Non-linear Mendelian randomization publications on vitamin D report spurious findings and require major correction (2024)

Zhou et al.¹ purport to show through Mendelian randomization (MR) that there is a markedly non-linear effect of vitamin D on cardiovascular disease (CVD) risk. They analyse data from UK Biobank (UKB) and report a null overall effect of vitamin D and CVD risk (odds ratio, OR, per 10 nmol/L higher vitamin D, 1.01; 95% confidence interval, CI, 0.98–1.04; presented in supplementary table 6). Applying the residual non-linear MR (NLMR) approach,² which aims to estimate the effect of vitamin D in different strata of vitamin D level, from low to high, they report effects in the protective direction across the entire range of vitamin D—e.g. ‘individuals with serum 25(OH)D at 25 nmol/L had 11% (95% CI 1.05–1.18) higher odds of CVD compared to those with 50 nmol/L’¹ with ‘further lowering in the odds of CVD with higher concentrations, and for example participants with 75 nmol/L had 2% lower odds (95% CI 0.97–0.99) compared with 50 nmol/L’.¹ Such findings from the NLMR analysis—with apparent increased risk of CVD at all baseline levels of vitamin D—are not compatible with the overall null. Combining a causal interpretation of Simpson’s paradox³ with the intention of MR⁴ to estimate causal effects, it is not possible to envisage situations in which the key assumptions of the MR analysis⁴ are not violated, the assumptions regarding other causal effects are consistent with what is known, and the data are distributed as presented.⁵

The Zhou et al. paper was published online after a very similar paper on vitamin D and CVD risk had appeared which involved the developers of the residual NLMR method.⁶ Using data from several studies, but with the analyses numerically dominated by UKB,⁶ they unsurprisingly reported very similar findings to Zhou et al. That the results they reported were literally impossible was pointed out,⁵ and the paper has, rightly, been formally retracted by the Lancet Diabetes and Endocrinology,⁷ as has a supportive comment published with it.⁸ A paper using a new, doubly-ranked, NLMR method has replaced the retracted paper.⁹ This reports a null overall finding, with no meaningful non-linearity, and includes discussion of the serious biases of the original NLMR method.⁹ The authors now consider the results they had reported as being ‘a logical impossibility’.⁹ Of note, the results for UKB alone are reported in the supplementary material; they are of course very similar to the results for the full sample.

Analyses using negative controls applied to UKB data demonstrates that the analytical strategy used by Zhou et al. would yield the nonsense ‘findings’ that vitamin D has marked non-linear ‘causal effects’ on the age and sex of study participants.¹⁰ The negative control approach also casts some doubt on the validity of the new doubly-ranked method, although it appears that it may in some circ*mstances be less misleading than the original residual approach. The Zhou et al. paper is as deeply flawed as the now retracted original NLMR vitamin D paper.⁶ The same applies to yet another overlapping paper on all-cause mortality¹¹ which naturally replicates the erroneous findings on all-cause mortality from the original, now retracted, paper.⁶

The conclusion in the abstract of the Zhou et al. paper reads ‘Vitamin D deficiency can increase the risk of CVD. Burden of CVD could be reduced by population-wide correction of low vitamin D status’.¹ There is no support for this from either randomized controlled trials (RCTs) or competently done MR studies; the latter having been recently summarized in an excellent systematic review and meta-analysis.¹² The Zhou et al. paper was accompanied by a press release in which the senior author claimed that it is now ‘not ethical’ to recruit those with ‘vitamin D deficiency’ to RCTs because of the (now clearly spurious) apparent benefit in terms of reduction in CVD.¹³ Vitamin D deficiency appears to refer to ∼50% of the population (the term ‘severe Vitamin D deficiency’ is used for a lower cut-off group), and this cut-off is also implied by the wording of the papers (including that used in the now retracted paper) regarding the level of baseline vitamin D at which meaningful benefit ceases. Mendelian randomization being weaponized to promote vitamins and claim that RCTs are unethical is bitterly ironic, being the antithesis of how the approach was intended to be applied.^14,15 It is hoped that the authors will as energetically promote the dissemination of the correction of their fallacious paper as they did when the original paper was, unfortunately, published.

Declarations

Disclosure of Interest

All authors declare no disclosure of interest for this contribution.

References

© The Author(s) 2024. Published by Oxford University Press on behalf of the European Society of Cardiology. All rights reserved. For commercial re-use, please contact reprints@oup.com for reprints and translation rights for reprints. All other permissions can be obtained through our RightsLink service via the Permissions link on the article page on our site—for further information please contact journals.permissions@oup.com.

Email alerts

Citing articles via

More from Oxford Academic