Body Mass Index and Risk of Breast Cancer: A Nonlinear Dose-Response Meta-Analysis of Prospective Studies

Abstract

The role of Body Mass Index (BMI) for Breast Cancer (BC) remains to be great interest for a long time. However, the precise effect of nonlinear dose-response for BMI and BC risk is still unclear. We conducted a dose-response meta-analysis to quantitatively assess the effect of BMI on BC risk. Twelve prospective studies with 4,699 cases identified among 426,199 participants and 25 studies of 22,809 cases identified among 1,155,110 participants in premenopausal and postmenopausal groups, respectively, were included in this meta-analysis. Significant non-linear dose-response (P < 0.001) association was identified between BMI and BC risk in postmenopausal women. Individuals with BMI of 25, 30, and 35 kg/m² yielded relative risks (RRs) of 1.02 [95% confidence interval (CI): 0.98–1.06], 1.12 (95% CI: 1.01–1.24), and 1.26 (95% CI: 1.07–1.50), respectively, when compared to the mean level of the normal BMI range. However, inverse result though not significant was observed in premenopausal women. In conclusion, the results of this meta-analysis highlighted that obesity contributed to increased BC risk in a nonlinear dose-response manner in postmenopausal women, and it is important to realize that body weight control may be a crucial process to reduce BC susceptibility.

Due to the high prevalence, obesity has been one of major public health burdens in the world. It is estimated that 10–20% men and 15–25% women in Europe are obese¹, while 32.2% adult men and 35.5% adult women in United States². In China, as a developing country with a large population, the total number of overweight and obese people is already close to one quarter of all population³. Epidemiological studies indicated that obesity may contribute to the increased incidence of various human cancers⁴, among which breast cancer (BC) is the most common and leading cause of cancer death among female in the world⁵. With increasing trend of obesity epidemic, much more attention should be paid on this public health problem and new intervention approaches should be proposed.

Current several epidemiological evidences suggested that higher body mass index (BMI) was positively associated with increased BC risk in postmenopausal women⁶,⁷,⁸ but inversely reduced BC risk in premenopausal women⁹,¹⁰. However, even restricted in same menopausal status population, conflicting results were obtained among different studies⁶,¹¹. Though obesity was considered as a risk factor for BC by the International Agency for Research on Cancer (IRAC)¹², meta-analysis is still the best way to integrate the all available data to better interpret the relationship pattern of obesity and BC risk. A meta-analysis conducted by Ursin G, et al¹³ identified that high BMI was inversely associated with BC risk in premenopausal women. In addition, another meta-analysis conducted by Suzuki, et al¹⁴ suggested that the relationship between body weight and risk of BC may vary based on the menopausal status or estrogen receptor (ER) and progesterone receptor (PR) status. However, to the best of our knowledge, those previous reported meta-analyses just considered the linear relationship between BMI and BC risk or just compared the highest BMI versus the reference category to evaluate the association of BMI and BC risk which did not take into account of the inconstant change of the BC risk per unit change of the BMI. Therefore, we conducted this nonlinear dose-response meta-analysis of prospective studies to quantitatively and precisely evaluate the relevance of BMI and BC risk in different menopausal status of participants.

Results

Studies characteristics

A total of 12 articles⁹,¹⁰,¹¹,¹⁵,¹⁶,¹⁷,¹⁸,¹⁹,²⁰,²¹,²²,²³ containing 12 studies of 4,699 cases identified among 426,199 women in premenopausal group and 20 articles⁶,⁸,¹⁰,¹¹,¹⁵,¹⁶,¹⁷,¹⁹,²⁰,²¹,²³,²⁴,²⁵,²⁶,²⁷,²⁸,²⁹,³⁰,³¹,³² containing 25 studies of 22,809 cases identified among 1,155,110 women in postmenopausal group were finally included in the meta-analysis. The duration of follow-up ranged from 2.14 to 24.1 years. Among these studies (Table 1), 24 studies were conducted in white population, 5 in Asian, 2 in black and 6 in mixed population. All studies were published in English except one published in Chinese.

Table 1. the characteristics of the cohort studies about the BMI and BC risk included in the meta-analysis.

First author	Published year	Country	Ethnicity	Follow-up (mean years)	No. of BC cases	Sample size, no.	Person-years	Menopausal status	Adjustmenta
van den Brandt PA[24]	1997	Netherlands	White	4.3	553	1716	6284	Postmenopausal	1,5,6,7,8
Galanis DJ[15]	1998	US	White, Asian	14.9	86	11460	170754	Premenopausal	1,8,11,20
			White, Asian		292	11344	169025.60	Postmenopausal
Kaaks R[11]	1998	Netherlands	White	10.6	147	5891	56646	Premenopausal	1,5,6,
		Netherlands	White	10.6	76	3521	34362	Postmenopausal	7,12,33
		Netherlands	White	10.6	52	2068	19870	Postmenopausal(hysterectomy/ovariectomy)
Sonnenschein E[16]	1999	US	White	6.6	109	4475	29535	Premenopausal	1,2,5,6, 22,21
			White		150	3941	26010.60	Postmenopausal
Manjer J[17]	2001	Sweden	White	13.1	112	3873	58079	Premenopausal	1,4,5,7,8,10,
			White		157	5865	60845	Pere/Postmenopausal	13,23,31
Weiderpass E[18]	2004	Sweden and Norway	White	6.7	716	96986	775888	Premenopausal	1,2,4,5,6,7,10,19,26,27,28
Sweeney C[8]	2004	US	White	16	428	NA	124202	Postmenopausal (55–64)	1,2,4,5, 6,7,11
			White		1297	NA	282749	Postmenopausal (65–74)
			White		561	NA	101319	Postmenopausal (75–84)
Michels KB[9]	2006	US	White	10.8	1398	113130	1225520	Premenopausal	1,2,3,4,5,6,7,8,9,10
Suzuki R[6]	2006	Sweden	White	8.3	1284	51823	430331	Postmenopausal	1,2,3,4,5,6,7,8,10,11,13,14,15,16,17
Mellemkjaer L[25]	2006	Denmark	White	6.7	217	11796	78120	Postmenopausal (HRT user)	3,6,7,8, 11,13
			White		416	11992	78690	Postmenopausal (HRT no-user)
Li HL[20]	2006	China	Asian	5.66	221	37827	214164	Premenopausal	1,2,3,6,11,
			Asian		211	35583	201638	Postmenopausal	12,15,27
Tehard B[10]	2006	France	White	3.6	212	20000	72000	Premenopausal	2,3,5,6,7,
			White	4.7	1037	41857	196728	Postmenopausal	8,9,11,34
Palmer JR[21]	2007	US	Black	10	495	42538	316637	Premenopausal	1,2,5,6,7,9
		US	Black	10	454	9542	112263	Postmenopausal	11,12,13,19
Song YM[26]	2008	Korea	Asian	8.75	713	154693	1491465	Postmenopausal	1,4,8,9,15,23
Kerlikowske K[27]	2008	US	White, Asian and Black	2.14	4446	287115	614562	Postmenopausal	1,18,20,30,
Lacey Jr JV[28]	2009	US	Asian, Black	4.98	2063	69756	384968	Postmenopausal	1,2,3,4,5,6,7,13,18,20,31
Phipps AI[29]	2011	US	White, Asian and Black	7.9	2898	152126	1201795.4	Postmenopausal	1,2,9,11,20,29,30
Manders P[22]	2011	Netherlands	White	10	155	609	3013	Premenopausal	9
Opdahl S[30]	2011	Norway	White	24.1	2178	44952	1002895	Postmenopausal	1,5,24,26,34,35
Cecchini RS[23]	2012	North American	White	4.1	126	5850	23985	Premenopausal	10,13,23,3236,37,
			White	4.1	126	6352	26043.2	Postmenopausal
			White	6.4	555	19357	123884.8	Postmenopausal
Canchola AJ[31]	2012	US	White, non-white, Missing	12.1	1938	52642	636968.20	Postmenopausal	2,5,6,7,8,9,13,20,22
Gaudet MM[32]	2014	US	White	11.58	1088	28965	279665	Postmenopausal	2,3,4,6,7,8,9,10,11,12,13,
									20,23,30,32
Wada K[19]	2014	Japan	Asian	11.93	721	62689	789388	Premenopausal	1,5,6,7,8,18,23
				11.93	996	111446	1301876	Postmenopausal	1,5,6,7,8,12,18,23

Abbreviation: US, United States; NA, unkown; UK, United kingdom; HRT, hormone replacement therapy; BC, breast cancer.

a: 1 = age; 2 = family history of BC; 3 = history of benign breast disease; 4 = height; 5 = age at menarche; 6 = age at first birth; 7 = parity; 8 = alcohol consumption; 9 = physical activity; 10 = current and past oral contraceptive use; 11 = education; 12 = age at menopause; 13 = use of postmenopausal hormones (duration); 14 = the reason of menopause; 15 = energy intake; 16 = dietary fiber; 17 = total fat intake; 18 = area (study centers); 19 = BMI at 18 yr; 20 = ethnicity; 21 = weight/height ratio; 22 = history of breast biopsy; 23 = smoking; 24 = occupation; 25 = BMI at 20 yr; 26 = country of residence; 27 = total duration of breast-feeding; 28 = body shape at age 7; 29 = income; 30 = history of mammography; 31 = weight; 32 = history of diabetes; 33 = menopausal status; 34 = marital status; 35 = urban or rural community; 36 = treatment; 37 = Gail score; 38 = waist circumference.

Overall dose-response association between BMI and BC risk

The random-effects model was applied due to significant heterogeneity (P <0.001) was observed between studies in postmenopausal group. We found a significant nonlinear dose-response association (P <0.001, Figure 1) between BMI and BC risk with an increasing trend of RRs with per 1 kg/m² increase in BMI. When compared with reference (mean level of the normal BMI range, BMI = 21.75), the pooled RRs for BC risk were 1.02 (95% CI: 0.98–1.06) for BMI at 25 kg/m², 1.12 (95% CI: 1.01–1.24) for BMI at 30 kg/m², and 1.26 (95% CI: 1.07–1.50) for BMI at 35 kg/m² (Table S1). The Egger's regression test showed no evidence of publication bias with the P value of 0.236 (Figure S2).

Figure 1. The overall nonlinear dose-response meta-analysis of BMI and BC risk in postmenopausal women.

The BMI contributed to increased BC risk in a nonlinear dose-response manner in postmenopausal women (The reference was 21.75 kg/m² and the vertical lines indicate 95% confidence intervals).

In premenopausal women, the random model was applied as significant heterogeneity (P = 0.061) was identified. And we did not find evidence of non-linear relationship of BMI and premenopausal BC risk (P = 0.608). Additionally, no significant association was found between BMI and BC risk in premenopausal women when we modeled the linear relation of BMI and BC risk (RR = 0.989, 95% CI = 0.977–1.002 for every 5 kg/m² increase in BMI).

Stratified analyses in postmenopausal group

We performed stratified analyses according to ethnicity and follow-up year to explore the possible source of between-study heterogeneity in postmenopausal women. When stratified by ethnicity, 3 studies and 20 studies were included in Asian and European subgroups, respectively. The between-study heterogeneity was still existed in both groups (P <0.001 for Asian and P = 0.027 for European group). Furthermore, significant nonlinear dose-response relationship of BMI and BC risk was observed in European (P for nonlinear < 0.001, Figure S3). However, in Asian subgroup, no significant nonlinear association of BMI and BC was presented. The stratified analysis was not performed in Black subgroup due to only one study was included in this meta-analysis.

When stratified by follow-up year, 15 studies and 17 studies were included in <10 subgroup and ≧−10 subgroup, respectively. The between-study heterogeneities were still remained in both subgroups (P = 0.020 for <10 subgroup and P = 0.001 for ≧−10 subgroup). Significant nonlinear dose-response association was identified in <10 subgroup but was not find in ≧−10 subgroup (data not shown).

Sensitivity analysis

Similar results were presented in postmenopausal and premenopausal women before and after elimination of each study in the meta-analyses. These indicated that our results were stable.

Discussion

The meta-analysis of 25 prospective studies with 22,809 cases identified among 1,155,110 participants indicated a significant non-linear dose-response (P <0.001) association between BMI and BC risk in postmenopausal females. However, in the meta-analysis of 12 studies with 4,699 cases identified among 426,199 premenopausal women, we found inverse association of increasing BMI on BC development but with no significance. In postmenopausal women, the association was significantly identified among the subjects with a BMI more than 29. It suggested that the risk to develop BC was increased in women with obesity when compared to women in normal.

Although the mechanisms of the heterogeneous association of BMI and BC risk in different menopausal status women are still poorly understood, it is believed that the endogenous estrogen may be involved in the diverse attribution of BMI on BC risk in different menopausal status women³³. In premenopausal women, it was reported that there were more frequent anovulatory cycles in obese³⁴ which possibly protect against BC risk; and the clearance of free estrogen in liver was faster in obese than in lean women³⁵. All of these may finally lead to lower levels of both progesterone and estrogen. In contrast, among postmenopausal women, the excess of adipose tissue may elevate the production of endogenous estrogen through the increased activity of enzymes aromatase and 17β-hydroxysteroid dehydrogenase (17β-HSD). In parallel, the decrease of sex-hormone-binding globulin (SHBG) caused by obesity along with the effect of increased formation of oestrone and testosterone may finally promote cellular proliferation and inhibit apoptosis in breast³³. No significant association between BMI and BC risk in premenopausal women in our study was consistent with Cheraghi E et al.³⁶ identified. However, it was different from the studies conducted by Renehan, et al.⁴ and Amadou A et al.³⁷ reported that BMI was significantly associated with decreased BC risk in premenopausal women. These may be due to different methods applied in different meta-analyses. Those previous meta-analyses considered the linear association of BMI and BC risk and did not take into account the inconstant change of BC risk per unit change of BMI. In our study, we first tested the nonlinearity of the association between BMI and BC risk and then identified that BMI was associated with BC risk in a manner of non-linear dose-response in postmenopausal women. For all we know, this was the first report about the non-linear association of BMI and BC risk in postmenopausal women.

Subgroup analyses stratified by ethnicity and follow-up year were performed to explore the possible reasons of heterogeneity in postmenopausal group. However, the between-study heterogeneity was still remained when stratified by ethnicity or follow-up year. Significant non-linear dose-response association of BMI and BC risk was identified in White women. But in Asian subgroup, no significant association was observed. This may due to only three studies were included. The heterogeneous association of BMI and BC risk in different ethnicity may partly due to different genetic background. In addition, the difference in fat distribution, life styles and other BC risk factors in different ethnicities may also modify the association of BMI and BC risk. As for the follow-up, significant non-linear dose-response association between BC risk and BMI was identified in <10 years subgroup. However, no significant association was observed in ≧−10 years subgroup. This may be possibly owing to more complex confounders introduced into the longer follow-up studies.

Despite the clear strength of the current study due to comprehensive analysis strategy, several limitations also should be acknowledged. Obesity could affect BC risk through impacting circulating endogenous estrogen levels. However, we did not assess the modifiable effect of ER and PR status or hormone replacement therapy on BC risk due to the insufficient data available in this meta-analysis. Additionally, the methods used in our study restricted the number of studies included. Furthermore, the definition of menopausal status varied between different cohorts may lead to misclassification bias. Moreover, it is considered that BMI may not be the valid indicator for some people to assess the adiposity. Such as for elderly, the waist-to-hip ratio or waist circumference may be more suitable to predict risk of disease³⁸ since the shift of fat from peripheral to central abdominal sites with an accompanying increase in waist-to-hip ratio was discerned in older people³⁹. In addition, the recent advance in genetic studies had identified many genes and variants were associated with BMI level or BC risk⁴⁰,⁴¹,⁴², and whether there were interactions between genes and BMI should be further investigated.

In conclusion, the quantitative summary of the accumulated prospective evidence suggest that obesity may act as a risk factor for BC incidence in the manner of non-linear dose-response in postmenopausal women. Although the mechanism of obesity contributes to BC risk is still unclear; it is important to realize that body weight control should be considered as one of the most effective methods to reduce the BC susceptibility.

Methods

Selection of studies

A systematic literature research updated to 30 June 2014 was performed in the PubMed, ISI Web of Science with the language restriction in English and Chinese using the combination of the following on each term: “breast cancer” and “body mass index, overweight, or obesity” to identify eligible studies. In addition, all references listed in the retrieved articles and reviews were also scanned to further identify possible relevant publications.

Studies were eligible for inclusion in the meta-analysis if they satisfied the following criteria: (a) prospective studies evaluated the association between BMI and BC incidence; (b) study population was restricted in female humans; (c) the participants were categorized according to the menopausal status; (d) the exposure of interest was BMI with 3 or more quantitative categories; (e) the studies provide the RRs with 95% CIs and the number of cases and person - years for each BMI categories. When multiple studies had the same or overlapping study populations, only the studies contained the largest sample size or mostly completed were finally included.

As shown in Figure S1, the systematic literature research identified 5073 records, of which 5007 articles were excluded after review the abstracts or roughly scanning the full texts. Among these articles, 15 articles reported overlapping samples and 5 articles together with another 15 articles lacked BMI level-specific RRs and cases or person-years of each BMI categories were further excluded. Moreover, 9 articles were removed due to the unclear menopausal status of the participants.

Data extraction

The following information was extracted from each of the eligible publications: the name of the first author, publication year, the country in which the study was conducted, the ethnicity of the major participants, years of follow-up and person-years, the number of total BC cases (BC patients identified in the cohort) and the sample size of cohort (sample size of the study), menopausal status (premenopausal-, postmenopausal-), the categories of BMI and RR with 95% CI for each category, the number of case and sample size of each BMI category, the covariates adjusted for in the multivariable analysis. In addition, the ER and PR status of cases which were expected to be extracted were not collected due to the unavailability of the information from each included study. We extracted the RRs with 95% CI that reflected the greatest degree of adjustment for potentially confounding variables. The adjusted RRs were included in the meta-analysis if the studies provided the crude RRs and adjusted RRs. Furthermore, if the results were reported for 2 or more multivariable models, we extracted the RRs that reflected the maximal adjustment for possible confounders. If data was reported separately by age category, study center or HRT use in one publication, they were considered as different studies⁸,²³,²⁵.

The midpoint of the upper and lower boundaries of each category was assigned as the mean BMI to each corresponding RRs of every study. If the upper boundary for the highest category (such as ≧−30) and the lower boundary for the lowest category (such as <18.5) were not provided in the articles, we assumed that the boundary had the same amplitude as the adjacent category⁴³.

Statistical analysis

The dose-response meta-analysis was performed to evaluate a potential non-linear relationship between BMI levels and BC risk. We first applied restricted cubic splines with three knots in settled percentiles (10%, 50%, and 90%) of the distribution to model the possible association⁴⁴. Cubic splines are defined as piecewise-polynomial line segment which was used to present the nonlinearity association of response variable and covariate. And the boundaries of these segments are called knots. Then, the GLST command with the generalized least-squares regression, which required the cases, person-years and mean level of BMI in each category, as well as the BMI level-specific RRs with variance estimated for at least three quantitative categories of each article⁴⁵ was used to carry out the dose-response meta-analysis. Testing the null hypothesis that the coefficient of the second spline was equivalent to zero was used to evaluate the nonlinearity association between BMI and BC risk⁴⁶,⁴⁷. Then, the procedure described by Orsini and Greenland was finally used to estimate the pooled RRs for specific exposure values (per 1 kg/m² increase from BMI at 18 to 38 kg/m²) when compared to the reference (mean level of the normal BMI range)⁴⁸. Since the normal BMI of 18.5 - 25 was recommended to maintain a healthy condition by IRAC¹², we selected the mean level of normal BMI as the reference.

Heterogeneity among studies was assessed by the Q statistic test and was considered significant when P < 0.1⁴⁹. The fixed-effects model was applied when the heterogeneity was negligible; otherwise, the DerSimonian and Laird random-effects model⁵⁰ was used in the meta-analysis. The meta-analysis was performed separately in different menopausal status patients. Besides, stratified analysis of ethnicity (Asian, white, or black) and years of follow-up (≧−10 or < 10) were separately conducted. Publication bias was evaluated by the Egger's regression test⁵¹. Additionally, sensitivity analysis was performed to assess the stable of results by removing one study each time. All statistical analyses were performed by Stata Software (version 10.0) and P values of two-sided less than 0.05 were considered statistically significant.

Author Contributions

Conceived and designed the study strategy: J.G.; Acquisition of data, statistical analysis and interpretation of data: W.C., J.Y.L., R.R.; Drafting or revision of the manuscript: W.C., L.L., X.P.X.; Reference collection and data management: X.Q.C., Y.S.; Prepared the tables and figures: W.C., C.L.; Study supervision: J.G., P.Y.; All authors reviewed the manuscript.

Supplementary Material

Supplementary Information

Supplementary information