Abstract
Background: Low plasma 25-hydroxyvitamin D [p-25(OH)D] is associated with increased risk of ischaemic heart disease and with the subgroup myocardial infarction. However, whether this association is causal or due to confounding or reverse causation is presently unknown. We tested the hypothesis that genetically reduced plasma 25(OH)D is associated with increased risk of ischaemic heart disease and myocardial infarction.
Methods: We used a Mendelian randomization design in the Copenhagen City Heart Study, the Copenhagen General Population Study, and the Copenhagen Ischaemic Heart Disease Study. Two 25(OH)D reducing genetic variants in the DCHR7 gene (rs7944926 and rs11234027) and two in the CYP2R1 gene (rs10741657 and rs12794714) were genotyped in 92 416 participants of Danish descent, of whom 14 455 developed ischaemic heart disease (ICD-8:410‐414; ICD-10:I20-I25) and 7061 myocardial infarction (ICD-8:410: ICD-10:I21-I22) from 1977 through 2011. P-25(OH)D was measured in 36 089 participants. APOE genotype was included as a positive control for risk of ischaemic heart disease.
Results: The multivariable adjusted hazard ratios for lowest vs highest quartile of 25(OH)D were 1.82 [95% confidence interval (CI): 1.42–2.32] for ischaemic heart disease. Each allele increase in a combined allele score was associated with a 1.9-nmol/l decrease in p-25(OH)D (P = 7 × 10−55; R2 = 0.9%). The genetic variants were, however, not associated with increased risk of ischaemic heart disease. In instrumental variable analysis, the odds ratio for ischaemic heart disease for a genetically 25-nmol/l decrease in p-25(OH)D was 0.98 (95% CI: 0.76‐1.26), with a corresponding observational hazard ratio by Cox regression of 1.07 (1.01–1.13). Similarly, with myocardial infarction as the outcome, observational analyses suggested an increased risk with lower 25(OH)D, whereas genetic analyses suggested no causal effect. For APOE genotype, the odds ratio for ischaemic heart disease for a 1-mmol/l genetic increase in plasma total cholesterol concentrations was 1.23 (1.08–1.41), with a corresponding observational hazard ratio of 1.08 (1.04–1.14).
Conclusion: We found no evidence to suggest that genetically reduced p-25(OH)D is associated with increased risk of ischaemic heart disease or myocardial infarction.
Key Messages
In epidemiological studies, low p-25(OH)D concentrations have been associated with increased risk of ischaemic heart disease. However, whether this association is causal is currently unclear.
A Mendelian randomization study approach can be used to make causal inference where such epidemiological associations may be confounded or due to reverse causation.
In the present study, we found no evidence to support that genetically reduced p-25(OH)D is associated with increased risk of ischaemic heart disease.
Introduction
Low p-25(OH)D concentrations have in numerous epidemiological studies been associated with increased risk of ischaemic heart disease and the subgroup myocardial infarction.1,–3 However, it is still unclear whether the observational association is causal or due to confounding or reverse causation. This question is an important public health issue, as ischaemic heart disease is a leading cause of death globally. The question is likewise important because the sale of vitamin D supplements is increasing annually without proper scientific evidence for an advantageous effect on the many diseases supplementation is claimed to prevent, including ischaemic heart disease. Although randomized clinical trials of vitamin D supplementation to reduce the risk of ischaemic heart disease are ongoing,4 trials of vitamin D supplementation have mainly been conducted to study bone health as the primary outcome,5 and meta-analyses of such studies have not been able to demonstrate a simultaneous cardioprotective effect.6,7 Also, randomized clinical trials with vitamin D supplementation have often been conducted with concomitant calcium supplementation. This makes it difficult to interpret the trials with respect to the isolated effect of vitamin D supplementation on risk of ischaemic heart disease: calcium supplementation has been shown to increase the risk of cardiovascular disease, especially myocardial infarction.8
With the lack of randomized controlled trial evidence, a Mendelian randomization study approach can potentially be used to make causal inference where epidemiological associations may be confounded or due to reverse causation.9 Mendelian randomization refers to the random assortment of genes at the time of gamete formation. As a consequence of this randomization, genetic variants known to influence p-25(OH)D concentrations can be used to study risk of ischaemic heart disease and myocardial infarction without reverse causation and largely without confounding.10,11 Nevertheless, there are several assumptions underlining the Mendelian randomization approach, including that: (i) the genotypes are associated with the exposure of interest; (ii) the genotypes are randomly distributed an thus independent of confounding factors; and (iii) the genotypes affect the outcome only through the exposure, e.g. no pleiotropy. These assumptions can partly be tested for; however, it is naturally not possible completely to exclude, for example, pleotropic effects of the genetic instrument.
We tested the hypothesis that genetically reduced p-25(OH)D is associated with increased risk of ischaemic heart disease and with the subgroup myocardial infarction. First, we tested whether low p-25(OH)D concentrations are associated with ischaemic heart disease and myocardial infarction observationally. Second, we tested whether genetic variants are associated with reduced p-25(OH)D concentrations. Third, we tested whether such genetic variants also are associated with increased risk of ischaemic heart disease and myocardial infarction. Finally, using instrumental variable analysis, we tested whether genetically reduced 25(OH)D concentrations were associated with increased risk of ischaemic heart disease and myocardial infarction, and compared the estimates with corresponding observational estimates using observational epidemiology. To demonstrate that the risk of ischaemic heart disease in the study cohorts follows established patterns and to test the predictive power, we included the APOE genotype as a positive control of the association with risk of ischaemic heart disease.
Methods
The studies were approved by Danish ethical committees (No. KF-V.100.2039/91, H-KF-01‐144/01, KA-93125, and KA-99039), and were conducted according to the Declaration of Helsinki. Written informed consent was obtained from participants. All participants were White and of Danish descent. Consanguinity may be present in some individuals in the study cohorts.
Study cohorts
The Copenhagen City Heart Study (CCHS) is a prospective study of the Danish general population initiated in 1976–78, with follow-up examinations in 1981–83, 1991–94 and 2001–03.12,13 The Copenhagen General Population Study (CGPS) is likewise a prospective study of the Danish general population initiated in 2003 with ongoing enrolment.14,15 The Copenhagen Ischaemic Heart Disease Study (CIHDS) comprises a total of 6805 participants referred for coronary angiography in 1991 through 2012 and with documented evidence of ischaemic heart disease. For details on study cohorts and information on included participants and availability on their measurements, see Supplementary Methods and Supplementary Figure 1 (available as Supplementary data at IJE online).
Outcomes
In all three studies, information on diagnoses of ischaemic heart disease (World Health Organization; International Classification of Diseases, 8th edition codes 410 to 414, and 10th edition codes I20 to I25) and myocardial infarction (ICD-8: 410; ICD-10: I21-I22) were collected and verified by reviewing hospital admissions and diagnoses entered in the national Danish Patient Registry, causes of death entered in the national Danish Causes of Death Registry, and medical records from hospitals and general practitioners. Ischaemic heart disease comprised angina pectoris and myocardial infarction, the latter based on characteristic chest pain, electrocardiographic changes and/or elevated cardiac markers following the changes in diagnostic criteria overtime.16
Biochemical analysis
25(OH)D concentrations were measured in plasma samples using a competitive chemiluminescent immunoassay, CLIA (DiaSorin, Stillwater, MN, USA) with intra- and inter-assay coefficients of variation of 10% and 11%, respectively, at concentrations of ∼40 nmol/l; 1.7% of samples had concentrations below the lower detection limit of 10 nmol/l and were assigned a concentration of 7.5 nmol/l for statistical analyses. We used plasma samples because these were available in both studies; however, 25(OH)D can be measured in both plasma and serum with the assay used in the present study. P-25(OH)D concentrations were measured during January 2010 through April 2012, using samples stored at −20°C (CCHS; samples from 1981–83), or −80°C (CGPS; samples from 2003–04) without previous thawing or exposure to sunlight, or measured in fresh samples (CGPS; samples from 2010–12). Colorimetric assays (Boehringer Mannheim, Mannheim, Germany or Konelab, Espoo, Finland) were used to measure total cholesterol, high-density lipoprotein (HDL) cholesterol and creatinine concentrations in plasma. Biochemical assays were followed for precision daily and for accuracy monthly using a Scandinavian external quality assessment program.
Genotypes
Genetic variants were selected based on the effect sizes of influence on p-25(OH)D concentrations reported in genome-wide association studies17,18 and from knowledge about the functionality of the gene products,19,20 either influencing plasma 25-hydroxyvitamin D concentrations via synthesis of pre-vitamin D from 7-dehydrocholesterol in the skin (marking endogenous production) or activation of vitamin D to 25(OH)D in the liver (marking liver conversion). We deliberately did not include polymorphisms in the vitamin D-binding protein, since we wanted to study polymorphisms influencing bioavailability of vitamin D, not just total plasma levels or affinity/functionality of the vitamin D-binding protein leading to unpredictable biological effects on the bioavailability of vitamin D. A flowchart indicating the selection procedure for the genetic variants included in the study is shown in Supplementary Figure 2 (available as Supplementary data at IJE online). Genotyping for genetic variants close to DHCR7 (rs7944926 and rs11234027) and close to CYP2R1 (rs10741657 and rs12794714) and for genetic variants within APOE (rs429358 and rs7412) were by TaqMan assays, the latter used as a positive control. Genotypes were verified by sequencing of randomly selected samples. After performing re-runs, all genotype call rates were more than 99% complete. Genotype distributions in all study cohorts were in Hardy–Weinberg equilibrium.
Other covariates
Covariates for adjustments were chosen based on known influence on p-25(OH)D concentrations or on risk of ischaemic heart disease and were assessed similarly in the three cohorts. Body mass index was expressed as measured weight in kilograms divided by measured height in metres squared. Weight was measured without shoes and in light clothing to the nearest 0.1 kg on Soehnle Professional scales and height was measured to the nearest 0.1 cm with a Seca stadiometer. Physical activity was during leisure time in four categories. Smoking status was classified as pack-years smoked. Alcohol consumption was based on average weekly consumption of beer, wine and liquor in units: 1 unit alcohol = ∼12 g. Systolic blood pressure was measured in mmHg with a sphygmomanometer or an automated digital blood pressure monitor (Kivex) after 5 min of rest with the individual in the sitting position. Diabetes mellitus was non-fasting blood glucose concentrations ≥11.0 mmol/l, and/or use of insulin or oral antidiabetic treatment. Estimated glomerular filtration rate21 was low (<60 ml/min per 1.73−m2) or high (≥6 0 ml/min per 1.73−m2). Hormone replacement therapy and lipid-lowering therapy were self-reported.
Statistical analyses
We used STATA/SE version 12.0 and NCSS-PASS. For trend tests, we used the Cuzick extension of the Wilcoxon rank sum test, in which groups of participants (classified by seasonally adjusted p-25(OH)D quartiles, genotypes, number of p-25(OH)D reducing alleles, or APOE genotype combination) were ranked according to decreasing p-25(OH)D concentrations (or increasing cholesterol for APOE) and coded as 0, 1, 2, 3, 4 etc. Alternative models using weighted allele scores gave similar results; results from a simpler unweighted allele score were reported. To adjust for seasonal variation in p-25(OH)D concentrations, we assigned participants to percentiles of p-25(OH)D concentrations by month of sample collection in each study.22 Participants were grouped according to p-25(OH)D concentrations into quartiles, with the highest quartile as the reference group.
First, in the prospective CCHS including 10 170 participants with a p-25(OH)D measurement, Cox proportional hazard regression models with age as a time scale, implying that age is automatically adjusted for, and the use of left truncation (delayed entry) at 25-hydroxyvitamin D measurement at study entry, were used to examine the association between seasonally adjusted quartile groups of p-25(OH)D concentrations and risk of ischaemic heart disease and the subgroup myocardial infarction. Totals of 3100 cases of ischaemic heart disease and 1625 cases of myocardial infarction were identified. Multivariable Cox regression analyses included age, body mass index, smoking in pack-years, alcohol consumption, total cholesterol concentration, HDL cholesterol concentration and systolic blood pressure on a continuous scale; and gender, physical activity during leisure time, diabetes mellitus, hormone replacement therapy, lipid-lowering therapy and estimated glomerular filtration as categorical covariates. Hazard ratios including 95% confidence intervals were corrected for regression dilution bias using a non-parametric method.23 We also analysed the data using a restricted cubic spline model.
Second, to test whether the individual genotypes or combined allele scores were associated with reduced plasma 25-hydroxyvitamin D concentrations, per allele effects of genotypes or the effects of number of p-25(OH)D reducing alleles in allele scores were calculated in the CCHS and CGPS combined, using one-way analysis of variance. We compared consistency of effect of each individual genetic variant on p-25(OH)D levels by visual inspection. A total of 31 158 participants (5459participants from the CCHS and 25 699 from the CGPS) had both p-25(OH)D measurement and genotype data available.
Third, the associations between individual genotypes or combined allele scores and observed risk for ischaemic heart disease and myocardial infarction were by logistic regression estimating the odds ratio in the CCHS, CGPS and CIHDS combined to obtain maximal power; both prevalent and incident cases were included as genotypes are present at birth and therefore precede all events. A total of 92 416 participants had genotype data available. In total, 14 455 cases of ischaemic heart disease and 7061 cases of myocardial infarction were identified in the three cohorts. Theoretically predicted risk was estimated from changes in p-25(OH)D concentrations and the known observational associations of plasma 25-hydroxyvitamin D concentrations in the CCHS, as done previously.24 As a positive control, similar calculations were made for APOE genotype, total cholesterol concentrations and risk of ischaemic heart disease. We performed power calculations to estimate the minimal detectable odds ratio with a one-sided P-value < 0.05 and with a 90% statistical power in the combined studies
Fourth, we performed an instrumental variable analysis with a two-stage least squared regression model using each genotype individually and combined allele scores as instruments to estimate the influence of a 10 nmol/l, a 25 nmol/l and a 50% genetic reduction in p-25(OH)D concentrations on risk of ischaemic heart disease and myocardial infarction, respectively. We calculated instrumental variable estimates of genetically determined odds ratios by using the Wald-type estimator, which involves taking the ratio of the outcome allele score log odds ratio to the exposure allele score coefficient and then exponentiating to express it as an odds ratio.11 We used Fieller’s theorem to derive the confidence intervals of Wald-type instrumental variable log odds ratios. For APOE genotype, we had complete data on total cholesterol concentrations, and thus used the control function to calculate instrumental variable estimates.
We also performed sensitivity analyses using a logistic regression model with a weighted allele score and a probit model as the second stage regression.
The observational analyses were adjusted for sampling month, whereas the genetic models including the instrumental variable analyses were not so adjusted. For further details on statistical analyses please refer to Supplementary Methods (available as Supplementary data at IJE online).
Results
Baseline characteristics of participants in the CCHS are shown by p-25(OH)D quartiles (Table 1), in the CCHS and the CGPS by vitamin D-lowering allele score (Supplementary Table 1) and by disease status (Supplementary Table 2, available as Supplementary data at IJE online). Collectively, these data show that all baseline characteristics are strongly associated with 25(OH)D quartiles and/or disease status, but not with allele score (compare P-values in Table 1). This illustrates that the allele score can be used as a largely unconfounded instrument for reduced 25(OH)D concentrations on the risk of ischaemic heart disease and myocardial infarction, at least for the measured potential confounders. The distribution of 25-hydroxyvitamin D concentrations in the CCHS and the CGPS are shown in Supplementary Figure 3 (available as Supplementary data at IJE online). The number of participants with 25(OH)D measurements, with genotyping data, or with both in the three cohorts are shown in Supplementary Table 3 (available as Supplementary data at IJE online).
Baseline characteristics of participants from the Danish general population, the Copenhagen City Heart Study
| Plasma 25-hydroxyvitamin D quartiles | |||||||
|---|---|---|---|---|---|---|---|
| 4th | 3rd | 2nd | 1st | Ptrend | Pcomparison, IHD | Ptrend, allele score | |
| Number of participants | 2,595 | 2,490 | 2,532 | 2,553 | |||
| Age, years | 51(43–59) | 51(43–59) | 52(44–59) | 53(45–60) | 1 × 10−4 | <1 × 10−300 | 0.11 |
| Women, % | 59 | 56 | 55 | 55 | 5 × 10−3 | 7 × 10−185 | 0.50 |
| Body mass index, kg/m2 | 24(22–26) | 25(22–27) | 26(23–28) | 26(25–29) | 7 × 10−36 | 3 × 10−127 | 0.57 |
| High physical leisure time activity, % | 31 | 30 | 27 | 22 | 3 × 10−34 | 7 × 10−185 | 0.60 |
| Smoking, pack-years | 19(3–29) | 19(3–29) | 21(7–30) | 24(10–32) | 1 × 10−31 | <1 × 10−300 | 0.58 |
| Alcohol consumption, units/week | 8(2–11) | 8(2–11) | 9(1–11) | 9(0–12) | 2 × 10−7 | 1 × 10−16 | 0.17 |
| Total cholesterol, mmol/l | 6.0(5.2–6.7) | 6.1(5.2–6.8) | 6.2(5.3–6.8) | 6.2(5.3–6.9) | 4 × 10−8 | 2 × 10−11 | 0.32 |
| HDL cholesterol, mmol/l | 1.2(1.0–1.4) | 1.2(0.9–1.3) | 1.1(0.9–1.3) | 1.1(0.9–1.3) | 1 × 10−22 | <1 × 10−300 | 0.01 N.S. |
| Systolic blood pressure, mmHg | 134(121–146) | 136(121–149) | 138(122–150) | 139(123–152) | 3 × 10−12 | 1 × 10−160 | 0.07 |
| Diabetes mellitus, % | 1 | 2 | 2 | 3 | 6 × 10−7 | 7 × 10−156 | 0.75 |
| Hormone replacement therapy, %a | 22 | 18 | 17 | 13 | 7 × 10−9 | 1 × 10−27 | 0.60 |
| Lipid-lowering therapy, % | 0 | 0 | 0 | 0 | 1.00 | <1 × 10−300 | 0.97 |
| Low eGFR, % | 14 | 13 | 16 | 15 | 0.20 | 1 × 10−198 | 0.13 |
| Plasma 25-hydroxyvitamin D quartiles | |||||||
|---|---|---|---|---|---|---|---|
| 4th | 3rd | 2nd | 1st | Ptrend | Pcomparison, IHD | Ptrend, allele score | |
| Number of participants | 2,595 | 2,490 | 2,532 | 2,553 | |||
| Age, years | 51(43–59) | 51(43–59) | 52(44–59) | 53(45–60) | 1 × 10−4 | <1 × 10−300 | 0.11 |
| Women, % | 59 | 56 | 55 | 55 | 5 × 10−3 | 7 × 10−185 | 0.50 |
| Body mass index, kg/m2 | 24(22–26) | 25(22–27) | 26(23–28) | 26(25–29) | 7 × 10−36 | 3 × 10−127 | 0.57 |
| High physical leisure time activity, % | 31 | 30 | 27 | 22 | 3 × 10−34 | 7 × 10−185 | 0.60 |
| Smoking, pack-years | 19(3–29) | 19(3–29) | 21(7–30) | 24(10–32) | 1 × 10−31 | <1 × 10−300 | 0.58 |
| Alcohol consumption, units/week | 8(2–11) | 8(2–11) | 9(1–11) | 9(0–12) | 2 × 10−7 | 1 × 10−16 | 0.17 |
| Total cholesterol, mmol/l | 6.0(5.2–6.7) | 6.1(5.2–6.8) | 6.2(5.3–6.8) | 6.2(5.3–6.9) | 4 × 10−8 | 2 × 10−11 | 0.32 |
| HDL cholesterol, mmol/l | 1.2(1.0–1.4) | 1.2(0.9–1.3) | 1.1(0.9–1.3) | 1.1(0.9–1.3) | 1 × 10−22 | <1 × 10−300 | 0.01 N.S. |
| Systolic blood pressure, mmHg | 134(121–146) | 136(121–149) | 138(122–150) | 139(123–152) | 3 × 10−12 | 1 × 10−160 | 0.07 |
| Diabetes mellitus, % | 1 | 2 | 2 | 3 | 6 × 10−7 | 7 × 10−156 | 0.75 |
| Hormone replacement therapy, %a | 22 | 18 | 17 | 13 | 7 × 10−9 | 1 × 10−27 | 0.60 |
| Lipid-lowering therapy, % | 0 | 0 | 0 | 0 | 1.00 | <1 × 10−300 | 0.97 |
| Low eGFR, % | 14 | 13 | 16 | 15 | 0.20 | 1 × 10−198 | 0.13 |
Continuous variables are reported as median and interquartile range. P-values for test of trend across quartile categories of p-25(OH)D in the Copenhagen City Heart Study (10,170 participants), the Copenhagen City Heart Study and the Copenhagen General Population Study (86 051 participants) for test of trend across number of plasma 25-hydroxyvitamin D-reducing alleles in a combined allele score based on all four genotypes (Supplementary Table 1) and for test of comparison between individuals with and without ischaemic heart disease (IHD) (Supplementary Table 2) are shown, all performed by the Cuzick extension of the Wilson rank-sum test. Information on covariates reported in this table were >98% complete. Data are from the 1981–83 examination of the Copenhagen City Heart Study. Conversion factors: to convert p-25(OH)D in nmol/l to ng/ml, divide by 2.496, and cholesterol in mmol/l to mg/dl, divide by 0.0259. eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; NS, not significant after Bonferroni correction for 14 parallel tests (required P-value: 0.05/14 = 0.004) (see Supplementary Table 1).
aWomen only.
Baseline characteristics of participants from the Danish general population, the Copenhagen City Heart Study
| Plasma 25-hydroxyvitamin D quartiles | |||||||
|---|---|---|---|---|---|---|---|
| 4th | 3rd | 2nd | 1st | Ptrend | Pcomparison, IHD | Ptrend, allele score | |
| Number of participants | 2,595 | 2,490 | 2,532 | 2,553 | |||
| Age, years | 51(43–59) | 51(43–59) | 52(44–59) | 53(45–60) | 1 × 10−4 | <1 × 10−300 | 0.11 |
| Women, % | 59 | 56 | 55 | 55 | 5 × 10−3 | 7 × 10−185 | 0.50 |
| Body mass index, kg/m2 | 24(22–26) | 25(22–27) | 26(23–28) | 26(25–29) | 7 × 10−36 | 3 × 10−127 | 0.57 |
| High physical leisure time activity, % | 31 | 30 | 27 | 22 | 3 × 10−34 | 7 × 10−185 | 0.60 |
| Smoking, pack-years | 19(3–29) | 19(3–29) | 21(7–30) | 24(10–32) | 1 × 10−31 | <1 × 10−300 | 0.58 |
| Alcohol consumption, units/week | 8(2–11) | 8(2–11) | 9(1–11) | 9(0–12) | 2 × 10−7 | 1 × 10−16 | 0.17 |
| Total cholesterol, mmol/l | 6.0(5.2–6.7) | 6.1(5.2–6.8) | 6.2(5.3–6.8) | 6.2(5.3–6.9) | 4 × 10−8 | 2 × 10−11 | 0.32 |
| HDL cholesterol, mmol/l | 1.2(1.0–1.4) | 1.2(0.9–1.3) | 1.1(0.9–1.3) | 1.1(0.9–1.3) | 1 × 10−22 | <1 × 10−300 | 0.01 N.S. |
| Systolic blood pressure, mmHg | 134(121–146) | 136(121–149) | 138(122–150) | 139(123–152) | 3 × 10−12 | 1 × 10−160 | 0.07 |
| Diabetes mellitus, % | 1 | 2 | 2 | 3 | 6 × 10−7 | 7 × 10−156 | 0.75 |
| Hormone replacement therapy, %a | 22 | 18 | 17 | 13 | 7 × 10−9 | 1 × 10−27 | 0.60 |
| Lipid-lowering therapy, % | 0 | 0 | 0 | 0 | 1.00 | <1 × 10−300 | 0.97 |
| Low eGFR, % | 14 | 13 | 16 | 15 | 0.20 | 1 × 10−198 | 0.13 |
| Plasma 25-hydroxyvitamin D quartiles | |||||||
|---|---|---|---|---|---|---|---|
| 4th | 3rd | 2nd | 1st | Ptrend | Pcomparison, IHD | Ptrend, allele score | |
| Number of participants | 2,595 | 2,490 | 2,532 | 2,553 | |||
| Age, years | 51(43–59) | 51(43–59) | 52(44–59) | 53(45–60) | 1 × 10−4 | <1 × 10−300 | 0.11 |
| Women, % | 59 | 56 | 55 | 55 | 5 × 10−3 | 7 × 10−185 | 0.50 |
| Body mass index, kg/m2 | 24(22–26) | 25(22–27) | 26(23–28) | 26(25–29) | 7 × 10−36 | 3 × 10−127 | 0.57 |
| High physical leisure time activity, % | 31 | 30 | 27 | 22 | 3 × 10−34 | 7 × 10−185 | 0.60 |
| Smoking, pack-years | 19(3–29) | 19(3–29) | 21(7–30) | 24(10–32) | 1 × 10−31 | <1 × 10−300 | 0.58 |
| Alcohol consumption, units/week | 8(2–11) | 8(2–11) | 9(1–11) | 9(0–12) | 2 × 10−7 | 1 × 10−16 | 0.17 |
| Total cholesterol, mmol/l | 6.0(5.2–6.7) | 6.1(5.2–6.8) | 6.2(5.3–6.8) | 6.2(5.3–6.9) | 4 × 10−8 | 2 × 10−11 | 0.32 |
| HDL cholesterol, mmol/l | 1.2(1.0–1.4) | 1.2(0.9–1.3) | 1.1(0.9–1.3) | 1.1(0.9–1.3) | 1 × 10−22 | <1 × 10−300 | 0.01 N.S. |
| Systolic blood pressure, mmHg | 134(121–146) | 136(121–149) | 138(122–150) | 139(123–152) | 3 × 10−12 | 1 × 10−160 | 0.07 |
| Diabetes mellitus, % | 1 | 2 | 2 | 3 | 6 × 10−7 | 7 × 10−156 | 0.75 |
| Hormone replacement therapy, %a | 22 | 18 | 17 | 13 | 7 × 10−9 | 1 × 10−27 | 0.60 |
| Lipid-lowering therapy, % | 0 | 0 | 0 | 0 | 1.00 | <1 × 10−300 | 0.97 |
| Low eGFR, % | 14 | 13 | 16 | 15 | 0.20 | 1 × 10−198 | 0.13 |
Continuous variables are reported as median and interquartile range. P-values for test of trend across quartile categories of p-25(OH)D in the Copenhagen City Heart Study (10,170 participants), the Copenhagen City Heart Study and the Copenhagen General Population Study (86 051 participants) for test of trend across number of plasma 25-hydroxyvitamin D-reducing alleles in a combined allele score based on all four genotypes (Supplementary Table 1) and for test of comparison between individuals with and without ischaemic heart disease (IHD) (Supplementary Table 2) are shown, all performed by the Cuzick extension of the Wilson rank-sum test. Information on covariates reported in this table were >98% complete. Data are from the 1981–83 examination of the Copenhagen City Heart Study. Conversion factors: to convert p-25(OH)D in nmol/l to ng/ml, divide by 2.496, and cholesterol in mmol/l to mg/dl, divide by 0.0259. eGFR, estimated glomerular filtration rate; HDL, high-density lipoprotein; NS, not significant after Bonferroni correction for 14 parallel tests (required P-value: 0.05/14 = 0.004) (see Supplementary Table 1).
aWomen only.
Risk of ischaemic heart disease and myocardial infarction: observational estimates
Lower concentrations of seasonally adjusted p-25(OH)D were associated with higher risk of ischaemic heart disease and with the subgroup myocardial infarction in the CCHS (Figure 1), as reported previously.3 The multivariable adjusted hazard ratios for lowest vs highest quartile were 1.82 (95% CI: 1.42‐2.32) for ischaemic heart disease and 1.83 (1.30–2.57) for myocardial infarction. By use of restricted cubic spline modelling (Supplementary Figures 4 and 5, available as Supplementary data at IJE online), we did not find any indication of a U-shaped association between seasonally adjusted p-25(OH)D concentrations and risk of either ischaemic heart disease or the subgroup myocardial infarction.
Risk of ischaemic heart disease: observational estimates. Risks of ischaemic heart disease and myocardial infarction as a function of quartiles of seasonally adjusted p-25(OH)D concentrations in the Copenhagen City Heart Study (the 1981–83 examination). All estimates are corrected for regression dilution bias. Multivariable adjustment was for age, body mass index, smoking in pack-years, alcohol consumption, total cholesterol concentration, high-density lipoprotein cholesterol concentration and systolic blood pressure on a continuous scale; and gender, physical activity during leisure time, diabetes mellitus, hormone replacement therapy, lipid-lowering therapy and estimated glomerular filtration as categorical covariates. Black dots represent hazard ratios, and error bars the 95% confidence intervals. P-values for trend are estimated by Cuzick’s extension of a Wilcoxon rank-sum test. IQR, interquartile range of the corresponding absolute concentrations; these values are not corrected for seasonal variation and therefore overlap among quartiles of seasonally adjusted values.
Genotypes and p-25(OH)D concentrations
A one-allele increase in a combined allele score of all four genotypes in the CCHS and CGPS combined was associated with a 1.9-nmol/l decrease in p-25(OH)D (P = 7 × 10−55; R2 = 0.9%)t (Figure 2). Total cholesterol, oestradiol, testosterone and cortisol are all potential substrates for the CYP2R1 enzyme, and their concentrations as a function of number of p-25(OH)D-reducing alleles in a combined CYP2R1-allele score in participants in the CCHS were examined to address potential pleiotropy (Supplementary Tables 1 and 4, available as Supplementary data at IJE online); we found no association of genotypes with concentrations of these four potential CYP2R1 substrates.
Genotypes and 25-hydroxyvitamin D concentrations. P-25(OH)D concentrations as a function of individual genotypes and combined allele scores with number of p-25(OH)D-reducing alleles in the Copenhagen City Heart Study (the 1991–94 examination) and the Copenhagen General Population Study combined. Columns show mean p-25(OH)D concentrations with standard error bars, percentage difference in p-25(OH)D concentrations, F-statistics as an evaluation of the statistical strength of the instruments from the first stage regression of the instrumental variable analysis, R2 estimating the contribution of genotype to variation in p-25(OH)D concentrations in percent, P-values for test of Hardy–Weinberg (HW) equilibrium, and P-values for trend test across individual genotypes and number of alleles in combined allele scores.
Genotypes and risk of ischaemic heart disease and myocardial infarction
Calculated as previously,24 under the assumption that p-25(OH)D is causally associated with risk of ischaemic heart disease, the theoretically expected risk of ischaemic heart disease with genetically reduced p-25(OH)D concentrations would be in the same direction and magnitude as observed in the observational epidemiological design (Figure 3). On the basis of this assumption, a 14% decrease in plasma 25-hydroxyvitamin D concentration as seen for six to eight p-25(OH)D-reducing alleles vs zero to one reducing allele in a combined allele score of all four genotypes, translates into a theoretically predicted hazard ratio of 1.05 (95% CI: 1.04‐1.07) for ischaemic heart disease. In contrast and after combining data from the CCHS, CGPS and CIHDS to achieve maximal statistical power, the observed odds ratio for ischaemic heart disease did not differ from 1.0 for individual genotypes or combined allele scores. For myocardial infarction, theoretically predicted and observed risks were similar to those for ischaemic heart disease (Supplementary Figure 6, available as Supplementary data at IJE online).
Genotypes and risk of ischaemic heart disease. Theoretically predicted and observed risk of ischaemic heart disease as a function of individual genotypes and combined allele scores with number of p-25(OH)D-reducing alleles and for plasma total cholesterol increasing APOE genotype combination. APOE genotype combination was included as a positive control. To achieve maximal statistical power, data from the Copenhagen City Heart Study (the 1991–94 examination), the Copenhagen General Population Study and the Copenhagen Ischaemic Heart Disease Study were combined. 25(OH)D = 25(OH)D. Δ = difference. The last column shows the odds ratio that can be detected with a one-sided P-value less than 0.05 with 90% statistical power.
To demonstrate that the risk of ischaemic heart disease in the study cohorts follows established patterns and to test the predictive power, we included as previously24 the APOE genotype as a positive control of the association with risk of ischaemic heart disease. In ε44 carriers vs ε32 carriers, the theoretically predicted risk of ischaemic heart disease was 1.12 (95% CI: 1.07–1.16), comparable to but less than the observed odds ratio of 1.29 (1.10–1.50) (Figure 3, lower part).
Risk for ischaemic heart disease and myocardial infarction: genetic and observational estimates
In an instrumental variable analysis, the association between genetically reduced p-25(OH)D concentrations and risk of ischaemic heart disease was examined (Figure 4). The odds ratio for ischaemic heart disease for a 25-nmol/l genetic decrease in p-25(OH)D concentration was 0.98 (95% CI: 0.76 -1.26), with a corresponding observational hazard ratio of 1.07 (1.01–1.13) for ischaemic heart disease. For myocardial infarction, the corresponding values were 1.15 (0.83–1.59) genetically and 1.16 (1.06–1.27) observationally. Results were similar for a 10-nmol/l and a 50% reduction in p-25(OH)D concentrations, for both ischaemic heart disease and myocardial infarction, for each genotype individually and for the combined allele scores (Supplementary Figures 7–9, available as Supplementary data at IJE online). Using the two-stage least squares regression estimator or the control function estimator10 in the instrumental variable analyses gave similar results (data not shown) as did use of weighted allele score in the instrumental variable analysis (Supplementary Figure 10, available as Supplementary data at IJE online). Furthermore, using a probit regression model in the second stage likewise did not change the magnitude or direction of the associations (Supplementary Figure 10).
Observational and genetic estimates for ischaemic heart disease. Observational and genetic estimates for risk of ischaemic heart disease and myocardial infarction as a function of reduced 25(OH)D and of elevated total cholesterol. Observed risk of ischaemic heart disease is shown as a multivariable adjusted hazard ratio for a 25 nmol/l decrease in p-25(OH)D concentrations from the Copenhagen City Heart Study (the 1981–83 examination) using Cox regression. Odds ratio for a 25-nmol/l genetic decrease in p-25(OH)D concentrations due to a combined allele score based on all four genotypes are from the Copenhagen City Heart Study (the 1991–94 examination), the Copenhagen General Population Study and the Copenhagen Ischaemic Heart Disease Study combined and estimated by instrumental variable analysis. Corresponding analyses were performed for APOE genotype combination, plasma total cholesterol concentrations and risk of ischaemic heart disease (lower panel). The P-values are for significance of risk estimates. HR/OR, hazard ratio/odds ratio.
For the APOE genotype combination, the odds ratio for ischaemic heart disease for a 1-mmol/l genetic increase in plasma total cholesterol concentrations was 1.23 (1.08–1.41), with a corresponding observational hazard ratio of 1.08 (1.04–1.14) (Figure 4, lower part).
Discussion
In the presence of lack of randomized controlled trial evidence, as is the case for vitamin D supplementation to reduce the risk of ischaemic heart disease, a Mendelian randomization study approach can potentially be used to make causal inference where epidemiological associations may be confounded or due to reverse causation.
Using the Mendelian randomization study approach, we found no evidence to support that genetically reduced p-25(OH)D is associated with increased risk of ischaemic heart disease or the subgroup myocardial infarction. In contrast, the positive control APOE genotype associated with elevated cholesterol levels was associated with increased risk of ischaemic heart disease. Taken together, this suggests that the corresponding observational association for reduced p-25(OH)D and increased risk of ischaemic heart disease may not represent a causal relationship, but more likely is due to confounding or reverse causation.
A recent meta-analysis including 18 observational prospective population studies showed a 39% (95% CI: 25%-54%) increase in risk of ischaemic heart disease comparing individuals in the lowest vs the highest quartile of p-25(OH)D concentrations,3 less than observed in the present study. Meta-analyses of randomized controlled trials aimed at improving bone health (and often with concomitant calcium supplementation masking the isolated effect of vitamin D supplementation) have not demonstrated a cardioprotective effect of vitamin D supplementation.6 Using a Mendelian randomization approach free from reverse causation and largely free from confounding, the present study shows no evidence that genetically reduced p-25(OH)D concentrations are associated with increased risk of ischaemic heart disease or myocardial infarction. On the other hand, the confidence intervals of the instrumental variable analyses did include a positive association, so we cannot completely exclude such an association. Nevertheless, this conclusion of no evidence of a causal association is in line with both a recently published study by Afzal et al., showing no evidence of a causal association between genetically low p-25(OH)D concentrations and risk of cardiovascular mortality25 and with results from two case-control studies by Jorde et al.26 and Kühn et al.27 One explanation of our findings could be that low p-25(OH)D concentration simply is a marker of poor health or an unhealthy lifestyle,28,29 well known to be associated with increased risk of ischaemic heart disease and myocardial infarction. It should be noted, however, that a recent Mendelian randomization study using the same studies and the same genetic variants as in the present study, found evidence that genetically low vitamin D levels using variants in DHCR7 related to the endogenous production was associated with type 2 diabetes.30 Also, a Mendelian randomization study by Vimaleswaran et al. found evidence for a causal association between genetically increased p-25(OH)D concentrations and reduced blood pressure and risk of hypertension.31
Strengths of the present study include recruitment of participants from a homogeneous population, detailed information on potential confounders, high participation rate, correction for regression dilution bias, long follow-up time without losses to follow-up (that is, all persons were accounted for during the entire period of observation), and inclusion of the APOE genotype as a positive control. The results from this positive control confirm the validity of our data and that the risk of ischaemic heart disease in our study cohorts follows established patterns.
We also believe that it is a strength of our study that we did not include genetic variants in the GC gene encoding the vitamin D-binding protein, as we wanted to study polymorphisms influencing bioavailability of vitamin D, not just total plasma levels. If asked to speculate, we would expect that inclusion of the GC polymorphisms would have resulted in a combined genotype score instrument, influencing the bioavailable p-25(OH)D in an unpredictable manner, which may have biased the results toward the null hypothesis.
Limitations include those of the Mendelian randomization design,11 with pleiotropy possibly being the most important: the genetic variants used may affect risk of ischaemic heart disease and myocardial infarction through mechanisms other than their effects on p-25(OH)D concentrations. Whereas the DHCR7 gene encodes the 3β-hydrocholesterol Δ7-reductase that catalyzes the final reaction in the endogenous cholesterol synthesis, and to our knowledge is selective towards 7-dehydrocholesterol and dehydrodesmosterol as its substrates,19 the CYP2R1 gene encodes a hepatic microsomal monooxygenase belonging to the cytochrome P450 superfamiliy20 and have many possible substrates such as other lipids, steroid hormones and drug metabolites leading to possible pleiotropy. However, we did not find any indication of influence of the CYP2R1 variants on plasma total cholesterol, oestradiol, testosterone or cortisol concentrations.
Also, comparing the risk estimates for the instrumental variable analysis for both ischaemic heart disease and the subgroup myocardial infarction for the combined DHCR7 allele-score with the combined CYP2R allele score, the risk estimates were nominally in opposite directions, and therefore we cannot exclude pleiotropy based on this finding; if the estimates had been completely consistent, this would have indicated that pleiotropy was unlikely as these genotypes influence p-25(OH)D by different pathways. One could argue that yet another limitation is that we did not include genetic variants in the GC gene (encoding vitamin D-binding protein) as done in other studies, reducing the comparability; however we deliberately did not include GC polymorphisms, since we wanted to study polymorphisms influencing bioavailability of vitamin D, not just total plasma levels or affinity/functionality of the vitamin D-binding protein leading to unpredictable biological effects on the bioavailability of vitamin D. Another potential limitation is linkage disequilibrium: the genetic variants used could be in linkage disequilibrium with genetic variants in other genes that may influence risk of ischaemic heart disease and myocardial infarction through mechanisms other than through p-25(OH)D concentrations; however, we did not detect linkage disequilibrium with genetic variants associated with cardiovascular disease in previous studies (Supplementary Figure 11, available as Supplementary data at IJE online). Also, we used two different 25(OH)D-reducing loci in complete linkage equilibrium and found similar results. Importantly, the genotypes were highly associated with p-25(OH)D concentrations and also independent of measured confounding factors, and thus the key assumptions of the Mendelian randomization approach did not seem to be significantly violated.
Other limitations of our study include insufficient statistical power: it is thus possible that our study is unable to discount non-ignorable effects of 25(OH)D on ischaemic heart disease or myocardial infarction. Moreover, we used p-25(OH)D measurements and genetic variants that influence p-25(OH)D concentrations, whereas it is 1,25-dihydroxyvitamin D that is the active vitamin D metabolite leading to biological effects, and moderate changes in p-25(OH)D concentrations (nmol/l range) may not necessarily translate into corresponding changes in 1,25-dihydroxyvitamin D concentrations (pmol/l range). Finally, our three studies have different designs, and therefore different limitations and potential biases. Also, we only included White persons, and therefore our results may not necessarily apply to all races.
In conclusion, we found no evidence that genetically reduced p-25(OH)D is associated with increased risk of ischaemic heart disease or the subgroup myocardial infarction. This suggests that the observed association of reduced 25(OH)D with ischaemic heart disease and myocardial infarction may not represent a causal relationship, but more likely is due to confounding by unaccounted influences or due to reverse causation. Our findings also question the value of widespread use of vitamin D supplementation for prevention of a plethora of non-skeletal diseases including cardiovascular prevention, the use of which should await results from ongoing large randomized controlled trials of vitamin D supplementation aimed at reducing risk of ischaemic heart disease.4 However, it is possible that our study is unable to discount non-ignorable effects of 25(OH)D on ischaemic heart disease or myocardial infarction.
Supplementary Data
Supplementary data are available at IJE online.
Funding
This work was supported by the Danish Heart Foundation, the Faculty of Health and Medical Sciences, University of Copenhagen, and by Herlev Hospital, Copenhagen University Hospital, Denmark. DiaSorin provided free kits for measurement of plasma 25-hydroxyvitamin D, but had no influence on the submitted work.
Acknowledgments
B.N. initiated the study which was designed in detail by P.B-J., B.N. and M.B. and they, with S.A., had full access to all data. B.N. collected raw data. Database handling and statistical analyses were by P.B-J., B.N. and S.A. and they, with M.B., analysed and interpreted the data. P.B-J. wrote the first draft of the paper, which was revised and fully accepted by the other authors. We are indebted to participants and staff of the Copenhagen City Heart Study and of the Copenhagen General Population Study for their important contributions.
Conflict of interest: All authors report no conflict of interest.
