In this analysis of data from the D-Health Trial we found some evidence that supplementation with 60 000 IU of vitamin D 3 per month for up to five years reduced the incidence of major cardiovascular events, particularly myocardial infarction and coronary revascularisation. The absolute differences were small, and the confidence intervals for total major cardiovascular events and coronary revascularisation were consistent with null findings. For total major cardiovascular events, there was some indication of a stronger effect in those who were using statins or other cardiovascular drugs at baseline, or who had higher predicted vitamin D status, although the interaction terms were not statistically significant. We found no evidence of interaction with age, sex, or body mass index.
The D-Health Trial has several strengths. Over 21 000 people were recruited from the general population and supplemented for five years, with extremely high retention and adherence. 17 Determination of cardiovascular events and mortality outcomes was achieved through comprehensive data linkage to population based administrative data sources. The lack of private hospital data for South Australia and Tasmania would have resulted in a small underestimate of events. However, the underestimation would have been low because only a quarter of participants came from these states, we captured public hospital data, and procedures were able to be identified through Medicare Benefits Schedule data. Importantly, any underestimate would probably not have differed between the study groups.
Comparison with other studies
A meta-analysis of randomised controlled trials, including the VITAL and ViDA studies that had major cardiovascular events or cardiovascular disease as the primary outcome, concluded that vitamin D supplementation does not prevent cardiovascular events.13 VITAL did not observe a protective effect for overall major cardiovascular events (including myocardial infarction, stroke, death from cardiovascular causes, and coronary revascularisation; hazard ratios ranged from 0.95 to 0.96).16 Similarly, the ViDA study concluded that vitamin D supplementation was not protective against total cardiovascular disease (hazard ratio 1.02, 95% confidence interval 0.87 to 1.20) or stroke (0.95, 0.55 to 1.62).15 The hazard ratio for myocardial infarction was similar to the D-Health Trial findings, although the confidence interval was wide (0.90, 0.54 to 1.50). The D-Health Trial has multiple outcomes, increasing the likelihood of chance findings. However, if the effect on myocardial infarction observed in the D-Health Trial is a true effect, and not due to chance, the reasons for the lack of consistency across studies are unclear. The discrepancy with VITAL might partly be caused by differences in study design and adherence. For example, VITAL excluded people with a history of cardiovascular disease (other than hypertension), and the cohort was more racially diverse. Whereas we used linked data to capture major cardiovascular events, VITAL captured events through participant report in annual surveys, followed by verification of reported events. Differential reporting between study groups might have masked any protective effect of vitamin D. Further, unlike D-Health and ViDA, VITAL used a daily dosing regimen of 2000 IU/day. While evidence is emerging to suggest that daily dosing is of greater benefit for health outcomes such as cancer mortality and infection, the monthly dosing regimen might have led to higher adherence in D-Health than in VITAL; in D-Health 80% of participants reported taking approximately 80% of study tablets, whereas in VITAL around 80% reported taking two thirds of study tablets.16
We did not observe a protective effect of vitamin D on stroke. However, the number of stroke events was relatively low, particularly when haemorrhagic stroke, which has different pathophysiology, was excluded; therefore, the confidence intervals were wide and consistent with benefit or harm. Moreover, there are several examples where associations with myocardial infarction and stroke differ,212223 so this finding is not entirely unexpected.
In prespecified subgroup analyses, we observed an effect of vitamin D on major cardiovascular events in people who were taking statins or cardiovascular drugs at baseline, but not in those who were not taking these drugs. The interactions were not significant at P<0.05, and it is plausible that these are chance findings. Nevertheless, given the lower power to detect interactions compared with main effects, and the observed strong protective effect in those taking these drugs, these interactions are of interest. There was high concurrent use of statins and other cardiovascular drugs (supplementary table 7), and the interaction could reflect an effect in people who are already at high risk of experiencing a cardiovascular event, rather than a synergistic effect between vitamin D and a particular drug. However, the exploratory analysis by self-reported history of major cardiovascular events was inconsistent with this hypothesis, and it is plausible that there is an interaction between vitamin D and the drugs examined. For example, a number of commonly used statins depend on the enzyme CYP3A4 for activation, and the CYP3A4 gene is responsive to calcitriol, suggesting that vitamin D might alter the effect of statin use.24 Further investigation of these potential interactions is warranted.
Although we observed a protective effect for vitamin D on major cardiovascular events among people predicted to be vitamin D sufficient at baseline, but not on those predicted to be insufficient, this finding needs to be interpreted with caution because we used predicted rather than measured vitamin D status. Because of the relatively low positive predictive value of the model (0.23), a considerable proportion of those predicted to be in the low group will have been vitamin D replete. While it is plausible that vitamin D supplementation becomes protective at higher serum 25(OH)D concentrations, we found that the 25(OH)D concentration attained in the vitamin D group was only slightly higher in those with predicted deseasonalised baseline serum 25(OH)D concentration ≥50 nmol/L than in those predicted to be deficient (supplementary fig 24).