to the time interval. The following time intervals between surgery and radiation therapy were compared: <42 days, 42-55 days and >55 days. In population 2, patients receiving chemotherapy before and after radiation therapy were studied. For patients treated with chemotherapy after radiation therapy, 42-55 days and >55 days were compared with <42 days. For patients treated with chemotherapy before radiation therapy, 112-140 days and >140 days were compared with <112 days. For all of these analyses it was shown that shorter time intervals did not improve recurrence and survival rates. Therefore, the quality indicator in the Netherlands that states that patients have to be treated with radiation therapy within 35 days following surgery can be questioned, and patients should be informed that a slightly longer time interval does not negatively affect their risk of recurrence and survival. 

PART II: Clinical risk prediction in breast cancer

Prediction models used in the shared (treatment) decision-making process are not routinely validated, which may result in suboptimal treatment recommendations and unrealistic risk perceptions. As the target population may be different from the development population with respect to patient-, tumour- and treatment-related characteristics, validation of currently available prediction models is crucial. The online prediction tool PREDICT estimates 5-year and 10-year overall survival, and the expected benefit of adjuvant systemic therapy in patients with primary operated stage I-III breast cancer. After inactivation of Adjuvant! Online, PREDICT became the most frequently used prediction tool in Dutch clinical practice. To assess its validity on the Dutch population, this model was validated in Chapter 9. This validation study showed that PREDICT accurately predicts overall survival in most Dutch breast cancer patients, but that results for both 5-year and 10-year overall survival should be interpreted cautiously in ER negative tumours, as PREDICT slightly underestimated survival outcomes for this group. In addition, 10-year overall survival was shown to be overestimated in patients 75 years and older, in T3 tumours and patients treated with both endocrine therapy and chemotherapy.

The relevance of breast cancer subtypes in clinical risk prediction is becoming increasingly recognised, as the underlying biology of the tumour is shown to greatly affect recurrence and survival rates. In Chapter 10, the effect of breast cancer subtypes on 10-year recurrence and survival rates in the Dutch population was described. It was shown that breast cancer subtypes were independently associated with these survival outcomes: luminal A disease was related to the lowest percentages of recurrences, while triple negative disease showed the highest percentages of recurrences. This knowledge further paves the way to increased patient-tailored treatment and individualised follow-up.

Accurate risk prediction is not only important for treatment decision-making and a patient’s risk perception. It is also of at least equal relevance for cancer survivors who are finding their way back into the society. In the Netherlands, a study showed that approximately 60% of cancer survivors are facing difficulties in the application for a life insurance. The data used by life insurers for these decisions were based on internationally available survival rates, which were considered to be outdated and not representative for the Dutch population. As most cancer survivors only apply for a life insurance a couple of years after their diagnosis, and it is known that survival rates increase as time passes by, it is of great importance to take into account the number of years survived in estimating a patient’s prognosis. Chapter 11 described the development of a model predicting the conditional annual extra risk on mortality compared to the general Dutch population. By accounting for the number of years survived, this model provides more accurate estimates for patients who apply for a life insurance several years after their diagnosis. The model was both internally and externally validated with satisfying results. The model was additionally tested on a random selection of patients from the Netherlands Cancer Registry, and compared to insurer’s guidelines that were valid before the introduction of this model. Test results showed that the newly developed model, in general, increased insurability of breast cancer patients. Further monitoring of the use of this model is necessary to identify any remaining problems in the application for a life insurance needing further research.

In Chapter 12, conditional 10-year recurrence and survival rates were assessed for patients with early stage breast cancer, according to T and N stage and breast cancer subtype as important prognostic factors. It was revealed that differences in outcomes between these prognostic subgroups attenuated as time passed by. For this reason, the most important message from this study is that conditional survival estimates specific for a risk profile should be used in patient communication. This should lead to a more accurate risk perception of patients. 

In conclusion, this thesis showed that breast-conserving therapy is at least equal to mastectomy in T1-2N0-2 stage breast cancer in terms of 10-year overall and breast cancer-specific survival. Combined with results of randomised controlled trials, this thesis contributes to increased awareness that a mastectomy does not reduce the risk of recurrence and mortality, as compared to breast-conserving therapy. Besides, this thesis revealed the importance of specific risk profiles and use of conditional survival in clinical risk prediction. Overall, communication to patients, based on relevant data, plays a central role and should form the basis of the shared decision-making process. Ultimately, this will lead to better treatment decisions and more realistic risk perceptions.