Discussions
In this meta-analysis, there was a significant positive association between BMI and breast cancer risk among postmenopausal women but no correlation among total premenopausal women. Non-linear dose-response analysis showed that the breast cancer risk increased by 3.4% for every 1 kg/m2 increment of BMI in postmenopausal women.
Heterogeneity was observed among the studies included in our meta-analysis. The diversity most likely comes from differences in study design, sample source and size, reference categories, different BMI cut-offs, different weight distribution, length of follow up, racial difference, ER status, HRT use or other potential confounders. We conducted a subgroup analysis to explore the heterogeneity and potential factors influencing the associations. The heterogeneity was zero (I
2 = 0) when stratified by ER status, suggesting ER status may be an interactive factor for the relationship between BMI and breast cancer risk. The lastest WCRF 2017 report showed that in postmenopausal women BMI was significantly positively associated with ER+ breast cancer, but not ER– breast cancers, it was consistent with our study.
The mechanism of the association between obesity and breast cancer remains unclear. Currently it is thought that adipose tissue of obese women produces excessive estrogen hormone, which in turn stimulates more estrogen-sensitive breast tissues that may already have a propensity for hyper-stimulation, ultimately promoting the formation and development of tumors [
46]. In premenopausal women, estrogen is mainly derived from the ovaries, while after menopause, most circulating estrogen is derived from the conversion of adrenal androgens by means of adipose aromatase [
47,
48]. Therefore, women with higher amounts of body fat tend to have higher levels of circulating estrogen. Moreover, studies have found a stronger relationship between obesity and estrogen receptor (ER) positive breast cancers than between obesity and ER-negative cancers [
42,
44]. Interestingly, a history of using hormone therapy attenuates the relationship between obesity and breast cancer risk among postmenopausal women [
3,
4]. These findings provide further evidence for the estrogen availability theory among postmenopausal women, and our analysis also confirmed this point. In addition, other hormonal factors including insulin resistance, adipocytokines, AMP-activated protein kinase and leptin, which are related to obesity, are also important factors for the formation and development of breast cancer [
47,
49‐
51].
Though hormonal etiology may partially explain the differential role of BMI in relation to breast cancer risk, it is increasingly considered that premenopausal breast cancer cases might have distinct etiologies from postmenopausal breast cancer cases. A recent study [
52] showed that obesity has divergent impacts on risk of aggressive subtypes of breast cancer in premenopausal versus postmenopausal women, and that some non-hormonal pathways may also mediate the association between obesity and TN breast cancers among premenopausal women. Obesity-related factors such as inflammation, elevated levels of insulin and insulin-like growth factors, or other carcinogens may play different roles on postmenopausal and premenopausal breast cancer risk.
The heterogeneity observed in this meta-analysis remained after stratification by a previous history of HRT use or not, and the positive association only existed in postmenopausal women who never used HRT. The results showed that HRT may be a factor influencing the relationship between BMI and postmenopausal breast cancer, and it is similar to the WCRF 2017 report. One study reported that the relationship between HRT and breast cancer risk was related to race, BMI and breast density among postmenopausal women [
53]. Compared with HRT non-users, HRT users were at increased risk of breast cancer; however, there was no statistical correlation between HRT and breast cancer when the women were obese. Considering the above results, it is likely that BMI and a history of HRT use may interact to affect the incidence of breast cancer.
When stratified by geographical location among postmenopausal women, a significant non-linear dose-response relationship between BMI and breast cancer risk existed in American and Asian women, but not among European women. These results were inconsistent with the previous report conducted by Xia Xiaoping, et al. [
54] Xia showed that significant non-linear dose-response association of BMI and BC risk was identified in White women, while in an Asian subgroup, no significant association was observed. This may be because the studies included in our study were divided into America, Europe and Asia, whereas in their study ethnicity was divided into White and Asian and did not take into account Black or African Americans, which make up a significant proportion of the US population. In our study, the breast cancer risk in Asia was considerably higher than in America, this may be because Asian people are more likely to have higher body fat with less lean mass and skeletal muscle [
55], which may increase breast cancer risk. In China, changes in reproductive patterns (such as family planning, less breastfeeding, or older age at first full-term pregnancy) are affecting the concentrations of sex hormones, and coupled with changing lifestyle and dietary factors may be related to an increasing risk of breast cancer [
56]. For Europe, a larger heterogeneity (
I
2 = 71%) between studies remained, and the relationship between BMI and BC risk was not statistically significant. In WCRF 2017 report, dose-response meta-analyses for postmenopausal breast cancer by geographical location showed a statistically significant increased risk in North American and European studies, and a stronger association in Asian studies. The AICR&WCRF 2017 report includes 19 cohort and case-control studies of Europe (
I
2 = 75%), while 9 cohort studies of Europe (
I
2 = 71%) were included in this study. Potential confounding factors such as the difference in fat distribution, life styles or a possible interaction effect might be the source of the high heterogeneity. It necessitates further studies in order to provide more insights into such effects among European women.
For premenopausal women, the AICR&WCRF report summarized the epidemiologic evidence available up to 2017 and concluded that there is a substantial dose response relationship for the inverse association between BMI and risk of breast cancer among premenopausal women. However, the non-significant association of BMI with risk of premenopausal breast cancer observed in this study is not consistent with the WCRF report. This may be because the studies included in this study differed from those included in AICR&WCRF 2017 report. The studies included in this article were all cohort studies which were more convincing, however, the number of studies included in AICR&WCRF 2017 report were more than this study, and more recent studies may be included. Therefore, the sample source and size, different fat distribution, follow-up year, race, ER status, HRT use or other confounders are possibly the reasons of differences.
A lack of data prohibited many subgroup analyses from being carried out. However, stratification by geographical location demonstrated an inverse significant association between BMI and breast cancer among European premenopausal women but not among Asian and American premenopausal women. This conclusion is consistent with the AICR&WCRF 2017 report when stratified by geographical location. When compared to women with a normal BMI (normal BMI was set as 21.5), women with a higher BMI were at a decreased risk of breast cancer risk by 2.3%. In Asian women, the results showed a positive non-linear relationship between BMI and breast cancer when BMI was above 25 kg/m
2. In American premenopausal women, there was no association between BMI and breast cancer. These results are comparable to those observed in a systematic review by Amadou et al. [
57], which explored the relationship between overweight, obesity and premenopausal breast cancer according to ethnicity. However, the results in this review are not completely consistent with our findings in premenopausal women, a significant inverse association remained among Africans and Caucasians after stratification, while a significant positive association was detected among Asian women. This is possibly due to the inclusion of 19 case control studies and 11 cohort studies, while all 18 studies within our meta-analysis were cohort. The similarities were the results of Europeans in our analysis and the Caucasians in the study by Amadou et al. [
57], in addition, Asian women in our study showed a positive correlation when BMI was above 25 kg/m
2
, which was similar to the report of Amadou et al. These results suggest that ethnicity or geographical location may be a factor influencing the relationship between BMI and breast cancer risk, but this conclusion is speculative. In this meta-analysis, breast cancer risk could be influenced by overweight or obesity according to different geographical locations, irrespective of menopausal status. The difference may be related to different genetic susceptibility among different ethnic groups, as demonstrated for breast cancer susceptibility genes (BRCA1 or BRCA2) mutations previously [
58].
BMI, adult weight gain, body weight and fat distribution may play a role in the prognosis of breast cancer separately or together in combination. IARC working group [
59,
60] reported that higher weight gain was associated with decreased risk for premenopausal women and increased risk for postmenopausal, and the WCRF also reported that adult weight gain is a probable cause of postmenopausal breast cancer. Weight gain is a widely used index of obesity in adult, and a dynamic measure unlike static measures such as BMI [
61], therefore, it may be better to combine BMI and adult weight gain to explore the relationship between obesity and breast cancer.
The studies included in this meta-analysis were all cohort studies, with large sample sizes, and adequate subgroup analyses. Moreover, the confirmation of the breast cancer cases was done by diagnosis, hospital medical record or linkage data. Thus, the results were persuasive, and were less likely to suffer from recall bias and selection bias. However, some limitations also exist for this meta-analysis; the measurement of BMI was determined mostly by questionnaire investigation or self-reported, and people tended to report their weight lower and height higher than reality. The adjustments factors of each study were not the same, thus the RRs we extracted from the studies were possibly affected by potential confounders even after selecting the adjusted RRs. In addition, the categorization of BMI in a number of articles included in this study was not in accordance with the WHO standard, thus misclassifications may be caused when we conduct the highest versus the lowest BMI meta-analysis.