Critique of the U-shaped serum 25-hydroxyvitamin D level-disease response relation

Abstract

Previous work has suggested an optimal serum 25-hydroxyvitamin D [25(OH)D] level near 20-30 ng/mL, above which disease risk may increase. Although based primarily on a prostate cancer study in Nordic countries, examples include colorectal, esophageal, and pancreatic cancer; cardiovascular disease; and all-cause mortality rate. However, these studies apparently are not representative of the findings in the literature for these diseases or disease outcome in general. The prostate cancer study was from Nordic countries and used serum 25(OH)D levels from more than 15 years prior to cancer diagnosis for about half of the cases. Most studies of prediagnostic serum 25(OH)D find no significant correlation with risk of prostate cancer. Many risk-modifying factors for prostate cancer exist that observational studies generally do not include. The esophageal cancer data were from a region of China with high incidence of esophageal cancer. The pancreatic study was conducted on smokers in Finland. Both the esophageal and pancreatic studies are at odds with many ecological and observational studies in various countries. When all available studies for colorectal cancer, cardiovascular disease, and all-cause mortality rate are combined in preliminary meta-analyses, the best fits to the data are power laws, which show a monotonic decrease of hazard ratio with increasing logarithm of serum 25(OH)D. Thus, little support exists for the U-shaped serum 25(OH)D dose-disease response relation, and such studies should not be used in forming public health policies regarding vitamin D and ultraviolet-B irradiance.

Introduction

The journal literature embodies an ongoing debate about the existence of a U-shaped 25-hydroxyvitamin D [25(OH)D] level-disease response relation at moderate serum 25(OH)D levels.1–6 This concern was initiated by a report that found higher prostate cancer incidence in Nordic countries for those with prediagnostic serum 25(OH)D levels below or above 40–59 nmol/L (16–24 ng/mL).1 Two subsequent studies have been noted to show evidence of a U-shaped serum 25(OH)D level-disease relation, one on cardiovascular disease (CVD) incidence rates7 and one on all-cause mortality rates.6 Still other reports directly correlate disease incidence with serum 25(OH)D for pancreatic cancer in Finnish smokers9 esophageal carcinoma in Chinese men10 and esophageal squamous dysplasia in Chinese men and women.11

Although these few studies are at odds with the vast majority of ecological, observational and randomized controlled trials regarding solar ultraviolet-B (UVB) doses, serum 25(OH) D levels, and vitamin D supplementation, others have proposed explanations for the U-shaped relation. Vieth proposed that at high latitudes, higher summertime 25(OH)D levels are followed by sharper declines in 25(OH)D, causing inappropriately low 1-hydroxylase and high 24-hydroxylase levels, resulting in tissue 1,25-dihydroxyvitamin D below its ideal set point.12 This mechanism was proposed for pancreatic and prostate cancer, but no explanation was given as why, if applicable to pancreatic and prostate cancer, it would not also apply to other cancers, none of which show a U-shaped relation with respect to vitamin D. Recently, the Tuohimaa group has proposed that 25(OH)D is an agonistic vitamin D receptor ligand, adding to the effect of 1,25-dihydroxyvitamin D.13 Although both proposed mechanisms are plausible, these effects may not have any notable influence if they are not real.

It is my position that these reports are not representative of findings for each disease outcome and are probably due either to statistical fluctuations associated with low numbers of cases or confounding factors not considered (the ecological fallacy). This report examines each report of a U-shaped serum 25(OH) D level-disease response curve in terms of possible reasons why the report should not serve as the basis of vitamin D health policies.

Prostate Cancer

The Tuohimaa et al.1 report found a higher risk of prostate cancer for those with serum 25(OH)D levels >32 ng/mL (Table 1). Sixty-seven men were in that range, 57 of whom were from Norway. More important, the lag time between serum 25(OH) D measurement and prostate cancer diagnosis was >10 years for 52 of these men. The OR for those 52 was 1.8 (95% confidence interval [CI], 1.1–2.9), whereas the OR for those with <10-year lag time was 1.4 (95% CI, 0.7–2.7). The long lag time is problematic because it is not clear that serum 25(OH)D levels at the time when prostate cancer might have been initiated or progressed were as measured. A randomized, controlled trial of calcium and vitamin D for postmenopausal women found a very significant benefit of vitamin D for cancer risk reduction between the ends of the first and fourth years.16 A recent study on prediagnostic serum 25(OH)D levels and risk of lymphoid cancers in Hawaii found a beneficial effect for up to a 7-year lag but a nonsignificant direct correlation with 25(OH)D for longer periods.17 Thus, there is reason to doubt that the long lag time results are meaningful.

Another reason to doubt the Nordic finding is that it has not been replicated elsewhere—even in Nordic countries. A recent review18 found that only one of seven observational studies reported a statistically significant correlation between serum 25(OH)D and risk of prostate cancer, and that was from Finland.19 A more recent study from Finland fo8und no association between serum levels of 25-hydroxyvitamin D and the subsequent risk of prostate cancer in Finnish men.14 In this study, the OR for prostate cancer for the highest quartile (>24 ng/mL) was 0.89 (95% CI, 0.49–1.62), which was higher than for the third quartile (19–24 ng/mL), whose OR was 0.59 (95% CI, 0.31–1.11), but the trend was not statistically significant (P_trend = 0.97) (Table 1). However, this study was mentioned as supporting the U-shaped serum 25(OH)D level-disease relation in ref. 4. A recent meta-analysis of studies on prostate cancer incidence resulted in a summary OR of 1.03 (95% CI, 0.96–1.11) associated with an increase of 25(OH)D by 10 ng/mL (p = 0.36).20

Considering the geographic variation of prostate cancer mortality rates in the US is also worthwhile. As shown in the Atlas of Cancer Mortality in the United States,21 the mortality rate pattern for prostate cancer differs from the well-documented vitamin D-sensitive cancers22–25 in that the highest rates for prostate cancer are in the northwest, whereas the highest rates for the vitamin D-sensitive cancers are in the northeast. Those cancers have been shown inversely correlated with July solar UVB doses.26 On the basis of the Tuohimaa et al.1 paper and the finding that prevalence of multiple sclerosis had a significant increase with latitude, as well as that this dependence was indicative of low solar UVB and serum 25(OH)D in winter,27 an ecologic study was conducted using latitude and summertime solar UVB.28 This study was presented in support of the Tuohimaa et al.1 paper. Later, it was thought that the latitudinal distribution was indicative of a vitamin D-sensitive viral infection in men.29 This hypothesis was supported in an ecological study finding that the annual prostate cancer incidence and mortality rates for white men correlated most strongly with UVB exposure levels in the fall and winter.30 More recent work noted that the mortality rate distribution was highly correlated with ancestry by county,31 indicating that ethnic background was an important risk-modifying factor for prostate cancer. This observation inspired another ecological study, this time involving apolipoprotein E ε4 (ApoE4) allele prevalence, diet, and per capita gross domestic product for 122 countries.32 A significant correlation was found for prostate cancer incidence and mortality rate with respect to ApoE4 allele prevalence, low supply of cereals, and per capita gross domestic product. ApoE4 increases cholesterol levels in the blood, an important risk factor for prostate cancer. Although the viral risk factor hypothesis does not seem to explain the geographical variation of prostate cancer mortality rates in the US, support for this hypothesis was published recently with the finding of the xenotropic murine leukemia virus-related virus in many prostate tumors.34 Dietary factors also play an important role in the etiology of prostate cancer.34

In summary, identifying the risk-modifying factors for prostate cancer incidence and mortality rates is difficult because many such factors exist and, in general, no single study uses them all. Ecological studies based solely on serum 25(OH)D levels are, thus, subject to the ecological fallacy.

Pancreatic Cancer

As noted, an increased risk of pancreatic cancer was found with increasing serum 25(OH)D for smokers in Finland.5 In a more recent study by the same author, “vitamin D concentrations were not associated with pancreatic cancer overall [highest versus lowest quintile, >82.3 versus <18.4 ng/mL: OR, 1.45 (95% CI, 0.66–3.15; P_trend = 0.49)]. However, positive associations were observed among subjects with low estimated annual residential solar UBV exposure, but not among those with moderate to high annual exposure (P_interaction = 0.015). We did not confirm the previous strong positive association between 25(OH)D and pancreatic cancer; however, the increased risk among participants with low residential UVB exposure is similar.”15 The data are presented in Table 1.

Again, looking at other studies is worthwhile. The results from two Harvard cohort studies are as follows: “Compared with participants in the lowest category of total vitamin D intake (<150 IU/d), pooled multivariate relative risks for pancreatic cancer were 0.78 (95% CI, 0.59–1.01) for 150 to 299 IU/d, 0.57 (95% CI, 0.40–0.83) for 300 to 449 IU/d, 0.56 (95% CI, 0.36–0.87) for 450 to 599 IU/d, and 0.59 (95% CI, 0.40–0.88) for ≥600 IU/d (P(trend) = 0.01). These associations may be stronger in men than women.”35

Ecological studies provide strong support for solar ultraviolet-B irradiance's reducing the risk of pancreatic cancer in Japan,36–38 the US,22,23,39 Spain40 and Australia.41 Although not all the ecological studies accounted for confounding factors, the fact that similar results were found in many different countries satisfies an important criterion for causality in a biological system as laid down by Hill.42,43

The best hypothesis for why Finnish smokers have higher incidence of pancreatic cancer with higher serum 25(OH)D levels is that the vitamin D-pancreatic cancer relation is different for smokers and nonsmokers.

Esophageal Carcinoma and Squamous Dysplasia

Two reports from China correlated esophageal carcinoma10 and squamous dysplasia11 prevalence with higher serum 25(OH)D levels (Table 1). The region where these studies were conducted, Linxian, China, has a population at high risk for developing esophageal squamous cell carcinoma.44 There has been a concerted effort to find a viral cause for esophageal cancer in this region, which has been unsuccessful to date.45 In an ecological study of cancer in China, esophageal and several other cancers had mortality rates directly correlated with latitude.46 Several other ecological studies inversely correlated solar UVB with esophageal cancer.22–25,39,40 Even though the high rate of esophageal cancer in Linxian cannot be readily explained, the finding should not be used as support for a U-shaped serum 25(OH)D level-disease response relation.

CVD

The first study showing an effect of serum 25(OH)D on risk of CVD8 was also singled out in support of the U-shaped 25(OH) D level-disease response relation. The tabulated data therein did not show a U-shaped serum 25(OH)D level-disease relation (see Table 1). Again, a meta-analysis should be performed to increase the number of cases in the study. When data in refs. 8 and 47–51 are graphed by overlaying the data and adjusting the hazard ratios (HRs) to align the data, noting that the HRs in the reports are sensitive to the mean serum 25(OH)D values, the power law and logarithmic fit to the data show a monotonically decreasing HR out to 43 ng/mL. A summary of the regression fit to the data is given in Table 2. Thus, the CVD data do not support the U-shaped serum 25(OH)D level-disease relation. A meta-analysis of cardiometabolic disorders with respect to high versus low serum 25(OH) D levels has just been published, finding very strong inverse correlations between the ratio of highest to lowest serum 25(OH)D quantile and CVD, diabetes and metabolic syndrome.52

All-Cause Mortality Rate

The report by Melamed et al. on all-cause mortality rate8 is singled out as additional support for the U-shaped 25(OH) D level-disease response relation (see Table 1). The data by quartiles are the fully adjusted mortality rate data from Table S4. The data by quintiles are from Figure S2. For both the quartile and quintile analyses, only the data for the lowest quantile are statistically significantly different from 1.0. Therefore, the suggestion that this data set supports the U-shaped relation is a stretch. Even if it did, singling out one study in a field is not a correct approach. A more proper method would be to look at all similar studies and do a meta-analysis of the data. At least three other articles exist that can be used in this regard.47–49 When the data from all four reports are graphed by overlaying the data and adjusting the hazard ratios (HRs) as for CVD (reviewed in ref. 9), the power law and logarithmic fits to the data decrease monotonically out to 43 ng/mL. There is one value from ref. 49 at 43 ng/mL that counters the value in Melamed et al.8 Thus, the observational data for all-cause mortality rate with respect to prediagnostic serum 25(OH)D do not support the U-shaped serum 25(OH)D level-disease response relation.

Vitamin D Recommendations

Several recent findings have suggested that the optimal serum 25(OH)D level is at least 40–60 ng/mL.14,53 Efforts to counter the positive findings regarding vitamin D, such as that by the International Agency for Research on Cancer,5 appear to be largely biased and misguided.54 Studies of the health benefits of raising serum 25(OH)D levels to 40–45 ng/mL at the population level on the basis of serum 25(OH)D level-disease outcome responses estimate about 15% reductions in all-cause mortality rate and reduction in the economic burden of disease treatment of 10%–20%.55,56 Another study estimated that if all those in the US were to double their solar ultraviolet-B irradiance to raise their serum 25(OH)D level to 45 ng/mL, the net result could be as many as 400,000 reduced deaths compared with only 11,000 increased deaths from melanoma and other skin cancer.57

Summary and Conclusion

The data used to suggest that higher serum 25(OH)D levels are associated with adverse health outcomes for several types of cancer, CVD, and all-cause mortality rate are inadequate representations of findings in the literature or are essentially taken out of context. When other findings are combined with those findings, one finds monotonically reducing disease rates with increasing vitamin D indices. Thus, these studies should not be used as support for an upper limit for serum 25(OH) D levels.

Figures and Tables

Table 1 Previous reports showing a U-shaped serum 25(OH)D level dose-disease response relation

	Quantile*
Disease outcome	1	2	3	4	5	Reference
Colorectal cancer	11.5	23	29.5	37	50	7
	1.00	0.48	0.25	0.21	0.73
Prostate Cancer (OR)	<7.6	8–15.6	16–23.6	24–31.6	>32	1
	2.6 (1.1–6.2)	1.2 (0.8–1.9)	1.0	1.1 (0.7–2.8)	1.4 (0.7–2.7)
	<14.8	14.8–18.8	18.8–24.0	>24.0		16
	1.00	0.88 (0.48–1.61)	0.59 (0.31–1.11)	0.89 (0.49–1.62)
Pancreatic cancer	<12.8	12.8–16.4	16.4–20.4	20.4–26.2	>26.2	10
	1.00	1.30 (0.70–2.40)	2.12 (1.15–3.90)	1.50 (0.81–2.76)	2.92 (1.56–5.18)
	<18.4	18.4–24.1	24.1–27.8	27.8–32.9	>32.9	17
	1.00	0.97 (0.47–1.98)	0.86 (0.40–1.84)	0.84 (0.39–1.80)	1.45 (0.66–3.19)
Esophageal cancer	<7.8	7.8–12.8	12.8–19.5	>19.5		11
	1.00	0.87 (0.64–1.17)	0.85 (0.63–1.16)	1.30 (0.97–1.73)
Esophageal squamous dysplasia	<9.6	9.6–13.2	13.2–19.3	>19.3		12
	1.00	1.43 (1.00–2.06)	1.47 (1.04–2.12)	1.86 (1.35–2.62)
CVD	<10	10–15	>15			8
	2.63 (1.57–4.38)	1.80 (1.05–3.08)	1.00
All-cause mortality rate	<17.8	17.8–24.3	24.4–32.1	>32.1		9
	1.26 (1.08–1.46)	1.06 (0.89–1.24)	(0.93 (0.79–1.10)	1.00

* [25(OH)D, OR] (ng/mL; OR, 95% CI).

Table 2 Summary of data showing that the U-shaped serum 25(OH)D level-disease outcome does not apply for cardiovascular disease and all-cause mortality rates

Disease outcome	Hazard or rate ratio for serum 25(OH)D levels					Reference
	7*	10*	20*	30*	40*
CVD (hazard ratio)	2.21	1.97	1.49	1.24	1.06	9, 49–53
All-cause mortality rate ratio	1.93	1.90	1.38	1.15	1.00	16

* ng/mL.

Acknowledgements

I receive funding from the UV Foundation (McLean, VA), the Sunlight Research Forum (Veldhoven), and Bio-Tech-Pharmacal (Fayetteville, AR) and have previously received funding from the Vitamin D Society (Canada).