Vitamin D Supplementation and Total Cancer Incidence and Mortality

A Meta-Analysis of Randomized Controlled Trials

N. Keum; D. H. Lee; D. C. Greenwood; J. E. Manson; E. Giovannucci


Ann Oncol. 2019;30(5):733-743. 

In This Article


This meta-analysis was conducted and reported in accordance with the PRISMA guideline.[30] Two authors (DL and NK) participated in the literature search, study selection, data abstraction independently and resolved any discrepancy through discussion.

Study Selection

PubMed and Embase were searched for relevant articles published up to November 2018. Detailed search terms are provided (Supplementary Table S1, available at Annals of Oncology online). Except for English language and human subjects, no other restrictions were imposed. Abstracts and unpublished results were not included. To identify additional papers, the reference lists of previous systematic reviews and meta-analyses were reviewed.

To be included, studies had to be a RCT that tested the effect of vitamin D supplementation (provided as cholecalciferol or ergocalciferol, with or without other nutrients) on total cancer incidence or mortality, with the results as relative risk (RR) (risk ratio or hazard ratio) and 95% CI (confidence interval); or as the number of incident cases of total cancer and/or total cancer death in each arm. We excluded RCTs when the total number of outcomes is ≤10, because effect size is unreliable and these were mostly small short-term RCTs (e.g. 1 year or less) in specialized populations (e.g. at risk for fractures). RCTs with ≤1 year of follow-up were also excluded because (1) latent cancers may be undiagnosed and a sizable proportion of cancer incidence in year 1 consists of undiagnosed cancers present at baseline; (2) cancer mortality within 1 year of follow-up is likely mostly from undiagnosed cancers that had metastasized already at the time of study inception; (3) it may take 3 to 6 months for 25(OH)D levels to reach homeostasis after initiation of supplementation. When there were several publications from the same trial, the publication fully covering the intervention period and directly reporting RR rather than the number of events alone was selected. This study selection process is summarized in Figure 1.

Figure 1.

Flowchart for study selection. *Two publications are from the same trial, Calcium plus vitamin D trial [34,35].

Data Abstraction

The following information was extracted: definition of intervention and control, RR and corresponding 95% CI or the number of participants and events in each arm, level of circulating 25(OH)D (at baseline, at follow-up), and important characteristics of the study population (Table 1). To conduct secondary analyses on total mortality, a robust end point that answers the ultimate question of whether vitamin D supplementation improves overall survival, we also extracted RR and corresponding 95% CI for all-cause death or the number of participants and total deaths in each arm.

Statistical Analyses

The summary RR and 95% CI on primary end points (total cancer incidence, total cancer death) and secondary end point (total mortality) were calculated using the DerSimonian–Laird random effects model.[31] Potential for small study effects, such as publication bias, was assessed using Egger's test.[32] For primary meta-analyses, several sensitivity analyses were performed. In some trials,[27,29,33] the benefit of vitamin D emerged approximately 1 year after randomization. To explore the potential for a latent effect, meta-analyses were conducted after replacing the results including all outcomes during the follow-up with the results excluding cases during the first year after randomization when reported. Two trials compared the intervention of vitamin D and calcium combined against placebo,[27,34,35] and a meta-analysis was conducted after excluding these trials. To understand the degree to which the effect of vitamin D supplementation on total mortality may be explained by its effect through total cancer mortality, we conducted a meta-analysis by including only trials that reported outcomes for both total mortality and total cancer mortality.

Heterogeneity in the relationship across trials was assessed by I2.[36] To explore potential sources of heterogeneity, subgroup analyses and meta-regression were conducted using a priori selected variables: potential effect modifiers [regimen of vitamin D supplementation, baseline and attained circulating 25(OH)D levels, contrast in circulating 25(OH)D levels between the intervention and control group, age, sex, body mass index of the population] and methodological characteristics (duration of follow-up, total number of cases, exclusion of active cancer at baseline, primary end point).

For statistical significance, two-sided α was set at P = 0.05. All statistical analyses were conducted using STATA 12 (StataCorp, College Station, TX).