In our study, higher 25(OH)D score was associated with a significantly lower risk of pancreatic cancer. This inverse association was consistent across gender and strata of other covariates. There are several mechanisms through which vitamin D may affect pancreatic cancer risk. Pancreatic cancer cells express VDRs (Colston et al, 1997
) and 25(OH)D3
-1a-hydroxylase (Schwartz et al, 2004
), which metabolises 25(OH)D to the active 1,25(OH)2
D vitamin D form. Binding of VDRs by 1,25(OH)2
D leads to increased differentiation and apoptosis as well as reduced proliferation, invasiveness, angiogenesis, and metastasis (Giovannucci, 2005
); and experimental studies have shown that 1,25(OH)2
D analogs inhibit growth of pancreatic cancer cells in vitro
and in vivo
(Colston et al, 1997
; Schwartz et al, 2004
). In addition, pancreatic islet cells express VDRs and 25(OH)D3
-1a-hydroxylase, and in vitro
and in vivo
evidence supports that vitamin D deficiency impairs endocrine pancreatic function (Pittas et al, 2007
). Observational studies have shown that vitamin D status is inversely associated with development of type 2 diabetes or metabolic syndrome (Mattila et al, 2007
; Pittas et al, 2007
; Forouhi et al, 2008
). As diabetes, hyperglycaemia, and insulin resistance have been linked to pancreatic cancer development, vitamin D may act to decrease pancreatic cancer risk by improving glucose metabolism and reducing insulin resistance.
Epidemiological studies have used four approaches to examine the association of pancreatic cancer with vitamin D status, and their results have been inconsistent. Sunlight exposure, a major source of vitamin D in humans, was inversely correlated with pancreatic cancer in ecological studies conducted in North America, Europe, and Japan (Mizoue, 2004
; Boscoe and Schymura, 2006
; Grant, 2007
). Higher dietary vitamin D intake as well as higher total vitamin D intake (from foods and supplements) was related to lower risk in the NHS and the HPFS (Skinner et al, 2006
) but not in a cohort of male Finnish smokers enrolled in the Alpha-Tocopherol, Beta-Carotene (ATBC) Cancer Prevention Study (Stolzenberg-Solomon et al, 2002
). Directly measured circulating 25(OH)D, which reflects vitamin D status from both sun and dietary sources, was positively associated with pancreatic cancer risk in male Finnish smokers (fifth vs
first quintile, RR=2.92, 95% CI=1.56–5.48) (Stolzenberg-Solomon et al, 2006
) but not in a cohort of the Prostate, Lung, Colorectal, and Ovarian (PLCO) Screening Trial (fifth vs
first quintile, RR=1.45, 95% CI=0.66–3.15) (Stolzenberg-Solomon et al, 2009
). In contrast, the vitamin D prediction score, which also accounts for sun exposure (by using residential state and physical activity as surrogates) and vitamin D intake, was inversely associated with pancreatic cancer risk in the present two US prospective cohort studies (NHS and HPFS) (Giovannucci et al (2006)
and this study).
However, participants in the ATBC study might not be comparable with those in the American studies, as they were all current smokers and lived at higher latitude. In this study, we observed a stronger inverse association with vitamin D among nonsmokers and among those living in southern states. The PLCO study also found a significant interaction by geographic region (Pinteraction
=0.015): they reported a positive association of risk with plasma vitamin D concentrations among those living in northern latitudes, but no association was observed among those living in southern latitudes (Stolzenberg-Solomon et al, 2009
). Another difference in the ATBC study population concerns dietary pattern: a major dietary source of vitamin D in the Finns does not tend to be from fortified dairy products or breakfast cereal as American populations but from vitamin D-rich fish that may contain some pancreatic carcinogen such as organochlorine compounds (Stolzenberg-Solomon et al, 2006
). In addition, both the ATBC study and the PLCO study were based on one measurement of 25(OH)D in blood, which most likely reflects recent exposure to sources of vitamin D rather than long-term average vitamin D level; whereas the 25(OH)D prediction score would track well over time because factors that influence the score, such as race and residential region, are immutable or relatively stable. As evidence of this supposition, in a previous study, the correlation between two direct plasma measurements 4 years apart was 0.70, whereas the correlation between the two 25(OH)D scores was 0.83 (Giovannucci et al, 2006
Retinol has been hypothesised to counteract the cancer prevention effects of vitamin D, possibly acting through competition with vitamin D for the retinoid X receptor (Giovannucci, 2005
). In this study, we observed that the inverse association between vitamin D status and pancreatic cancer was more pronounced among those with lower intake of retinol; we also observed a stronger inverse association among those who did not use multivitamins or supplemental vitamin D. We have previously reported an increased pancreatic cancer risk associated with multivitamin use (Skinner et al, 2004
), so it is possible that some factor in multivitamin supplements other than vitamin D, potentially retinol, antagonises the protective effect of vitamin D or increases risk independently.
One concern of our approach is that the 25(OH)D score may act as a surrogate for other potential risk factors, such as BMI or physical activity. A higher BMI and less physical activity have previously been associated with increased pancreatic cancer risk in our cohorts (Michaud et al, 2001
). However, our results for predicted 25(OH)D did not change when adjusted for BMI or physical activity, which suggests that total vitamin status, rather than simply low BMI or physical activity, is driving the significant inverse association. We are also aware that the prediction score has been developed in men; however, given the prospective design, any misclassification of the 25(OH)D score among women tended to be non-differential, and therefore would only bias the results towards the null. In addition, when we stratified the analyses by gender, similar RRs were obtained for men and women, with no significant interactions. To rule out the possibility that vitamin D status might be changed by preclinical pancreatic cancer at baseline, we excluded the first 2 years of follow-up for all participants in sensitivity analyses. The results were unchanged. Residual confounding by smoking was not likely because a stronger association was observed among never smokers. Residual confounding by other measured factors might be of minor importance in this study, as our age/sex-adjusted models and multivariate models yielded very similar results.
The strengths of our study include its prospective design, use of a validated prediction score taking into account both diet and sun exposure, comprehensive information on many potential confounders, a large sample size that allowed us to stratify the data by potential effect modifiers, and 20 years of follow-up with a high follow-up rate. Because the participants were health professionals, the accuracy of self-reported data is likely to be high; moreover, any misclassification of vitamin D status is likely to be random and would therefore have attenuated rather than exaggerated a true association.
In conclusion, our data suggest that higher 25(OH)D levels may significantly decrease the risk of pancreatic cancer. Given the growing epidemic of vitamin D insufficiency in the US population (Ginde et al, 2009
), more epidemiological research is needed, particularly prospective studies with repeated measures of plasma 25(OH)D. If the association between vitamin D and pancreatic cancer is causal, many with low vitamin D levels might benefit from increased vitamin D status for pancreatic cancer risk reduction.