Prospective Population-Based Study of the Association between Serum 25-Hydroxyvitamin-D Levels and the Incidence of Specific Types of Cancer

Vitamin D is a fat soluble vitamin produced in sun-exposed skin and can be ingested from the diet and dietary supplements. In addition to its traditional role in bone metabolism and remodeling, vitamin D has numerous biologic functions ranging from antiproliferative and antiangiogenic effects to modulation of the immune system (1). The vitamin D receptor is found in most cells of the body and many tissues can convert vitamin D to its active form (1). Vitamin D insufficiency and deficiency are common worldwide and are associated with a number of common diseases, such as cardiovascular risk factors (e.g., hyperlipidemia and albuminuria), diabetes, cardiovascular disease, and mortality (1–7).

The role of vitamin D in cancer is largely unresolved. Cancer is a broad group of diseases resulting from unregulated cell growth. Six main features of cancer cells have been suggested: self-sufficiency of growth signals, evasion of apoptosis, insensitivity to antigrowth signals, sustained angiogenesis, limitless replicative potential, and tissue invasion and metastasis (8). Through its ability to induce apoptosis and prevent angiogenesis and migration in malignant cells, vitamin D could play an important role in some of the common pathways of cancer (1).

From a public health point of view, both the impact of vitamin D status on specific cancer types and the impact of vitamin D on the total incidence of cancer are important. In observational studies, vitamin D deficiency was associated with an increased risk of cancers, such as colorectal cancer (9), lung cancer (10), and breast cancer (11). Also, a recent meta-analysis by Yin and colleagues found a moderate inverse association between vitamin D status and total cancer incidence and mortality (12).

We investigated the prospective association between vitamin D status, as assessed by serum 25-hydroxyvitamin D (25-OH-D), and the specific type of cancer according to The International Classification of Disease in three cohorts from the general Danish population.

Participants gave their informed written consent. The studies were approved by the Ethics Committee of Copenhagen and the Danish Data Protection Agency, and the recommendations of the Declaration of Helsinki were followed.

We included the population-based studies, Monica10, Inter99, and Health2006, all recruited from the Danish Central Personal Register as random samples of the population in the southern part of the former Copenhagen County. The Monica10 study (1993–1994) included 2,656 individuals, which was 64.3% of the 4,130 individuals that were invited, between 40 and 71 years old (13).

The Inter99 study was conducted between 1999 and 2001 and included examination of 6,784 individuals aged 30 to 60 years from the general population (14). The baseline participation rate was 52.5%. The Inter99 study was a population-based randomized controlled trial (CT00289237, ClinicalTrials.gov) performed to investigate the effects of lifestyle intervention on cardiovascular disease (14). Only participants with a Northern European origin (Danish, Norwegian, Swedish, Icelandic, or Faroese nationality) were included in the current study (N = 6,405).

In the Health2006 study, 7,931 Danish citizens aged 18 to 69 years and born in Denmark were invited to participate in a health examination (15). A total of 3,471 (43.8%) individuals were examined between 2006 and 2008. In the present study, we included 2,649, 6,146, and 3,409 participants from the Monica10, the Inter99, and the Health2006 studies, respectively, with measurements of vitamin D status, yielding a total of 12,204 persons. The three studies included questionnaires, physical examinations, and blood tests.

Measurements of serum 25-OH-D in the Monica10 study were performed by the IDS-SYS 25-Hydroxy Vitamin D method with the IDS-iSYS Multi-Discipline System (IDS Nordic A/S; ref. 16). In the Inter99 study, measurements of 25-OH-D were done by high-performance liquid chromatography as previously described (17). In the Health2006 study, 25-OH-D was measured by immunoassay using Cobas e411 (Roche Diagnostics; ref. 18).

All residents in Denmark have a unique and permanent personal civil registration number that allows data linkage from national registers on an individual level. Participants were followed until July 11, 2011. Information on fatal and nonfatal cancers was obtained from the Danish Cancer Registry (19, 20) according to the International Classification of Diseases (ICD). Reporting to the Cancer Registry has been mandatory since 1987. From 1943 to 1978, the Registry was classified according to the modified ICD-7, and starting in 1978, the diagnoses were coded in accordance with the ICD-10 (19). Information on deaths and emigration status was obtained from the Danish Civil Registration System (21). The grouping of codes used for ICD-10 is not unambiguously translated to ICD-7 codes (or vice versa). However, as we are only using ICD-7 codes for excluding people who had cancer before the study, and since most of these cases are, in fact, ICD-10 coded, the inconsistency is likely to be negligible.

Our classification of the main types of cancer (our end points) according to the ICD-7 and the ICD-10 codes is displayed in Table 1. For each cancer type, we excluded the persons with a history of that particular cancer at baseline in the regression analyses.

Questionnaires were used to obtain the following: information on education/vocational training [no education (only basic education), education beyond basic including students]; intake of fish (<twice a week, ≥twice a week); physical activity during leisure time (sedentary, light, or moderate/vigorous); smoking habits (never smoked, ex-smoker, occasional smoker, current smoker <15 g/day; 15–<25 g/day, or ≥25 g of tobacco/day; 1 cigarette = 1 g, 1 cheroot = 2 g, 1 cigar = 3 g, pipe = stated in g); and alcohol consumption (0, >0–7, >7–14, or >14 drinks per week). Weight and height were measured with no shoes and light clothes. Body mass index (BMI) was calculated as weight divided by height squared (kg/m²).

Analyses were performed using SAS, version 9.2 (SAS Institute Inc.). P values were two-tailed, and statistical significance was defined as P < 0.05. Descriptive characteristics (number) and vitamin D status (mean and SD) of the study participants according to the study population are presented and compared with the Kruskal–Wallis test in Table 2. Table 3 shows the distribution of cancers (% and number) and corresponding vitamin D status (mean and SD) according to the three cohorts and overall.

In Tables 4 and 5, the associations between vitamin D status and incidence of type-specific cancer are shown. For each regression analysis, only participants for which we had complete information for all considered variables were included (complete case analysis). Vitamin D status was used both as a continuous variable and in season-specific quartiles (divided before pooling of data, with the lowest quartile used as reference; ref. 22). We pooled the data from the three population-based studies. Persons with the cancer of interest before the baseline examination were excluded from the analyses. In the Cox regression analyses, we used age as the underlying time and delayed entry where participants entered the analysis at the baseline age and exited the analysis at their event or censoring age. The first model was adjusted for study and gender. The second model was further adjusted for education, time of year when the blood sample was drawn (March–May, June–August, September–November, or December–February), physical activity, smoking habits, alcohol intake, intake of fish, and BMI. P_trend was the P value for a linear trend across quartiles. Only women were included for the analysis of breast and uterine cancers. Likewise, only men were included in the analysis of prostate cancer. There were no interactions between study population and vitamin D status, so there was no evidence of vitamin D having differential effects in the different study populations. The assumption that there was a linear relationship between vitamin D status and cancer was tested by adding vitamin D status squared and checking for significance.

In additional analyses, we excluded persons who developed a cancer of interest in the first two years after examination. Therefore, participants started to contribute risk time at baseline + two years (Supplementary Table S1). We also stratified the analyses by gender and BMI, respectively (Supplementary Table S2). In the analyses stratified by BMI, we excluded persons with BMI < 18.5 kg/m².

In further analyses, we used Stata, version 12.1 (StataCorp LP) for meta-analyses of the study-specific estimates of the association between vitamin D status and cancer using both fixed and random effects models, and the results are summarized in Fig. 1. Heterogeneity between the studies was assessed by the I² test.

Vitamin D levels were highest in the Monica10 study and lowest in the Health2006 study, with the Inter99 study showing intermediate levels. In addition, vitamin D levels were highest in the summer and autumn. Levels were highest in individuals with normal BMIs and were lower in both underweight and overweight individuals. Likewise, physically active participants had higher vitamin D levels than those that were physically inactive. Abstinent individuals had lower vitamin D status than nonabstinent (Table 2).

The overall mean (SD) vitamin D levels among individuals with a history of cancer before baseline (N = 471), incident (N = 1,248), and no cancer (N = 10,485) were 53.7 (27.6), 57.4 (27.0), and 51.8 (26.5) nmol/L, respectively. The vitamin D levels according to study population are shown in Tables 2 (history of cancer) and 3 (no cancer and incident cancer). Almost 11% of those without cancer before baseline developed some type of cancer during the follow-up (Table 3). The most frequent cancer, non-melanoma skin cancer (NMSC), affected 3.3% of the persons, followed by breast cancer (1.4%), colorectal cancer (1.2%), prostate cancer (1.1%), and cancer of bronchus and lung (1.0%; Table 3). The least abundant cancers included in this study were malignant melanoma (0.5%), cancer of the urinary organs (0.4%), head and neck cancer (0.4%), and cancer of the uterus (0.2%; Table 3).

The mean baseline vitamin D status was 57.4 nmol/L among persons developing any type of cancer during follow-up. Having a mean vitamin D status below this value were, in ascending order, persons developing head and neck cancer, person without incident cancer and persons with colorectal cancer, breast cancer, cancer of bronchus and lung, and all cancers excluding NMSC. Having a mean vitamin D status above the mean of any cancer were persons developing malignant melanoma, prostate cancer, NMSC, cancer of the urinary organs, and cancer of the uterus (Table 3).

The overall person-years at-risk was 120,680 years for all cancers. The median follow-up time was 11.3 years: 16.8, 11.5, and 4.0 years in the Monica10, Inter99, and Health2006 studies, respectively. The overall HR [95% confidence interval (CI)] for total cancer incidence was HR = 1.02 (95% CI, 0.99–1.04) per 10 nmol higher vitamin D status in the fully adjusted model (Table 5). For all cancers excluding NMSC, we found that HR = 1.00 (95% CI, 0.97–1.03). We found a statistically significant positive association between vitamin D status and NMSC, with an HR = 1.06 (95% CI, 1.02–1.10) per 10 nmol/L higher baseline vitamin D status in both the partly and the fully adjusted model (Table 5). The association between vitamin D status and development of colorectal cancer was statistically significant, with an HR = 0.93 (95% CI, 0.87–1.00) for a 10 nmol/L higher vitamin D status in the partly adjusted model, but was not significant in the fully adjusted model, although the estimate only changed little. On the other hand, the statistically insignificant inverse association between vitamin D status and head and neck cancer in the partly adjusted model changed when adjusted for smoking habits: the HR changed from 0.88 (Table 5, model 1) to 0.97 (Table 5, model 2) per 10 nmol/L higher vitamin D status. Except for the abovementioned associations, we found no statistically significant associations between baseline vitamin D status and other types of cancers. In general, the CIs were relatively narrow for many of the cancers (Tables 4 and 5). When we excluded individuals who developed the cancer of interest in the first two years after baseline (Supplementary Table S1), the results were similar. There were no major changes when we stratified by gender and BMI, respectively (Supplementary Table S2). The association between vitamin D status and NMSC were, however, no longer statistically significant for persons with BMI ≤ 25 kg/m².

The results from meta-analyses of the study-specific estimates are summarized in Fig. 1. They were very similar to the individual-based analyses (Table 5). As in the individual-based analyses, there was a statistically significant higher risk of NMSC with higher vitamin D status. Although there were signs of heterogeneity across studies in some of the outcomes, the estimates change little. At most, the estimate differed between 0.95 (Table 5) and 0.99 (Fig. 1) for vitamin D and colorectal cancer when using the random effects model.

We explored the association between vitamin D status and total and specific cancers in the same populations. We found a statistically significant positive association between vitamin D status and the incidence of NMSC. Except for this, there were no statistically significant associations between vitamin D status and total cancer as well as specific cancers in the fully adjusted models. However, the inverse associations between vitamin D status and colorectal cancer and head and neck cancer were statistically and borderline significant, respectively, in the partly adjusted models.

Regarding other general population studies on vitamin D status and total cancer incidence, the results were inconclusive. Ordonez-Mena and colleagues found no statistically significant association between vitamin D status and the incidence of total cancer in a general German population cohort, except in subgroup analyses (23). Thus, there was a significantly increased overall cancer risk for low vitamin D status among men, the non-obese, and individuals reporting low fish consumption, and for high vitamin D status among non-smokers and non-obese individuals (23). Afzal and colleagues performed an analysis on 9,791 participants from the Copenhagen City Heart study and found an HR = 1.06 (95% CI, 1.02–1.11) for a 50% reduction in vitamin D for all cancers (24). Yin and colleagues, however, recently summarized the results from prospective studies of vitamin D and cancer in a meta-analysis and found a moderate inverse association between vitamin D status and total cancer incidence [risk ratio (RR) = 0.89; 95% CI, 0.81–0.97], and mortality (RR = 0.83; 95% CI, 0.71–0.96) per 50 nmol/L higher vitamin D status (12).

Although NMSCs are the most common form of cancer, accounting for about one-third of all cancers worldwide, they are often excluded from cancer statistics because they are easily treated and are almost always cured, often in a short, outpatient procedure. It is worth noting that we chose to exclude carcinoma in situ in general and performed analyses both with and without NMSCs in the analyses. Neither the analyses on all cancers nor the analyses on all cancers excluding NMSC showed statistically significant associations with vitamin D status.

The observed positive association between vitamin D status and NMSC may be explained by the fact that vitamin D status is a marker for UV exposure. Our results are in agreement with a study by Eide and colleagues who found that not only was an increased baseline vitamin D status significantly associated with an increased risk of NMSC, but also that the association was attenuated on less exposed body sites, indicating that UV exposure is a likely confounding factor (25). Supporting this view is a study by Tang and colleagues, who performed post hoc analyses of the women's health initiative randomized controlled trials (RCT) and found that vitamin D plus calcium supplementation did not reduce the overall incidence of NMSC (26).

A meta-analysis on vitamin D and colorectal cancer by Ma and colleagues revealed an inverse association between blood 25-OH-D levels and the risk of colorectal cancer (9). They included nine prospective studies with objective measurements of vitamin D levels and the risk for colorectal, colon, or rectal cancer, which yielded a pooled risk ratio of 0.74 (95% CI, 0.63–0.89) for a 10 ng/mL higher vitamin D. In comparison, we found an HR = 0.95 (95% CI, 0.88–1.02) for a 10 nmol/L higher vitamin D status.

Our results for vitamin D and head and neck cancer are partially in line with a study by Orell-Kotikangas and colleagues, who investigated 57 patients with head and neck cancer and found subnormal vitamin D levels in a significant proportion of the patients (27). However, a study by Meyer and colleagues of 522 patients with head and neck cancer revealed that vitamin D status before treatment did not influence the disease outcomes (28).

Our results for vitamin D and prostate cancer are in agreement with previous studies. In a meta-analysis, Yin and colleagues summarized the evidence from existing longitudinal studies on the association between vitamin D and the risk of prostate cancer. Yin and colleagues included 11 original reports and found a summary OR of 1.03 (95% CI, 0.96–1.11) per 10 ng/mL higher vitamin D. They concluded that vitamin D status is not associated with the incidence of prostate cancer based on the available evidence.

A meta-analysis on 25-OH-D levels and breast cancer by Yin and colleagues found no statistically significant association in the cohort studies measuring vitamin D status at baseline before cancer diagnosis (29), whereas a meta-analysis of case-control studies by Chen and colleagues found a statistically significant lower risk of breast cancer for the women in the highest vitamin D quartile when compared with the lowest (30). A recent meta-analysis of prospective studies reported a dose–response relationship with a statistically significant 3.2% reduction in breast cancer risk per 10 ng/mL increment in serum 25-OH-D concentration (31). In comparison, we found no statistically significant association between vitamin D status and incident breast cancer.

The strengths of our study include the large random sample of the Danish population many of whom have low vitamin D levels (17) and the longitudinal population-based design minimizing risk of reverse causation (i.e., that the disease modifies lifestyle and vitamin D status); a long-term follow-up and the use of standardized registry-based diagnoses with almost no individuals lost to follow-up; and the available prospectively collected information on several important potential confounders, which allowed us to minimize confounding. To further reduce the risk of reverse causation, we performed additional analyses, where we excluded individuals that developed the cancer of interest in the first two years after the initial examination. Another strength of our study is that we included three cohorts with baseline data from two decades.

The limitations of the study include the different methods of measuring vitamin D levels in the merged studies, the risk of residual confounding inherent in an observational study, and the single vitamin D measurement, which likely loses predictive power over time. Also, the types of cancer in the main groups may not share the same potential vitamin D-dependent pathway. The vitamin D levels differed between the studies, which could have been due to evaporation during storage. Therefore, we chose to use vitamin D both as a continuous variable and in season-specific quartiles, classified before pooling of the data. We included meta-analyses of the study-specific estimates. Also, we may have lacked ability to detect small to medium effect sizes. The relatively low participation in some of the included cohorts may result in selection bias and limit the generalizability of the results. Of note, however, another similar Danish study using registry-based outcomes with almost complete follow-up of participants, but with lower baseline participation than in the present study was shown not to be biased by baseline nonparticipation (32). We used complete case analysis for missing covariates in the present study due to the simplicity and comparability across analyses. An alternative approach would be multiple imputations. A study by White and Carlin examined bias and efficiency of multiple imputation compared with complete case analysis for missing covariates (33). On the basis of theoretical results and simulation studies of different scenarios including uni- and bivariate analyses with covariates missing completely at random and at random, respectively, they concluded that none of the methods was universally applicable although the multiple imputation method seemed appropriate across a wider range of settings. In general, both methods were valid when covariates were missing completely at random.

Regarding RCTs, a recent meta-analysis by Bjelakovic and colleagues reported a significantly decreased cancer mortality in a total of 44,492 person from 4 trials using vitamin D₃ supplementation (34). They estimated the quality of the evidence as moderate quality which means that further research is likely to change the estimate and to affect the confidence in the estimate. In particular, the validity of results may be questionable because a large number left the trials before completion. Also, more RCTs are needed in younger, healthy persons and in elderly with no apparent vitamin D deficiency. In comparison, a systematic review by Lazzeroni and colleagues reported a statistically nonsignificant lower risk of cancer mortality in the supplementation group (35), possible due to several things. First, the studies were neither designed, nor sufficiently powered, to investigate the cancer end points. Second, most studies supplemented with both calcium and vitamin D, thus preventing distinguishing between the calcium and the vitamin D effects and raising the possibility that combination therapy is required to prevent cancer. In view of existing evidence from RCTs, Lazzeroni and colleagues suggested that vitamin D could be a risk marker for cancer rather than a risk factor (35). In support of this, Thuesen and colleagues previously reported that vitamin D status was significantly associated with several known cancer-related risk factors (17). There are several ongoing phase II and III RCTs investigating the effect of vitamin D supplementation alone and when combined with calcium on total and different cancers (35).

Currently, the evidence regarding a potential role for vitamin D in the prevention of cancer is inconclusive (36). We found no statistically significant associations between vitamin D status and incidence of total or specific cancers, except for the statistically significant positive association between vitamin D status and the incidence of NMSC. Our results do not indicate that there is an impact of vitamin D on total cancer incidence. However, since cancer is a class of several diseases rather than a single one, vitamin D could be beneficial for the prevention of some cancers while having no or even a detrimental effect on others. We are still awaiting the results of large RCTs to help elucidate this point.

No potential conflicts of interest were disclosed.

Conception and design: T. Skaaby, B.H. Thuesen, A. Linneberg

Development of methodology: T. Skaaby

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.L.N. Husemoen, C. Pisinger, T. Jørgensen, A. Linneberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Skaaby, L.L.N. Husemoen, B.H. Thuesen, T. Jørgensen, A. Linneberg

Writing, review, and/or revision of the manuscript: T. Skaaby, L.L.N. Husemoen, B.H. Thuesen, C. Pisinger, T. Jørgensen, N. Roswall, S.C. Larsen, A. Linneberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Skaaby, C. Pisinger, A. Linneberg

Study supervision: L.L.N. Husemoen, A. Linneberg

T. Jørgensen received a grant from the Health Insurance Foundation (grant No. 2010 B 131).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.