Introduction

Breast cancer is an important public health problem in developed countries as it is one of the most common cancers, being the most if only the female population is considered1. The incidence is decreasing every year, which is partly due to early detection programs2.

In the last decades, cellular in vitro experiments and in vivo models have evaluated the role of vitamin D in the development of breast cancer, finding a protective anticancer role of 1,25(OH)D33. It has been demonstrated that treating breast cancer cells with 1,25(OH)D3 induces two beneficial effects: an anti-proliferative effect4 and a pro-apoptotic effect5,6. The former is linked to the suppression of growth stimulatory signals and the potentiation of growth inhibitory signals, whilst the second one is explained by the bcl-2 family proteins. The interaction between vitamin D and its receptors induces an increase in the expression of pro-apoptotic family member (bax and bak protein) and simultaneously a decrease of anti-apoptotic (bcl-2/bcl-XL)6. In addition, the breast tissue contains the 1-α-hydroxylase, allowing for the generation of the active vitamin D metabolite (1,25 dihydroxyvitamin D) from the circulating precursor (25 hydroxyvitamin D). As vitamin D receptors are found in the breast6, an autocrine role of vitamin D has been suggested7.

Despite this biological background, literature shows inconsistent results8,9,10,11,12,13,14,15,16 (Table 1). Several additional observational studies have appeared since the last meta-analysis publication (including articles until 2013). The main purpose of the present meta-analysis is to update the relationship between vitamin D exposure and breast cancer risk by adding the studies published more recently. Thus sixty-eight observational studies: thirty of these were case-control, twenty-one were nested case-control and the remaining were cohort studies.

Table 1 RR of breast cancer and vitamin D in previous meta-analysis.

Methods

Search strategy

Firstly, the following inclusion criteria were defined: we looked for cohort or case-control studies performed in humans, which reported, at least, one relative risk (RR) or odds ratio (OR) with confidence interval at 95%. (95% CI)

We began our search in Pub-Med, Scopus and Web of Science database using “breast cancer” and “vitamin D” as keywords, finding 2313 articles. After having read the title and abstract, 2123 articles that did not meet the above criteria were eliminated. Next, we carried out a more exhaustive and complete reading, which allowed us to reject another additional 69 articles (Fig. 1). Finally, sixty eight studies meeting our inclusion criteria were identified: fifty one case-control10,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65 and seventeen cohort studies65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81. Tables 2 and 3 summarize the main characteristics of the included articles.

Figure 1
figure 1

Flowchart which describes the methodology of selection of the articles.

Table 2 Studies included in our meta-analyses of blood 25-hydroxyvitamin D and breast cancer risk.
Table 3 Studies included in our meta-analyses of dietary or supplements vitamin D and breast cancer risk.

Data extraction

The following step was to create a database to gather all relevant information extracted from each article: year of publication, author, journal, follow up, country, sample size, exposure levels, units of measure, data for the creation of the contingency table and RR/OR with 95% CI; as well as a section to assess the quality of the study using the STROBE scale82.

Statistical analysis

Statistical analysis was performed separately for cohort and case-control studies. In the case control studies a sensitivity analysis was also carried-out including only nested case-control studies. We performed separate analyses for any type of vitamin D exposure reported in at least three studies: 25(OH)D, dietary intake of vitamin D, 1,25(OH)2D and vitamin D supplements.

The ways that doses or levels of vitamin D were reported in each individual article were not standardized across studies (for instance, some papers reported vitamin D levels in quartiles; others in tertiles, and so on), making it difficult to extract them in an analyzable form. Therefore, in order to provide a consistent criterion of comparability, we selected the OR or RR reported for the highest category compared to the lowest one.

Regarding the type of breast cancer, we analyzed all invasive breast cancers together, and breast cancer stratified according to the cancer estrogen receptor status and woman’s menopausal status. Pooled OR or RR were estimated by weighting individual OR/RR by the inverse of their variance. OR or RR heterogeneity was measured using Q and I2 statistics83. A fixed-effect model was preferred if the Q statistic was higher than 0.1 or I2 lower than 25%, indicating no relevant heterogeneity; a random-effect model was otherwise chosen84. The presence of small-study bias was explored with Rosenthal model and with Egger test85; due to the low sensitivity of Egger test, the cut-off was set at p = 0.1. Funnel plots86 were applied to detect publication bias.

An analysis of influence was performed via the re-estimation of pooled OR/RR by removing one study at a time. Studies that, when removed, strongly changed the OR/RR would be considered as highly influential. Results are displayed as forest plots showing OR/RR and their 95% confidence intervals for each individual study and for the pooled result. Cumulative meta-analyses were carried out to deem the stability of the OR/RR estimates. In order to do that, all studies considered were arranged from oldest to neweest. Then an OR/RR estimate was obtained for the two eldest studies; another for the three eldest, and so on, adding a study each time. Results are reported as forest plots.

All the statistical analyses were carried out with the package Stata 14/SE (Stata Corporation, College Station, TX, US).

Results

Relationship between 25(OH) D and breast cancer

Twenty-nine case control studies were analyzed to study the relationship between 25 (OH) D and breast cancer10,19,20,21,22,25,27,29,30,31,32,33,34,35,38,42,44,45,46,48,49,51,55,56,58,59,60,61,62,63 obtaining a pooled OR of 0.65 (95%CI: 0.56–0.76) (Fig. 2a, Table 4). This value was calculated using the random effects model because of the high heterogeneity (I2 = 77.76%) of the fixed-effect. Although Egger test cannot rule out a small-study effect (p = 0.001), no study shows a relevant influence. The funnel plot shows asymmetry (Supplementary Fig. 1a), indicating either publication bias or heterogeneity that cannot be explained by a random-effect meta-analysis. Rosenthal model shows that 1194 negative studies would be needed to lose statistical significance. In order to further clarify the heterogeneous result, we carried out a sensitivity analysis including only nested case-control studies21,22,25,31,32,33,34,42,45,46,51,55,56,59 reaching a pooled OR = 0.92 (95%CI: 0.83–1.01) (Fig. 2b) with I2 = 15.87%, Q-based p value = 0.22 and a very symmetrical-looking funnel plot (Supplementary Fig. 1b).

Figure 2
figure 2

(a) Forest plot for the relationship between 25(OH)D and breast cancer in case control studies. (b) Forest plot for the relationship between 25(OH)D and breast cancer in nested case control studies.

Table 4 Results from the meta-analysis.

Four cohort studies75,78,79,80 provided results on 25(OH)D and breast cancer relationship, from which we obtained a pooled RR of 0.85 (95% CI:0.74–0.98).

We also analyzed the relationship between 25(OH) D and breast cancer, stratifying results by hormonal receptors (ER+/ER−) and menopausal status (postmenopausal or premenopausal). Regarding hormonal receptors (Table 4), we have found only one cohort study80 and five case-control studies19,32,33,42,45. In both cases (ER+ and ER− tumors) statistical significance was not reached. With respect to menopausal status (Table 4), we obtained a protective effect in both groups: nineteen case-control studies targeted postmenopausal women18,21,28,30,34,35,36,38,41,47,49,51,55,60,81 with a pooled OR of 0.74 (95%CI: 0.59–0.93), and nine focused on premenopausal21,30,34,35,38,49,51,55,60 obtaining a pooled OR of 0.63 (95%CI: 0.49–0.80) (Fig. 3a). When the sensitivity analysis was carried out including only nested case-control studies, the protective vitamin D – breast cancer association persisted only in the premenopausal group (Fig. 3b, Supplementary Table 1). On the other hand three cohorts studies analyzed separately postmenopausal women79,81 without reaching statistical significance (OR = 1.15 (0.59–2.23)).

Figure 3
figure 3

(a) Forest plot for the relationship between 25(OH)D and premenopausal breast cancer in case control studies. (b) Forest plot for the relationship between 25(OH)D and premenopausal breast cancer in nested case control studies.

Relationship between 1,25(OH)2D and breast cancer

Three case-control studies25,37,39 examined the relationship between circulating 1,25(OH)2D and breast cancer; significant association was not found either in the whole analysis (pooled OR = 0.61 (0.33–1.16)) or in postmenopausal women (combined OR = 1.28 IC 95%: 0.98–1.67)36,37.

Relationship between dietary vitamin D and breast cancer

We found eight case-control studies24,38,40,50,52,53,57,64 on the relationship between dietary vitamin D and breast cancer with a pooled OR of 0.91 (95%CI: 0.72–1.17) (Table 4, Supplementary Fig. 2a). In addition, by combining five cohort studies66,68,70,71,72 we obtained a RR of 1.00 (95% CI 0.93–1.07) (Table 4, Supplementary Fig. 2b).

When stratifying by menopausal status, four case-control38,40,53,64 and five cohort studies66,73,74,76,77 assessed the risk of breast cancer in postmenopausal women. The pooled OR for case-control studies was 0.78 (95%CI: 0.68–0.90) and the pooled RR for cohort studies was 0.95 (95%CI: 0.83–1.09) (Table 4). In both analyses, Egger test rejected the possibility of small study bias (p = 0.536 in case-control studies and p = 0.68 in cohort studies). On the other hand, five case-control studies17,38,40,53,63 and three cohort studies66,73,77 targeted premenopausal women; the pooled OR was 0.65 (95%CI: 0.52–0.82) for case-control studies and the RR for cohort studies was 1.01 (95% CI: 0.86–1.18) (Table 4).

Relationship between supplements of vitamin D and breast cancer

We identified five case-control studies23,24,43,52,65 and two cohort studies67,71 that had evaluated the association between supplements of vitamin D and breast cancer risk. The pooled OR and RR were 0.78 (95% CI: 0.63–0.98) and 1.06(95% IC: 0.90–1.25) respectively (Table 4). Regarding menopausal status, Kim et al.41 published a study on five different populations of postmenopausal women; when combining all five results, we found no significant association (OR: 0.82 95%CI: 0.49–1.35).In addition, we found two case-control studies26,54 focused on premenopausal women obtaining a weak protection (pooled OR 0.89 95%CI (0.84–0.95)).

Relationship between total vitamin D intake (dietary and supplements) and breast cancer

Finally, we found two cohort studies69,71 and four case control studies23,24,38,64 on vitamin D intake (dietary plus supplements) and breast cancer risk, providing no separate results on dietary/supplemented vitamin D origin. We obtained a combined RR = 0.93 (95% CI: 0.86–1.02) for cohort studies, and a combined OR = 0.84 (95% CI: 0.68–1.05) for case-control studies. Five cohort studies69,73,74,76,77 provided results on postmenopausal women (RR = 0.94 95% CI: 0.87–1.00) and three cohort studies69,73,77 on about premenopausal women (RR = 0.90 95% CI: 0.72–1.12) (Table 4). Only two case-control studies provided results according menopausal status38,64 without being significant in both groups.

Discussion

According to our results, 25(OH)D levels were associated with smaller risk of breast cancer in both case-control and cohort studies; these results were consistent on premenopausal women for case-control studies but could not be analyzed for cohort studies. Results for the relationships between breast cancer and dietary vitamin D or between breast cancer and vitamin D supplements, however, showed a protective association only in case-control studies.

In relation to the influence of vitamin D on breast cancer development prospective (cohort and nested case-control) and case control studies tend to show discrepant results: case-control studies usually show a protective effect while prospective studies rarely find it87. This discrepancy might be the result of several factors: Firstly, it is well known that prospective studies are less prone to be affected by both information and reverse-causation bias. Secondly, several authors highlight the season when the vitamin D measurement was made as a potential limitation of case-control studies. Eliassen et al.33 in a nested case-control study found an inverse association between serum 25(OH) D levels and breast cancer limited only to summer measures. It can be assumed that people with low vitamin D levels in summer would also have low levels year-round; therefore, vitamin D levels in summer would be more adequate for analyzing vitamin D – breast cancer relationship than vitamin D levels in any other moment of the year.

When stratifying by menopausal status, our meta-analysis shows a consistent protective effect of 25(OH) D in both case-control and nested case-control studies, but only in premenopausal women. There are different explanations for the influence of menopausal status in the relationship between vitamin D and breast cancer. One of them may be related to the joint relationship between vitamin D and insulin-like growth factors (IGFs). IGF-I is a mitogenic and antiapoptotic peptide that can stimulate the proliferation of breast epithelial cells, increasing the risk of neoplastic transformation88,89. The active vitamin D metabolite is able to block the mitogenic effects of IGF-I, leading to a decrease in proliferation and an increase in apoptosis90. As there is a physiological decline of the IGF with aging91, the interaction between IGF pathways and vitamin D is likely to be stronger for premenopausal than for postmenopausal women, leading to greater risk reduction in premenopausal breast cancer73,92. Finally, high levels of vitamin D may reduce progesterone and estradiol, providing a potential mechanism for reducing breast cancer risk in young women93.

Previous meta-analyses of prospective studies showed contradictory results. Kim et al.13 (who included 24 studies, 14 of those having measured serum 25(OH)D) found a slightly stronger inverse association among premenopausal than among postmenopausal women but without significant differences, whereas in the meta-analysis of Bauer et al.8 (nine studies included) the inverse association was only observed in postmenopausal women. In our meta-analysis, new prospective studies31,33,41,56,58,59,67,78,79,80,81,94 not included in previous reviews, were added and this fact may explain the differences in the results.

Concerning hormonal receptors (ER+/ER−), the relationship with breast cancer remains controversial. On the one hand, a decreased risk in ER+ would be expected, since it seems that sensitivity to 1,25(OH)2D is generally reported as being higher in breast cancer cells that express the estrogen receptor than in those that do not93,95. It has been demonstrated that treating breast cancer cells ER+ with 1,25(OH)D3 induces a cell cycle shutdown in GO/G13,96. On the other hand, two-thirds of triple negative tumors express VDR97 and it has been demonstrated that VDR expression is inversely associated with more aggressive breast cancer98. In consonance with previous epidemiological studies32,33,42,45, our study does not reach significant differences when the analysis was performed separately in ER+ or ER− subgroups. However, other studies found a decreased risk of ER− breast cancer regarding the serum levels of 25 (OH) D18,60.

No relationship is found between the level of circulating 1,25(OH)2D and breast cancer. This result is consistent with previous studies9, while Janowsky et al.39 found an inverse association. Several authors consider that 1,25(OH)2D is not a good indicator of vitamin D status: First, 1,25(OH)2D’s half-life is only 4–6 h, whereas 25(OH)D’s half-life is 3 weeks; second, 1,25(OH)2D is influenced by many factors10, for instance, it can be elevated in patients with vitamin D deficiency as a result of hyperparathyroidism12,99; finally, as 1,25(OH)2D is metabolized by 1-α -hydroxylase in breast tissue, plasma levels may not adequately represent breast tissue levels12,100.

We do not find a relationship between vitamin D intake and breast cancer in the overall analysis. In contrast, when stratifying by menopausal status, a protective effect is observed in case-control studies in both premenopausal and postmenopausal women, whereas this association is not present in cohort studies. On the other hand, when analyzing the influence of vitamin D supplements on breast cancer risk, we find a borderline protective effect.

In the relationship between vitamin D intake (dietary and/or supplements) and breast cancer, most observational studies showed non-significant differences; only two articles17,53 found a protective association. In a previous meta-analysis13, this association was not significant for either vitamin D intake or supplements.

A probable explanation for the lack of association observed in the analysis of dietary intake or supplements compared to the 25(OH)D levels may be that the main source of vitamin D is sunlight rather than food or supplements.

In addition, the French E3N Cohort Study12 reported that high vitamin D intake is associated with lower breast cancer risk in regions with high ultraviolet solar radiance. These results suggested that the total amount of vitamin D needed to reach a protective effect on breast cancer is too high to be achieved in regions with low ultraviolet radiance. Under these circumstances, as the vitamin D intake has to be higher than the usually recommended, it could eventually lead to side effects such as hypercalcemia, constipation or muscle weakness.

Our study has some limitations; firstly each article uses different cutoff points according to serum levels of vitamin D. To analyze it we restricted our analysis to the comparison between the highest vs. lowest category of exposure. This analysis strategy does not allow for a dose-response analysis. Moreover, we carried out a sensitivity analysis excluding one study at a time, showing that no single study substantially affected the pooled RR/OR. Secondly, there is huge variability in the literature on the type of vitamin D studied, which makes it difficult to perform the analysis. In addition, levels of vitamin D depend on the season, so it would be advisable to take all samples at the same time, or at least refer to when they were collected75. Thirdly, case-control studies are more prone to methodological issues, such as recall and selection biases, which limits the strength and quality of evidence. However, about half of the case-control studies included in our meta-analysis are nested in cohort studies, which minimizes the possibility of introducing biases. Finally, breast cancer is a heterogeneous disease and it is possible that vitamin D only affects certain breast cancer subtypes. However, this aspect has been scarcely studied in primary articles, so we have not been able to analyze it in the present meta-analysis.

Despite these limitations, our study also has several strengths; first, we have gathered all the observational studies published in the last twenty years. In addition, we have focused the analysis on different types of vitamin D exposure (diet, supplements and blood-levels of 25(OH) D and 1,25(OH)2D) whereas other meta-analyses are only focused on 25(OH)D levels9,10,16,99 or vitamin D intake12. This strategy allows us to obtain a more detailed analysis of the relationship between vitamin D and breast cancer.

In conclusion, our meta-analysis supports the hypothesis that high serum levels of 25(OH) vitamin D has a protective effect on breast cancer risk in premenopausal women; we cannot draw the same conclusion regarding vitamin D intake or supplements of vitamin D since the number of studies are still limited and publication biases cannot be excluded.