Skip to main content
NCBI home page
UKPMC Funders Author Manuscripts logoLink to UKPMC Funders Author Manuscripts
. Author manuscript; available in PMC: 2024 Feb 1.
Published in final edited form as: Lancet Diabetes Endocrinol. 2023 Dec 1;12(1):e2–e11. doi: 10.1016/S2213-8587(23)00287-5

Estimating dose—response relationships for vitamin D with coronary heart disease, stroke, and all-cause mortality: observational and revised Mendelian randomization analyses

Emerging Risk Factors Collaboration/EPIC-CVD/Vitamin D Studies Collaboration26, Eleni Sofianopoulou 1,#, Stephen K Kaptoge 1,#, Shoaib Afzal 2,3, Tao Jiang 1, Dipender Gill 4, Thomas E Gundersen 5, Thomas R Bolton 1,6, Elias Allara 1,6, Matthew G Arnold 1, Amy M Mason 1, Ryan Chung 1,7, Lisa AM Pennells 1, Fanchao Shi 1, Luanluan Sun 1, Peter Willeit 1,8, Nita G Forouhi 9, Claudia Langenberg 9,10, Stephen J Sharp 9, Salvatore Panico 11, Gunnar Engström 12, Olle Melander 12,13, Tammy YN Tong 14, Aurora Perez-Cornago 14, Margareta Norberg 15, Ingegerd Johansson 16, Verena Katzke 17, Bernard Srour 17, María José Sánchez 18,19,20,21, Daniel Redondo-Sánchez 18,19,20, Anja Olsen 22,23, Christina C Dahm 23, Kim Overvad 23, Magritt Brustad 24,25, Guri Skeie 24, Conchi Moreno-Iribas 25, N Charlotte Onland-Moret 27, Yvonne T van der Schouw 27, Konstantinos K Tsilidis 28,29, Alicia K Heath 28, Claudia Agnoli 30, Vittorio Krogh 30, Ian H de Boer 31, Camilla Jannie Kobylecki 2, Yunus Çolak 2, Armin Zittermann 32, Johan Sundström 33, Paul Welsh 34, Elisabete Weiderpass 35, Elom K Aglago 35, Pietro Ferrari 35, Robert Clarke 36, Marie-Christine Boutron 37, Gianluca Severi 37,38, Conor MacDonald 37, Rui Providencia 39, Giovanna Masala 40, Raul Zamora-Ros 41, Jolanda Boer 42, WM Monique Verschuren 27,42, Peggy Cawthon 43,44, Louise L Schierbeck 45, Cyrus Cooper 46,47, Matthias B Schulze 48,49, Manuela M Bergmann 48, Anke Hannemann 50, Stefan Kiechl 51,52, Hermann Brenner 53,54,55,56, Natasja M van Schoor 57, Juan R Albertorio 58, Carlotta Sacerdote 59, Allan Linneberg 60, Line L Kårhus 60, José María Huerta 20,61, Liher Imaz 62,63, Christel Joergensen 64, Yoav Ben-Shlomo 65, Annamari Lundqvist 66, John Gallacher 67, Naveed Sattar 34, Angela M Wood 1,6,7,68,69, Nicholas J Wareham 9, Børge G Nordestgaard 2,3,70, Emanuele Di Angelantonio 1,6,7,68, John Danesh 1,6,7,68,71, Adam S Butterworth 1,6,7,68, Stephen Burgess 1,7,72,†
PMCID: PMC7615586  EMSID: EMS193231  PMID: 38048800

Abstract

Background

Randomized trials of vitamin D supplementation for cardiovascular disease and all-cause mortality have generally reported null findings. However, generalizability of results to individuals with low vitamin D status is unclear. We characterized dose—response relationships between 25-hydroxyvitamin D [25(OH)D] concentrations and risk of coronary heart disease (CHD), stroke, and all-cause mortality in observational and Mendelian randomization frameworks.

Methods

Observational analyses were conducted in 33 prospective studies comprising 500,962 individuals with no known history of CHD or stroke at baseline. Mendelian randomization analyses were conducted in four population-based cohort studies comprising 386,406 middle-aged individuals of European ancestries, including 33,546 CHD cases, 18,166 stroke cases, and 27,885 deaths.

Findings

Observational analyses suggested threshold relationships for incident CHD, stroke, and mortality outcomes. In population-wide genetic analyses, there were no associations of genetically-predicted 25(OH)D with CHD (odds ratio [OR] per 10 nmol/L higher genetically-predicted 25(OH)D concentration 0·98, 95% confidence interval [CI] 0·95-1·01), stroke (OR 1·01, 95% CI 0·97-1·05), or all-cause mortality (OR 0·99, 95% CI 0·95-1·02). Null findings were also observed in genetic analyses for cause-specific mortality outcomes, and in stratified genetic analyses for all outcomes at all observed levels of 25(OH)D concentrations.

Interpretation

Stratified Mendelian randomization analyses suggest a lack of causal relationship for 25(OH)D concentrations with both cardiovascular and mortality outcomes for individuals at all levels of 25(OH)D. Our findings suggest that substantial reductions in mortality and cardiovascular morbidity due to long-term low-dose vitamin D supplementation are unlikely even if targeted at individuals with low vitamin D status.

Funding

British Heart Foundation, Medical Research Council, National Institute for Health Research, Health Data Research UK, International Agency for Research on Cancer.

Keywords: 25-hydroxyvitamin D, Vitamin D deficiency, Mendelian randomization, cardiovascular disease, all-cause mortality

Introduction

Vitamin D is an essential nutrient obtained from sunlight, dietary intake, and supplementation.1 Observational epidemiological studies have consistently found that low concentrations of circulating 25-hydroxyvitamin D [25(OH)D] (a metabolite used as a clinical indicator of vitamin D status) are associated with higher risk of cardiovascular disease (CVD) and all-cause mortality, as well as other chronic diseases.2,3 However, several large randomized trials of vitamin D supplementation have reported null results,4–6 casting doubt on the observational evidence. However, as trials have typically recruited participants irrespective of baseline 25(OH)D concentration, they have had limited power to test supplementation effects among subgroups with low 25(OH)D concentrations.7

An efficient approach for assessing the potential causal effect of vitamin D supplementation is Mendelian randomization. Mendelian randomization uses genetic variants specifically related to a particular exposure to compare genetically-defined population subgroups with different average levels of the exposure. The independent segregation of alleles at conception means that these genetically-defined subgroups should not differ systematically with respect to confounding variables, creating a natural experiment analogous to a randomized trial.8 Therefore, compared with conventional observational analyses, Mendelian randomization analyses can provide more reliable insights into causal relationships between risk factors and disease outcomes. Previous Mendelian randomization analyses have reported null associations of genetically-predicted 25(OH)D concentrations with coronary heart disease (CHD)9,10 and ischaemic stroke.11,12 An inverse association has been observed between genetically-predicted 25(OH)D and all-cause mortality.13 Null findings have been observed for several further outcomes, including other cardiovascular diseases and cancers.14

Most previous Mendelian randomization analyses assumed a linear dose—response relationship between genetically-predicted 25(OH)D and CVD. However, some observational analyses have reported non-linear associations,15,16 suggesting methods that assume linearity may not provide an accurate picture of the dose—response relationship. Several non-linear Mendelian randomization analyses have been performed (see Supplementary Table 1 for list). However, such analyses can be severely biased if genetic effects on the exposure vary in the population17. This manuscript is a revised version of a previous publication that was affected by this bias18.

Here, we conduct the largest observational analysis to date to characterize the shape of association between 25(OH)D concentrations and CVD outcomes in an individual-participant data meta-analysis of 33 prospective studies. We then perform stratified Mendelian randomization analyses using the doubly-ranked method19, which is more robust to variation in the genetic effects on the exposure, to assess evidence for potential causal effects of 25(OH)D concentrations on risk of major CVD outcomes, all-cause mortality, and cause-specific mortality for population subgroups with different average 25(OH)D concentrations.

Methods

Study populations

Observational analyses were conducted in UK Biobank, the European Prospective Investigation into Cancer and Nutrition Cardiovascular Disease study (EPIC-CVD), and 31 studies from the Vitamin D Studies Collaboration (VitDSC). Genetic analyses were conducted in UK Biobank, EPIC-CVD, and two Danish population-based studies. Only baseline measurements of 25(OH)D were used in analyses.

UK Biobank is a prospective cohort study of around 500,000 people aged 40 to 69 years at baseline, recruited in 2006-2010 from the United Kingdom and followed-up for a median of 10·9 years.20 For observational analyses, we analysed 384,721 individuals without previous known CVD at baseline. For genetic analyses, we included data on 333,002 unrelated individuals of European ancestries. EPIC-CVD is a case-cohort study derived from a cohort of over 500,000 individuals recruited at 23 centres across 10 European countries.21,22 Participants were followed-up for a median of 9·5 years. We analysed 26,336 individuals without previous known CVD at baseline in observational analyses, and 22,142 individuals of European ancestries in genetic analyses. VitDSC comprises 89,915 participants from 31 mostly population-based, prospective studies across 11 countries. Individual-participant data on 25(OH)D concentrations, conventional cardiovascular risk factors, and major incident cardiovascular morbidity and mortality were available for 67,992 individuals without previous known CVD. The Copenhagen City Heart Study (CCHS) and Copenhagen General Population Study (CGPS) are prospective cohort studies in the Danish population.23,24 CCHS was initiated in 1976 and participants were followed up periodically until 2018. Median follow-up was 28·6 years. CGPS was initiated in 2003 and has a median follow-up of 8·8 years. For genetic analyses, we analysed 31,262 individuals from both studies with genetic data and a 25(OH)D measurement. For all studies, written informed consent was obtained from participants and approval was obtained from relevant ethics committees.

Vitamin D measurement and classification

Concentrations of 25(OH)D in blood were measured using the DiaSorin Liaison immunoassay analyser (UK Biobank and Copenhagen studies) or liquid chromatography-tandem mass spectrometry (EPIC-CVD). In VitDSC, concentrations were measured by radioimmunoassay, direct chromatographic approaches, or other immunoassays (Supplementary Table 2). Measurements were seasonally adjusted in each study to correspond to a measurement taken in autumn by subtracting the study-specific mean 25(OH)D concentration for the season the measurement was taken in and then adding the study-specific mean 25(OH)D concentration for autumn measurements. In EPIC-CVD, centre-specific means were used rather than study-specific means.

Outcomes

Outcomes were ascertained using International Classification of Diseases, Tenth Revision (ICD-10) codes. Primary outcomes were CHD, defined as fatal ischaemic heart disease (ICD-10 code: I20-I25) or non-fatal myocardial infarction (I21-I23); stroke, defined as any cerebrovascular disease (I60-I69); and all-cause mortality. We performed secondary analyses for cause-specific mortality divided into cardiovascular mortality, cancer mortality, or non-cardiovascular non-cancer mortality using ICD-10 codes (Supplementary Table 3). Observational analyses included incident events only. Supplementary genetic analyses were performed restricting to incident CHD and stroke events, and separating ischaemic stroke (I63-I64) and haemorrhagic stroke (I60-I61).

Genetic variants

To minimize potential bias due to horizontal pleiotropy, we considered genetic variants from four gene regions previously shown to be strongly associated with 25(OH)D and implicated in the transport, metabolism, and synthesis of vitamin D25 – GC, DHCR7, CYP2R1, and CYP24A1. The GC gene encodes vitamin D binding protein. The DHCR7 gene product converts 7-dehydrocholesterol to cholesterol, reducing 7-dehydrocholesterol available for conversion to previtamin D3 by solar radiation. The CYP2R1 gene encodes vitamin D 25-hydroxylase, a regulator of 25(OH)D synthesis through 25-hydroxylation of vitamin D in the liver. The CYP24A1 gene product inactivates the active form of vitamin D (1α25(OH)2D). To maximize the variance explained by the genetic instrument, we considered available variants at each genetic locus and selected variants associated with 25(OH)D concentrations using a stepwise selection method (Supplementary Methods). For UK Biobank and EPIC-CVD, 21 variants were included in the analysis (Supplementary Table 4). For the Copenhagen studies, due to limited availability, analyses were restricted to three variants: two from the CYP2R1 locus (rs12794714 and rs117913124) and one from the DHCR7 locus (rs7944926).

We also considered a score based on 71 genetic variants from across the genome (“genome-wide score”) previously demonstrated to be associated with 25(OH)D concentrations at a genome-wide level of statistical significance26.

Statistical methods

Observational analysis

Observational associations were assessed by inverse-variance weighted random-effects meta-analysis of study-specific hazard ratios (HRs), calculated using Cox proportional hazards regression models stratified by sex and (where appropriate) centre or trial arm. Principal analyses were adjusted for age at blood draw for 25(OH)D measurement, calendar month of blood draw, smoking status (current versus other), total cholesterol, high-density lipoprotein (HDL) cholesterol, systolic blood pressure, known history of diabetes, and body mass index; all measured at study baseline.

The primary dose—response analyses assessed the shape of association between 25(OH)D and outcomes by meta-analysis of fractional polynomials adjusted for the conventional risk factors. Supplementary analyses combined study-specific hazard ratios by tenths of 25(OH)D and plotted the pooled hazard ratios against the pooled mean 25(OH)D within each tenth.

Genetic analysis

We constructed a genetic risk score (GRS) weighted by the conditional associations of the genetic variants with 25(OH)D concentration in UK Biobank (Supplementary Table 4). Mendelian randomization estimates were calculated using the ratio method by dividing the genetic association with the outcome by the genetic association with 25(OH)D concentration, and scaling the estimate to a 10 nmol/L difference in genetically-predicted 25(OH)D concentration. Genetic associations were estimated using logistic regression for disease outcomes, and using linear regression for 25(OH)D concentrations. All regression models were adjusted for age at baseline, sex, centre (for UK Biobank and EPIC-CVD), and ten genetic principal components of ancestry. We assessed specificity of the GRS by testing its associations with a range of cardiovascular risk factors in the UK Biobank study.

In addition to analyses conducted in the overall study sample to estimate population-averaged causal effects, we also conducted stratified analyses in strata of the population constructed using the doubly-ranked stratification method19, a statistical method for constructing strata of the population with different average levels of the exposure such that stratum membership is independent of the genetic instrument. Stratification on 25(OH)D levels directly would induce collider bias, meaning that the distribution of the genetic instrument would vary between strata, and the instrumental variable assumptions would be violated in the strata. A previously-proposed method that stratifies on residual values of 25(OH)D (that is, the non-genetic component of 25(OH)D) requires the genetic effect on 25(OH)D concentrations to be linear and homogeneous in the population to provide unbiased estimates; an assumption that is generally implausible and is violated in this case17. We calculated Mendelian randomization estimates for 10 equally-sized strata using the ratio method with the GRS as an instrumental variable, and combined stratum-specific estimates across studies using fixed-effect meta-analysis. All estimates from the doubly-ranked method were averaged across 100 iterations of the method (see Supplementary Material for details).

All statistical analyses were performed in R version 4·0·5 (R Foundation for Statistical Computing, Vienna, Austria), Stata/SE 15·1 (College Station, TX, USA), or BOLT-LMM version 2·3·4 (Broad Institute, MS, USA).

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

Table 1 shows baseline characteristics of 386,406 participants in the four studies included in genetic analyses (details of 500,962 participants included in observational analyses are in Supplementary Tables 5 and 6). Mean age of participants ranged from 54·8 to 57·5 years, with similar numbers of men and women in each study. Mean season-shifted 25(OH)D concentrations (corresponding to an autumn measurement) were 54·5 nmol/L in UK Biobank, 46·9 nmol/L in EPIC-CVD, and 53·8 nmol/L in the Copenhagen studies (Supplementary Table 7). Mean 25(OH)D estimates did not notably differ by assay type (Supplementary Figure 1). The focused GRS explained 4·7% of the variance in 25(OH)D concentrations in UK Biobank, 5·8% in EPIC-CVD, and 1·8% in the Copenhagen studies. This GRS was not associated with a range of cardiovascular risk factors in UK Biobank, with the exception of body mass index and HDL-cholesterol, although these associations were small in magnitude (Supplementary Figure 2). The genome-wide score was strongly associated with LDL-cholesterol and triglycerides (Supplementary Figure 3), and so Mendelian randomization estimates using this score are unreliable.

Table 1. Characteristics of participants in the genetic analyses by study.

UK Biobank EPIC-CVD Copenhagen studies
Participants 333,002 22,142 31,262
Age at baseline 57·1 ± 8·1 54·8 ± 9·4 57·5 ± 12·9
Sex    
   Female 177,733 (53·4) 11,426 (51·6) 17,311 (55·4)
   Male 155,269 (46·6) 10,716 (48·4) 13,951 (44·6)
25(OH)D concentration (nmol/L) 54·5 ± 19·6 46·9 ± 16·4 53·8 ± 25·9
CHD events 22,363 (6·7) 5942 (26·8) 5241 (16·8)
Stroke events 10,489 (3·1) 5478 (24·7) 2199 (7·0)
Deaths 20,340 (6·1) - 1 7545 (24·1)
BMI (kg/m2) 27·3 ± 4·8 26·7 ± 4·3 25·9 ± 4·2
SBP (mmHg) 137·5 ± 18·6 137·4 ± 21·3 140·1 ± 21·0
Smoking    
   Current 34,085 (10·2) 6867 (31·0) 8387 (26·9) 
   Other 298,940 (89·8) 15,275 (69·0) 22,784 (73·1)
Diabetes    
   Known history 15,822 (4·8) 1234 (5·6) 1244 (4·0)
   No known history 317,203 (95·2) 20,908 (94·4) 30,018 (96·0)
Open in a new tab

Values represent mean ± standard deviation for continuous traits or N (%) for categorical traits. 25(OH)D concentrations are season-shifted to correspond to a measurement taken in autumn.

Abbreviations: 25(OH)D = 25-hydroxyvitamin D, BMI = body mass index, CHD = coronary heart disease, SBP = systolic blood pressure.

1

EPIC-CVD was specifically designed as a case-cohort study of cardiovascular disease outcomes, and therefore does not contribute to the analysis of non-cardiovascular disease or all-cause mortality outcomes.

Observational associations had a similar non-linear shape for all outcomes (Figure 1 and Supplementary Figure 4): at low concentrations of 25(OH)D, there was an inverse association; whereas at higher concentrations of 25(OH)D, the association was null (for CVD outcomes) or weakly positive (for mortality outcomes). For CHD and stroke, there was no strong association for 25(OH)D concentrations above 50 nmol/L, but a progressively steeper association was observed below this threshold. For all-cause mortality, the strength of the inverse association at lower 25(OH)D concentrations was stronger and began at a higher 25(OH)D concentration.

Figure 1. Observational associations of 25(OH)D concentration with outcomes.

Figure 1

Open in a new tab

Reference value is 50 nmol/L, which is close to the mean concentration of 25(OH)D in most contributing studies. The shaded area represents the 95% confidence interval for the dose—response curve. Study-specific analyses involved fractional polynomial modelling of continuous associations of 25(OH)D and outcomes using Cox regression stratified by sex, and (where appropriate) trial arm and centre, and adjusted for conventional cardiovascular risk factors, namely: age at blood draw for 25(OH)D measurement, calendar month of blood draw, smoking status (current versus other), total cholesterol, high-density lipoprotein (HDL) cholesterol, systolic blood pressure, known history of diabetes, and body-mass index (all measured at baseline), followed by random effects meta-analysis (see Methods).

The shapes of the observational associations in the three primary data sources were broadly similar (Supplementary Figure 5). Dose—response findings were also similar in supplementary analyses that combined study-specific hazard ratios by deciles of 25(OH)D or according to the four 25(OH)D categories (Supplementary Figure 6 and Supplementary Table 8).

Mendelian randomization estimates, representing the odds ratio (OR) per 10 nmol/L higher genetically-predicted 25(OH)D concentration, are presented in Figure 2 (top row) for primary outcomes (study-specific estimates in Supplementary Table 9). In overall analyses (that is, population-averaged estimates across the full range of the 25(OH)D concentration distribution), there was no association between genetically-predicted 25(OH)D and CHD (OR 0·98, 95% confidence interval [CI] 0·95-1·01, p=0·18), stroke (OR 1·01, 95% CI 0·97-1·05, p=0·61), or all-cause mortality (OR 0·99, 95% CI 0·95-1.02, p=0·39). However, there was some evidence of an overall inverse association with all-cause mortality in the Copenhagen studies (OR 0·89, 95% CI 0·80-0·99, p=0·030). In stratified analyses (that is, stratum-averaged estimates for strata of the population with different average 25(OH)D concentrations), there were no clear associations between genetically-predicted 25(OH)D and the main outcomes in any stratum, and no discernible trends in the stratum-specific estimates. The distributions of 25(OH)D concentrations in strata are provided in Table 2.

Figure 2. Mendelian randomization estimates in overall population and strata of residual 25(OH)D concentrations.

Figure 2

Open in a new tab

Estimates (95% confidence intervals) represent odds ratios per 10 nmol/L higher genetically-predicted concentration of 25(OH)D in strata of the population defined by residual concentration of 25(OH)D.

Table 2. Distribution of 25(OH)D concentrations in strata by study.

UK Biobank EPIC-CVD Copenhagen studies
Stratum 1 (lowest) 28.6 (22.7, 38.0) 25.3 (19.5, 34.2) 25.3 (18.3, 36.2)
Stratum 2 35.7 (27.0, 45.0) 31.9 (24.1, 40.1) 33.6 (23.1, 44.8)
Stratum 3 41.1 (31.9, 51.0) 36.4 (28.5, 45.1) 40.1 (29.1, 51.7)
Stratum 4 46.0 (36.1, 56.5) 40.3 (32.2, 49.3) 46.0 (34.7, 57.9)
Stratum 5 50.7 (40.2, 61.8) 44.1 (35.7, 53.4) 51.6 (39.9, 64.1)
Stratum 6 55.5 (44.3, 67.4) 47.9 (38.7, 57.2) 57.4 (44.7, 70.7)
Stratum 7 60.7 (48.6, 73.3) 51.8 (41.9, 61.6) 63.7 (50.0, 78.1)
Stratum 8 66.5 (53.3, 80.2) 56.3 (45.7, 67.6) 71.1 (55.8, 87.5)
Stratum 9 73.8 (59.0, 89.5) 62.2 (49.3, 75.4) 81.1 (62.9, 100.8)
Stratum 10 (highest) 86.2 (67.1, 98.6) 72.7 (56.2, 83.0) 100.0 (73.5, 116.4)
Open in a new tab

Mean season-shifted 25(OH)D concentration in each stratum (nmol/L); in brackets, 10th and 90th percentiles of the distribution, or 20th and 80th percentiles for the lowest and highest strata.

Similar results were observed for supplementary analyses that considered incident stroke outcomes only, ischaemic stroke only (Supplementary Table 10), and incident CHD outcomes only (Supplementary Table 11). Estimates using the pleiotropic genome-wide score are presented in Supplementary Table 12. The precision of estimates varied between strata; this is because genetic associations with 25(OH)D were much stronger in individuals with high 25(OH)D concentrations than in individuals with low 25(OH)D concentrations (Supplementary Figure 7). In UK Biobank, genetic associations with 25(OH)D were 5.0 times stronger in the highest decile compared with the lowest decile; in EPIC-CVD, this ratio was 4.6, and in the Copenhagen studies, this ratio was 9·7.

Mendelian randomization estimates for cause-specific mortality in UK Biobank and the Copenhagen studies are presented in Figure 2 (bottom row; study-specific estimates in Supplementary Table 13). In overall analyses, there was no association between genetically-predicted 25(OH)D and cardiovascular mortality (OR 1·01, 95% CI 0·95-1·08, p=0·71), cancer mortality (OR 0·98, 95% CI 0·93-1·02, p=0·29), or non-cardiovascular non-cancer mortality (OR 1·00, 95% CI 0·94-1.06, p=0·99). Similarly, null findings were obtained for mortality outcomes in all stratum-specific analyses.

A comparison between estimates from the doubly-ranked method and the residual method for primary outcomes in the UK Biobank is presented in Figure 3 (Supplementary Figure 8 for mortality outcomes). Estimates differ substantially at low levels of 25(OH)D, with the residual method giving estimates in the protective direction, with confidence intervals that exclude the null for stroke and all-cause mortality, and the doubly-ranked method giving estimates in the harmful direction for primary outcomes, although compatible with the null in all cases.

Figure 3. Stratified Mendelian randomization estimates for primary outcomes in UK Biobank from residual and doubly-ranked methods.

Figure 3

Open in a new tab

Estimates (95% confidence intervals) represent odds ratios per 10 nmol/L higher genetically-predicted concentration of 25(OH)D in strata of the population defined by residual concentration of 25(OH)D. Estimates from the residual method (unreliable when the genetic effect on the exposure vary in the population) are shown as blue triangles. Estimates from the doubly-ranked method (more reliable when the genetic effect on the exposure vary in the population) are shown as red triangles.

Discussion

In observational analyses, we found evidence for non-linear dose—response relationships of 25(OH)D concentrations with CVD and all-cause mortality. However, population-averaged estimates from our genetic analyses suggest that interventions to increase 25(OH)D concentrations are unlikely to translate into substantial risk reductions for CVD or all-cause mortality in the population overall. Similarly, genetic analyses conducted in individuals with low 25(OH)D concentrations provided no evidence supporting a causal relationship between 25(OH)D concentrations and cardiovascular or mortality outcomes. Our results suggest that substantial reductions in mortality and cardiovascular morbidity due to long-term low-dose vitamin D supplementation are unlikely even from trials targeted at individuals with low vitamin D status.

The majority of large prior trials for CVD and mortality were conducted in broadly-selected groups of the population, so had limited reliability to assess evidence for causality in individuals with low 25(OH)D concentrations. Most previous Mendelian randomization analyses did not consider estimates for strata of the population defined according to baseline 25(OH)D concentrations, and hence have not considered the shape of the causal relationship between 25(OH)D concentrations and CVD or all-cause mortality. Previous non-linear Mendelian randomization investigations conducted by ourselves18 and by others27 used the residual stratification method, which assumes that the effect of the genetic instrument on the exposure is linear and homogeneous in the population. These analyses suggested a protective causal effect of 25(OH)D at low 25(OH)D concentrations, as has been suggested for respiratory tract infections28. However, the residual method cannot reliably detect when the constant genetic effect assumption is violated19, and when it is violated, the method can produce biased results that reflect observational confounded associations rather than causal relationships17.

We have several reasons for greater trust in the updated null results from the doubly-ranked method over the previous results from the residual stratification method. First, genetic associations with the exposure varied strongly between strata, violating the assumption required by the residual method. There are several plausible reasons why genetic associations with 25(OH)D may be smaller in individuals with lower 25(OH)D concentrations: if genetic variants act via biological mechanisms relating to 25(OH)D synthesis or metabolism, then we may expect smaller magnitudes of effect in individuals with low levels of 25(OH)D synthesis and metabolism. Genetic variants may have similar proportional effects on 25(OH)D concentrations, rather than constant additive effects. Secondly, in the UK Biobank dataset, significant associations with confounders have been observed in 25(OH)D strata defined by the residual method17. This provides empirical evidence that the Mendelian randomization assumptions are violated in strata defined by the residual method, even if they are not violated in the overall population. Thirdly, a sensitivity analysis for the residual method log-transforming 25(OH)D concentrations before stratification provided substantially attenuated estimates in low 25(OH)D strata17. Log-transformation reduces the difference between genetic associations with 25(OH)D at the top and bottom of the distribution of 25(OH)D concentrations, and so should reduce bias due to violation of the constant genetic effect assumption. Fourthly, theoretical investigations have shown greater bias in estimates from the residual method compared to the doubly-ranked method in a wide range of simulated scenarios when the constant genetic effect assumption does not hold17,19. Finally, there were discrepancies in results from the residual method in our previous analysis; namely, the overall estimate for cardiovascular mortality was positive (but non-significant), even though all stratum-specific estimates were negative29. While this is possible for observational associations (it is known as “Simpson’s paradox”), this is a logical impossibility if all estimates have a causal interpretation. Hence, we have strong empirical and methodological reasons to doubt our previously published finding supporting a protective effect of 25(OH)D in low vitamin D status individuals18. However, all statistical methods, particularly those for inferring causal relationships, make untestable assumptions, which cannot be empirically verified. Hence, even for the doubly-ranked method, there remains intrinsic uncertainty in the validity of results: in the validity of the genetic variants as instrumental variables, in the assumptions required for non-linear Mendelian randomization, and so on.

Our revised investigation has several strengths. The Mendelian randomization design means that estimates are less susceptible to bias from confounding and reverse causation than those from conventional observational analyses. Our focused genetic instrument for vitamin D afforded strong statistical power and biological specificity, minimizing the potential for bias due to horizontal pleiotropy arising from use of variants that do not have specific effects on vitamin D pathways. The focused score was not associated with major cardiovascular risk factors, providing empirical evidence to support the Mendelian randomization assumptions. While the genetic instrument only explained a limited proportion of the variance in 25(OH)D levels, vitamin D supplementation in a randomized controlled trial would only explain a limited proportion of the variance in 25(OH)D levels. In the VITAL trial, supplementation only explained 26% of the variance in 25(OH)D levels4. While the limited proportion of variance explained affects the power to detect a causal effect, it does not preclude causal inferences either in a trial or a Mendelian randomization investigation.

However, there are also potential limitations. Firstly, the Mendelian randomization assumptions state that the only causal pathway from the genetic variants to the outcome is via 25(OH)D concentrations. While our variants are all from gene regions specifically relevant to vitamin D biology, variants in the CYP24A1 gene region are known to associate with circulating calcium levels. Secondly, while weaker than the assumptions required by the residual stratification method, non-linear Mendelian randomization analyses using the doubly-ranked method require additional assumptions beyond standard population-based Mendelian randomization analyses; namely the “rank preserving assumption” that the ranking of participants by their exposure values would be the same for all values of the genetic instrument. This assumption is generally plausible, but cannot be empirically tested. Thirdly, to reduce the scope for confounding by ethnicity (“population stratification”), our analyses were limited to middle-aged participants of European ancestries. This means that our findings may not be applicable to other populations. In particular, further analyses are needed to assess the potential effect of vitamin D supplementation in individuals with dark skin, as this correlates with lower 25(OH)D concentrations. Fourthly, UK Biobank and EPIC-CVD are not fully representative samples of the UK and European populations, further limiting the applicability of findings. Selection into these studies is dependent on age, sex, and other covariates, which can lead to bias in Mendelian randomization estimates30. While the effect of moderate selection bias on Mendelian randomization estimates is often slight, one specific concern in non-linear Mendelian randomization is that selection bias may affect stratum-specific estimates non-differentially; that is, bias may be greater in some strata than in others. Fifthly, fewer genetic variants were available in the Copenhagen studies, limiting comparability between datasets. However, while this will reduce power for analysis, it should not lead to bias. Sixthly, we do not have information from all studies on the accuracy of 25(OH)D measurements from external quality control programs. However, there was no indication that mean 25(OH)D estimates varied by assay type. As any such variation is likely non-differential to morbidity and mortality outcomes, it would bias the results toward the null. Seventhly, our primary genetic analyses for CVD considered both prevalent and incident events. Stratification into categories according to residual 25(OH)D concentration may therefore be affected by reverse causation. However, genetic associations with disease outcomes within each of the strata will not be affected by reverse causation, as genotype is fixed from conception. Finally, while our Mendelian randomization analyses provided confidence intervals that overlapped the null in all strata, it is not possible to prove a null finding; we may not have sufficient power to detect a small causal effect, particularly in the stratified analyses, as the sample size is naturally reduced for these analyses.

In conclusion, while observational analyses found a threshold association between 25(OH)D and cardiovascular and mortality outcomes, this was not confirmed in genetic analyses. This suggests the absence of a causal effect of 25(OH)D on cardiovascular and mortality outcomes, even at low 25(OH)D concentrations.

Supplementary Material

Supplementary material

Research in Context.

Evidence before this study

We searched for randomized trials investigating the effects of vitamin D supplementation on cardiovascular disease or all-cause mortality published in any language up until April 16, 2021 (with no specified earliest date), in PubMed, Scientific Citation Index Expanded, and Embase using the search terms listed in the appendix. However, no study had combined the following key features required to achieve reliable estimates of the effect of vitamin D supplementation on cardiovascular disease or mortality in vitamin D deficient individuals: availability of individual-participant data on large sample sizes (e.g., >1000 individuals with 25(OH)D <25 nmol/L) with moderate numbers of cardiovascular disease or all-cause mortality events (e.g., > 100 events). In summary, large randomized controlled trials have not been able to demonstrate beneficial effects of vitamin D supplementation on risk of cardiovascular disease or mortality.

Added value of this study

We characterized dose—response relationships for 25-hydroxyvitamin D [25(OH)D], a metabolite used as a clinical indicator of vitamin D status, with cardiovascular disease and mortality in the largest such epidemiological analysis performed to date, using statistical methods tailored to the investigation of non-linear risk factor—disease relationships. Our observational analyses indicate threshold relationships between 25(OH)D concentrations and risk of both cardiovascular disease and all-cause mortality. However, genetic analyses did not support a causal effect of 25(OH)D concentrations on cardiovascular disease or mortality at any level of 25(OH)D concentrations.

Implications of all the available evidence

Taken together, these findings suggest that long-term low-dose vitamin D supplementation is unlikely to reduce cardiovascular disease or mortality risk even in individuals with low 25(OH)D levels.

Acknowledgements

We thank all EPIC participants and staff for their contribution to the study, the laboratory teams at the Medical Research Council Epidemiology Unit for sample management, Cambridge Genomic Services for genotyping, Matthew Walker and Sarah Spackman (BHF Cardiovascular Epidemiology Unit) for data management, Nicola Kerrison (MRC Epidemiology Unit, Cambridge) for managing the data for the InterAct Project, the laboratory team at VITAS Analytical Services, especially Ms. Anne Marte Haug, for vitamin D assays in EPIC-CVD, and the team at the EPIC-CVD Coordinating Centre for study coordination and administration. We acknowledge the use of data and biological samples from the EPIC-Ragusa cohort and the EPIC-Asturias cohort. A full list of Emerging Risk Factors Collaboration / Vitamin D Studies Collaboration investigators is provided in the Supplementary Materials. For the purpose of open access, the author has applied a CC-BY public copyright licence to any Author Accepted Manuscript version arising from this submission.

This research has been conducted using the UK Biobank Resource under Application Number 11833.

Sources of funding:

No specific funding was obtained for this manuscript. EPIC-CVD was funded by the European Research Council (268834) and the European Commission Framework Programme 7 (HEALTH-F2-2012-279233). The coordinating centre was supported by core funding from the: UK Medical Research Council (G0800270; MR/L003120/1), British Heart Foundation (SP/09/002; RG/13/13/30194; RG/18/13/33946) and NIHR Cambridge Biomedical Research Centre (BRC-1215-20014). The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.

AMM is funded by the EC-Innovative Medicines Initiative (BigData@Heart). LS is Funded by a BHF Programme Grant (RG/18/13/33946), TRB is Funded by the NIHR Blood and Transplant Research Unit in Donor Health and Genomics (NIHR BTRU-2014-10024) and a BHF Programme Grant (RG/18/13/33946). SB is supported by the Wellcome Trust (225790/Z/22/Z), and by the UKRI Medical Research Council (MC_UU_00002/7). RC is funded by a BHF PhD studentship (FS/18/56/34177). EA was funded by the EC-Innovative Medicines Initiative (BigData@Heart) and a BHF Programme Grant (RG/18/13/33946). JD holds a British Heart Foundation Professorship and a NIHR Senior Investigator Award. TJ is funded by the NIHR Cambridge Biomedical Research Centre (BRC-1215-20014). AMW is supported by a BHF-Turing Cardiovascular Data Science Award and by the EC-Innovative Medicines Initiative (BigData@Heart). LAMP is funded by a BHF Programme Grant (RG/18/13/33946). TYNT is supported by a Nuffield Department of Population Health Intermediate Fellowship. MA is funded by a BHF Programme Grant (RG/18/13/33946). NJW, CL and NGF acknowledge funding from the Medical Research Council Epidemiology Unit MC_UU_12015/1 and MC_UU_12015/5. FS is funded by the Chinese Scholarship Council, and Cambridge Commonwealth European and International Trust. NJW and NGF received support from the NIHR Biomedical Research Centre Cambridge: Nutrition, Diet, and Lifestyle Research Theme (IS-BRC-1215-20014). DG is supported by the British Heart Foundation Centre of Research Excellence (RE/18/4/34215) at Imperial College London and a National Institute for Health Research Clinical Lectureship (CL-2020-16-001) at St George's, University of London, CM is supported by the Federation Française de Cardiologie, RZ-R was supported by the “Miguel Servet type II” program (CPII20/00009) from the Institute of Health Carlos III (Co-funded by the European Social Fund (ESF) - ESF investing in your future). YBS is partially funded by the NIHR School of Population Health Research.

The InterAct project was funded by the EU FP6 programme (grant number LSHM_CT_2006_037197). Measurements for 25(OH)D were funded jointly by the InterAct project, the EPIC-CVD project and the MRC Cambridge Initiative (RG71466, SJAH/004). EPIC-Oxford is supported by Cancer Research UK (C8221/A29017) and UK Medical Research Council (MR/M012190/1). The Longitudinal Aging Study Amsterdam (LASA) is supported by a grant from the Netherlands Ministry of Health, Welfare and Sport, Directorate of Long-Term Care. The Multi-Ethnic Study of Atherosclerosis is supported by grant R01HL096875 and contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159,75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS). The Osteoporotic Fractures in Men (MrOS) Study is supported by National Institutes of Health funding. The following institutes provide support: the National Institute on Aging (NIA), the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), the National Center for Advancing Translational Sciences (NCATS), and NIH Roadmap for Medical Research under the following grant numbers: U01 AG027810, U01 AG042124, U01 AG042139, U01 AG042140, U01 AG042143, U01 AG042145, U01 AG042168, U01 AR066160, and UL1 TR000128. MrOS data is available online: http://mrosdata.sfcc-cpmc.net. The Study of Osteoporotic Fractures (SOF) is supported by National Institutes of Health funding. The National Institute on Aging (NIA) provides support under the following grant numbers: R01 AG005407, R01 AR35582, R01 AR35583, R01 AR35584, R01 AG005394, R01 AG027574, and R01 AG027576. The Study of Health in Pomerania-1 (SHIP-1) is part of the Community Medicine Research network of the University of Greifswald, Germany, which is funded by the Federal Ministry of Education and Research (grants 01ZZ9603, 01ZZ0103 and 01ZZ0403), the Ministry of Cultural Affairs, as well as the Social Ministry of the Federal State of Mecklenburg–West Pomerania.

The coordination of EPIC is financially supported by International Agency for Research on Cancer (IARC) and also by the Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London which has additional infrastructure support provided by the NIHR Imperial Biomedical Research Centre (BRC). The national cohorts are supported by: Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l’Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM) (France), German Cancer Aid, German Cancer Research Center (DKFZ), German Institute of Human Nutrition Potsdam-Rehbruecke (DIfE), Federal Ministry of Education and Research (BMBF) (Germany); Associazione Italiana per la Ricerca sul Cancro-AIRC-Italy, Compagnia di SanPaolo and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund (WCRF), Statistics Netherlands (The Netherlands); Austrian Research Promotion Agency (Austria); Health Research Fund (FIS) - Instituto de Salud Carlos III (ISCIII), Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, and the Catalan Institute of Oncology - ICO (Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden); Cancer Research UK (14136 to EPIC-Norfolk; C8221/A29017 to EPIC-Oxford), Medical Research Council (1000143 to EPIC-Norfolk; MR/M012190/1 to EPIC-Oxford) (United Kingdom).

Footnotes

Authors and contributors:

Conception and design: ES, SKK, EA, MGA, AMM, RC, LAMP, LS, PW, JD, AMW, EDA, ASB, SB; Acquisition and interpretation of data: all authors; Performed analyses: ES, SKK, SA, TJ, TRB, SB; Drafted the manuscript: ES, SKK, DG, AMW, EDA, ASB, JD, SB; Critically revised paper: all authors; Approved of final submission: all authors. ES, SKK, SA, and SB have verified the underlying data. SB is responsible for the decision to submit the manuscript. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of interest:

The authors declare the following conflicts of interest: AB reports grants outside of this work from AstraZeneca, Biogen, BioMarin, Bioverativ, Merck, Novartis, Pfizer and Sanofi and personal fees from Novartis; JD reports grants, personal fees and non-financial support from Merck Sharp & Dohme (MSD), grants, personal fees and non-financial support from Novartis, grants from Pfizer and grants from AstraZeneca outside the submitted work; YÇ reports personal fees from Boehringer Ingelheim, AstraZeneca, and Sanofi Genzyme outside the submitted work; BGN reports consultancies and talks sponsored by AstraZeneca, Sanofi, Regeneron, Akcea, Amgen, Kowa, Denka, Amarin, Novartis, Novo Nordisk, Esperion, and Silence Therapeutics outside submitted work; MA is now an employee of AstraZeneca; DG is employed part-time by Novo Nordisk. PW reports grant income from Roche Diagnostics, AstraZeneca, Boehringer Ingelheim, and Novartis, outside the submitted work. NS reports personal fees from Afimmune, Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Hanmi Pharmaceuticals, Merck Sharp and Dohme, Novartis, Novo Nordisk, Pfizer and Sanofi, and grant funding paid to his institution from AstraZeneca, Boehringer Ingelheim and Roche Diagnostics. All other authors report no relevant conflict of interest.

Data sharing

Data from UK Biobank is available to any bona fide scientific research on application. Applications to access data from EPIC-CVD should be addressed to the steering committee: https://www.phpc.cam.ac.uk/ceu/epic_cvd/. Data from the Vitamin D Studies Collaboration is available at the discretion of the principal investigators of the individual studies.

References

  • 1.Cashman KD, Dowling KG, Škrabáková Z, et al. Vitamin D deficiency in Europe: pandemic? The American Journal of Clinical Nutrition. 2016;103(4):1033–44. doi: 10.3945/ajcn.115.120873. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Wang TJ, Pencina MJ, Booth SL, et al. Vitamin D deficiency and risk of cardiovascular disease. Circulation. 2008;117(4):503–11. doi: 10.1161/CIRCULATIONAHA.107.706127. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Heath AK, Kim IY, Hodge AM, English DR, Muller DC. Vitamin D status and mortality: A systematic review of observational studies. International Journal of Environmental Research and Public Health. 2019;16(3):383. doi: 10.3390/ijerph16030383. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Manson JE, Cook NR, Lee I-M, et al. Vitamin D Supplements and Prevention of Cancer and Cardiovascular Disease. New England Journal of Medicine. 2019;380(1):33–44. doi: 10.1056/NEJMoa1809944. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Zittermann A, Ernst JB, Prokop S, et al. Effect of vitamin D on all-cause mortality in heart failure (EVITA): a 3-year randomized clinical trial with 4000 IU vitamin D daily. European Heart Journal. 2017;38(29):2279–86. doi: 10.1093/eurheartj/ehx235. [DOI] [PubMed] [Google Scholar]
  • 6.Scragg R, Stewart AW, Waayer D, et al. Effect of Monthly High-Dose Vitamin D Supplementation on Cardiovascular Disease in the Vitamin D Assessment Study : A Randomized Clinical Trial. JAMA Cardiol. 2017;2(6):608–16. doi: 10.1001/jamacardio.2017.0175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Zhang Y, Fang F, Tang J, Jia L, Feng Y, Xu P, Faramand A. Brit Med J. Vol. 366. l4673: 2019. Association between vitamin D supplementation and mortality: systematic review and meta-analysis. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Thanassoulis G, O’Donnell CJ. Mendelian randomization: nature’s randomized trial in the post-genome era. JAMA. 2009;301(22):2386–8. doi: 10.1001/jama.2009.812. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Brøndum-Jacobsen P, Benn M, Afzal S, Nordestgaard BG. No evidence that genetically reduced 25-hydroxyvitamin D is associated with increased risk of ischaemic heart disease or myocardial infarction: a Mendelian randomization study. International Journal of Epidemiology. 2015;44(2):651–61. doi: 10.1093/ije/dyv078. [DOI] [PubMed] [Google Scholar]
  • 10.Manousaki D, Mokry LE, Ross S, Goltzman D, Richards JB. Mendelian randomization studies do not support a role for vitamin D in coronary artery disease. Circulation: Cardiovascular Genetics. 2016;9(4):349–56. doi: 10.1161/CIRCGENETICS.116.001396. [DOI] [PubMed] [Google Scholar]
  • 11.Larsson SC, Traylor M, Mishra A, Howson JMM, Michaëlsson K, Markus HS, Consortium MPotISG Serum 25-Hydroxyvitamin D Concentrations and Ischemic Stroke and Its Subtypes. Stroke. 2018;49(10):2508–11. doi: 10.1161/STROKEAHA.118.022242. [DOI] [PubMed] [Google Scholar]
  • 12.Huang T, Afzal S, Yu C, et al. Vitamin D and cause-specific vascular disease and mortality: a Mendelian randomisation study involving 99,012 Chinese and 106,911 European adults. BMC Medicine. 2019;17(1):160. doi: 10.1186/s12916-019-1401-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Afzal S, Brøndum-Jacobsen P, Bojesen SE, Nordestgaard BG. Genetically low vitamin D concentrations and increased mortality: mendelian randomisation analysis in three large cohorts. BMJ : British Medical Journal. 2014;349:g6330. doi: 10.1136/bmj.g6330. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Ye Y, Yang H, Wang Y, Zhao H. A comprehensive genetic and epidemiological association analysis of vitamin D with common diseases/traits in the UK Biobank. Genetic Epidemiology. 2021;45(1):24–35. doi: 10.1002/gepi.22357. [DOI] [PubMed] [Google Scholar]
  • 15.Zhang R, Li B, Gao X, et al. Serum 25-hydroxyvitamin D and the risk of cardiovascular disease: dose-response meta-analysis of prospective studies. Am J Clin Nutr. 2017;105(4):810–9. doi: 10.3945/ajcn.116.140392. [DOI] [PubMed] [Google Scholar]
  • 16.Crowe FL, Thayakaran R, Gittoes N, Hewison M, Thomas GN, Scragg R, Nirantharakumar K. Non-linear associations of 25-hydroxyvitamin D concentrations with risk of cardiovascular disease and all-cause mortality: Results from The Health Improvement Network (THIN) database. The Journal of steroid biochemistry and molecular biology. 2019;195:105480. doi: 10.1016/j.jsbmb.2019.105480. [DOI] [PubMed] [Google Scholar]
  • 17.Burgess S. Violation of the constant genetic effect assumption can result in biased estimates for non-linear Mendelian randomization. Human Heredity. 2023 doi: 10.1159/000531659. (in press) [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Sofianopoulou E, Kaptoge SK, Afzal S, et al. Estimating dose-response relationships for vitamin D with coronary heart disease, stroke, and all-cause mortality: observational and Mendelian randomisation analyses. Lancet Diabetes & Endocrinology. 2021;9(12):837–46. doi: 10.1016/S2213-8587(21)00263-1. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  • 19.Tian H, Mason AM, Liu C, Burgess S. Relaxing parametric assumptions for non-linear Mendelian randomization using a doubly-ranked stratification method. PLOS Genetics. 2023;19(6):e1010823. doi: 10.1371/journal.pgen.1010823. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Sudlow C, Gallacher J, Allen N, et al. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLOS Medicine. 2015;12(3):e1001779. doi: 10.1371/journal.pmed.1001779. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Danesh J, Saracci R, Berglund G, et al. EPIC-Heart: the cardiovascular component of a prospective study of nutritional, lifestyle and biological factors in 520,000 middle-aged participants from 10 European countries. European Journal of Epidemiology. 2007;22(2):129. doi: 10.1007/s10654-006-9096-8. [DOI] [PubMed] [Google Scholar]
  • 22.Langenberg C, Sharp S, Forouhi NG, et al. Design and cohort description of the InterAct Project: an examination of the interaction of genetic and lifestyle factors on the incidence of type 2 diabetes in the EPIC Study. Diabetologia. 2011;54(9):2272–82. doi: 10.1007/s00125-011-2182-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Appleyard M. The Copenhagen City Heart Study. Østerbroundersøgelsen. A book of tables with data from the first examination (1976-78) and a five year follow-up (1981-83) Scandinavian Journal of Social Medicine. 1989;41:1–160. [PubMed] [Google Scholar]
  • 24.Nordestgaard BG, Palmer TM, Benn M, Zacho J, Tybjaerg-Hansen A, Smith GD, Timpson NJ. The effect of elevated body mass index on ischemic heart disease risk: causal estimates from a Mendelian randomisation approach. PLOS Medicine. 2012;9(5):e1001212. doi: 10.1371/journal.pmed.1001212. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Wang TJ, Zhang F, Richards JB, et al. Common genetic determinants of vitamin D insufficiency: a genome-wide association study. Lancet. 2010;376(9736):180–8. doi: 10.1016/S0140-6736(10)60588-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Manousaki D, Mitchell R, Dudding T, et al. Genome-wide association study for vitamin D levels reveals 69 independent loci. The American Journal of Human Genetics. 2020;106(3):327–37. doi: 10.1016/j.ajhg.2020.01.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Zhou A, Selvanayagam JB, Hyppönen E. Non-linear Mendelian randomization analyses support a role for vitamin D deficiency in cardiovascular disease risk. European Heart Journal. 2022;43(18):1731–9. doi: 10.1093/eurheartj/ehab809. [DOI] [PubMed] [Google Scholar]
  • 28.Martineau AR, Jolliffe DA, Hooper RL, et al. Vitamin D supplementation to prevent acute respiratory tract infections: systematic review and meta-analysis of individual participant data. British Medical Journal. 2017;356:i6583. doi: 10.1136/bmj.i6583. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Davey Smith G. Mendelian randomisation and vitamin D: the importance of model assumptions. The Lancet Diabetes and Endocrinology. 2023;11(1):P14. doi: 10.1016/S2213-8587(22)00345-X. [DOI] [PubMed] [Google Scholar]
  • 30.Gkatzionis A, Burgess S. Contextualizing selection bias in Mendelian randomization: how bad is it likely to be? International Journal of Epidemiology. 2018;48(3):691–701. doi: 10.1093/ije/dyy202. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary material
EMS193231-supplement-Supplementary_material.docx (934.2KB, docx)

Data Availability Statement

Data from UK Biobank is available to any bona fide scientific research on application. Applications to access data from EPIC-CVD should be addressed to the steering committee: https://www.phpc.cam.ac.uk/ceu/epic_cvd/. Data from the Vitamin D Studies Collaboration is available at the discretion of the principal investigators of the individual studies.

RESOURCES