Comparison between de novo and metachronous metastatic breast cancer: the presence of a primary tumour is not the only difference—a Dutch population-based study from 2008 to 2018

The aim of this study was to compare characteristics and survival of patients with de novo and metachronous metastatic breast cancer. Data of patients with metastatic breast cancer were obtained from the Netherlands Cancer Registry. Patients were categorized as having de novo metastatic breast cancer (n = 8656) if they had distant metastases at initial presentation, or metachronous metastatic disease (n = 2374) in case they developed metastases within 5 or 10 years after initial breast cancer diagnosis. Clinicopathological characteristics and treatments of these two groups were compared, after which multiple imputation was performed to account for missing data. Overall survival was compared for patients treated with systemic therapy in the metastatic setting, using Kaplan Meier curves and multivariable Cox proportional hazards models. The hazard ratio for overall survival of de novo versus metachronous metastases was assessed accounting for time-varying effects. Compared to metachronous patients, patients with de novo metastatic breast cancer were more likely to be ≥ 70 years, to have invasive lobular carcinoma, clinical T3 or T4 tumours, loco-regional lymph node metastases, HER2 positivity, bone only disease and to have received systemic therapy in the metastatic setting. They were less likely to have triple negative tumours and liver or brain metastases. Patients with de novo metastases survived longer (median 34.7 months) than patients with metachronous metastases (median 24.3 months) and the hazard ratio (0.75) varied over time. Differences in clinicopathological characteristics and survival between de novo and metachronous metastatic breast cancer highlight that these are distinct patients groups.


Introduction
In the Netherlands, around 14,500 patients annually are diagnosed with invasive breast cancer [1]. Around 5% of these patients present with de novo distant metastases at the time of initial diagnosis [2]. Moreover, in 15-20% distant metastases are diagnosed in the years following their initial breast cancer diagnosis (metachronous metastases) [3,4]. While systemic treatment, with a palliative intent, is the standard of care for both de novo and metachronous metastatic breast cancer (MBC) [5,6], there are specific therapeutic considerations for each group. For instance, in de novo MBC the best approach regarding the primary tumour is still unclear. Many studies suggested an overall survival (OS) benefit of local treatment [7][8][9], but recent randomized studies have refuted this [10,11]. Unlike patients with de novo MBC, many patients with metachronous MBC have already received (neoadjuvant or adjuvant) systemic treatment in addition to loco-regional treatment following diagnosis of the primary tumour. Recurrence following these previous systemic therapies could reflect resistance to these drugs or mean that the maximum tolerated cumulative dose of these drugs was already reached. Moreover, patients can suffer from lasting side effects and therefore be less fit for further systemic treatment. These specific considerations illustrate the importance of understanding differences between patients with de novo and metachronous MBC.

Data source
The Netherlands Cancer Registry (NCR) is a nationwide cancer registry hosted by the Netherlands Comprehensive Cancer Organisation (IKNL) and includes all patients with newly diagnosed cancer, with an estimated coverage of 96% [40]. Cancer diagnoses are notified through the nationwide Pathology Archive (PALGA) and the National Registry of Hospital Discharge Diagnoses. Trained data managers register data on diagnosis, clinicopathological characteristics and primary treatment directly from the patient files. Tumour location and morphology are coded according to the International Classification of Diseases for Oncology (ICD-O, third edition) and tumour stage is coded following the Tumour Node Metastasis (TNM) classification. Estrogen receptor (ER) and Progesterone receptor (PR) positivity of the primary tumour are set at ≥ 10% according to Dutch nationwide guidelines. Additional data on recurrences (including local, regional recurrences and distant metastases) were collected by the NCR retrospectively. Specifically, for patients with a primary diagnosis in 2003 and 2005 all recurrences up to 10 years after initial breast cancer diagnosis were identified. In addition, for half of the patients diagnosed in 2008, the first recurrence up to 5 years afterwards was identified and for patients diagnosed in the first quartile of 2012 (Q1 2012), all recurrences up to 5 years after diagnosis were recorded (Fig. 1). Recurrence locations were also registered following the ICD-O third edition. After diagnosis of metastases, the first systemic and/or local treatment is registered. Information on vital status was derived from the official municipal population database.

Patient selection
All patients diagnosed with de novo MBC between 2008 and 2018, as well as registered patients diagnosed with distant metachronous metastases between 2008-2018 were selected from the NCR. Patients with distant metastasis within 3 months of the primary diagnosis were excluded, because they are often considered as de novo MBC. For the survival analyses we excluded a small group of patients (n = 333, 3.0%) with tumour morphologies other than ductal carcinoma NOS or invasive lobular carcinoma (ILC). Moreover, we excluded patients from the survival analyses who had not received systemic therapy in the metastatic setting, because untreated patients likely have severe comorbidities and inherently have a different prognosis. Moreover, knowledge about patients eligible for systemic treatment can support medical oncologists' clinical decision-making.

Data definition
Patient characteristics (age at diagnosis of metastasis, sex, performance status), primary tumour characteristics (morphology, multifocality, tumour grade, clinical T stage, receptor and HER2 status), clinical N stage at primary diagnosis and location of metastases and treatment (local and systemic therapy after diagnosis of primary tumour and metastases) were analysed. Receptor and HER2 status of the metachronous metastases were not available. Period of metastasis diagnosis was categorized in 2008-2011, 2012-2015 and 2016-2018. Within metachronous MBC, metastasis free interval (MFI), defined as time between primary diagnosis and distant recurrence, was categorized in 3-12 months, 12-24 months, 24-60 months and MFI > 60 months. OS was analysed using time between diagnosis of distant metastasis and death or end of follow-up. If patients were alive at the end of follow-up (January 31st 2022), they were censored.

Statistical analysis
Descriptive statistics were used to depict clinicopathological and treatment characteristics and to describe missing data. Chi-squared tests or Fisher's exact tests were used to test difference in characteristics between de novo and metachronous MBC. To describe OS of the groups, Kaplan Meier curves were plotted and multivariable Cox proportional hazards (PH) analyses were performed including important confounders. Because some confounders included missing data, we used multiple imputation by chained equations (MICE) [41]. Data were incomplete for any of the chosen variables in 19% of patients, therefore we considered 19 simulated datasets to have a sufficiently reliable estimation of missing values [42,43]. The imputation model included MFI, which was used to categorize de novo versus metachronous MBC, and the following confounders: year of metastasis diagnosis; sex; age at metastasis diagnosis; tumour morphology; tumour multifocality; clinical T stage; clinical N stage; receptor status; metastasis location; tumour grade; performance status (assessed at primary diagnosis) and therapy variables, plus the outcome variables (vital status and the Empty stars indicate a possible diagnosis of distant metastasis that was not included in our study because it was outside the study period (green) or because it was not registered (red) Nelson-Aalen estimator) as was recommended in literature [44][45][46][47].
The multivariable Cox PH model comparing survival of patients with de novo and metachronous MBC included the same variables included in the imputation model, except for tumour grade and performance status, which were considered confounders but contained too many missings. For the variable 'age at metastasis diagnosis' we used a restricted cubic spline with four knots. We also excluded therapy variables for several reasons. First, therapy choices are partly determined by the variable of interest (de novo versus metachronous MBC influences therapy choices) and are therefore not a confounder but an intermediary variable. Second, specifically in de novo MBC a RCT reported that local therapy of the primary tumour did not improve OS [10,11], so local therapy is not a confounder either. Likelihood ratio tests of Cox PH models were used to compare OS between the groups. The PH assumption is the most important assumption underlying the Cox model [48]. The assumption was tested for each variable included in the Cox model using Schoenfeld residuals plots [49]). Confounding variables that did not meet the PH assumption were added as strata to the model. For the variable of interest (de novo versus metachronous MBC), we visualised the time-varying effect of the hazards by plotting the hazard ratio (HR) against time. For this purpose, we generated a time dependent Cox model with an interaction between the variable (de novo vs metachronous) and a restricted cubic spline of survival time with five knots [50]. This time dependent model was based on one imputation dataset.
In addition to the described analyses we performed supplementary analyses to study patients with de novo and metachronous MBC separately in more detail, to test the robustness of our results and to explore hypotheses behind the differences in OS (Supplementary methods 1).
A p value < 0.05 was considered significant. Statistical analyses were performed in Stata/SE 17.0 and R version 4.0.3.  Table 1 lists patient, tumour and treatment characteristics in patients with de novo and metachronous MBC. Most notable differences were those in T and N stages, metastasis locations and receptor status. Patients with de novo MBC were more likely than metachronous to have T3 or T4 tumours and positive loco-regional lymph nodes, while the majority (68%) of metachronous MBC had N0 at time of primary diagnosis. Notably, de novo metastases were more commonly limited to the bone with less frequent involvement of the liver or central nervous system (CNS) than in metachronous disease. However, in young de novo patients (≤ 40 years, n = 489) we saw more liver metastases (39%, versus 25% in de novo patients of all ages and 38% in young metachronous patients). CNS involvement was the same (3%) in de novo patients above or below 40 years of age. ER-negative/HER2-negative tumours were observed less in de novo MBC. Although ER positivity did not differ between de novo patients and the entire group of metachronous MBC, supplementary analyses showed that ER-positive (HER2 negative/unknown) tumours were more common in patients with metachronous MBC with a MFI > 60 months (83%), while patients with shorter MFI's had less ER positive tumours (39 to 65%) than patients with de novo MBC (67%) (supplementary Table 1). Supplementary Table 2 shows changes in patients with de novo MBC over time.

Treatment characteristics
Local treatment of the primary breast tumour was performed in 26% of patients with de novo MBC, in 43% consisting of surgery combined with radiotherapy, and in all patients with metachronous MBC at initial diagnosis. 74% of patients with metachronous metastases had received systemic treatment after primary tumour diagnosis.
Systemic therapy for metastatic disease was administered in 89% of de novo and in 79% of metachronous stage IV patients. Chemotherapy (without HER2 targeting agents) was administered less often to patients with de novo MBC (24% vs 29%) (Table 1). Meanwhile, de novo ER positive patients received endocrine treatment more often (67% vs 48%) and de novo HER2 positive patients received HER2 targeted therapy more often (75% vs 41%). Radiotherapy was the preferred locoregional treatment in de novo patients while surgery was more common in metachronous patients.
Supplementary Table 2 shows changes in treatment of patients with de novo MBC over time, these changes could not reliably be compared between de novo and metachronous MBC due to the method of registration for patients with metachronous metastases (from just four primary tumour years, Fig. 1.).

Discussion
In this large population-based study, we compared clinicopathological features, treatment and OS between all patients with de novo MBC diagnosed in the Netherlands between 2008 and 2018 to a group of patients with metachronous MBC diagnosed in that same period. Our study shows that these are not only two very distinct groups but also that patients with de novo MBC survive longer.
A number of differences in characteristics between the groups are notable. Our data corroborate earlier reports of frequent bone and lymph node metastases and less involvement of viscera and brain in patients with de novo MBC [4,30,31]. However, in our comparison of both groups, we did not observe more ER positive tumours in de novo patients [28], most likely caused by the 43% of metachronous patients with a MFI > 60 months, who are more often ER positive. Stage T3 and T4 and multifocal tumours were more often encountered in patients with de novo MBC, possibly reflecting delay in time to diagnosis. In de novo MBC, metastases limited to the bone were more common than in metachronous MBC resulting in increased use of endocrine treatment in this group, as was local radiotherapy on painful bone metastases [5]. The higher prevalence of triple negative tumours, liver and CNS metastases reflect unfavourable tumour characteristics and biology in metachronous patients. For example, triple negative tumours are known to metastasize hematogenously more often.
Regarding outcome, our data support a better OS in de novo MBC compared to patients with metachronous MBC, partly explained by differences in disease characteristics. Even after correction for known confounding characteristics, improved survival persists in patients with de novo MBC. This finding is consistent with previous literature [4,21,28,30,[33][34][35] showing that absolute differences in median OS range from 1.5 months (when comparing de novo to MFI > 24 months) to 20.3 months (comparing to MFI < 24 months) [31]. Without selection for MFI the reported median OS differences are similar to the 10.4 months found in our study [28,30,32,33]. In our sensitivity analysis we found a similar effect of MFI on OS compared to the literature: longer MFI was associated with longer OS after diagnosis of metastases. However, even MFI > 60 months did not have better survival than de novo MBC, which is not a consistent observation [21,31,34].
There are a number of possible explanations for the longer survival observed in de novo MBC. The most likely hypothesis is that shorter survival of patients with metachronous MBC is related to previous (neo)adjuvant systemic therapies. Recurrence despite previous therapy could reflect (1) disadvantageous tumour characteristics, (2) patient comorbidity/fitness and (3) primary or acquired therapy resistance. In our study, we could not quantify to what extent these factors played a role because we did not have data on the exact regimens and duration of systemic therapy nor could we account for fitness because data on performance score was only assessed at primary breast cancer diagnosis and contained too many missing values to include in the model. Nevertheless, the difference in use of systemic therapy in the metastatic setting (any systemic therapy in 89% of de novo and 79% of metachronous MBC and specifically targeted therapy in 75% of de novo and 41% of metachronous HER2 positive patients) may indicate that metachronous patients were less fit or had less treatment options for other reasons. As mentioned before, previous therapy can decrease treatment options in the metastatic setting due to acquired resistance to a drug, reaching a maximum tolerated cumulative dose or lasting side effects such as peripheral neuropathy or cardiotoxicity. Of note, due to the method of registration our data on systemic therapy could be an underestimation, as therapies administered not directly after metastasis diagnosis, but for example when symptoms did arise, may have been missed. In addition, the difference in targeted therapy among HER2 positive patients could be an overestimation because HER2 status was determined on the primary tumour and (a small percentage of) metachronous patients may have converted to HER2 negative. The same might be true for endocrine treatment, as metastases of an ER positive primary may be ER negative.
Still, we did find some evidence to corroborate the hypothesis that previous systemic therapy plays a role in the survival difference between de novo and metachronous MBC. In a supplementary exploratory analysis we observed a longer OS among patients with metachronous MBC who were not systemically treated for their primary tumour. While this metachronous group had favourable characteristics at primary tumour diagnosis, the difference remained after correcting for baseline characteristics.
We also hypothesize that differences in metastatic burden could contribute to the observed differences in survival. Possibly, clinicians are inclined to perform more (and perhaps more sensitive) diagnostic imaging in a patient presenting with de novo MBC than in those diagnosed with recurrent disease. This would lead to detection of smaller, asymptomatic or oligo metastases in de novo MBC, associated with longer survival and possibly even curative treatment options. Although we have no data on number or volume of metastases, our data do support this hypothesis (Supplementary Table 2) as we saw an increase in patients with de novo distant metastases limited to lymph nodes and increased use of anthracycline and taxane treatment (first choice in the neoadjuvant curative setting [5] and used for curative treatment of oligometastases).
In this study we extensively studied OS of patients with de novo and metachronous MBC using Kaplan Meier curves and Cox PH analysis. In the literature it is seldom reported whether the PH assumption was met and time-varying effects are often overlooked [51]. In our study, the variable of interest (de novo versus metachronous MBC) did not meet the PH assumption and therefore we additionally estimated the time-varying effects on OS. Overall, it appeared that the OS difference between de novo and metachronous MBC persisted over the years. The relatively small difference in OS between de novo and metachronous MBC in the first year could mean that a group of patients progress and die quickly despite any beneficial characteristic. Apparently, differences between de novo and metachronous MBC start to count after surviving longer than a year.
This study is unique as it presents a complete overview of patients with de novo MBC diagnosed in 2008-2018 in the Netherlands and the comparison to patients with metachronous MBC. The data convincingly shows that patients with de novo and metachronous MBC are distinct patient groups.
However, there are some limitations of our data. It would be relevant to also study metachronous patients in more detail using nationwide data. The MFI of our patients was probably not an accurate representation of all patients with metachronous MBC in 2008-2018 because the majority of metachronous patients in our cohort had their initial diagnosis in 2003 or 2005 (thus MFI at least 5 or 3 years, respectively) (Fig. 1). Another limitation is that for patients with an initial diagnosis in 2008, distant metastases were only Fig. 2 Overall survival in patients with versus patients metachronous metastatic breast cancer (treated with systemic therapy in the metastatic setting). The 95% confidence interval is indicated by colour around the line, number of patients at risk is noted below each year of follow-up. Overall survival is significantly longer in de novo MBC patients compared to metachronous MBC patients registered if they did not have a local or regional recurrence preceding the occurrence of metastases. Due to this registration difference, we probably missed approximately 20% of patients with metachronous MBC and an initial diagnosis in 2008 (i.e. of patients with a primary tumour in 2003/2005/ Q1 2012, about 20% had a local or regional preceding their distant metastases). Additional patients were missed because recurrences from 2008 had only been registered in half of the hospitals. Nationwide data including all patients with metachronous metastases in a given period would have allowed a more accurate comparison of the two groups. Ideally, such data would also include information on metastatic burden (e.g. oligometastases), receptor and HER2 status of the metastases and information about treatment administered in the metastatic setting in more detail and beyond those given as initial therapy.

Conclusion
Dutch patients with de novo MBC survive longer compared to patients with metachronous metastases, also following correction for different clinicopathological characteristics. Our data show that de novo and metachronous MBC represent two distinct groups, the presence of a primary tumour being not the only difference.

Acknowledgements
The authors thank the registration team of the Netherlands Comprehensive Cancer Organisation (IKNL) for the collection of data for the Netherlands Cancer Registry as well as IKNL staff for scientific advice. We thank Peter Zuithoff of the Julius Center for his statistical advice in the early stages of the project and we thank Linda McPhee for her helpful writing comments.  Fig. 3 Overall survival in patients with de novo versus metachronous metastatic breast cancer (treated with systemic therapy in the metastatic setting), hazard ratio over time in multivariable analysis performed on the first multiple imputed dataset. The 95% confidence interval is indicated by colour around each line. The difference in overall survival was not proportional over time, with a lower HR in favour of de novo MBC in the first years of follow-up in the multivariable model, while the HR starts to rise towards 1.0 after about 5 years. Note that the confidence interval widens after about eight years