Comment to: ‘Meta-analysis of long-term vitamin D supplementation on overall mortality’

Vitamin D supplementation to prevent mortality

Goran Bjelakovic,1,2 Dimitrinka Nikolova,1 Jørn Wetterslev,3 Christian Gluud1,3

1 The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Department 7812, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark
2 Department of Internal Medicine - Gastroenterology and Hepatology, Medical Faculty, University of Nis, Nis, Serbia
3 The Copenhagen Trial Unit, Centre for Clinical Intervention Research, Department 7812, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark

Corresponding author:
Professor of Medicine Goran Bjelakovic, MD, Dr. Med. Sc.
The Cochrane Hepato-Biliary Group, The Copenhagen Trial Unit,
Centre for Clinical Intervention Research, Department 7812,
Rigshospitalet, Copenhagen University Hospital
Blegdamsvej 9, DK-2100, Copenhagen, DENMARK,
And Department of Internal Medicine - Gastroenterology and Hepatology,
Medical Faculty, University of Nis
Boulevard Dr Zorana Djindjica 81, 18000 Nis, SERBIA
Telephone: +381 18 532381 Facsimile: +381 18 4238770
E-mail: goranb@junis.ni.ac.rs

Short running head: Vitamin D and mortality

Zheng et al. published a meta-analysis of long-term vitamin D supplementation effect on overall mortality. [1] The conclusions they reached were that vitamin D (i.e., D3 and D2) is effective in preventing all-cause mortality in trials with duration of follow-up longer than 3 years but not in trials with shorter duration.
Generally, Zheng et al.[1] followed methodology similar to The Cochrane Collaboration methodology that we used to conduct our Cochrane systematic review, published in 2011.[2] Zheng et al. included and analyzed almost the same set of trials that were included in our Cochrane review[2], but they applied two restrictions, namely they included only trials of oral administration and they included only trials assessing vitamin D3 or vitamin D2.[1] At the time when Zheng et al. review was published, we have been updating our Cochrane review.[3] Following our protocol inclusion criteria, we could include 6 newly published trials, with another 1,138 participants.[3] Our updated review is now based on 56 trials with 95,286 participants in total.[3] Our present conclusions on the evidence on vitamin D3 and D2 seem less optimistic than those reached by Zhen et al.[1] and originally ourselves.[2] Therefore, we would like to point out several important caveats in the meta-analysis of Zheng et al.[1] that should have influenced their results and conclusions.
One of the issues is related to the different effects of vitamin D3 and vitamin D2 on all-cause mortality. Zheng et al. meta-analyzed together the trials on vitamin D3 and vitamin D2. Our Cochrane review suggested a difference between the two forms of supplemental vitamin D. Vitamin D3 seemed to significantly decrease mortality, while the effects of vitamin D2 seemed to be neutral or even detrimental.[3] Therefore, we decided here to reanalyze the set of 49 trials (37 vitamin D3 trials; 11 vitamin D2 trials; and 1 vitamin D3 and vitamin D2 trial) out of the 56 trials included in our updated Cochrane review.[3] We excluded 7 trials that tested active forms of vitamin D (4 alfacalcidol trials and 3 calcitriol trials). We performed random-effects model meta-regression analyses in order to identify potential covariates that could predict intertrial heterogeneity, that is, the covariates that were statistically associated with the estimated intervention effects. The included covariates were: percentage of women; mean age; duration of treatment; and dose of vitamin D as continuous co-variates; the form of vitamin D (D3 or D2); vitamin D status at trial entry (adequate or deficient); and co-administered calcium (vitamin D singly or vitamin D combined with calcium) as binary co-variates. In addition, we performed subgroup analyses comparing trials of vitamin D3 (cholecalciferol) to trials of vitamin D2 (ergocalciferol). We compared the intervention effects in subgroups of trials using the method described by Bornstein et al.[4] which is implemented in RevMan 5.2 for all types of meta-analyses.
The univariate meta-regression analyses including 49 trials with 94,276 participants revealed that the duration of treatment was significantly associated with a significantly lower estimated intervention effect on mortality (RR 0.98, 95% CI 0.95 to 0.999, P = 0.04) and vitamin D2 was associated with a significantly higher estimated intervention effect on mortality (RR 1.13, 95% CI 1.03 to 1.25, P = 0.01). None of the other covariates, i.e., sex of participants, dose of supplement, vitamin D status, and co-administered calcium were significantly associated with the estimated intervention effect on mortality. In a multivariate meta-regression analysis including all covariates, the mean age of participants was significantly associated with a significantly lower estimated intervention effect on mortality (RR 0.98, 95% CI 0.96 to 0.99, P = 0.02), the duration of treatment was significantly associated with a significantly lower estimated intervention effect on mortality (RR 0.92, 95% CI 0.86 to 0.98, P = 0.01), and vitamin D2 was associated with a significantly higher estimated intervention effect on mortality (RR 1.13, 95% CI 1.00 to 1.28, P = 0.048).
In addition, we also performed random-effects model meta-regression analyses using the same covariates but now only including the 38 vitamin D3 trials with 75,927 participants. Univariate and multivariate meta-regression analyses revealed that none of the covariates were significantly associated with the estimated intervention effect on mortality.

We also performed meta-regression analyses using the same covariates but only including the 12 vitamin D2 trials. Univariate meta-regression analyses (RR 0.90, 95% CI 0.83 to 0.99, P = 0.02) and multivariate meta-regression analysis (RR 0.87, 95% CI 0.77 to 0.98, P = 0.02) revealed that duration of follow-up was associated with a significantly lower estimated intervention effect on mortality.
Our subgroup analysis in which we compared the effect of vitamin D3 to vitamin D2 on mortality, using random-effects model meta-analysis, showed that vitamin D3 significantly decreased mortality (RR 0.94, 95% CI 0.91 to 0.98, P = 0.002, I2 = 0%) while vitamin D2 had no statistically significant effect on mortality (RR 1.02, 95% CI 0.96 to 1.08; P = 0.54; I2 = 4%). The difference between the estimate of the effect of vitamin D on mortality in the trials using vitamin D3 and the trials using vitamin D2 was statistically significant by the test of interaction (P = 0.02) (Figure 1).
These results indicate that the observed beneficial effect of vitamin D in trials with duration of follow-up longer than 3 years in Zheng et al. meta-analysis may be related to the different forms of vitamin D used. Most trials with duration of follow-up longer that 3 years used vitamin D3 for supplementation. A number of recently published clinical trials,[4-7] and a systematic review[8] found evidence that vitamin D3 increases serum 25-hydroxyvitamin D more efficiently than vitamin D2. However, we do not know yet if this may influence clinical outcomes.
Another issue is related to the different kind of biases that may influence the Zheng et al. meta-analysis results. Zheng et al. meta-analysis is potentially severely misleading due to the risk of attrition bias (originating from substantial dropout of participants), outcome reporting bias due to a number of trials not reporting on mortality, and other weaknesses, like variable inclusion criteria. Most of the included trials in the Zheng et al. meta-analysis as well as in our updated Cochrane review[3] had a substantial number of missing outcome data. Our analyses of the 56 trials (37 vitamin D3, 11 vitamin D2, 1 vitamin D3 and vitamin D2, 4 alfacalcidol, and 3 calcitriol trials) with 95,286 participants pointed to the fact that outcome data reporting was lacking for more than 8% of the randomized participants.[3] This proportion is too high when mortality in our Cochrane review amounts to only 12% to 13% in the placebo or no intervention groups. Accordingly, our intention-to-treat 'best-worst’ case and 'worst-best’ case scenario analyses revealed that our results were compatible with both a very large beneficial effect and a very large detrimental effect of vitamin D3 on mortality.[3] Although the results of these extreme sensitivity analyses are unlikely, they are still indicative how few unaccounted for patients should have died to substantially shift the findings from a modest benefit to nil effect, or maybe even harm.[3]
We abstained from conducting 'uncertainty' analysis.[9] The latter analysis accepts the point estimate from the complete-participant analysis, assuming that the distribution of deaths among the participants lost to follow-up is equal to the distribution of deaths among all participants. But the distribution of dead participants among the lost to follow-up participants may indeed be different from the distribution of dead participants among participants actually followed up through the whole observation period, making the 'uncertainty' analyses themselves uncertain.
A meta-analysis of randomized clinical trials increases the power and precision of the estimated intervention effect, but this effect may be influenced by systematic errors or random errors and can lead to a report of false significant results.[10,11] It is probable that the results of Zheng et al. meta-analysis[1] are influenced by both random errors and systematic errors. By dividing the meta-analysis of 42 trials into two meta-analyses, Zheng et al. significantly increased the risk of random error (‘play of chance’). A cumulative meta-analysis runs the risk of random errors due to analysis of sparse data and repetitive testing of data.[11-13] In our Cochrane review[3] we conducted trial sequential analyses to control the risk of random errors and to prevent premature statements of superiority of the experimental or control intervention or probably falsely declarations of absence of effect in cases for which we have too few data.[11-13] We performed trial sequential analyses with a type I error of 5%, a type II error of 20% (80% power), and a diversity-adjusted required information size.[11-14] We assumed an event proportion of 10% of deaths in the control group and an anticipated intervention effect of 5% relative risk reduction. The diversity was zero. Trial sequential analysis of 20 vitamin D3 trials with low risk of bias showed that the required information size had not yet been reached and that the cumulative Z-curve did not cross the trial sequential monitoring boundaries for benefit, harm, or futility. Consequently, the trial sequential analysis does not exclude risk of random errors (Figure 2). This finding is in agreement with the findings of Bolland et al. trial sequential meta-analysis.[15]
Another issue is related to the trial inclusion criteria. Zheng et al. included 2 trials in patients with chronic kidney disease[16,17] but have left out, without any explanation, many other randomized trials in patients with chronic kidney disease supplemented with vitamin D.[18,19] This way of selection increases the selection bias (i.e., bias from unrepresentative sample). In our opinion these two trials[16,17] should be excluded from the meta-analysis by Zheng et al.[1] due to the fact that patients with chronic kidney diseases have deranged vitamin D metabolism. In a similar vein, Zheng et al. included one randomized trial in patients with tuberculosis,[20] while few other randomized trials in these patients have again been omitted.[21] In addition, Zheng et al. included one quasi-randomized clinical study[22] and omitted some randomized trials that fulfill their inclusion criteria.[23-25] These facts additionally threat the reliability of their meta-analysis.

In conclusion, the meta-regression analyses and subgroup analyses of vitamin D trials included in our updated Cochrane review revealed that the effect of vitamin D3 and vitamin D2 on all-cause mortality seems to be opposite although we lack direct head-versus-head comparisons. Our results indicate that the observed beneficial effect of vitamin D in trials with duration of follow-up longer than 3 years in Zheng et al. meta-analysis may actually not be related to the longer duration of follow-up but rather to the different forms of vitamin D used, simply because most of the trials with follow-up longer than 3 years were vitamin D3 trials.
Furthermore, the results of Zheng et al. meta-analysis are influenced by different kind of biases and did not thoroughly consider the icreased risks of random errors. We therefore draw the attention to our findings of high risks of attrition bias in vitamin D trials, which may invalidate results of meta-analyses.[2,3] We do not know yet whether vitamin D affects mortality.[15,26-30]

Reference List

(1) Zheng Y, Zhu J, Zhou M, Cui L, Yao W, et al. (2013) Meta-analysis of long-term vitamin D supplementation on overall mortality. PLoS ONE 8: e82109.
(2) Bjelakovic G, Gluud LL, Nikolova D, Whitfield K, Wetterslev J, et al. (2011) Vitamin D supplementation for prevention of mortality in adults. Cochrane Database Systematic Reviews CD007470.
(3) Bjelakovic G, Gluud LL, Nikolova D, Whitfield K, Wetterslev J, et al. (2014) Vitamin D supplementation for prevention of mortality in adults. Cochrane Database Systematic Reviews CD007470.
(4) Bornstein M, Hedges LV, Higgins T, Rothstein HR. (2009) Introduction to meta-analysis.Chichester: John Wiley & Sons.
(5) Heaney RP, Recker RR, Grote J, Horst RL, Armas LAG. (2010) Vitamin D3 is more potent than vitamin D2 in humans. The Journal of Clinical Endocrinology & Metabolism 96: E447-E452.
(6) Lehmann U, Hirche F, Stangl GI, Hinz K, Westphal S, et al. (2013) Bioavailability of vitamin D2 and D3 in healthy volunteers, a randomized placebo-controlled trial. The Journal of Clinical Endocrinology & Metabolism 98: 4339-4345.
(7) Logan VF, Gray AR, Peddie MC, Harper MJ, Houghton LA. (2013) Long-term vitamin D3 supplementation is more effective than vitamin D2 in maintaining serum 25-hydroxyvitamin D status over the winter months. British Journal of Nutrition 109: 1082-1088.
(8) Tripkovic L, Lambert H, Hart K, Smith CP, Bucca G, et al. (2012) Comparison of vitamin D2 and vitamin D3 supplementation in raising serum 25-hydroxyvitamin D status: a systematic review and meta-analysis. The American Journal of Clinical Nutrition 95: 1357-1364.
(9) Gamble C, Hollis S. (2005) Uncertainty method improved on best-worst case analysis in a binary meta-analysis. Journal of Clinical Epidemiology 58: 579-588.
(10) Gluud C. (2006) The culture of designing hepato-biliary randomised trials. Journal of Hepatology 44: 607-615.
(11) Wetterslev J, Thorlund K, Brok J, Gluud C. (2008) Trial sequential analysis may establish when firm evidence is reached in cumulative meta-analysis. Journal of Clinical Epidemiology 61: 64-75.
(12) Thorlund K, Engstrøm J, Wetterslev J, Brok J, Imberger G, Gluud C. User manual for trial sequential analysis (TSA). www.ctu.dk/tsa . 2011. Copenhagen, Denmark, Copenhagen Trial Unit, Centre for Clinical Intervention Research.
Ref Type: Internet Communication
(13) Thorlund K, Imberger G, Walsh M, Chu R, Gluud C, et al. (2011) The number of patients and events required to limit the risk of overestimation of intervention effects in meta-analysis- a simulation study. PLoS ONE 6: e25491.
(14) Wetterslev J, Thorlund K, Brok J, Gluud C. (2009) Estimating required information size by quantifying diversity in random-effects model meta-analyses. BMC Medical Research Methodology 9: 86.
(15) Bolland MJ, Grey A, Gamble G, Reid IR. (2014) The effect of vitamin D supplementation on skeletal, vascular, or cancer outcomes: a trial sequential meta-analysis. Lancet Diab Endo .
(16) Alvarez JA, Law J, Coakley KE, Zughaier SM, Hao L, et al. (2012) High-dose cholecalciferol reduces parathyroid hormone in patients with early chronic kidney disease: a pilot, randomized, double-blind, placebo-controlled trial. The American Journal of Clinical Nutrition 96: 672-679.
(17) Wasse H, Huang R, Long Q, Singapuri S, Raggi P, et al. (2012) Efficacy and safety of a short course of very-high-dose cholecalciferol in hemodialysis. The American Journal of Clinical Nutrition 95: 522-528.
(18) Palmer SC, McGregor DO, Craig JC, Elder G, Macaskill P, et al. (2009) Vitamin D compounds for people with chronic kidney disease requiring dialysis. Cochrane Database Systematic Reviews CD005633.
(19) Palmer SC, McGregor DO, Craig JC, Elder G, Macaskill P, et al. (2009) Vitamin D compounds for people with chronic kidney disease not requiring dialysis. Cochrane Database Systematic Reviews CD008175.
(20) Wejse C, Gomes VF, Rabna P, Gustafson P, Aaby P, et al. (2009) Vitamin D as supplementary treatment for tuberculosis. Am J Respir Crit Care Med 179: 843-850.
(21) Martineau AR, Timms PM, Bothamley GH, Hanifa Y, Islam K, et al. (2015) High-dose vitamin D3 during intensive-phase antimicrobial treatment of pulmonary tuberculosis: a double-blind randomised controlled trial. The Lancet 377: 242-250.
(22) Meyer HE, Smedshaug GB, Kvaavik E, Falch JA, Tverdal A, et al. (2002) Can vitamin D supplementation reduce the risk of fracture in the elderly? A randomized controlled trial. J Bone Miner Res 17: 709-715.
(23) Bischoff HA, Hannes Stähelin B, Dick W, Akos R, Knecht M, et al. (2003) Effects of vitamin D and calcium supplementation on falls: a randomized controlled trial. J Bone Miner Res 18: 343-351.
(24) Cherniack EP, Florez HJ, Hollis BW, Roos BA, Troen BR, et al. (2011) The response of elderly veterans to daily vitamin D3 supplementation of 2,000 IU: A pilot efficacy study. Journal of the American Geriatrics Society 59: 286-290.
(25) Glendenning P, Zhu K, Inderjeeth C, Howat P, Lewis JR, et al. (2012) Effects of three-monthly oral 150,000 IU cholecalciferol supplementation on falls, mobility, and muscle strength in older postmenopausal women: A randomized controlled trial. J Bone Miner Res 27: 170-176.
(26) Autier P, Boniol M, Pizot C, Mullie P. (2014) Vitamin D status and ill health: a systematic review. Lancet Diab Endo 2: 76-89.
(27) Guallar E, Miller III, Ordovas JM, Stranges S. (2010) Vitamin D supplementation in the age of lost innocence. Annals of Internal Medicine 152: 327-329.
(28) Guallar E, Stranges S, Mulrow C, Appel LJ, Miller III. (2013) Enough is enough: stop wasting money on vitamin and mineral supplements. Annals of Internal Medicine 159: 850-851.
(29) Kupferschmidt K. (2012) Uncertain verdict as vitamin D goes on trial. Science 337: 1476-1478.
(30) Nicholas CH, Cyrus C. (2012) Vitamin D: some perspective please. BMJ 345.



Figure 1. Intervention effect of vitamin D3 and vitamin D2 vs placebo or no intervention on mortality

Figure 2. Trial sequential analysis of 20 vitamin D3 trials with low risk of bias. The diversity-adjusted required information size (DARIS) was calculated based on mortality in the control group of 10%; relative risk reduction of 5% in the experimental group; type I error of 5%; and type II error of 20% (80% power). No diversity was noted. The required information size was 110,505 participants. The cumulative Z-curve (blue line) did not cross the trial sequential monitoring boundaries for benefit (red line) at any time. Accordingly, the crossing of the conventional statistical 5% boundary (the horizontal brown line) may be due to random errors.