Introduction

Intrauterine environmental exposures to endogenous or exogenous hormones, notably estrogens, may influence the subsequent development of breast cancer in offspring [1]. During pregnancy, levels of circulating estrogens and other hormones with growth-enhancing properties are at least 10 times higher than those in nonpregnant women, with increases seen with advancing gestational age [2–4]. The hypothesis that breast cancer in daughters may be influenced by the intrauterine environment is receiving increased attention [5]. Perinatal factors, including birth weight, birth order, maternal age, gestational age, twin status, and parental smoking, have been postulated as risk factors for breast cancer through altered hormonal influences on the developing fetal mammary glands [1]. Despite ample biologic plausibility, this hypothesis is difficult to evaluate directly [5], and previous epidemiologic studies have reported conflicting results [6, 7].

Here we review the epidemiologic studies that have assessed the association between perinatal factors and breast cancer risk in daughters. A meta-analytical approach was applied in order to clarify further the possible role played by the intrauterine environment in the etiology of breast cancer.

Materials and methods

Identification of studies

The data retrieved for the systematic review were based on searches of all published papers, letters, abstracts, and review articles on birth weight, birth order, maternal age, gestational age, twin status, and maternal or paternal smoking and breast cancer using the MEDLINE database from January 1966 through February 2007. We used keywords combining text words, with terms for six perinatal factors combined with terms for breast cancer (Table 1). We also manually searched the reference lists of all studies that fulfilled the inclusion criteria for further relevant publications. Articles were included in our systematic review if they fulfilled the following three criteria: the exposure status of at least one of six perinatal risk factors of interest was compared with nonexposure status; the outcome focused on the daughter's breast cancer morbidity or mortality using an epidemiologic study design (case-control design, data linkage study, or cohort study design); and the article was written in English language. We excluded animal studies, investigations focusing on male breast cancer, reviews, and studies that did not provide separate relative risks for breast cancer. We also excluded studies if odds ratios (ORs) or relative risks (RRs) were not specifically provided, raw data were not available for calculation of risks, or the emphasis of analyses was on hazard ratios or standardized incidence ratios.

Table 1 Search terms used in systematic review
Full size table

Statistical analyses

For the purposes of meta-analysis, birth weight was classified in three different ways: five categories (<2,500 g, 2,500 to 2,999 g [referent], 3,000 to 3,499 g, 3,500 to 3,999 g, and ≥4,000 g); three categories (<3,000 g [referent], 3,001 to 3,999 g, and ≥4,000 g); and two categories (<3,000 g [or ≤3,000 g; referent] and ≥3,000 g [or >3,000 g]). Birth order was examined using two different categorical schemes: 1 (referent) versus ≥2; and 1 (referent), 2 to 4, and ≥5. Maternal age was classified into three categories: <25 years old (referent), 25 to 29 years old, and ≥30 years old. Gestational age was also analyzed in two ways: ≤36 weeks versus ≥37 weeks (referent); and ≤32 weeks versus ≥33 weeks (referent). To examine twin status, three classification schemes were employed: twin versus singleton (referent); monozygotic twin (or sister twins if zygosity was not reported) versus singleton (referent); and dizygotic twin (or sister-brother twins if zygosity was not reported) versus singleton (referent). Maternal or paternal smoking was considered as follows: no smoking during pregnancy (referent) versus smoking during pregnancy. If the criteria utilized in an article were slightly different from our criteria, then we included the data and described the difference in a footnote.

ORs and 95% confidence intervals (CIs) were recalculated from published frequency tables of individual studies using the Mantel-Haenszel common OR estimate. However, the reported OR (95% CI) was used when the published studies did not provide further details as to the frequencies of the exposure variables. If the manuscript reported the results after performing a stratified analysis, then we re-calculated the crude OR by combining across strata. A random-effects model was used to obtain summary ORs and 95% CIs.

Heterogeneity was assessed by heterogeneity test using Cochran Q statistics [8]. Publication bias was assessed according to the Egger regression asymmetry test, and the Begg and Mazumdar adjusted rank correlation tests [9, 10]. The Egger test is a simple linear regression of the natural log of ORs or RRs against its precision (the inverse of its standard error) [10]. The Begg and Mazumdar rank correlation test reports the rank correlation (Kendall's tau) between the standardized effect size and the standard errors of these effects. If asymmetry is caused by publication bias, then we would expect that high standard errors (small studies) would be associated with larger effect sizes (ORs or RRs) or low standard errors (large studies) would be associated with smaller effect sizes [9]. The Begg and Mazumdar test makes fewer assumptions than does the Egger test, but it is insensitive to many types of bias (lower power) that the Egger test is sensitive to [11].

When significant heterogeneity or publication bias was found, we performed subgroup analyses by study design (case-control study versus cohort study), and the source of information (data linkage versus self-report) to assess the impact on between-study variations (heterogeneity). Many of the large-scale studies, especially cohort studies, were published after 2000; thus, we classified publication years into before 2000 versus 2000 or later in order to assess publication biases associated with small study sizes. Few studies focused on either Asians or African-Americans, and so it was not possible to examine ethnicity effects. All statistical analyses were conducted using STATA (Version 8.2 [special edition]; Stata Corp., College Station, TX, USA).

Results

We identified 34 studies that assessed the association between birth weight and breast cancer risk (Table 2): 19 case-control studies (eight population based, three nested, six record linkage based, and two twin-based) and 15 cohort studies (seven population based and eight record linkage based). Many studies showed positive association between heavier birth weight and breast cancer risk [12–31], and some of the studies observed stronger effects among younger (<45 years) or premenopausal women [14, 22, 29, 30]. In contrast, studies observing no association [32–41] or a negative one [42–45] also have been reported. Additionally, some authors reported a J-shaped relationship between birth weight and breast cancer risk [12, 14, 15, 18, 25, 33, 35, 37, 45], particularly for early-onset cancers [18].

Table 2 Studies assessing the association of birth weight and the risk for breast cancer
Full size table

Among 34 studies of birth weight and breast cancer, we selected studies that employed the same categories of birth weight. To evaluate whether a J-shaped relationship existed, we grouped birth weight into more than three categories. The findings of meta-analysis of eight studies that utilized five categories of birth weight (<2,500, 2,500 to 2,999, 3,000 to 3,499, 3,500 to 3,999, and ≥4,000 g) and 11 studies that used three categories (<3,000, 3,000 to 3,999, and ≥4,000 g) are shown in Figures 1 and 2. To include more studies, we also categorized birthweights as <3,000 g (or ≤3,000 g) and ≥3,000 g (or >3,000 g; Figure 3). Sixteen studies among all 34 studies were included in the meta-analyses for birth weight and breast cancer: seven studies [13, 14, 18, 32, 33, 35, 44] were included in the all three meta-analyses; four studies [22, 23, 27, 39] were included in the two of the three meta-analyses; and five studies were included in only one meta-analysis [12, 16, 28, 34, 38]. There was no significant heterogeneity across studies (P_Q test > 0.05 for all categories). In the five-category meta-analysis, ORs were 1.11 (95% CI 0.90 to 1.33) for birth weight <2,500 g, 1.11 (0.99 to 1.25) for 3,000 to 3,499 g, 1.15 (1.04 to 1.26) for 3,500 to 3,999 g, and 1.24 (1.04 to 1.48) for ≥4,000 g relative to the referent category of 2,500 to 2,999 g. In the three-category meta-analysis, ORs were 1.06 (95% CI 0.98 to 1.14) for 3,000 to 3,999 g and 1.15 (1.01 to 1.31) for ≥4,000 g relative to the referent category of <3,000 g. In the two-category meta-analysis, ORs were 1.09 (95% CI 1.02 to 1.18) for the category of >3,000 g (or ≥3,000 g) relative to the referent category of ≤3,000 g (or <3,000 g).

We identified 17 studies (15 case-control and two cohort) that assessed the association between birth order and breast cancer risk (Table 3). Eight of the studies reported an inverse relationship [14, 20, 31, 35, 36, 42, 46, 47]. Some studies found significantly lower risks for second or later born children versus first-born children [31, 46]. Some studies found significantly or marginally significantly reduced risk among women whose birth had been preceded by the birth of at least five siblings [20, 35]. Other several studies noted an increased risk associated with higher birth order [15, 37, 48, 49], whereas some studies failed to observe such an association [12, 18, 32, 50]. One study did not supply the estimated risk but describe the P value by the mean difference of birth order [51].

Table 3 Studies assessing the association of birth order and the risk of breast cancer
Full size table
Figure 1
figure 1

Meta-analysis of the association between birth weight (five categories) and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Category I (birth weight <2,500 g) versus reference: Q = 9.66 (8 degrees of freedom), P = 0.29; Begg test, P = 0.75; Egger test, P = 0.66. Category II (2,500 to 2,999 g) is the reference. Category III (3,000 to 3,499 g) versus reference: Q = 6.53 (8 degrees of freedom), P = 0.59; Begg test, P = 0.25; Egger test, P = 0.46. Category IV (3,500 to 3,999 g) versus reference: Q = 4.17 (8 degrees of freedom), P = 0.84; Begg test, P = 0.60; Egger test, P = 0.93. Category V (≥4,000 g) versus reference: Q = 11.18 (8 degrees of freedom), P = 0.19; Begg test, P = 0.25; Egger test, P = 0.30. 1We used adjusted odds ratios (ORs) for meta-analysis because the numbers of cases and controls were not represented in the original article. CI, confidence interval.

Full size image
Figure 2
figure 2

Meta-analysis of the association between birth weight (three categories) and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Category I (birth weight 3,000 to 3,999 g) versus reference (<3,000 g): Q = 4.97 (11 degrees of freedom), P = 0.93; Begg test, P = 0.54; Egger test, P = 0.27. Category II (≥4,000 g) versus reference: Q = 13.44 (11 degrees of freedom), P = 0.27; Begg test, P = 0.54; Egger test, P = 0.53. CI, confidence interval; OR, odds ratio.

Full size image
Figure 3
figure 3

Meta-analysis of the association between birth weight (two categories) and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as folows. Reference (<3,000 g [or ≤3,000 g]) versus ≥3,000 g (or >3,000 g): Q = 11.57 (15 degrees of freedom), P = 0.93; Begg test, P = 0.15; Egger test, P = 0.50. CI, confidence interval; OR, odds ratio.

Full size image

For the meta-analysis, we included 14 studies (13 case-control studies and one cohort) that used two birth order categories: 1 (referent) and ≥2. There was significant heterogeneity across all studies (P_Q test < 0.01), although there was no significant heterogeneity across the case-control studies (P_Q test = 0.90). As shown in Figure 4, there was no difference in risk according to birth order across all studies (OR 0.97 [95% CI 0.91 to 1.04]) or within the case control studies (OR 0.99 [95% CI 0.94 to 1.04]). We calculated the crude odds ratio from the cohort study [31], and the result was very different from the summary OR (calculated crude OR 0.28 [95% CI 0.21 to 0.36]). The results of all case-control studies were near null, whereas the cohort study found a significant risk reduction in birth orders of 2 or greater. We also examined the seven studies that classified individuals according to three birth order levels (1 [referent], 2 to 4, ≥5; Figure. 4). There was significant heterogeneity across studies (all of which were case-control studies) for the highest birth order category (P_Q test = 0.03) Women with a birth order of ≥5 were at nonsignificantly reduced risk compared with first-born women (OR 0.88 [95% CI 0.75–1.01]). There was no difference in risk for women of birth orders 2 to 4 (OR 0.97 [95% CI 0.91–1.03]).

Figure 4
figure 4

Meta-analysis of the association between birth order and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Category I (birth order 2+) versus reference (birth order 1): Q = 87.79 (13 degrees of freedom), P < 0.01; Begg test, P = 0.44; Egger test, P = 0.46. Category II (birth order 5+ and 2 to 4) versus reference: Q = 4.56 (6 degrees of freedom), P = 0.60; Begg test, P = 0.37; Egger test, P = 0.44. Category II (birth order ≥6, 2 to 5) versus reference: Q = 14.42 (6 degrees of freedom), P = 0.60; Begg test, P = 0.37; Egger test, P = 0.44. 1Category I of birth order was 2+ vs 1. 2Category II of birth order was composed of two conditions: 5+ and 2 to 4; and ≥6 and 2 to 5 vs 1. 3We used adjusted odds ratios (ORs) for meta-analysis because the numbers of cases and controls were not represented in the original article. CI, confidence interval.

Full size image

We identified 28 studies (22 case-control and six cohort) that assessed the association between maternal age and breast cancer risk (Table 4). Seven studies observed modestly increased risks for daughters born to older mothers [15, 31, 32, 36, 42, 46, 52]. A pattern of slight decrease after modest increase in risk was found in five other studies [50, 53–56]. Fourteen studies, however, no association was observed [12, 14, 18, 20, 35, 37, 38, 47, 49, 57–60]. Two studies did not estimate the risks [51, 61].

Table 4 Studies assessing the association of maternal age with risk for breast cancer
Full size table

In our meta-analyses, we included the 18 studies that reported categorical data and examined three age categories (≤24 [referent], 25 to 29, and ≥30 years; Figure 5). There was, however, significant study heterogeneity (P_Q test < 0.01 for 25 to 29 years and for ≥30 years). Heterogeneity was also present across case-control studies and studies published after 2000 (P_Q test < 0.01). The ORs (95% CI) were 1.18 (1.05 to 1.11) for 25 to 29 years and 1.23 (1.07 to 1.15) for ≥30 years across all studies.

Figure 5
figure 5

Meta-analysis for the association between maternal age and the risk of breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Maternal age 25 to 29 years: Q = 39.40 (17 degrees of freedom), P < 0.01; Begg test, P = 0.85; Egger test, P = 0.38, Maternal age 30+ years: Q = 67.34 (17 degrees of freedom), P < 0.01; Begg test, P = 0.88; Egger test, P = 0.07. 1The reference for maternal age is ≤24 years old. CI, confidence interval; OR, odds ratio.

Full size image

We identified 15 studies (10 case-control and five cohort) that assessed the association between prematurity and breast cancer risk (Table 5). Most studies did not observe a significant relationship [13–16, 20, 25, 29, 31, 33, 40, 42, 45]. Two studies found that extreme prematurity was associated with an increased risk (OR 3.96 [95% CI 1.46 to 10.81] for ≤32 weeks relative to ≥33 weeks [32], and SIR (standardized incidence ratio) 6.7 [95% CI 1.4 to 19.5] for <31 weeks [62]). In contrast, another study [34] found that longer gestation was associated with a significantly increased risk (OR 8.4 [95% CI 1.3 to 54.4] for ≥40 weeks relative to ≤32 weeks).

Table 5 Table 5 Studies assessing the association of premature birth and the risk of breast cancer
Full size table

There was no significant heterogeneity across studies (P-Q test = 0.55), whereas we found no association between prematurity (≤36 weeks) and risk (OR 1.04 [95% CI 0.92 to 1.18]; Figure 6). However, a strong publication bias was observed (P-Egger test = 0.03 and P-Begg test = 0.11; Figure 7). A significant publication bias occurred because three studies with smaller standard errors of log RR [15, 16, 34] reported RRs near 1.0, whereas five studies with larger standard errors [13, 20, 31, 33, 45] reported substantially reduced RRs. When the analysis was performed for extreme prematurity (≤32 weeks), heterogeneity was also evident across the studies (P-Q test = 0.04), and the association was not significant (OR 1.20 [95% CI 0.74 to 1.95]).

Figure 6
figure 6

Meta-analysis of studies assessing the association of prematurity and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Category 36+: Q = 5.91 (7 degrees of freedom), P = 0.55; Begg test, P = 0.11; Egger test, P = 0.03. Category 32+: Q = 10.10 (4 degrees of freedom), P = 0.04; Begg test, P = 0.09; Egger test, P = 0.40. 1Category of prematurity (week): ≤36 versus ≥37 (reference). 2Category of prematurity (week): ≤32 versus ≥33 (reference). CI, confidence interval; OR, odds ratio.

Full size image
Figure 7
figure 7

Begg's funnel plot for publication bias in meta-analysis of premature birth and breast cancer risk. Premature birth (gestational age ≤36 weeks) was compared with gestational age ≥37 weeks. Egger test, P = 0.03; Begg test, P = 0.11. rr, relative risk; s.e., standard error.

Full size image

We examined 13 studies (eight case-control and five cohort) that assessed the association between twin status and risk (Table 6). Most studies identified a slightly increased risk among twins [15, 31, 32, 43, 45, 58, 63–66], with five studies demonstrating significant associations [31, 58, 64–66]. In contrast, some studies observed a slightly reduced risk [14, 20, 67], with one of the risks being marginally significant [67]. Seven studies [20, 32, 58, 63, 65, 66] had information on zygosity. Of these studies, two [58, 63] used the twins' sex as a proxy for zygosity. For monozygotic twins, a reduction in risk was significant in one study [68]. Most studies failed to observe an association [20, 32, 58, 63, 65, 66]. Three studies reported a significantly increased risk associated with being a dizygotic twin [58, 65, 66], whereas other studies reported no association [20, 32, 63, 67] (Figure 8).

Table 6 Studies assessing the association of twinship with risk for breast cancer
Full size table
Figure 8
figure 8

Meta-analysis for the association between twinship and risk for breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows. Twinship: Q = 18.79 (13 degrees of freedom), P = 0.13; Begg test, P = 0.78; Egger test, P = 0.24. Monozygote twin: Q = 5.79 (6 degrees of freedom), P = 0.45; Begg test, P = 0.55; Egger test, P = 0.85. Dizygote twin: Q = 12.53 (6 degrees of freedom), P = 0.06; Begg test, P = 1.0; Egger test, P = 0.3. 1The authors used the female twins as the proxy for the monozygote twin and the female twin with male twin as the proxy for the dizygote twin. 2Women aged 21 to 45 years. 3Women aged 50 to 64 years. CI, confidence interval; OR, odds ratio.

Full size image

The Q test for heterogeneity was not significant (P-Q test = 0.13), and the meta-analysis of 13 studies examining twin status (without regard to zygosity) found an OR of 1.22 (95% CI 1.01 to 1.11). There was no evidence of any publication bias (P-Egger test or P-Begg test >0.1). There were little evidence of heterogeneity (P-Q test > 0.1 for monozygotic or dizygotic twins), and breast cancer risk was not significantly increased among either monozygotic (OR 0.95 [95% CI 0.85 to 1.07]) or dizygotic (OR 1.17 [95% CI 0.99 to 1.37]) twins, albeit based on limited statistical power. In subgroup analysis by study design, cohort studies identified significantly increased risk (OR 1.23 [95% CI 1.00 to 1.11]) for breast cancer in twins versus singletons, with no study heterogeneity (P-Q test = 0.07). Case-control studies showed no association with twin status (OR 1.39 [95% CI 0.91 to 1.12]). There was no evidence of any publication bias (P-Egger test or P-Begg test > 0.05) among the case-control or cohort studies. In subgroup analysis by study design and zygosity, there were no heterogeneity in studies (P-Q test > 0.1). In subgroup analysis by study year, significant heterogeneity by publication year was identified (P = 0.01), and the OR (95% CI) for studies published before 2000 was 1.06 (0.97 to 1.47), whereas the OR (95% CI) for studies published in 2000 or later was 1.27 (1.03 to 1.58).

We identified nine studies that assessed the association between maternal or paternal smoking and risk (Table 7). Two cohort studies reported nonsignificantly reduced risks associated with maternal smoking (OR 0.49 [95% CI 0.29 to 1.03] [68]; OR 0.8 [95% CI 0.5 to 1.1] [31]), whereas a case-control study [43] identified a significant positive association (age-adjusted OR 2.7 [95% CI 1.1 to 6.3]), although its crude OR was not statistically significant (OR 1.1 [95% CI 0.7 to 1.7]). The majority of studies, however, identified no associations with maternal [14, 20, 33, 35, 58, 69] or paternal [20, 35, 69] smoking during pregnancy.

There was no heterogeneity or publication bias (P-Q test > 0.05, P-Egger test and P-Begg test > 0.1 among all studies, case-control or cohort). The meta-analysis for maternal smoking (Figure 9) found no significant association with risk (OR 0.98 [95% CI 0.86 to 1.13]), although cohort studies [40, 68] noted a significant negative association with maternal smoking (OR 0.59 [95% CI 0.41 to 0.85]).

Table 7 Studies assessing the association of maternal or paternal smoking and the risk of breast cancer
Full size table
Figure 9
figure 9

Meta-analysis for the association of maternal smoking during pregnancy with risk for subsequent breast cancer. The tests for homogeneity and for publication bias in the studies analyzed are as follows: Q = 16.90 (9 degrees of freedom), P = 0.06; Begg test, P = 0.59; Egger test, P = 0.31. 1Titus-Ernstoff and coworkers [35] classified three categories: nonparental smoking, paternal or maternal smoking only or both parents smoking during pregnancy. The odds ratios (ORs) of father smoking on breast cancer risk was almost unity (OR 1.0, 95% confidence intrval [CI] 0.9 to 1.1). Thus, in this study, the mother smoking and both parents smoking versus nonparental smoking can be considered to the maternal smoking versus no maternal smoking. 2Women aged 21 to 45 years. 3Women aged 50 to 64 years.

Full size image

Discussion

The main finding of our meta-analysis was that heavier birth weight was associated with increased breast cancer risk (18% increased risk for the heaviest weight). Twin status was associated with 1.2-fold higher risk for breast cancer relative to a singleton birth. Although we found some evidence of increased risk associated with older maternal age (OR 1.16 for maternal age ≥30 years), there were heterogeneous findings across study designs.

Most studies identified an increased risk for breast cancer with heavier birth weight, with the association being particularly strong for premenopausal or early-onset breast cancers [14, 22, 29, 30]. Our result was similar to the findings of a recent meta-analysis of 26 studies, which revealed that high birth weight was associated with a RR of 1.23 and restricted to premenopausal women (OR 1.25 [95% CI 1.14 to 1.38) [70]. This analysis grouped birth weight into two categories (classified into high and low birth weight in each study, regardless of specific weight in terms of grams), preventing evaluation of dose-response relationships. We did in fact observe evidence of a dose-response relationship of risk with birth weight, although this was based on a relatively small number of studies involving three or four categories.

Although some studies identified a J-shaped relationship between birth weight and breast cancer risk [12, 14, 15, 18, 25, 33, 35, 37, 45], others failed to note an increased risk associated with very low birth weights. A recent study involving 3,066 breast cancer patients and 106,504 comparison individuals in a Danish cohort also found no elevated risk among those with very low birth weights [71]. Similarly, our meta-analysis provided little evidence of increased risk for very low birth weights.

Although the mechanisms underlying the association between high birth weight and breast cancer risk remain unclear, it has been suggested that heavier birth weights may result from increased in utero exposuresto factors such as insulin-like growth factor-I or estrogens [72–76]. These substances may act as mitogens by increasing the likelihood of genetic mutations [75, 77]. However, several studies have failed to find any correlation between umbilical cord estrogen levels and birth weight [78, 79]. One study, however, reported a significant positive relationship with estriol [80]. Further studies should be undertaken to assist in the resolution of these conflicting data.

Our analysis found no association of breast cancer risk with birth orders between 2 and 4, but we did note a somewhat reduced risk associated with higher birth orders (at least 5), although the results were heterogeneous across studies. Biologically, pregnancy estrogen levels appear to be higher during first pregnancies and decline in successive pregnancies [81]. Furthermore, cord blood levels of estradiol, estrone, and progesterone are lower for later born than first born children [82]. These findings suggest that the reduced risk associated with higher birth orders may relate to lower estrogen levels. However, evidence supporting birth order as a risk factor for breast cancer is limited, with further investigations needed to evaluate dose-response relationships more fully.

In our meta-analysis, we found some evidence that having been born to an older mother was associated with higher breast cancer risk, although the results were heterogeneous across studies. Our data failed to support the previous studies that suggested a J-shaped relationship between maternal age and breast cancer risk. It was previously suggested that older maternal age may have an adverse effect on the primordial mammary gland of their daughters because of altered hormonal profiles [37] or may linked to the epigenetic change of mtDNA which can lead to breast carcinogenesis by oocyte inheritance [83]. However, the two studies that examined pregnancy estrogen levels according to maternal age found that both total estrogen and estradiol levels were lowest in youngest mothers (<20 years of age), highest in those aged 20 to 24 years, and intermediate in mothers over 25 years of age [78, 81]. Thus, it remains unclear from both our meta-analysis as well as from biologic data whether maternal age is a proxy for estrogen or estradiol exposure to fetus. Although it has been suggested that older paternal age may cause germ cell mutations, previous epidemiologic studies have failed to support an association [35, 20, 69, 82, 84, 85]. Because the purpose of this study was to evaluate whether the intrauterine hormone environment affects subsequent breast cancer risk, our meta-analysis did not include paternal age.

We observed no association between prematurity and breast cancer risk. Biologically, women having abruptio placentae or an extremely premature birth (<32 week) have been shown to have elevated levels of human chorionic gonadotropin and α-fetoprotein, which could inhibit the differentiation of stem cells in human breast tissue cells [15]. Gestational age is related to birth weight, of course, because birth weights in infants born prematurely are lower than those in infants born at term [13].

Twin pregnancies are associated with an approximate doubling of estrogen levels compared with singleton pregnancies [86, 87]. Dizygotic twin pregnancies have elevated levels of estrogens and gonadotropins [88–90]. It has therefore been postulated that twins, especially dizygotic twins, could be at an elevated risk for breast cancer. In general, our results did not support differences in risk between monozygotic and dizygotic twins, and there was evidence that risk estimates published after 2000 were qualitatively different from those of earlier studies.

Studies of parental smoking, especially maternal smoking, and daughter's breast cancer risk have yielded inconsistent results. Biologically, maternal smoking, rather than paternal smoking, has a greater impact on the fetus. In the meta-analytic results, both factors failed to exhibit a significant association with risk. Some studies have reported that maternal smoking in pregnancy reduces serum estrogen levels [91, 92]. A recent experimental study reported that both estradiol-17β levels and progesterone:estradiol-17β ratios were reduced in pregnant mice exposed to cigarette smoke [93]. However, the relevance of these findings to humans is unclear.

These meta-analyses are based on results from studies involving heterogeneous designs and methodology. We did note between-study heterogeneity for the associations of birth order, maternal age, and twin status. To resolve the heterogeneous findings, we considered the influence of study design and the date of study publication on the results by subgroup analyses. However, heterogeneity in studies could only be explained partially.

Effects of maternal age, birth order, prematurity (cut-off value 32 weeks), and maternal smoking were found to be heterogeneous across study designs, but birth weight and twinning were comparable. Self-reported measures of perinatal factors may be vulnerable to misclassification biases, with differential or nondifferential effects [94, 95]. Because studies based on data linkage to medical records have a lower chance of misclassification bias, we conducted subgroup meta-analyses stratified by source of information (data linkage versus self-report) and found no substantial differences in the results. Although the completeness of records is a critical factor in evaluating biases in studies based on data linkage, most papers did not provide details about the completeness of records. We also conducted subgroup meta-analyses stratified by publication year. Only twin status exhibited significant heterogeneity according to publication year (<2000 versus ≥2000).

Our findings may be somewhat inflated because of our dependence on crude rather than adjusted ORs or RRs. A possible misclassification bias for zygosity might have resulted in studies that used sex as a proxy for zygosity [96]. Because this bias would probably attenuate associations, additional investigations are needed to determine the extent of any true association of risk with twin status.

Conclusion

It has been hypothesized that certain perinatal factors, including birth weight and order, twin pregnancies, prematurity, maternal age, and smoking, may reflect higher estrogenic environments in utero, thereby increasing the subsequent risk of breast cancer. Findings of an increase in breast cancer risk among daughters exposed to diethylstilbestrol in utero supports this hypothesis [97, 98]. Although the current meta-analysis found evidence that higher birth weights are associated with increased breast cancer risk, older maternal age and twin status were less convincingly related, and birth order and prematurity appeared unrelated. Greater birth weights have been attributed to higher maternal estrogens levels, which could affect fetal development [72–74] through epigenetic modifications of breast stem cells [1, 99, 100]. Although our findings regarding birth weight support the hypothesis that higher estrogen exposures in utero may be involved in the subsequent development of breast cancer, further biologic data are needed to elucidate the relationship fully.