Plasma 25-Hydroxyvitamin D and Risk of Pancreatic Cancer Free

This article is featured in Highlights of This Issue, p. 1

Epidemiologic evidence has suggested that adequate circulating vitamin D is associated with health benefits, leading to proposals to increase vitamin D intake in the general population (1–3). Based principally on studies of bone health, a recent report from the Institute of Medicine recommended dietary intake to achieve circulating 25-hydroxyvitamin D (25[OH]D) of >20 ng/mL (50 nmol/L; ref. 4). Laboratory studies suggest that vitamin D and its analogues may inhibit cancer development and growth, including pancreatic adenocarcinoma (5–7). Despite these data, epidemiologic investigations of vitamin D and incident pancreatic cancer have been inconsistent. Ecological studies of latitude variation and ultraviolet B radiation have suggested a possible inverse relationship between vitamin D and pancreatic cancer risk and mortality (8, 9). Studies of vitamin D intake and predicted vitamin D status have also suggested an association between higher vitamin D and reduced pancreatic cancer risk and mortality (10–12). However, studies evaluating circulating 25(OH)D have been conflicting, with some studies reporting an elevated risk for pancreatic cancer among those with the highest levels of plasma 25(OH)D (13–15).

Given the strong laboratory data and conflicting epidemiologic studies, we prospectively examined the association between plasma 25(OH)D and pancreatic cancer risk in 5 large cohorts of initially healthy men and women, in which plasma samples were collected prior to cancer diagnosis.

Our study population included pancreatic cancer cases and controls from 5 large, prospective studies: Health Professionals Follow-Up Study (HPFS), Nurses' Health Study (NHS), Physicians' Health Study (PHS), Women's Health Initiative-Observational Study (WHI), and Women's Health Study (WHS). HPFS was initiated in 1986 when 51,529 U.S. men aged 40 to 75 years working as dentists, veterinarians, pharmacists, optometrists, osteopathic physicians, or podiatrists responded to a mailed questionnaire. NHS was established in 1976 when 121,700 female registered nurses aged 30 to 55 years responded to a mailed questionnaire. The participants in these cohorts have subsequently responded to biennial questionnaires to update exposures and disease diagnoses. PHS is a completed randomized trial initiated in 1982 of aspirin and β-carotene among 22,071 male physicians between 40 and 84 years of age. After completion of the randomized components, study participants were followed as an observational cohort. WHI consists of 93,676 postmenopausal women, aged 50 to 79 years, enrolled between 1994 and 1998. The health of these participants is tracked via annual health forms and clinic visits. WHS is a completed randomized trial, initiated in 1992, of low-dose aspirin and vitamin E among 39,876 female professionals 45 years of age or older. The trial was closed in 2004, with all study participants followed as an observational cohort. Further details for these cohorts have been provided previously [HPFS (ref. 16), NHS (ref. 17), PHS (ref. 18), WHI (ref. 19), and WHS (ref. 20)]. The study was approved by the Human Research Committee at the Brigham and Women's Hospital, Boston, MA, and participants provided informed consent.

We included all cases of incident pancreatic adenocarcinoma diagnosed through 2008 with available plasma and no prior history of cancer, except non-melanoma skin cancer. Eligible controls were alive, free of cancer at of the date of the case's diagnosis, and provided a blood sample. We randomly selected 2 or 3 controls for each case, matching on cohort, year of birth, smoking status, fasting status, and month of blood draw. Median follow-up among controls was 14.1 years in HPFS, 18.3 years in NHS, 25.3 years in PHS, 12.2 years in WHI, and 14.4 years in WHS.

Incident cases of pancreatic cancer were identified by self-report or during follow-up of a participant's death. Deaths were ascertained from next-of-kin or the postal service and by searching the National Death Index; this method has been shown to capture >98% of deaths (21). For all cases, medical records were reviewed by study physicians blinded to exposure data. Of the 451 pancreatic cancer cases in the analysis, 448 (99.6%) were confirmed by review of medical records, tumor registry data, or death certificates.

Our pooled analysis of nested case–control studies included 501 cases of pancreatic adenocarcinoma and 1,175 matched controls from 5 prospective cohorts. Two cases and 5 controls were removed because of failure of the assay. One case (and its 2 matched controls) and one control were removed because of outlier values for 25(OH)D (>300 nmol/L). Because of concern about the possible influence of subclinical malignancy on lifestyle choices and plasma 25(OH)D, we excluded 47 cases diagnosed within 2 years of blood draw. To maximize power, matched controls for these cases were maintained in the data set, resulting in 451 cases and 1,167 controls for analysis.

Individual characteristics and habits were obtained from baseline questionnaires at study enrollment in PHS, WHI, and WHS and from questionnaires preceding the date of blood draw in HPFS and NHS. In all cohorts, data were available for age at blood draw, sex, race/ethnicity, weight, height, smoking status, history of diabetes, multivitamin use, and region of residence. We used body mass index (BMI) as a measure of total adiposity. For HPFS, NHS, WHI, and WHS, energy-adjusted intakes (22) of vitamin D, calcium, and retinol were available from validated food frequency questionnaires (23–25).

In HPFS, NHS, and WHI, alcohol intake was calculated in g/d. In PHS and WHS, participants were asked number of drinks consumed by prespecified categories. Because 1 standard drink equals 13.7 grams of alcohol, harmonized categories included: rarely/never, 1–3 drinks/mo, 1–6 drinks/wk, and 1+ drinks/d. In HPFS, NHS, WHI, and WHS, physical activity was calculated in metabolic equivalent task-h/wk (MET-h/wk), as previously described and validated (26, 27). In PHS, participants were asked number of times per week they exercised vigorously (rarely/never, 1–3/mo, 1/wk, 2–4/wk, 5–6/wk, 7/wk). Assuming a MET score of 6 for vigorous activity and 60 minutes of exercise per episode, we assigned 6 MET-h/wk for each episode of activity to determine the MET-h/wk for PHS participants.

Blood samples were collected from 18,225 men in HPFS from 1993 to 1995, 32,826 women in NHS from 1989 to 1990, 14,916 men in PHS from 1982 to 1984, 93,676 women in WHI-OS from 1994 to 1998, and 28,345 women in WHS from 1992 to 1995. All blood samples were continuously stored in well-monitored freezers. Details on blood draw procedures, transportation, and storage of plasma samples in these cohorts have been described previously [HPFS (ref. 28), NHS (ref. 29), PHS (ref. 30), WHI (ref. 31), and WHS (ref. 32)].

Plasma levels of 25(OH)D were assayed in the laboratory of Dr. Nader Rifai (Children's Hospital, Boston, MA), using the 25-Hydroxyvitamin D Enzyme Immunoassay Kit from Immunodiagnostic Systems, as per manufacturer's instructions. All samples were handled identically in a single batch and laboratory personnel were blinded to case, control, or quality control sample. The mean intra-assay coefficients of variance were ≤9% for blinded, replicate, quality control samples.

Quintiles of 25(OH)D were created from control subjects in all 5 cohorts jointly and in each cohort separately. In our primary analysis, we computed ORs to estimate relative risks and 95% confidence intervals (CI) using unconditional logistic regression, as the case–control match was broken due to exclusion of cases with short lag time between blood draw and cancer diagnosis. We also conducted conditional logistic regression among all subjects and after exclusion of cases and matched controls with blood drawn within 2, 3, and 5 years of pancreatic cancer diagnosis. Furthermore, we evaluated cohort-specific quintiles by meta-analysis in age-adjusted models, using a random-effects model (33). Of note, because of small numbers in one cohort (WHS: 29 cases/69 controls), multivariable adjustment led to an unstable logistic regression model, precluding meta-analysis of the multivariable-adjusted data. To evaluate whether one cohort unduly influenced the pooled results, we conducted a sensitivity analysis, whereby risk estimates were assessed after individually excluding each cohort in turn. Tests for trend using two-sided P values were calculated by entering the pooled quintile-specific median values for 25(OH)D as a variable in logistic regression models. Given that 25(OH)D was not normally distributed, we also evaluated a “global” trend test by including log-transformed 25(OH)D as a continuous variable in logistic regression models.

By adding each covariate in turn to age-adjusted models, we investigated possible confounders, including sex, cohort, BMI (World Health Organization categories), smoking status (never, past, current), history of diabetes, multivitamin use, physical activity (continuous, MET-h/wk), alcohol intake (never/rare, 1–3 drinks/mo, 1–6 drinks/wk, ≥1 drink/d), and total and dietary intakes of calcium (quartiles), retinol (quartiles), and calories (quartiles). In no instance, did addition of the covariate change the point estimate of risk by more than 10%. Given that no strong confounders of the relationship between 25(OH)D and pancreatic cancer risk were found, we investigated 3 models incorporating covariates, as in previous studies (14, 15, 34). The first model adjusted for age. The second adjusted for covariates thought a priori to have the greatest potential for confounding, including age, cohort (which also adjusts for sex), BMI (35), smoking status (36), history of diabetes (37), and multivitamin use (38), plus month of blood draw. The third additionally adjusted for fasting time, physical activity (39), alcohol intake (40), and dietary intakes of calcium (41), retinol (41), and calories (41). In our analysis of multivariable-adjusted quintiles of plasma 25(OH)D and pancreatic cancer risk, no change was noted in risk estimates between the second and third models, so results from the second model are presented.

Although optimal levels of 25(OH)D have not been definitively determined, plasma 25(OH)D <50 nmol/L has been defined as “insufficiency,” 50 to <75 nmol/L as “relative insufficiency,” and ≥75 nmol/L as “sufficient” (42, 43). Therefore, we investigated the risk of pancreatic cancer for relatively insufficient and sufficient subjects compared with insufficient subjects, using the above cutoff points. We also investigated risk among subjects with 25(OH)D ≥100 nmol/L, given a previous study suggesting increased risk specifically among these subjects (15).

We assessed the association of pancreatic cancer risk with plasma 25(OH)D in specific subgroups using cohort-specific quartiles of 25(OH)D, given fewer subjects within these subgroups. Because plasma levels of 25(OH)D may differ by race/ethnicity and gender, we conducted separate analyses in Whites, men, and women. In stratified analyses, we assessed for statistical interaction, by entering the main effect terms and a cross-product term of the 25(OH)D quintile trend and binary stratification variable into the model, evaluating likelihood ratio tests. We assessed heterogeneity in the association between 25(OH)D and pancreatic cancer risk across the cohorts using Cochran's Q-statistic (44); the Cochran's Q-statistic P value was 0.12 for the comparison of the fifth versus the first quintile of plasma 25(OH)D across the cohorts. All statistical analyses were conducted with the SAS 9.1 statistical package (SAS Institute Inc.) and all P values are two-sided.

Baseline characteristics of controls by quintile of 25(OH)D and by cohort are shown in Table 1 and Supplementary Table S1, respectively. Mean 25(OH)D was 64.5 nmol/L in control subjects and 61.3 nmol/L in case subjects (Wilcoxon rank-sum: P = 0.005). The global tests for trend were P_trend = 0.01 in the age-adjusted logistic regression model and P_trend = 0.03 in the multivariable-adjusted model, evaluating log-transformed, continuous 25(OH)D. Compared with participants in the bottom quintile of 25(OH)D, those in the second, third, fourth, and fifth quintiles had multivariable-adjusted ORs (95% CI) of 0.79 (0.56–1.10), 0.75 (0.53–1.06), 0.68 (0.48–0.97), and 0.67 (0.46–0.97), respectively (Table 2).

In both a pooled analysis of the raw data (Table 2) and a meta-analysis of the risk estimates, this inverse association was highly similar using cohort-specific quintiles. In the meta-analysis, the age-adjusted OR for the fifth versus first quintile of plasma 25(OH)D was 0.70 (0.43–1.14; P_trend = 0.03). Finally, the use of conditional logistic regression also resulted in highly similar inverse associations, even after exclusion of cases with successively greater time between blood draw and pancreatic cancer diagnosis and their matched controls (Table 3).

We next investigated clinically defined cutoff points for 25(OH)D (Table 4; refs. 42, 43). Compared with those with insufficient levels of vitamin D, ORs were 0.75 (0.58–0.98) for those with relative insufficiency and 0.71 (0.52–0.97) for those with sufficient levels. After exclusion of each cohort in turn, ORs for pancreatic cancer were similar comparing those with sufficient levels of 25(OH)D with those with insufficient levels: 0.70 (0.51–0.97) when excluding HPFS, 0.63 (0.45–0.88) when excluding NHS, 0.73 (0.53–1.00) when excluding PHS, 0.77 (0.54–1.11) when excluding WHI, and 0.65 (0.48–0.88) when excluding WHS. Given prior data suggesting a possible elevated risk for pancreatic cancer among subjects with high 25(OH)D (13, 15), we conducted an analysis using cutoff points as defined in the Cohort Consortium Vitamin D Pooling Project of Rarer Cancers (Table 5; ref. 15). No increased risk for pancreatic cancer was noted among subjects with 25(OH)D levels ≥100 nmol/L.

Similar inverse associations were noted in analyses limited to women, men, and Whites (Table 6). No statistically significant interaction of 25(OH)D and pancreatic cancer risk was seen by other potential predisposing factors for pancreatic cancer, or by month of blood draw, region of residence, or fasting status (Supplementary Table S2; all P interaction ≥0.18).

In this pooled analysis of 5 nested case–control studies, we noted a statistically significant inverse association between prediagnostic plasma 25(OH)D and the subsequent risk of pancreatic cancer. A similar inverse relationship was seen across a wide variety of subgroups including Whites, women, and men. Furthermore, the inverse association was similar regardless of whether blood was drawn in months with low or high ultraviolet light exposure. Analyses of clinically defined categories of 25(OH)D showed that those subjects with “sufficient” 25(OH)D had an approximate 30% lower risk for pancreatic cancer than those who were vitamin D “insufficient.” No increase in risk for pancreatic cancer was seen among subjects with 25(OH)D ≥100 nmol/L.

Laboratory studies have shown that pancreatic cancer cells express 25(OH)D-1α-hydroxylase (5) which metabolizes 25(OH)D to the active form of 1,25-dihydroxyvitamin D (1,25[OH]₂D). These cells also express vitamin D receptors (VDR; ref. 6), which translocate to the nucleus and bind vitamin D response elements to regulate gene expression, upon engagement by 1,25(OH)₂D. Thus, VDR target genes may be regulated locally at the site of a developing cancer, with important implications for cellular proliferation, apoptosis, and angiogenesis (45). Lending support to a role for vitamin D in tumorigenesis, vitamin D and its analogues can slow proliferation of pancreatic cancer cells both in cell culture and in xenograft mouse models (5, 6). Clinical studies are underway to exploit these effects of vitamin D and its analogues in patients with pancreatic cancer (46, 47).

Several epidemiologic approaches have investigated vitamin D and the risk of pancreatic cancer. Ecological studies have suggested an increased risk of pancreatic cancer with greater distance from the equator and lesser exposure to ultraviolet B radiation, both within countries and worldwide (8, 9). A study of vitamin D intake showed a reduced risk for pancreatic cancer in subjects with greater total and dietary vitamin D (10). Two studies have examined the risk of pancreatic cancer in relation to predicted vitamin D status, as calculated using a validated predictive model for plasma level of 25(OH)D (11, 12). This model integrated 5 independent predictors of plasma 25(OH)D, including race/ethnicity, geographic region, vitamin D intake, body mass index, and physical activity. In the larger of the 2 studies (12), the top quintile of predicted 25(OH)D was associated with a relative risk of 0.66 (95% CI, 0.46–0.96), compared with the bottom quintile.

In contrast to the above work, two initial studies examining serum 25(OH)D have generated inconsistent results (13, 14). A nested case–control study of male, Finnish smokers from Alpha-Tocopherol, Beta-Carotene (ATBC) Cancer Prevention Study reported a statistically significant elevated risk of pancreatic cancer among subjects in quintiles 3 and 5 of plasma 25(OH)D, compared with quintile 1 (13). In contrast, a nested analysis of men and women within the U.S. Prostate, Lung, Colorectal and Ovarian (PLCO) Screening trial showed no statistically significant difference in risk of pancreatic cancer by quintile of plasma 25(OH)D (14). Nevertheless, these studies were relatively small in size (200 cases in ATBC and 184 cases in PLCO).

A pooled analysis has also been conducted of 952 pancreatic cancer cases, including subjects previously examined from ATBC and PLCO and cases and controls added from several other prospective cohorts in United States and China (15). In contrast to the two previous studies, clinically defined cutoff points were used for 25(OH)D (<25, 25 to <37.5, 37.5 to <50, 50 to <75, 75 to <100, and ≥100 nmol/L), rather than quintiles, and the referent group consisted of subjects with 25(OH)D levels of 50 to <75 nmol/L. As a result of the post hoc pooled approach of this study, 25(OH)D was measured in multiple batches and in two separate laboratories. This pooled analysis found no association between circulating 25(OH)D and pancreatic cancer risk. However, in the small subset of subjects with 25(OH)D levels ≥100 nmol/L (39 cases and 30 controls), an increase in the risk of pancreatic cancer was observed (OR, 2.12; 95% CI, 1.23–3.64), when compared with those subjects with 25(OH)D levels between 50 and <75 nmol/L.

In light of these conflicting data, the current study has a number of important strengths. Its prospective design and exclusion of cases diagnosed within 2 years of blood draw greatly reduced the likelihood of bias due to reverse causation. The large number of pancreatic cancer cases and controls from diverse geographic locations across the United States allowed for the investigation of a wide spectrum of plasma 25(OH)D values, including subjects with 25(OH)D ≥100 nmol/L. All plasma assays for 25(OH)D were conducted in a single batch in a single laboratory and blinded to any identifiers of case and control status. Furthermore, coefficients of variance were low for repeated measures of quality control samples interspersed within participant samples. Pancreatic cancer cases were rigorously coded, with greater than 99% confirmed by review of medical records, state tumor registries, or death certificates. Data on covariates were also rigorously collected from all 5 cohorts, allowing for robust control of confounding and evaluation of effect modification.

Limitations of the current study include having only a single measure of 25(OH)D. However, we have previously shown a correlation of 0.70 for repeated measures of plasma 25(OH)D within individuals over time (48), suggesting that a single measurement is a reasonable proxy for long-term levels of 25(OH)D. We cannot rule out that our findings may be influenced in part by residual confounding. This may be more likely in an analysis involving multiple cohorts, as not all covariate data are collected identically. However, we evaluated multiple possible confounding covariates and did not observe meaningful changes in our risk estimates. Our sample consisted primarily of White subjects and we did not have adequate power to examine the association of 25(OH)D in non-Whites. Nevertheless, the use of 5 large cohorts with participants from across the United States allows for reasonable generalizability of our results to individuals of European descent. Further studies in non-Whites are warranted.

In this pooled analysis of 5 nested case–control studies, we observed a statistically significant inverse association between plasma 25(OH)D levels and the subsequent risk of pancreatic cancer. As compared with subjects who were vitamin D insufficient (<50 nmol/L, <20 ng/mL), those with relatively insufficient or sufficient levels (≥50 nmol/L, ≥ 20 ng/mL) experienced an approximate 30% lower risk for pancreatic cancer. In light of the high prevalence of vitamin D insufficiency in the population, further studies should examine whether increasing vitamin D levels impacts the incidence of this highly lethal malignancy.

The sponsors had no role in the design, data collection, data analysis, interpretation, or writing of the report. The authors had full access to all data in the study and had final responsibility for decision to submit for publication. No potential conflicts of interest were disclosed.

The authors thank the subjects of the Health Professionals Follow-up Study, Nurses' Health Study, Physicians' Health Study, Women's Health Initiative, and Women's Health Study for their participation and steadfast dedication. A full listing of WHI investigators is available (http://www.whiscience.org/publications/WHI_investigators_shortlist.pdf).

The NHS and HPFS are supported by the National Cancer Institute, NIH (grants no. P01 CA87969, P01 CA55075, P50 CA127003, and R01 CA124908). B.M. Wolpin, MD, MPH, is supported by NCI K07 CA140790, an American Society of Clinical Oncology Career Development award, and a Howard Hughes Medical Institute Early Career Physician-Scientist award. The PHS is supported by grants CA97193, CA34944, CA40360, HL26490, and HL34595 from the NIH (Bethesda, MD). The WHI program is funded by the National Heart, Lung, and Blood Institute, NIH, U.S. Department of Health and Human Services through contracts N01WH22110, 24152, 32100-2, 32105-6, 32108-9, 32111-13, 32115, 32118-32119, 32122, 42107-26, 42129-32, and 44221. The WHS is supported by grants CA047988, HL043851, and HL080467 from the NIH (Bethesda, MD).

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.