25-Hydroxyvitamin D status, vitamin D intake, and skin cancer risk: a systematic review and dose–response meta-analysis of prospective studies

Sun exposure is a major environmental risk factor for skin cancers and is also an important source of vitamin D. However, while experimental evidence suggests that vitamin D may have a protective effect on skin cancer risk, epidemiologic studies investigating the influence of 25-hydroxyvitamin D (25(OH)D) level and/or vitamin D intake on skin cancer risk are conflicting. A systematic review and dose–response meta-analyses of prospective studies was conducted to clarify these associations. Relevant studies were identified by searching the PubMed database up to 30th August 2019. Random effects dose–response meta-analyses were used to estimate summary relative risks (SRRs) and 95% confidence intervals (CIs). Overall, thirteen prospective studies were included. Circulating level of 25(OH)D was associated with higher risks of melanoma (SRR (95% CI) per 30 nmol = 1.42 (1.17–1.72)) and keratinocyte cancer (KC) (SRR (95% CI) per 30 nmol/L = 1.30 (1.13–1.49)). The SRR (95% CI) per 30 nmol/L increase in 25(OH) D level was 1.41 (1.19–1.67), and 1.57 (0.64–3.86), for basal cell carcinomas (BCCs) and squamous cell carcinomas (SCCs), respectively. However, while we found that vitamin D intake (from diet, supplemental and total) was not associated with risks of melanoma and SCC, vitamin D intake was associated with slightly increased BCC risk, albeit with no heterogeneity across skin cancer type. This meta-analysis suggests positive associations between circulating 25(OH)D level and risk of melanoma and KC, however, this finding is most likely confounded by sun exposure. We found no associations between vitamin D intake skin cancers, except positive associations with BCC risk.

Search strategy for study identification. A systematic search using several databases, such as PubMed, Embase, CAB Abstracts, ISI Web of Science, up to September 20th 2018, was performed by several reviews for eligible studies as part of the Continuous Update Project of the World Cancer Research Fund (WCRF-CUP) at Imperial College London. The protocol used for the search strategy can be accessed at https ://www.wcrf.org/ sites /defau lt/files /skin-cance r-proto col.pdf. However, since all the relevant studies were initially identified by the PubMed search, we have changed the protocol and searches were updated to 30th August 2019 using the same search strategy only in the PubMed database. We additionally conducted the search in Google Scholar using the same specific key terms, and no new relevant articles were identified compared to PubMed search. In addition, we searched the reference lists of the relevant publications, reviews and meta-analysis for further studies.
Selection criteria. We included in this meta-analysis prospective cohort or nested case-control studies investigating the association between either 25(OH)D level or vitamin D intake (dietary, supplemental, and total) and risk of melanoma or KC. Estimates of the relative risk (RR) (such as hazard ratios, risk ratios or odds ratios) with the 95% confidence intervals (CIs) had to be available in the publication, and for the dose-response analysis, a quantitative measure of intake or level of 25(OH)D, the total number of cases and person-years or non-cases had to be available in the publication. If several articles were published using the same study population on the topic, the one with the largest number of cases was selected 34,36,39 . Case-control studies, ecological studies, case reports, reviews and editorials were excluded from the meta-analysis. Studies which focused on the relationship between vitamin D exposure and survival from melanoma, recurrence or prognostic factor for melanoma (e.g. Tumor thickness, ulceration) were also excluded. The present study focused only on vitamin D exposure and risk of primary skin cancer.
Data extraction. For each relevant study included in this meta-analysis, the following information was extracted by two reviewers: first author's last name, publication year, country where the study was conducted, study name, study design, follow-up period, sex, age, number of cases, case ascertainment, exposure assessment, outcome, comparison, RRs and 95% CIs and adjustment factors (Table S1 and S2).

Risk of bias.
To assess the risk of bias and quality of the included studies, we used the Cochrane risk of bias tool ROBINS-I, which grades studies on a scale from critical risk of bias to low risk of bias due to confounding, selection, outcome and exposure assessment, classification as well as missing data 40 .
Scientific RepoRtS | (2020) 10:13151 | https://doi.org/10.1038/s41598-020-70078-y www.nature.com/scientificreports/ Statistical methods. Dose-response and highest versus lowest meta-analyses were conducted to summarize the associations between vitamin D status or intake and risk of skin cancer by using random-effects models that consider both within study and between-study variation 41 . We used the method described by Greenland and Longnecker 42 for the dose-response analysis to compute the trend from the correlated RRs and 95% CIs across categories of exposure when not provided in the publications. This method required the distribution of person-years, cases, median exposure, RRs, and 95% CIs for at least three categories. For studies that did not provide the distribution of person-years or the number of cases per categories, we estimated the distribution by dividing total person-years or case by the number of categories. The median of the exposure in each category was used if provided in the articles, and if not reported, the midpoint of the upper and lower boundaries was estimated as range in each category. When the highest and lowest categories were open-ended or had extreme upper or lower values, we used the width of the adjacent interval to estimate the upper and lower boundaries for the category. For studies that reported plasma or serum 25(OH)D level in ng/ml 24,26,32 , we converted the data to nmol/L by multiplying the concentration in ng/ml by 2.5, and for studies of vitamin D intake 32,43 , data in µg/day were divided by 0.025 to convert the data to IU/day. For studies that reported results separately for BCC and SCC 24,25 , we combined the results by the Hamling procedure to obtain an overall estimate for KC 44 . The dose-response meta-analyses were conducted in increments of 30 nmol/L for 25(OH)D level 45 and 100 IU/day for vitamin D intake 46 based on the previous published studies 45,46 .
To explore the shape of the association between vitamin D levels and/or dietary vitamin D and incidence of skin cancer, we conducted non-linear dose-response meta-analysis using restricted cubic splines with 3 knots at 10th, 50th, and 90th percentiles of the distribution 47,48 . We also conducted subgroup and meta-regression analyses to investigate potential sources of heterogeneity by study characteristics such as sex, duration of followup, geographic location, risk of bias and adjustment for confounding factors. Statistical heterogeneity between studies was assessed by the Cochran Q test and the I 2 statistic 49 . Small-study effects, such as publication bias, were assessed by visual inspection of funnel plot and with Egger's test 50, and the results were considered to indicate potential small-study bias when P values were < 0.10. Sensitivity analyses excluding one study at a time were conducted to clarify whether the results were simply due to one large study or a study with an extreme result. Stata version 14 software (Stata Corp, College Station, TX) was used for the statistical analyses.

Subgroup analyses. Associations between vitamin D status and melanoma risk persisted in most subgroup
analyses including analyses by sex, duration of follow-up, number of cases, geographical location, risk of bias and adjustment for confounding factors, including sun exposure. While there was no evidence of between-subgroup heterogeneity, heterogeneity within subgroup analyses was still present (Table 1). In sensitivity analyses exclud-   Figure S3A). In stratified analyses according to sex, the association between 25(OH)D level and KC risk was positive among women, whereas an inverse association was observed among men with heterogeneity detected across sex (P heterogeneity = 0.01); however only one study was available for men. There was heterogeneity between subgroup analyses of KC including those by geographical location with stronger associations among the European studies (P heterogeneity = 0.01), by number of cases with stronger associations for studies with more than 500 cases (P heterogeneity = 0.04), and by risk of bias with higher association for studies with moderate compared to those with serious risk of bias (P heterogeneity = 0.02). In addition, there was heterogeneity by adjustment for hair color (P heterogeneity = 0.02), skin color (P heterogeneity = 0.01), adjustment for physical activity (P heterogeneity = 0.05), smoking status (P heterogeneity = 0.05) and BMI (P heterogeneity = 0.05). However, in the remaining subgroup analyses, Table 1. Subgroup analyses of circulating 25-hydroxyvitamin D level and skin cancer risk. RR, summary relative risk; CI, confidence interval. I 2 (%) is a measure of the proportion of the heterogeneity attributed to between study variation rather than due to chance. I 2 -values of 25%, 50% and 75% indicates low, moderate and high between study heterogeneity, respectively. a P-value for heterogeneity among the studies within each cancer type. b P-value for between subgroup or category heterogeneity generated from meta-regression analysis.  Figure  S3B). Subgroup analyses of vitamin D intakes and risks of melanoma and KC could not be done because of the limited number of studies.

Discussion
This meta-analysis of thirteen prospective studies suggests that vitamin D status was associated with greater risks of melanoma and KC. Linear dose-response meta-analysis revealed that each 30 nmol /L increment in 25(OH) D was respectively associated with 42%, 30% and 41% increase risks of developing melanoma, KC and BCC. In non-linear dose-response meta-analysis, the association was significant for KC, with higher relative risk observed at a level of approximately 60 nmol/L of 25(OH) D, and a weaker association beyond this level. However, while dietary, supplemental and total intakes of vitamin D were not associated with melanoma and SCC risks, these latter were slightly associated with higher risk of BCC, although with no heterogeneity across skin cancer type. Similar findings were observed in high versus low meta-analysis. Previous research has suggested that high plasma/serum 25(OH)D concentration could have a protective effect against several chronic diseases 54,55 , including cancer risk 46,56,57 . Animal and human studies have indeed demonstrated that vitamin D status may influence some cancers, such as colon, stomach, kidney as well as skin through down regulation of cell growth 58,59 and modulation of the immune system 60 . Vitamin D is mainly synthesized in the skin after exposure to UV radiation, and it has been estimated 80-90% of vitamin D are from sun exposure and the remainder amount is from the diet and supplements intake 61 . However, although UV radiation is recognized as a major skin carcinogen, the same spectrum of UV exposure which can lead to DNA damage in skin cells also induce vitamin D synthesis 62,63 . This latter has been suggested to have anticancer properties in normal and skin cancer cells [64][65][66] . Consequently, this has led to a strong debate among scientific and public communities about the balance between the potential benefits and risks of UV-induced vitamin D production and skin cancer prevention. A meta-analysis, based on 4 case-control or cohort studies published between 2009 and 2013, suggested no statistically significant association between serum 25-(OH)D levels and the risk of melanoma, while a positive association was found with KC risk 30 . Unfortunately, this previous meta-analysis did not conduct dose-response meta-analyses or subgroup analyses by confounding factors that are critical to consider due to the conflicting results among published studies to date. The dose-response curve for vitamin D status and skin cancer provides insights regarding an optimal value for vitamin D status, which can be identified for skin cancer prevention, and for proposing an optimal level.
To our knowledge, this is the first meta-analysis to explore the linear and non-linear dose-response association between vitamin D status and the risk skin cancer, including only prospective studies, and the first to investigate subgroup analyses of the associations. Additionally, this is an updated meta-analysis to investigate the highest versus lowest serum or intake of vitamin D in relation to skin cancer 30 . In the present meta-analysis, we found that high vitamin D status was associated with greater risks of melanoma and KC. We found some evidence of a non-linear dose-response association between circulating 25(OH)D levels and KC risk, with the strongest summary relative risk observed at a level of approximately 60 nmol/L of 25(OH) D, which may be considered as insufficient vitamin D status. While levels of 25(OH)D below 50 nmol/L have suggested to be associated with severe vitamin D deficiency, levels between 50 and 74 nmol/L have been described as moderate vitamin D deficiency or insufficiency, and levels between 75-99 are considered as sufficient. Serum 25(OH)D levels between 100 and 150 nmol/L are indicated as adequate and healthy 67,68 . However, consistent with our findings, a recent prospective study suggested that higher levels of 25(OH)D was significantly associated with a higher incidence of Table 2. Summary results of vitamin D exposure and skin cancer risk, dose-response and high vs low and meta-analysis. RR summary relative risk; CI confidence interval. I 2 (%) is a measure of the proportion of the heterogeneity attributed to between study variation rather than due to chance. I 2 -values of 25%, 50% and 75% indicates low, moderate and high between study heterogeneity, respectively. a P-value for heterogeneity among the studies within each cancer type. www.nature.com/scientificreports/ skin cancer, with the hazard ratio seemed to peak around a vitamin D status of 80 nmol/L in the non-linear trend analysis 33 . Because of the limited number of studies included in our non-linear dose-response meta-analyses, mainly due to the missing information on the range of vitamin D levels, further studies should aim to clarify the shape of the dose-response relationship. The positive association between 25(OH)D levels and skin cancer risk is likely due to increased UV exposure causing both higher vitamin D levels and skin cancer risk. The current findings lend to support previous metaanalysis, published in 2014, that estimated a SRR of 1.64 (95% CI: 1.02-2.65) for the relationship between high serum 25(OH) D levels and KC risk, and of 1.46 (95% CI: 0.60-3.53) for the association with melanoma risk 30 . UV exposure induces both DNA lesion and immune suppression through production of reactive oxygen species www.nature.com/scientificreports/ which can alter the cellular redox equilibrium leading to premature skin aging and lipid peroxidation 69,70 . DNA damage caused by both UVB and UVA, together with UV-induced immune suppression contributes to the development of skin cancer 71 . Although, UVA and UVB are both implicated in skin cancer development 72 , studies reported that UVA plays a larger role in melanoma, whereas keratinocyte cancers are mainly related to UVB exposure 73 , which known to induce vitamin D production. Several studies suggested that UVB exposure induces genetic alterations, which in turn promote skin carcinogenesis [74][75][76] . UVB is mainly absorbed within the epidermis and upper dermis and is associated with skin burning and cause damage to keratinocytes in vitro 77 . However, while UVB radiation contributes to increase skin cancer risk, ecological studies reported an inverse correlation between solar UVB doses and other cancer risk [78][79][80][81] . Another hypothesis to explain the positive association between vitamin D status and skin cancer risk may be attributed to genetic factors. Several studies have addressed the issue of whether single nucleotide polymorphisms (SNPs) of the vitamin D receptor (VDR) gene are associated with the risk of developing different types of cancer 82,83 , and several VDR variants have been investigated in relation to skin cancer risk 84,85 , with three meta-analyses supporting a positive association, particularly with FokI and BsmI [86][87][88] . Since vitamin D exerts a function of engagement of its receptor VDR, it is likely that some SNPs of the VDR gene affect the ability of interacting with its ligand, which ultimately would lead to different levels of the biologic activity of vitamin D and increase skin cancer risk. Additional research is required to confirm the potential role of VDR gene in skin cancer incidence and to explore their interaction with sun exposure. A recent study based on the UK Biobank data, using a methodically robust Mendelian randomization, found no evidence of a causal association between genetic determinants of vitamin D concentrations and risk of melanoma 89 . However, while a Mendelian randomization study found no evidence for a causal association between genetically predicted vitamin D concentration and overall cancer risk 90, another one reported a weak evidence for linear causal associations between genetic determinants of circulating vitamin D levels and risk of several cancers cancers 91 . Given the contrasting and confounding effects of sun exposure and others factors such as pigmentary trait and dietary vitamin D on vitamin D status, it is, to date, difficult to examine an independent influence of vitamin D status on skin cancer risk. In our subgroup analysis, the positive association between circulating 25(OH)D levels www.nature.com/scientificreports/ and risk of melanoma was observed across all subgroup analyses. For KC, the positive association was stronger among studies conducted in Europe, in those with more than 500 cases and with moderate risk bias and in those adjusted for hair color, whereas association were not significant among studies adjusted for skin color, physical activity and BMI. In addition, our findings suggested a moderate and a substantial heterogeneity among studies for melanoma and KC, respectively. Part of heterogeneity for melanoma appeared to be attributed to study duration of follow-up; number of cases; risk of bias and some adjustment for confounders. This heterogeneity could also be related to melanoma localization or histologic type. Previous studies suggested the model of heterogeneity for melanoma by showing distinct etiologies for different body sites and tumor types 92,93 . However, this aspect has been largely overlooked in previous studies. Although there was a high level of heterogeneity across studies investigating vitamin D status in relation to SCC, no heterogeneity was detected for BCC, suggesting that difference in KC type may lead to important study heterogeneity. Nevertheless, anti-proliferative and pro-apoptotic properties of vitamin D have indeed been demonstrated in animal and in vitro studies 94,95 , and 1,25(OH)D has been shown to regulate multiple signaling pathways involved in differentiation, inflammation, invasion, angiogenesis and metastasis 17 . Recently, a pooled analysis of 25 studies showed that lower vitamin D levels were associated with higher Breslow thickness and mortality rates in patients with melanoma 96 . Several studies suggested that higher vitamin D levels may confer better prognosis from melanoma including Breslow thickness 30,97 . UV exposure has also been reported to be associated with a better prognosis and survival rate in several cancers sites 98, and some authors proposed that these may include melanoma 99,100 , which could be explained by UV exposure-induced high serum levels of vitamin D and lead to a better prognosis. Holiday sun exposure before melanoma diagnosis has been reported to be associated with lower thickness and the exposure after melanoma diagnosis was also associated with reduce melanoma recurrence 101 . Furthermore, increasing evidence has been suggested that circulating levels of vitamin D may play a protective role against several types of cancer such a bladder 102 and colorectal 103 . Also, a recent pooled analysis of two randomized controlled trials (RCTs) and a prospective cohort found that higher 25(OH)D levels were inversely associated with breast cancer risk with levels ≥ 60 ng/ml ((≥ 150 nmol/L) being most protective 104 .
Regarding intake of vitamin D, our finding suggested that intake of vitamin D from diet or supplement was not associated with melanoma and SCC risk, whereas there was a weak positive association with BCC risk, although with no heterogeneity across skin cancer type. Two studies from the NHS and HPFS were indeed included in these analyses. As mentioned by the authors, the positive association was mainly attributed to the vitamin D-rich foods such as fish and cereal which were associated with BCC after adjustment for several known risk factor of www.nature.com/scientificreports/ skin cancer. It could be argue that arsenic present in fish and breakfast cereals, including rice, may be responsible for the positive association. These findings were based on few studies, and thus, the results need to be interpreted with caution. Despite this research, mounting evidence reported a beneficial effect of vitamin D supplementation in reducing cancer incidence and mortality 105,106 . A meta-analysis of RCT suggested that evidence is stronger for cancer mortality rather than cancer incidence 107 .
Our meta-analysis has several strengths including the prospective studies, the relatively large sample size, and the linear and nonlinear dose-response analyses. As many previous published meta-analysis, the current analysis has several limitations. First, it is important to notice that all included studies have considered only a single baseline measure of 25 (OH)D, which may not necessarily represent long-term status. Also, most of studies lacked sensitivity analyses excluding skin cancer cases diagnosed within 2, 4, or 6 years of blood draw in order to assess whether reverse causation could have influenced the findings. Thus, we could not investigate a subgroup analysis by time of vitamin D measurement. However, one measurement of plasma 25(OH)D has been suggested to reflect long-term sun exposure and seemed to predict skin cancer risk 24 . In support of this supposition, previous study have reported a correlation of 0.70 for repeated measures of plasma 25(OH)D within individuals over time, suggesting that a single measurement is a reasonable proxy for long-term levels of 25(OH)D 108 . Further limitations of this meta-analysis include risk of bias of the primary studies, including potential measurement error in the assessment of exposure, residual confounding, especially regarding sun exposure. In addition, some of our meta-analyses were based on a low number of primary studies, and thus, subgroup and analyses by e.g. sun exposure level, season and pigmentary traits were not possible or relied on small numbers of studies.
In conclusion, our finding suggests that high vitamin D status was associated with increased risks of melanoma and KC. Given that 25 (OH)D level is mainly from sun exposure, higher risk of skin cancer may be confounded by sun exposure, data for which is lacking in most studies. However, while we found that intakes of Figure 6. Dose-response meta-analysis of each 100 IU/day increase in vitamin D intake (from diet, supplement and total) and the risk of skin cancer.
Scientific RepoRtS | (2020) 10:13151 | https://doi.org/10.1038/s41598-020-70078-y www.nature.com/scientificreports/ dietary or supplemental vitamin D were not associated with risk of melanoma and SCC, high intakes of vitamin D from diet and supplements were slightly associated with BCC risk, albeit with no heterogeneity across skin cancer. Overall, the current evidence suggests that unprotected sun exposure should be avoided in order to achieve high vitamin D status, and that an adequate amount of vitamin D should be obtained from a healthy diet.
Received: 10 December 2019; Accepted: 14 June 2020 Scientific RepoRtS | (2020) 10:13151 | https://doi.org/10.1038/s41598-020-70078-y www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creat iveco mmons .org/licen ses/by/4.0/.