Vitamin D and Clinical Cancer Outcomes: A Review of Meta‐Analyses

ABSTRACT The relationship between vitamin D status or supplementation and cancer outcomes has been examined in several meta‐analyses. To address remaining knowledge gaps, we conducted a systematic overview and critical appraisal of pertinent meta‐analyses. For meta‐analyses of trials, we assessed their quality using AMSTAR‐2 (A Measurement Tool to Assess Systematic Reviews), strength of associations using umbrella review methodology and credibility of evidence using GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) criteria. Meta‐analyses of observational studies reported inverse associations of 25OHD with risk of cancer incidence and cancer mortality and, particularly for colorectal cancer, fulfilled some of Bradford‐Hill's causation criteria. In meta‐analyses of trials, vitamin D supplementation did not affect cancer incidence. However, we found credible evidence that vitamin D supplementation reduced total cancer mortality risk, with five out of six meta‐analyses reporting a relative risk (RR) reduction of up to 16%: RR, 0.84 (95% CI, 0.74–0.95). The strength of the association, however, was classified as weak. This was true among meta‐analyses of high, moderate, and lower quality (AMSTAR‐2–rated). Trials did not include large numbers of vitamin D‐deficient participants; many tested relatively low doses and lacked sufficiently powered data on site‐specific cancers. In conclusion, meta‐analyses show that, although observational evidence indicates that low vitamin D status is associated with a higher risk of cancer outcomes, randomized trials show that vitamin D supplementation reduces total cancer mortality, but not cancer incidence. However, trials with larger proportions of vitamin D‐insufficient participants and longer durations of follow‐up, plus adequately powered data on site‐specific common cancers, would provide further insight into the evidence base. © 2020 The Authors. JBMR Plus published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research.

Introduction E cological studies of cancer incidence and mortality have shown that sun exposure, especially solar UV B (and hence vitamin D production), is associated with reduced risk of many cancer types. (1)(2)(3) This supports the hypothesis that vitamin D may have a beneficial impact on cancer outcomes, which is further strengthened by the identification of biological mechanisms that may explain these associations (4) and prospective observational studies, which show that dietary intake of vitamin D (including vitamin D supplementation) is associated with a reduced risk of cancers, although confounding is important to consider. (5)(6)(7)(8) Meta-analyses provide accurate, succinct, credible, and comprehensive summaries of the best available evidence, and act as a key tool for health care professionals to achieve evidence-based decisions. (9) Several meta-analyses have synthesized data on the association of vitamin D status with cancer outcomes from observational studies (5)(6)(7)(10)(11)(12)(13)(14) and the effect of vitamin D supplementation on cancer outcomes in randomized clinical trials. (15)(16)(17)(18)(19)(20)(21)(22) However, the evidence base is inconsistent and fragmented into various meta-analyses that combine different study populations and outcomes, making assessment of the evidence using a similar methodological framework difficult. Further, as the methodologic quality of vitamin D-cancer meta-analyses may vary, uncritically accepting their results carries risk. (9) Thus, we decided to conduct a systematic overview and critical appraisal of pertinent meta-analyses to better characterize the evidence on vitamin D status or supplementation in relation to cancer outcomes. In our critical appraisal, particular attention was given to intervention studies as this study design, as the gold standard for effectiveness research, provides the highest

Cancer incidence
With total cancer incidence as the outcome, one of these metaanalyses combined data from eight prospective cohort studies (70,018 participants and 7511 events). (13) The summary relative risk (RR) estimate of the highest 25OHD category compared with the lowest was 0.86 (95% CI, 0.73-1.02), indicating a marginal relationship in the inverse direction, with a significant between-study heterogeneity (I 2 = 70.8%). Dose-response analysis indicated a 7% reduction in risk (RR, 0.93; 95% CI, 0.91-0.96) per 20 nmol/L increment in 25OHD.

Cancer mortality
A recent meta-analysis had total cancer mortality as the outcome. (13) This report combined data from 16 prospective cohort studies on 101,794 participants without cancer at baseline, 8729 of whom had a cancer-related death. The pooled RR-of a highest-versus-lowest 25OHD group comparison-was 0.81 (95% CI, 0.71-0.93), indicating an inverse association, with moderate heterogeneity (I 2 = 48.8%). Dose-response analysis (of studies with ≥three categories of 25OHD) revealed that the risk of cancer mortality was reduced by 2% (RR, 0.98; 95% CI, 0.97-0.99) with each 20 nmol/L increment of 25OHD.
Four meta-analysis articles of prospective studies focused on patients with cancer. (10,(54)(55)(56) They reported sizeable inverse associations between 25OHD quantile (highest versus lowest) and cancer-specific mortality: pooled hazard ratios (HRs) were 0.50 to 0.65 (in patients with breast cancer, colorectal cancer, and lymphoma) in one study, (54) summary RR was 0.58 for breast cancer mortality in another study; (56) the pooled RR ranged from 0.57 (for breast cancer mortality) to 0.65 (for colorectal cancer mortality) in another (10) ; and the summary HR was 0.73 (for colorectal cancer mortality) in the fourth study. (55) One of these studies reported an inverse dose-response association. (56) Meta-analyses of randomized clinical trials We identified eight meta-analyses of clinical trials that evaluated the impact of vitamin D supplementation on cancer outcomes (incidence and mortality; Table 2). Appendix Table AA3 summarizes their study selection criteria, which excluded studies based on several factors such as coadministration of calcium that differed across the treatment groups, (16,19) treatment with hydroxylated vitamin D or vitamin D analogs, (15,17,20,21) duration of follow-up (<1 year (19,20) ) or intervention (<3 years (22) , age (<18 (21) or <60 (19) years), number of outcomes (<10 (20) ), and pregnancy. (15,16,21) Quality of meta-analyses The AMSTAR-2 ratings of the eight meta-analyses are summarized in Table 2 and detailed in Appendix Table AA4. Five meta-analyses had shortcomings in critical domains (17)(18)(19)(20)22) : All five did not cite a predefined, registered protocol, four did not cite excluded primary studies, (18)(19)(20)22) three did not use a satisfactory technique to assess risk of bias (17,18,20) , and two used a literature search that was not fully comprehensive. (17,18) All studies had deficiencies in noncritical domains, with the most common being reporting funding sources of primary studies (absent in seven meta-analyses (15,(17)(18)(19)(20)(21)(22) ) and explaining selection of study designs for inclusion in the review (absent in six metaanalyses (15)(16)(17)19,21,22) ). Based on weaknesses in the critical and noncritical domains, the overall confidence in the eight metaanalyses was deemed critically low or moderate for seven of these. (15,(17)(18)(19)(20)(21)(22)

Strength and credibility of meta-analysis effects
Five intervention meta-analyses of total cancer mortality had weak strength of association according to umbrella review criteria, (16,18,(20)(21)(22) with all scoring high with GRADE and with NNT values ranging from 86 to 381. This was consistent among meta-analyses of high, moderate, and lower quality (AMSTAR-2-rated). The remaining meta-analyses reported nonsignificant effects, with GRADE credibility that was moderate (16,17) to high (18)(19)(20)22) for total cancer incidence and low (15,16) to moderate (16) for site-specific cancer incidence. A breakdown of these scores is provided in Appendix Tables AA5 and AA6.

Differences between trial meta-analyses
As pooled effects across meta-analyses on the same outcome were mostly similar (with overlapping 95% CIs; Table 2) despite variable study selection criteria (Appendix Table AA3), this enhances the external validity of our findings that vitamin D reduces total cancer mortality (in nearly all meta-analyses), but not cancer incidence. Where pooled effects for the same outcome variations did vary, this can be explained by differences in the number of primary studies in meta-analyses, which is influenced by publication year ( Table 2) and study selection criteria. The most important variation occurred for cancer mortality, in which the meta-analysis by Goulão and colleagues (19) did not find a significant overall effect, but the remaining five metaanalyses did ( Table 2). Goulão and colleagues (19) excluded primary studies in which calcium was coadministered with vitamin D, but not with placebo, which allowed it to disentangle calcium and vitamin D effects. This restriction meant that the Women's Health Initiative, (59) in which calcium was given with vitamin D in the intervention group only, was not included in this meta-analysis (unlike nearly all other ones (16,18,20,22) ). However, this exclusion may not have been important for validity if calcium did not influence cancer mortality, as suggested by prior studies. (60) Also, the number of participants (n = 11202) and events (n = 320) in the Goulão and colleagues (19) meta-analysis was substantially lower than those in the other meta-analyses (n = 44,290 to 75,239 and n = 939 to 1192 for number of participants and events, respectively; Table 2). This is because two large (n > 5000) trials in addition to the Women's Health Initiative (59) -the VIDA (Vitamin D Assessment) (61) and VITAL (Vitamin D and Omega-3) (62) studies-were included in other meta-analyses (showing vitamin D benefits), (20)(21)(22) but not in the meta-analysis by Goulão and colleagues (19) as they were published after it. (19) Thus, the Goulão and colleagues' (19) metaanalysis did not capture effects from some large trials and had reduced statistical power.

Limitations of meta-analyses and their primary studies
Owing to several limitations, the meta-analyses and their primary studies have boundaries of applicability. These limitations influence the strength of associations and are key targets for future research. We discuss these for observational and intervention studies separately.

Observational studies
A limitation of the observational studies we reviewed is that, although they adjusted for multiple confounders, they are vulnerable to residual confounding; more so when the confounders (eg, smoking behavior, BMI, physical activity, diet) are measured less well. Adding to this issue is that vitamin D status is related to multiple diseases, besides cancer. (63) Thus, 25OHD-cancer associations do not fulfill one of Bradford-Hill's causation criteria, specificity, (64) raising the possibility that confounding may be important. On the other hand, confounding may not entirely explain these associations as risk factors can cause more than one disease, (64) and vitamin D has pleiotropic cellular effects. (63) A second limitation is that several observational studies, especially those on breast cancer incidence, (14,29,30,32,33,35) were case-control investigations-and are thus susceptible to reverse causation because 25OHD measurement was performed in those already diagnosed with cancer, and low vitamin D status may be a consequence of the disease rather than a cause. For instance, during cancer therapy or when symptoms are severe, sunlight exposure, physical activity, and dietary habits are likely to change (because of hospitalizations, disability, or lifestyle changes), and cancer-associated inflammation may depress 25OHD. (65) However, this problem was avoided in prospective studies where blood samples were collected well before cancer diagnosis. The studies in the meta-analyses we reviewed were mainly prospective (nested case-control or cohort) and we observed inverse associations in them, suggesting that 25OHD   (Table 1) rather than vice versa. Third, several observational studies are prone to 25OHD measurement error because they did not measure vitamin D status with the gold standard laboratory method for 25OHD measurement: liquid chromatography-tandem mass spectrometry. (66) Fourth, whereas many observational studies were of high quality, as assessed by the Newcastle-Ottawa Scale, several were of suboptimal (medium) quality, which may have attenuated 25OHDcancer associations. In support of this, a recent meta-analysis reported that 25OHD was inversely associated with total cancer incidence and mortality in studies of high quality, but not in those of medium quality. (13) Fifth, though most meta-analyses had low or moderate between-study heterogeneity (I 2 < 50% or P values >0.05), (6,31,36,38,39,42,44,45,47,49,50,(52)(53)(54)56,57) some did not; thus, their results should be interpreted with caution. These include metaanalyses on colorectal adenoma and cancer occurrences of the breast, rectum, and all types combined, (5,6,13,14,29,30,32,35,36) which reported high between-study heterogeneity (I 2 ≥ 50%). Finally, most participants in the observational studies were of White ethnicity, which limits generalizability of findings to non-White ethnic groups.

Randomized clinical trials
An important issue for intervention studies is the growing randomized controlled trial (RCT) evidence that health benefits of vitamin D supplementation are greatest in vitamin D-deficient people. (67) This may also apply to cancer as a recent metaanalysis of cohort studies revealed that 25OHD was inversely associated with cancer incidence at approximately <30 nmol/L only, and had an inverse relationship with cancer mortality that was strongest at low 25OHD levels (especially at approximately <50 nmol/L). (13) Thus, any cancer-related benefits of vitamin D supplementation may be greatest in vitamin D-deficient individuals. However, trials have not contained large numbers of such people. For example, the average baseline 25OHD of 10 trials a DerSimonian-Laird Q statistic. CC = Case control; CRC = colorectal cancer; CS = cross sectional; HR = hazard ratio; MR = Mendelian randomization; NCC = nested case-control; NR = not reported; NS = not significant; OR = odds ratio; PC = prospective cohort; POR = Peto odds ratio; RR = relative risk; RSP = retrospective; SMD = standardized mean difference. examined in the recent meta-analysis by Keum and colleagues (20) we reviewed was approximately 60 nmol/L. Insufficient vitamin D dosage may be another limitation. For example, in the Goulão and colleagues meta-analysis of cancer incidence, (19) the daily dose equivalent (dose divided by days between each dose) in multiple primary studies (n = 7) was relatively low (<1000 IU/day), raising a question of whether a higher dose may have produced a different (stronger) effect. A related issue is the frequency of the dosing regimens. Most trials have utilized daily dosing; evidence is limited for supplementation administered monthly or weekly. (20) Third, as carcinogenesis is a long-term and gradual process (often spanning decades), the need for a long follow-up period is particularly great. (68) The importance of this is reflected in the long follow-up periods of 25OHD-cancer cohort studies, such as those in the Han and colleagues' meta-analysis (of total cancer incidence and mortality), (13) which were 12 to 13 years on average and up to 28 years. However, most of the primary studies included in our meta-analyses of trials had follow-up periods of no more than 5 years; this may have been insufficient to detect effects on cancer. (68) In support of this, the meta-analysis by Zhang and colleagues (21) we reviewed found that benefit of supplementation on reduced cancer mortality was observed in trials with longer follow-up (>3 years), but not in those with a shorter follow-up.
A fourth limitation is that, as shown in the meta-analyses for observational studies, 25OHD was more consistently associated with colorectal and breast cancers than other cancers (Table 1). Thus, there may be stronger effects of supplementation against certain cancer types or, possibly, residual confounding may be greater for some cancers than others (eg, confounding by BMI, physical activity, and diet may be of particular relevance to studies of vitamin D and colorectal cancer). However, whereas the cancer outcomes that have dominated our RCT findings are those based on all cancers combined (Table 2), results for sitespecific cancers are lacking. Almost all trials that evaluated impacts on site-specific cancers did not include these as primary endpoints, and data on rarer cancers (eg, kidney) are missing.
Fifth, the vast majority of participants in the intervention studies were White, which restricts applicability of findings to non-White populations.
Finally, there were quality-related shortcomings of the metaanalyses of intervention studies we reviewed. The most common ones, detected by AMSTAR-2, were the lack of (i) information on funding sources of primary studies (noteworthy as many vitamin D trials are industry-funded and have a high risk of "for-profit" bias (16) ), (ii) an explanation of study design selection, and (iii) a review protocol describing prespecified methodology. Most of these meta-analyses were published prior to the availability of AMSTAR-2 reporting standards (in 2017 (9) ), which may have contributed to their lower AMSTAR-2 ratings. However, as mentioned above, our vitamin D-cancer findings for meta-analyses of high and moderate quality (AMSTAR-2-rated) were similar to those of lower quality.

Limitations of current review
A potential limitation of this review is that, although a comprehensive and systematic literature search was performed, we may have missed some meta-analyses. Second, our study was a meta-review, and although this provides an overarching perspective on a research topic, we did not provide granulate analyses at the primary study level. Third, our review focused on meta-analyses; thus, some primary studies may not have been included either because the meta-analysis did not identify them or they were too recent to be included. Finally, we did not critically appraise the quality of all primary studies individually. This should have been done in each meta-analysis; doing this here was beyond the scope of our review.

Future research
Several areas of future research would strengthen our understanding of vitamin D effects on cancer. First, given the emerging evidence for threshold effects related to vitamin D status, future trials should aim to recruit participants with vitamin D insufficiency (25OHD <50 nmol/L). There are major logistical and practical barriers to doing this in populations that are vitamin D replete, and trials could be undertaken more easily and cheaply in populations with a high prevalence of vitamin D insufficiency. These trials should have longer follow-up periods, include more adequately powered data on site-specific cancers (feasible for common cancers), and study more non-White populations. However, ethical issues can arise with the conduct of long-term trials in vitamin D-deficient participants, as 50% will be randomly assigned to placebo and remain deficient for a prolonged period.
Second, cells that express the cell-surface receptor proteins megalin and cubulin (eg, those in the kidney, lung, thyroid, mammary gland, gall bladder, and thyroid) can internalize 25OHD bound to vitamin D-binding protein, with subsequent unbinding of 25OHD intracellularly and conversion to 1,25-dihydroxyvitamin D, which can exert anticancer effects by activating the vitamin D receptor. (69,70) In contrast, 25OHD entry in cells not expressing the megalin-cubulin receptor is proposed to occur via diffusion of unbound, free 25OHD across the cell membrane. (69,70) However, the studies we reviewed measured total 25OHD, comprising not only free (and bioavailable) 25OHD, but mostly (approximately 90%) 25OHD, which is bound tightly to vitamin D-binding protein, and thus may not get into some cancer cells easily. (71) As evidence of importance, a large, prospective cohort study reported that higher bioavailable, rather than total, 25OHD levels were independently associated with improved survival in patients with hepatocellular carcinoma. (72) Further, in a recent RCT of patients with digestive tract cancer, vitamin D supplementation improved 5-year relapse-free survival in those with low bioavailable 25OHD, but not in those with high bioavailable 25OHD (71) or in those with low total 25OHD (<20 ng/mL). (73) These studies suggest that for some cancer types, free and bioavailable 25OHD may better assess true vitamin D status (and deficiency) than total 25OHD in future trials.
Third, articles that reported inverse, longitudinal associations between 25OHD and cancer mortality (10,(54)(55)(56) (Table 1) were studies of patients with cancer, and thus suggested a potential role of vitamin D in cancer therapy as opposed to cancer prevention, the focus of the other articles we reviewed. RCT data investigating vitamin D as a cancer treatment are scarce, with one relatively small RCT (n = 139) reporting an improvement in median progression-free survival or death (HR, 0.64) over 22.9 months (median) in patients with colorectal cancer (74) and another (n = 417) reporting an age-adjusted benefit on relapsefree survival (HR, 0.66) over 3.5 years (median) in patients with digestive tract cancers. (73) However, further trials are required in this research area, ideally with longer follow-up. (75) Including biological measurements would promote our understanding of underlying mechanisms: This requires considering not only antineoplastic influences, but broad biological effects too, as 25OHD Table 2.   is inversely related to all-cause mortality in patients with or without cancer. (75) Fourth, cancer incidence and mortality, and overall survival, though considered the gold standard endpoint in oncology trials, require a large sample size and long follow-up time to achieve adequate statistical power. (76) In comparison, other clinical endpoints can be assessed earlier, and thus could be measured in parallel. One of these is health-related quality of life, which is considered an outcome in assessing clinical benefit and has emerged as a primary endpoint in oncology clinical trials. (76) Encouragingly, observational cohort studies found that, in patients with cancer, vitamin D intake and 25OHD predict improved quality of life, (77)(78)(79) but RCT data are needed to evaluate effects on this outcome. Another is tumor-centered endpoints, such as progression-free survival, disease-free survival, tumor response, and circulating tumor cells. (76) As mentioned, two RCTs reported beneficial effects on median progression-free survival (74) and relapse-free survival, (73) but further RCTs are needed. (75) Thus, adding these other clinical endpoints (alongside gold standard outcomes) to future vitamin D trials would provide an earlier assessment and more comprehensive evaluation of efficacy. (76) Finally, a novel area of research is investigating whether vitamin D pathway genes may alter health effects on vitamin D supplementation. A meta-analysis of eight prospective studies found that colorectal cancer risk was lower in participants with the BB genotype of the BsmI vitamin D receptor singlenucleotide polymorphism. (6) A recent RCT found that vitamin D receptor genotypes modified the effect of vitamin D supplementation on the prevention of advanced colorectal adenomas. (80) Specifically, vitamin D supplementation reduced risk by 64% among those with the rs7968585 genotype and increased risk by 41% among those with one or two G alleles. (80) Such work helps identify who may benefit from supplementation for cancer prevention based on vitamin D-related genotypes. Given the knowledge gap in investigating vitamin D pathway genotypes as modifiers of effects on cancer outcomes, assessment of these in recent trials (possibly as IPD meta-analyses), and further trials in this research area are both warranted. As recognition of the importance of this, the VITAL trial, (62) for example, is in the process of conducting such analyses of gene variants (JE Manson, personal communication, 2020).

Conclusion
Observational studies showed that, in many cases, low vitamin D was inversely associated with cancer outcomes. For this, the associations for some outcomes, particularly colorectal cancer, seem to fulfill some (but not all) of Bradford-Hill's criteria for causation, (64) including consistency of findings across different meta-analyses and primary studies, temporality (prospectivestudy associations), biological gradient (dose-response associations), and strength of associations (strong in some cases).
To our knowledge, this review is the first report to systematically compile and appraise clinical evidence-by concurrently using AMSTAR-2, umbrella review, and GRADE assessment tools-of vitamin D supplementation in relation to cancer outcomes from meta-analyses. We found highly credible RCT evidence that vitamin D supplementation reduces risk of total cancer mortality, but the magnitude of effect was classified as weak. Our finding of a highly credible weak effect on total cancer mortality is in line with that of a 2017 systematic review of meta-analyses, which reported that vitamin D supplementation reduces risk of all-cancer mortality. (81) We extend that review by including more recent meta-analyses in our assessment, (13,(19)(20)(21) and by critically evaluating the evidence for this outcome using the above-mentioned appraisal tools.
The available research, however, is not without limitations. To address these limitations and to provide clearer and further insight into the role of vitamin D in cancer incidence and related mortality, future research should include trials with more vitamin D-insufficient participants and of longer follow-up duration, plus adequately powered data on site-specific cancers (where feasible).