Plasma 25-hydroxyvitamin D and premenopausal breast cancer risk in a German case-control study
Article first published online: 6 OCT 2008
Copyright © 2008 Wiley-Liss, Inc.
International Journal of Cancer
Volume 124, Issue 1, pages 250–255, 1 January 2009
How to Cite
Abbas, S., Chang-Claude, J. and Linseisen, J. (2009), Plasma 25-hydroxyvitamin D and premenopausal breast cancer risk in a German case-control study. Int. J. Cancer, 124: 250–255. doi: 10.1002/ijc.23904
- Issue published online: 27 OCT 2008
- Article first published online: 6 OCT 2008
- Manuscript Accepted: 29 JUL 2008
- Manuscript Received: 12 JUN 2008
- Deutsche Krebshilfe e. V. Grant Number: 70492
- Deutsche Forschungsgemeinschaft, Graduiertenkolleg 793
- Environmental Cancer Risk, Nutrition and Individual Susceptibility (ECNIS). Grant Number: 513943
- breast cancer;
- case-control study;
- vitamin D;
- 25-hydroxyvitamin D
Laboratory and epidemiological data have linked vitamin D to breast cancer prevention. Beside dietary intake, endogenous production of vitamin D substantially contributes to a subject's vitamin D status. Most studies, however, have assessed dietary intake only. Although differential effects of vitamin D on premenopausal and postmenopausal breast cancer have been discussed, this is the first study to investigate the association of plasma 25-hydroxyvitamin D [25(OH)D], as indicator of the overall vitamin D status, with breast cancer risk with restriction to premenopausal women only. We used data of a population-based case-control study comprising 289 cases and 595 matched controls. Information on sociodemographic and breast cancer risk factors was collected by questionnaire and plasma 25(OH)D was measured by enzyme immunoassay. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using conditional logistic regression. We observed a significant inverse association between breast cancer risk and plasma 25(OH)D concentrations. Compared with the lowest category (<30 nmol/L), the ORs (95% CI) for the upper categories (30–45, 45–60, ≥60 nmol/L) were 0.68 (0.43–1.07), 0.59 (0.37–0.94) and 0.45 (0.29–0.70), respectively (ptrend = 0.0006). The association was shown to be nonlinear (pnonlinearity = 0.06) in fractional polynomial analysis with a stronger effect in women at low plasma 25(OH)D levels, providing some evidence of a threshold effect (at circa 50 nmol/L). The association was stronger in progesterone receptor negative tumors, with suggestive evidence of effect heterogeneity (pheterogeneity = 0.05, case-only model). Our findings support a protective effect of vitamin D for premenopausal breast cancer. © 2008 Wiley-Liss, Inc.
The anticarcinogenic potential of vitamin D in various cell types, including normal and malignant breast cells, by influencing the induction of cell differentiation, inhibition of cell growth and regulation of apoptosis is well-established.1–6 Most of the epidemiologic studies regarding breast cancer risk have assessed the effects of vitamin D only for dietary intake yielding inconsistent results.7–15 However, beside dietary intake, the main source of vitamin D is cutaneous production via sun exposure. Studies assessing the association of breast cancer risk and measures of sun exposure, as a proxy of endogenous vitamin D synthesis, have observed inverse associations.16, 17 However, an appropriate biomarker to measure vitamin D status in humans, accounting for both vitamin D from diet and endogenous production, is 25-hydroxyvitamin [25(OH)D]. 25(OH)D is converted to 1,25-dihydroxyvitamin D [1,25(OH)2D], the biologically active form of vitamin D, in the liver and other tissues including breast.18–20
Few studies have assessed the relationship between serum or plasma 25(OH)D and breast cancer risk.21–25 Of 2 small hospital-based case-control studies in predominantly postmenopausal women, one study found a significantly inverse association23 whereas the other study revealed no association between plasma 25(OH)D and breast cancer risk.25 To date, 2 studies assessed the association of vitamin D metabolites with breast cancer risk with restriction to postmenopausal women only.21, 22 A study nested in the prostate, lung, colorectal and ovarian cancer screening trial reported no association between both higher plasma 25(OH)D and 1,25(OH)2D concentration and breast cancer risk.22 However, a strong statistically significant inverse association between postmenopausal breast cancer risk and serum 25(OH)D was found in a large case-control study.21 A case-control study nested in the Nurses' Health Study (NHS) is the only study that so far reported results stratified by menopausal status.24 The authors reported a nonsignificant decrease in breast cancer risk with both higher plasma 25(OH)D and 1,25(OH)2D concentration for overall breast cancer risk. Stratification by menopausal status suggested an inverse association for postmenopausal women only, however power was low for their analysis in premenopausal women and risk estimates were not presented. As reviewed recently, prospective studies on dietary intake of vitamin D suggest a more pronounced effect for premenopausal breast cancer risk.26 Furthermore, we recently reported a significant inverse association of dietary vitamin D with premenopausal breast cancer risk in the present study population.13 To our knowledge, no study so far assessed the association of plasma 25(OH)D with restriction to premenopausal breast cancer only. Following our previous report on the association between postmenopausal breast cancer risk and serum 25(OH)D,21 we here describe the results on plasma 25(OH)D and premenopausal breast cancer risk from another population-based case-control study conducted in broadly the same geographical region. We further assessed differential effects by receptor status of the tumor.
Material and methods
Participants in this population-based case-control study were recruited from 2 study regions in southern Germany (Freiburg and Rhein-Neckar-Odenwald) as previously described.27 Cases were eligible if they were German speaking, diagnosed by the age of 50 years with an incident in situ or invasive breast cancer between January 1992 and December 1995. Cases were identified through frequent monitoring of hospital admissions, surgery schedules and pathology reports in 38 hospitals. Two controls per case matched by exact age and study region were selected from a random list of residents provided by the population registries. All study participants gave informed consent. The study was reviewed and approved by the ethics committee of the University of Heidelberg. Overall 706 of 1,020 eligible cases (69.2 percent) and 1,381 of 2,257 eligible controls (61.2 percent) participated, including 494 cases and 957 controls from the Rhein-Neckar-Odenwald study region.
Information on risk factors and sociodemographic data was obtained by a self-administered questionnaire. Information was truncated at the reference date, which was date of diagnosis for cases and date of study questionnaire completion for controls. Women were defined as postmenopausal if they reported a natural menopause 6 months before reference date or a bilateral oophorectomy. The menopausal status of women with previous hysterectomy not accompanied by bilateral oophorectomy was classified as unknown. Information on hormone receptor status was obtained from pathology reports from the respective hospitals.
Among the 1,451 participants from Rhein-Neckar-Odenwald, plasma samples were available for 1,131 women (77.9 percent). With restriction to premenopausal women, 289 cases (including 18 in situ cases) and 595 controls were included in the present analysis.
We measured 25(OH)D in plasma with the 25-hydroxyvitamin D enzyme immunoassay (IDS, Immunodiagnostic Systems Limited, Boldon, UK). Samples were stored in aliquots at −80°C until measurement and analyzed in a single batch in October 2007. The coefficient of variation was 3.1% for intra-assay and 3.3% for interassay determination. Recovery ranged from 96 to 105% as previously reported.21 We measured 88 random samples (10.0%) in duplicate with an average absolute deviation from the mean of 1.7%.
Because of the skewed distribution of plasma 25(OH)D, geometric means adjusted for time of blood collection and body mass index (BMI) are calculated and compared among cases and controls using analysis of variance. We assessed the association of plasma 25(OH)D with risk of premenopausal breast cancer by means of conditional logistic regression with stratification by age (continuous, in years). Odds ratio (OR) estimates and 95% confidence intervals (CIs) were calculated assessing 25(OH)D concentration both as continuous (per 10 nmol/L increment) and as categorical variable divided into 4 categories (<30, 30–45, 45–60, ≥60 nmol/L). Variables were entered as covariates in the logistic regression model if (i) including the covariate considerably changed the OR for the main variables of interest (>5% change in the ORs), (ii) if the covariate is a known or potential breast cancer risk factor or, (iii) if the covariate was likely to be associated with the main variable of interest (biological plausibility). OR estimates were both adjusted for time of blood collection in four categories, January–March, April–June, July–September, October–December only and with additional adjustment for first-degree family history of breast cancer (yes/no), number of births (0, 1–2, ≥3), duration of breast feeding (continuous, in months), age at menarche (<13, 13–14, ≥15), BMI (continuous, in kg/m2) and alcohol consumption (0, 1–18, ≥18 g ethanol/day). Age at first birth, smoking and education did not affect the estimates considerably and were therefore not included in the final model.
Linear trend test was performed using the median values in each category as an ordinal variable. We further assessed dose-response-relation and departure from linearity of the log OR function for predictive 25(OH)D concentrations by using fractional polynomials.28 The continuous 25(OH)D variable was entered into the multivariate logistic regression model via a set of defined transformations [x−2, x−1, x−0.5, x0.5, x2, x3 and log(x)], allowing a maximum of 2 terms (including the untransformed variable) in the model. The function that best fitted the data was selected by the −2 log likelihood of the respective model. Test for nonlinearity was performed using the Wald statistics of the transformed term (best model selected by the −2 log likelihood) including both the transformed and the linear term in the model. Effect modification was evaluated by including a cross-product term of the continuous 25(OH)D variable and potential interaction variables and testing for multiplicative interaction with the likelihood ratio test. Analysis stratified by receptor status of the tumor was performed by polytomous logistic regression adjusting for the same covariates as in the overall model with additional adjustment on age. Effect heterogeneity of 25(OH)D status by estrogen receptor (ER) or progesterone receptor (PR) status of the tumor was assessed by means of a case-only analysis. ER or PR status was used as the dependent variable (outcome) and 25(OH)D (continuously) as independent variable in the logistic regression model. Differences in cases and controls were assessed with χ2-tests for categorical variables. All tests were two-sided with a significance level of p ≤ 0.05. Statistical analyses were performed with the software SAS 9.1 (SAS Institute, Cary, NC).
Characteristics of the 289 cases and 595 controls are shown in Table I. Mean age was 42.1 and 41.6 years for cases and controls, respectively. Statistically significant differences between cases and controls were observed with respect to family history of breast cancer in at least one first-degree relative, number of births and alcohol consumption.
|Characteristics||Cases (N = 289)||Controls (N = 595)||p2|
|Age at diagnosis/recruitment (years)||0.78|
|First degree family||<0.01|
|History of breast cancer|
|Number of births||0.01|
|Age at menarche (years)||0.65|
|Duration of breast feeding (months)||0.54|
|Alcohol consumption (g/day)||0.02|
|Physical activity (MET; hr/week)3||0.32|
|Hormonal receptor status of the tumor4|
|Time of blood collection||0.36|
Geometric means of plasma 25(OH)D adjusted for time of blood collection and BMI were 45.4 nmol/L and 51.3 nmol/L in cases and controls, respectively (p < 0.0001). Median (25th–75th percentile) serum 25(OH)D was 46.7 (31.7–63.7) 52.8 (36.4–71.5), in cases and controls, respectively. We observed a statistically significant inverse association between plasma concentration of 25(OH)D and risk of premenopausal breast cancer. Compared with the lowest category (<30 nmol/L), the ORs [OR (95% CI)] for higher plasma concentrations of 30–45, 45–60, and ≥60 nmol/L were 0.68 (0.43–1.07), 0.59 (0.37–0.94) and 0.45 (0.29–0.70), respectively (ptrend = 0.0006) (Table II). We found a significantly reduced risk of breast cancer with an OR of 0.90 (0.84–0.96) per 10 nmol/L increment when considering 25(OH)D as a continuous variable. Using fractional polynomial analysis, a nonlinear association between the plasma 25(OH)D concentration and the log OR with borderline statistical significance was observed (pnonlinearity = 0.06). The function that best fitted our data was OR(x) = exp[28.6 × (1/(x + 1))] (−2 log likelihood = 946.7 as compared to 950.0 for the linear model). For graphical illustration we used the median of the lowest category from the categorical analysis, i.e. 25.2 nmol/L 25(OH)D in the controls, as reference (Fig. 1). The graph indicates a more pronounced inverse association in the lower concentration ranges of plasma 25(OH)D, while the risk function flattens with increasing 25(OH)D concentrations. The function fitted the categorical risk estimates fairly well. For comparison between the linear and nonlinear model, the linear risk function is displayed as a dotted line.
|Plasma 25-hydroxyvitamin D||Cases||Controls||Crude model1 OR (95% CI)||Adjusted model2 OR (95% CI)|
|30–45||72||24.9||131||22.0||0.70 (0.45–1.09)||0.68 (0.43–1.07)|
|45–60||68||23.5||140||23.5||0.66 (0.42–1.04)||0.59 (0.37–0.94)|
|≥60||83||28.7||237||39.9||0.48 (0.31–0.74)||0.45 (0.29–0.70)|
|per 10 nmol/L increment||289||595||0.91 (0.85–0.96)||0.90 (0.84–0.96)|
There was no significant interaction between plasma 25(OH)D and first-degree family history of breast cancer, age at menarche, duration of breast feeding, number of births, alcohol intake or BMI. We further examined the association of plasma 25(OH)D with breast cancer risk by receptor status of the tumor. The association was somehow stronger in estrogen receptor negative (ER−) (N = 89) than in ER+ tumors (N = 128), but heterogeneity was not significant (pheterogeneity = 0.27 for the continuous variable in a case-only model) (Table III). Heterogeneity reached statistical significance for the PR status of the tumor. The association was stronger in PR− (N = 96) [OR (95% CI) = 0.29 (0.15–0.56)] than in PR+ tumors (N = 117) [OR (95% CI) = 0.73 (0.39–1.37)] comparing the highest (≥60 nmol/L) with the lowest (<30nmol/L) plasma 25(OH)D category (pheterogeneity = 0.05) (Table III). However, there were few cases for this analysis (N = 43, 22, 42 and 22 in the highest category for ER+, ER−, PR+, and PR− tumors, respectively).
|Plasma 25-hydroxyvitamin D||Co||ER-positive tumors||ER-negative tumors||PR-positive tumors||PR-negative tumors|
|OR (95% CI)||OR (95% CI)||OR (95% CI)||OR (95% CI)|
|30–45||131||28||0.66 (0.36–1.21)||24||0.76 (0.39–1.49)||28||0.85 (0.44–1.61)||22||0.52 (0.27–1.01)|
|45–60||140||30||0.64 (0.35–1.17)||22||0.62 (0.31–1.25)||26||0.72 (0.37–1.39)||25||0.52 (0.27–1.01)|
|≥60||237||43||0.56 (0.31–1.00)||22||0.40 (0.20–0.81)||42||0.73 (0.39–1.37)||22||0.29 (0.15-0.56)|
|Continuous||595||128||0.93 (0.86–1.01)||89||0.86 (0.77–0.96)||117||0.95 (0.87–1.03)||96||0.84 (0.76–0.94)|
|per 10 nmol/L increment|
Median difference between time of diagnosis and time of blood collection in cases was 189 days (25th–75th percentile = 25–471). To evaluate potential effects of diagnosis or cancer therapy on circulating 25(OH)D concentration we analyzed the data stratified by time of blood collection since diagnosis (<6 months or ≥6 months). Comparing the highest with the lowest 25(OH)D category, the inverse association remained significant with ORs (95% CI) of 0.45 (0.29–0.70) for all cases and 0.33 (0.18–0.60) and 0.57 (0.32–0.99) when including cases with time of blood collection <6 months and ≥6 months since diagnosis, respectively. However, the test for trend was no longer significant for the latter group (ptrend = 0.0006, 0.0001 and 0.15, respectively).
There is ample evidence from cellular and animal studies linking 25(OH)D to breast cancer, e.g. the known anticarcinogenic effects of vitamin D with regard to apoptosis, cell differentiation and proliferation and growth inhibition of human mammary epithelial cells by both 1,25(OH)2D and 25(OH)D. In addition, the presence and expression of the vitamin D receptor and CYP27B1—the enzyme that converts 25(OH)D to active 1,25(OH)2D—as well as the uptake of the vitamin D binding protein-25(OH)D complex in mammary cells provide further support for possible anticarcinogenic effects in breast tissue.29, 30 Confirming our a priori hypothesis, we observed in this population-based case-control study a significantly inverse association between plasma 25(OH)D and premenopausal breast cancer risk. The result fits well with the recent findings in a case-control study on postmenopausal breast cancer recruited in broadly the same geographical region.21 A consistent effect of vitamin D on premenopausal and postmenopausal breast cancer was found by dose-response analysis for 25(OH)D concentrations using fractional polynomials, which showed very similar functions for premenopausal and postmenopausal breast cancer risk (Fig. 1). Vitamin D exerts its anticarcinogenic effects by different mechanisms which involve the regulation of a variety of genes acting on growth regulation. These genes include the proto-oncogenes c-fos and c-myc, cyclin-dependent kinase inhibitors p21 and p27, as well as different genes involved in apoptosis, e.g. the bcl-2 family, insulin-like growth factor I/II, and tumor necrosis factor-alpha.2, 31, 32 As these general mechanisms of cellular growth equally apply to premenopausal and postmenopausal women, a protective effect of vitamin D on both premenopausal and postmenopausal breast cancer is likely.
To our knowledge, this is the first study that assesses plasma 25(OH)D and breast cancer risk in premenopausal women only. So far the relationship between vitamin D metabolites and breast cancer has been assessed in studies including postmenopausal or predominantly postmenopausal women without adequate statistical power to address differences by menopausal status.21–25, 33 The largest cohort to date nested in the prostate, lung, colorectal and breast cancer screening trial assessing postmenopausal breast cancer and vitamin D metabolites reported no association.22 In the NHS stratification by menopausal status did not suggest an inverse association in premenopausal women but power was low in their analyses and neither subgroup specific risk estimates nor p-values of test for statistical interaction were presented.24 Prospective studies on dietary intake tend to indicate a more pronounced effect in premenopausal women as reviewed recently.26 The NHS reported a significant association between dietary vitamin D and premenopausal but not postmenopausal breast cancer risk.7 Similar observations of an inverse association for total vitamin D intake in premenopausal but not postmenopausal women was also recently reported by the Women's Health Study.15 One proposed explanation lies in the interaction of vitamin D and insulin-like growth factors, thus the potential of vitamin D to suppress IGF-stimulated cell growth.34, 35 Because of the decline of IGF levels with age, the effect of vitamin D on IGF-related tumor development may be more pronounced in premenopausal compared to postmenopausal women. Furthermore, vitamin D metabolism is impaired with aging, including a declined cutaneous synthesis as well as reduced renal production of 1,25(OH)2D.36 Thus higher amounts of vitamin D in older people may be necessary to achieve the same beneficial effects in postmenopausal than in premenopausal women. However, these hypotheses could not be confirmed in our analyses in premenopausal and postmenopausal21 women.
As reported also for postmenopausal breast cancer,21 the analysis using fractional polynomials indicate a more pronounced inverse association in the lower concentration ranges of 25(OH)D. These findings contribute to further understanding the dose-response relationship of vitamin D and breast cancer risk. In addition, our data may provide additional information in explaining diverging study results, since the range of 25(OH)D levels vary widely among studies. Median plasma 25(OH)D concentrations were 52.8 nmol/L in our control population as compared to much higher levels ranging from 67 to 80 nmol/L in the UK or US.22–24 Differences in 25(OH)D concentrations may in part be explained by seasonal differences in 25(OH)D levels that may be more or less pronounced in different countries. As expected, we observed much higher levels of 25(OH)D in cases and controls in summer (April–September) as compared to winter (October–March) (Median 25(OH)D in nmol/L = 48.5 and 42.2 for cases and 59.5 and 49.3 for controls, in summer and winter respectively). However, stratifying the main logistic regression model in samples collected in summer and winter time did not result in considerable changes of the risk estimates. Comparing the highest with the lowest 25(OH)D category, the ORs (95% CI) were 0.40 (0.20–0.82), 0.44 (0.23–0.84) and 0.45 (0.29–0.70), for summer, winter and all samples respectively. Moreover, the collected blood samples were more or less equally distributed over the quarters of the year. We therefore, conclude that potential seasonal effects did not influence our results.
Our findings of a significant inverse association in PR− and ER− tumors only confirm the recent observation for dietary intake of vitamin D in the same premenopausal study population.13 However, effect heterogeneity reached statistical significance only for PR status of the tumor. Data from the NHS also suggests an inverse association in ER− tumors only.24 However, we did not find any evidence for effect modification in postmenopausal breast cancer by ER or PR status of the tumor.21 Anticarcinogenic effects of vitamin D could be mediated via the estrogen pathway by downregulation of the ER and therefore disruption of estrogen mediated mitogenic signals.37–39 Yet responsiveness to the growth inhibitory effects of vitamin D in ER− tumor cells have also been reported.40, 41 The observations on effect modification by receptor status of the tumor could have been a chance finding since the number of cases was very small, and test for heterogeneity in a case-only model was borderline significant only for PR status of the tumor.
Our study results raise the concern of bias due to the retrospective study design. A cancer diagnosis may change plasma 25(OH)D levels due to less sun exposure or less dietary vitamin D ingestion or as a consequence of cancer treatment. Therefore, reverse causality, i.e. the lower 25(OH)D levels were a consequence of the cancer diagnosis and not vice versa, cannot be excluded in our study. However, short-term changes in diet or behavioural habits are less likely to influence plasma concentration, because the half-life of 25(OH)D is rather long (about 3 weeks)42 and plasma levels of 25(OH)D have been reported to be fairly consistent over time.43 Although prospective studies with measurement in blood samples collected prior to diagnosis are the preferred study design, it is, however, still unclear at what time in life 25(OH)D should be measured to best predict breast cancer risk.26 Effects of vitamin D and analogues have been reported in early initiation and promotion phases of cancer development44, 45 as well as in late phases when used as a chemotherapeutic agent.46 To overcome these problems, repeated measurements at different time points before diagnosis would be most valuable.
The effects of chemotherapy or hormonal therapy on 25(OH)D concentration are not profoundly studied. However, a notable change in 25(OH)D concentration after chemotherapeutic treatment was not observed in 2 small studies.47, 48 Chemotherapy is commonly given within the first 6 months after diagnosis. We therefore stratified our analysis by time of blood collection since time of diagnosis using 6 months as cutpoint. The association was weaker in cases whose blood samples were collected ≥6 months since diagnosis than that for those collected <6 months since diagnosis, suggesting a potential overestimation of the protective effect in the main analysis. However, the risk estimate for the highest category of plasma 25(OH)D remained significant in both groups.
Selection bias is of further concern in the this study. However, comparison between the study group with information on 25(OH)D and the original study population provided only minor differences with respect to sociodemographic variables and breast cancer risk factors. In addition, 25(OH)D status in the control group was fairly comparable to data from the representative German National Health Survey for women in this age range.49 Another concern in our study lies in the validity of the 25(OH)D measurement since various publications reported variation within different assays and laboratories.50–52 However, the IDS enzyme immunoassay used in our study gave comparable results to the gold standard, high performance liquid chromatography, in the international Vitamin D Quality Assessment Scheme (DEQAS).53 Moreover, the low intra- and interassay variation as well as the low deviation from the mean between 2 duplicates give assurance on the validity of the assay.
A major strength of our study is the population-based study design and the ability to account for potential breast cancer risk factors and confounders. Furthermore, 25(OH)D status was assessed in detail using a categorical, a continuous and a fractional polynomial approach to assess dose-response relationships.
In summary, high concentrations of plasma 25(OH)D were associated with a significantly reduced premenopausal breast cancer risk. In a dose-response analysis, the association was found to be nonlinear. There was a stronger inverse association at low plasma 25(OH)D concentration ranges and a flattening of the risk function with higher 25(OH)D concentrations. The findings of a stronger inverse association in PR− tumors deserve further investigation. Our results suggest a protective effect of vitamin D as measured by plasma 25(OH)D for premenopausal breast cancer.
The authors are grateful to all participants for their help and patience as well as to the many gynecologists and oncologists in the clinics of the study regions for allowing us to contact their patients. They also thank U Eilber for data coordination and management.
- 42Vitamin D: biosynthesis, metabolism, and mode of action. In: DeGrootLJ,CahillGFJr,MartiniL,NelsonDH,OdellWD,PottsJT,Jr,SteinbergerE,WinegradAI, eds. Endocrinology. New York: Grune and Stratton, 1989. 902–26..