Prognosis and Expanded Definition of Pregnancy-associated Breast Cancer: A Dose-Response Meta-Analysis

Background Pregnancy-associated breast cancer (PABC) is defined as breast cancer that is diagnosed during pregnancy and/or postpartum period. Deﬁnitions of the duration in the postpartum have controversial and this variability may led to diverse results on the prognosis. Moreover, evidence on the dose-response association between the time from last pregnancy to breast cancer diagnosis and overall mortality has not been synthesized. Methods We systematically searched PubMed, Embase, and the Cochrane Database for observational studies about the prognosis of PABC published up to June 1, 2019. We estimated summary adjusted hazard ratios (HRs) and the corresponding 95% confidence intervals (CIs). Subgroup analyses based on diagnosed time, PABC definition, geographic region, year of publication and the way of HR estimate were performed. Additionally, the dose–response analysis was conducted by using the variance weighted least-squares regression (vwls) trend estimation. Results A total of 54 articles (76 studies) were included in our study. PABC is associated with poor prognosis on overall survival (OS), disease-free survival (DFS) and cause-special survival (CSS), and the pooled HRs with 95% CIs were 1.45 (1.30-1.63), 1.39 (1.25-1.54) and 1.40 (1.17-1.68), respectively. According to subgroup analyses, the varied definition of PABC led to diverse results. The dose-response analysis indicated a non-linear association between the time from last pregnancy to breast cancer diagnosis and overall mortality ( P <0.001). Compared with nulliparous women, PABC had a HR for all cause death that peaked at 1.62 (95% CI, 1.46-1.80) at 28 months since last pregnancy, starting from 78 months since last pregnancy became insignificant 1.07(95%CI, 0.99-1.26). It suggested that the

definition of PABC should be extended to include cases diagnosed up to about sixyears postpartum (78 months since last pregnancy) in order to capture this ongoing increased risk.
Conclusion This meta-analysis suggests that PABC is associated with poor prognosis, and the definition of PABC should be extended to include cases diagnosed up to about six-years postpartum.

Introduction
Breast cancer is the second most common cancer worldwide and the most commonly occurring malignancy in women [1]. Because of a trend of delayed childbearing, the number of women with breast cancer during a pregnancy or in the subsequent few years after a pregnancy is expected to increase [2].Breast cancer occurring during pregnancy presents a challenging clinical situation since the welfare of both the mother and the fetus must be taken into account in any treatment planning.
Conventionally, pregnancy-associated breast cancer (PABC) is defined as breast cancer that is diagnosed during pregnancy or in the postpartum period. Definitions of how many years after delivery a breast cancer can be diagnosed under this definition have ranged from 0.5 to 5 years, and sometimes even longer [3,4]. PABC is viewed as a clinically and biologically special type of breast cancer and only comprises 0.2-0.4% of all breast cancers [5,6]. However, it is the most common cancer in pregnancy, and is diagnosed in approximately 15 to 35 per 100,000 deliveries, with fewer breast cancer cases diagnosed during pregnancy than during the postpartum period [7][8][9][10].
The literatures prove pregnancy itself may transiently increase the risk of developing breast cancer, despite its long-term protective effect on the development of breast cancer [11,12]. But whether PABC has a worse prognosis is currently controversial. A meta-analysis published on 2016 has shown that risk of death increased in women with PABC compared with non-PABC (pooled hazard ratio (HR), 1.57; 95% confidence interval (CI), 1.35-1.82) [13]. However, other recent studies found there was no significant difference in prognosis of PABC and non-PABC [14][15][16][17]. Meanwhile, the specific definition of PABC has varied throughout the literatures and this variability may led to diverse results on the relationship among pregnancy, postpartum and breast cancer. So it is necessary to specify the definition of PABC through summarizing epidemiological evidence. Moreover, evidence on the dose-response association between the time from last pregnancy to breast cancer diagnosis and prognosis has not been synthesized. This study was initiated to understand prognosis of PABC and examine the dose-response relationship to provide quantitative evidence on defining PABC.

Search Strategy
This meta-analysis were performed in accordance with the preferred reporting items for systematic reviews and meta-analysis (PRISMA) guidelines. We did our best to include studies published until date, regarding prognosis of PABC. Eligible studies were found by searching PubMed, Embase, and Cochrane library for relevant reports published before June 1, 2019. The keywords used for the search were ("pregnan*" OR "gestation*" OR "childbirth" OR "postpartum" OR "parity") AND "breast" AND ("cancer" OR "neoplasia" OR "carcinoma"). The reference lists of all retrieved articles and previous systematic reviews were manually searched for additional relevant studies.

Inclusion and Exclusion Criteria
All eligible studies should meet the following criteria: (1) observational prognosis studies with follow-up period longer than 6 months; (2) participants were diagnosed with breast cancer on clinical diagnosis and/or histologically; (3) case group were diagnosed as PABC, control group were non-PABC or nulliparity; (4) the outcomes were in terms of overall survival (OS), disease-free survival (DFS) and cause special survival (CSS); (5) the risk point estimate was reported as an HR with 95% CI, or the data were presented such that an HR with 95% CI could be calculated. The exclusion criteria were as follows: (1) duplicated or irrelevant articles; (2) reviews, letters, case reports; (3) non-human studies; (4) study with inappropriate data for metaanalysis.

Data Extraction
Two reviews extracted data independently using a predefined data extraction form.

Assessment of Study Quality
The methodological quality of studies was assessed by the Newcastle-Ottawa scale (NOS) for observational studies [18]. A score of 0-9 was allocated to each study, with higher scores indicating higher quality.

Meta and Statistical Analysis
We used adjusted HRs and 95%CIs which are most appropriate for time-to-data events. If HRs were not reported directly, we could estimate HRs from crude data or Kaplan-Meier curve [19]. Fixed or random effect model was used to summarize the study-specific estimates. I 2 was used to assess heterogeneity across studies. Visual inspection of the funnel plot, Egger's and Begg's tests were performed to assess publication bias. Subgroup analyses were performed according to diagnosed time, PABC definition, geographic region, year of publication and the way of HR estimate.
The dose-response association between the time from last pregnancy to breast cancer diagnosis and overall mortality was assessed using the variance weighted least-squares regression (VWLS) model [20]. Restricted cubic splines were used to examine the time from last pregnancy to breast cancer diagnosis as a continuous, nonlinear exposure, defining time with knots at the 5th, 35th, 65th, and 95th percentiles for the distribution [21]. The time from last pregnancy to breast cancer diagnosis reported in each study were converted to month. We used the average value of the lower and upper limits of each category. If the lowest category was open ended, the average value of the upper limit and 0 was used. If the highest category was open, the average value was defined as 1.5 times of the lower limit.
All the statistical analyses were performed using STATA Version 13.0. P<0.05 was considered significant.

Search Results and Study Characteristics
We initially identified 12414 articles and screened their titles and abstracts, Figure   1. After duplicated and irrelevant articles were excluded, 54 articles with 76 studies met the inclusion criteria and were thus included in our meta-analysis. Quality of studies was assessed based on the 9-stars Newcastle-Ottawa Scale, ranged from 6 to 9 (mean of 7.2). The characteristics of the studies were summarized in Table 1.   3.6 Dose-response association between the time from last pregnancy to breast cancer diagnosis and overall mortality As the meta-analysis included studies reporting the HRs with their 95% CIs of overall mortality relating to three or more categories of time since last pregnancy, all the studies were eligible to be included in the dose-response analysis. A total of ten studies were included in dose-response meta-analysis, nulliparous were taken as comparison,

Publication Bias
As shown in Figure 6, each point represents an independent study of the indicated association, visual inspection of the funnel plot did not suggest an evidence of publication bias among the articles (Egger's test, P=0.451; Begg's test, P=0.077).

Discussion
We reviewed and meta-analyzed the existing scientific literatures on the prognosis of PABC to obtain the powerful conclusion that PABC is associated with a poor prognosis. This is the largest and latest meta-analysis in this field. It included a larger number of participants, thus reducing small-study effect to a great degree.
And the studies included in our meta-analysis were with relatively high quality. The mean of Newcastle-Ottawa score of studies was 7.2.
There are two explanations that may account for the results. On one hand, mammary gland involution following pregnancy has been suggested to explain the poor prognosis [71]. Breast involution occurs as a tissue remodeling process to the state indistinguishable from the nulliparous gland through wound healing, inflammatory milieus and immune infiltrate [72,73], which supposedly promotes tumor progression. On the other hand, pregnancy and breastfeeding lead less timely detection and clinical examination. The delayed diagnosis that allows the tumor more time to grow, increasing the metastatic potential of the disease [52,74].
Pregnancy also makes treatment strategy more conservative to ensure the safety of the fetus [10,75]. However, the exact reasons of the poor prognosis of PABC are needed to explore in future.
To the best of our knowledge, this is the first dose-response meta-analysis providing comprehensive insights into the association between the time from last pregnancy to breast cancer diagnosis and overall mortality of PABC. The scientific value of dose-response meta-analysis is higher than meta-analysis with exposure classified as two categories [20,76]. Using this method, we were able to provide a detailed and more flexible description of the occurrence of outcomes throughout the observed range of exposure. Through the vwlr with random effects model, we found a non-linear direct association between the time from last pregnancy to breast cancer diagnosis and the overall mortality. Compared with nulliparous, PABC women had an elevated all-cause death risk that peaked at 28 months since last pregnancy and lasted to 78 months. We propose that the definition of PABC should include cases diagnosed up to at least 6-years postpartum to better delineate the increased risk imparted by a postpartum diagnosis. These findings also provide valuable insights into further research. Callihan's cohort demonstrate that a breast cancer diagnosed within 5-years postpartum has a significant higher risk for metastasis and mortality risk compared to nulliparous cases [58].Compared to his cohort, our doseresponse meta-analysis provide a higher-level quality of evidence to expand the Thirdly, high between-study heterogeneity is another limitation of the current metaanalysis. It was likely due to significant differences in sample size, the definition of PABC and/or treatment interventions. Lastly, language of studies was limited to English, which may result in potential language bias.

Conclusions
In summary, this meta-analysis suggests that PABC is associated with a poor prognosis on overall OS, DFS and CSS compared to non-PABC cases. Acknowledgements: Not applicable Figure 1 Schematic representation of the study selection process Hazard ratios and 95% CIs of studies included in meta-analysis of CSS Figure 5 Dose-response relation between the time from last pregnancy to breast cancer diagnosis and Figure 6 Funnel plot to explore the presence of publication bias.

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download.