Prevention of Nonvertebral Fractures With Oral Vitamin D and Dose Dependency: A Meta-analysis of Randomized Controlled Trials

Background  Antifracture efficacy with supplemental vitamin D has been questioned by recent trials.

Methods  We performed a meta-analysis on the efficacy of oral supplemental vitamin D in preventing nonvertebral and hip fractures among older individuals (≥65 years). We included 12 double-blind randomized controlled trials (RCTs) for nonvertebral fractures (n = 42 279) and 8 RCTs for hip fractures (n = 40 886) comparing oral vitamin D, with or without calcium, with calcium or placebo. To incorporate adherence to treatment, we multiplied the dose by the percentage of adherence to estimate the mean received dose (dose × adherence) for each trial.

Results  The pooled relative risk (RR) was 0.86 (95% confidence interval [CI], 0.77-0.96) for prevention of nonvertebral fractures and 0.91 (95% CI, 0.78-1.05) for the prevention of hip fractures, but with significant heterogeneity for both end points. Including all trials, antifracture efficacy increased significantly with a higher dose and higher achieved blood 25-hydroxyvitamin D levels for both end points. Consistently, pooling trials with a higher received dose of more than 400 IU/d resolved heterogeneity. For the higher dose, the pooled RR was 0.80 (95% CI, 0.72-0.89; n = 33 265 subjects from 9 trials) for nonvertebral fractures and 0.82 (95% CI, 0.69-0.97; n = 31 872 subjects from 5 trials) for hip fractures. The higher dose reduced nonvertebral fractures in community-dwelling individuals (−29%) and institutionalized older individuals (−15%), and its effect was independent of additional calcium supplementation.

Conclusion  Nonvertebral fracture prevention with vitamin D is dose dependent, and a higher dose should reduce fractures by at least 20% for individuals aged 65 years or older.

The antifracture benefits of vitamin D have been questioned by several recent trials,^1-4 leading to uncertainty among patients and physicians regarding recommendations for vitamin D supplementation. Two 2007 meta-analyses^5,6 included most of these trials and concluded that vitamin D may not reduce fractures significantly or may do so only in combination with calcium and primarily among institutionalized older individuals. A third 2007 meta-analysis⁷ concluded that calcium with or without vitamin D may reduce total fracture risk by 12%, a result that was questioned by a more recent meta-analysis⁸ of high-quality trials of calcium supplementation alone in which calcium had a neutral effect on nonvertebral fractures and a possible adverse effect on hip fracture risk. Apart from the mixed data on calcium, the recent meta-analyses with vitamin D did not consider heterogeneity by received dose (incorporating adherence) or achieved level of 25-hydroxyvitamin D.

A dose-response relationship between vitamin D and fracture reduction is supported by epidemiologic data showing a significant positive trend between serum 25-hydroxyvitamin D concentrations and hip bone density⁹ and lower extremity strength.^10,11 In addition, greater antifracture efficacy with higher achieved 25-hydroxyvitamin D levels was documented in an earlier meta-analysis of high-quality primary prevention trials with supplemental vitamin D.¹² Factors that may obscure a benefit of vitamin D are low adherence to treatment,² low dose of vitamin D, or the use of less potent ergocalciferol (vitamin D₂).^13,14 Furthermore, open study design trials¹ may bias results toward the null because vitamin D supplementation is available over the counter.

Our primary goal was to determine the antifracture efficacy of oral vitamin D supplementation among individuals aged 65 years or older by performing a systematic review of the literature and meta-analysis of high-quality, double-blinded RCTs. In addition, we specifically addressed antifracture efficacy by received dose, achieved 25-hydroxyvitamin D levels, and in predefined subgroups.

We conducted a systematic review of all English and non-English articles using MEDLINE (Ovid, PubMed) and the Cochrane Controlled Trials Register from January 1960 through August 2008 and EMBASE from January 1991 through August 2008. Additional studies were identified from reference lists, contacts with experts in the field, and abstracts presented at the American Society for Bone and Mineral Research from 1995 through 2007.

The medical subject headings (MeSH terms) included trials (randomized-controlled-trial or controlled-clinical-trial or random-allocation or double-blind-method or single-blind-method or uncontrolled-trials), vitamin D (cholecalciferol or hydroxycholecalciferols or calcifediol or dihydroxycholecalciferols or calcitriol or vitamin D/aa [analogs and derivates] or ergocalciferol or vitamin D/bl [blood]), fractures (femoral fractures, femoral neck fractures, hip fractures, humerus fractures, forearm fractures, radius fractures, ankle fractures, nonvertebral fractures), accidental falls or falls, humans, elderly, bone density. Identical data were extracted independently by three of us (H.A.B.-F., A.T., and J.H.).

We included only RCTs that studied oral vitamin D supplementation (cholecalciferol [vitamin D₃] or ergocalciferol with a minimum follow-up of 1 year and required more than a total of 1 fracture in each trial. Because our target population consisted of older persons, the mean age of study subjects had to equal or exceed 65 years. To be included in the primary analysis, we required a double-blinded study design, a report of adherence, and a statement explaining how fractures were ascertained. We also conducted a separate pooled analysis of trials using 1-α-hydroxylated vitamin D, which included trials that used 1-α-hydroxyvitamin D₃, 1,25-hydroxyvitamin D₃, and 1α,25-dihydroxy-2β(3-hydroxypropoxy) vitamin D₃ (ED-71).

We excluded uncontrolled trials, observational studies, and animal studies. Because health conditions that place patients at high risk for falls and fractures could have confounded our analysis, we excluded studies that focused on patients following organ transplantation or stroke and those receiving steroid therapy or care for Parkinson disease or having unstable health states.

Our primary outcome measure was the relative risk (RR) of a first or repeated nonvertebral fracture or hip fracture in persons receiving supplemental vitamin D, with or without calcium supplementation, compared with those receiving placebo or calcium supplementation alone. To determine the effect of dose, we calculated the received dose of supplemental vitamin D by the cross-product of dose and percentage of adherence, which is a more comprehensive assessment of dosage, accounting for the low adherence in the latest trials. According to predefined criteria, heterogeneity by received dose of vitamin D was explored as a continuous variable plus using the same cutoff as in the 2005 meta-analysis¹² (≤400 IU/d compared with >400 mg/d). For the Women's Health Initiative trial,³ we also added the mean personal intake (365 IU/d) to the 400 IU/d provided in the trial to account for the significant vitamin D intake outside the study protocol. We chose achieved serum 25-hydroxyvitamin D levels, measured in at least a subgroup of the study population assessed during the trial period, because this reflects the starting 25-hydroxyvitamin D level (baseline risk for vitamin D deficiency) and the vitamin D dose received.

To be included as trials with minimal bias, studies had to be randomized and masked to treatment allocation. Trials that met all features but had an open study design were included in sensitivity analyses. Twelve studies (listed in Table 1) for supplemental vitamin D were identified through our MeSH term search. Four additional studies with an open study design were identified for the sensitivity analysis^1,24-26 (Table 2 and Figure 1).

Outcomes were analyzed on an intention-to-treat basis with random effects models.²⁷ We calculated the risk difference to determine the number needed to treat (NNT) to prevent 1 fracture. Heterogeneity among studies was explored by predefined covariates using the Q-statistic as a test (considered significant for P < .10).²⁸ The presence of heterogeneity suggests that the studies should not be pooled because of significant differences in results.²⁹ In such cases, we explored heterogeneity by received dose (dose × adherence: ≤400 IU/d vs >400 IU/d of vitamin D)¹² and achieved 25-hydroxyvitamin D level using visual inspection, and random-effects metaregression analysis. Predefined subgroup analyses included age, type of dwelling, and additional calcium supplementation. To evaluate publication bias, we used Begg and Egger tests with all 12 trials from the primary analysis or all 16 trials from the sensitivity analysis. Although the Begg funnel plot suggested a possible absence of negative studies involving small sample sizes, the trim and fill analysis did not confirm this suggestion.³⁰ Statistical analysis was performed with Stata software (version 8.0; StataCorp LP, College Station, Texas).

Table 1 shows characteristics of the 12 double-blind RCTs that were included in the primary analysis for the prevention of nonvertebral fractures, 8 of which were also included in the primary analysis for hip fracture. The 12 trials included 42 279 individuals with a mean age of 78 years, and 89% were women. The received dose of vitamin D (dose × adherence) was 400 IU/d or less in 3 trials,^2,19,22 whereas the other 9 RCTs had mean intakes of 482 to 770 IU/d. A total of 500 to 1200 mg/d of calcium supplementation was used in combination with vitamin D supplementation in 7 RCTs. Treatment duration varied from 12 to 84 months.

In the 12 high-quality RCTs (listed in Table 1) (n = 42 279 participants), the pooled RR for any dose of vitamin D preventing nonvertebral fractures was 0.86 (95% CI, 0.77-0.96). However, heterogeneity in results was seen among studies (Q test: P = .04). After stratifying trials by received dose, heterogeneity was resolved. For the 3 high-quality trials^2,19,22 (9014 individuals) with a received low dose of 400 IU/d or less of vitamin D (340-380 IU/d; all cholecalciferol), the pooled RR was 1.02 (95% CI, 0.92-1.15; Q test: P = .64) suggesting that 380 IU/d of vitamin D or less did not reduce nonvertebral fracture risk (Table 3).

For 9 trials with a higher received dose of more than 400 IU/d of vitamin D (482-770 IU; 33 265 individuals; Table 4), the pooled RR was 0.80 (95% CI, 0.72-0.89; Q-test: P = .31) suggesting that 482 to 770 IU/d of vitamin D reduced nonvertebral fracture risk by 20% (Figure 2A). The pooled risk difference for the higher received dose was 1.1% (95% CI, 0.6%-1.5%, P < .001), so the NNT was 93 [95% CI, 66-160] for 12 to 84 months.

In metaregression analyses, a greater reduction in nonvertebral fractures was seen both with a higher received dose (P = .003; Figure 3A) and with higher achieved 25-hydroxyvitamin D levels (P = .04; Figure 3B).

In subgroup analyses (Table 4), the pooled RR for nonvertebral fractures was 10% in trials that used ergocalciferol compared with 23% for trials that used cholecalciferol (for metaregression, P = .07). The effect of vitamin D was significant among all subgroups according to age and dwelling, with a somewhat greater effect among younger persons and those living in the community, but the differences were not significant (see Table 4 for Q-test P values). The combined effect of calcium plus vitamin D compared with placebo was tested in 4 trials (all using cholecalciferol) with a pooled RR reduction of 21%. The effect of vitamin D alone (either vitamin D vs placebo or vitamin D plus calcium compared with calcium alone) was tested in 5 trials, also with a pooled RR reduction of 21%. Thus, the addition of calcium to adequate intakes of vitamin D does not seem to enhance the effect of vitamin D in reducing nonvertebral fractures. Based on limited data for men (n = 2037),¹⁷ there was no significant heterogeneity by sex (Q test: P = .93).

For any received dose, after adding the 4 open study design trials^1,24-26 to the 12 double-blinded trials, the pooled RR for vitamin D preventing any nonvertebral fracture, including 56 459 individuals, was 0.88 (95% CI, 0.80-0.97), and significant heterogeneity among studies remained (Q test: P = .02). We therefore again stratified studies by dose received.

For the lower received dose, after adding 1 open study design trial to the 3 double-blind trials (16 087 individuals), the pooled RR was 0.96 (95% CI, 0.84-1.10; Table 3). For the higher received dose, after adding 3 open study design trials to the 9 double-blind trials (40 372 individuals), the pooled RR was 0.83 (95% CI, 0.74-0.95; Table 4). However, variation in results was seen between open study design (summarized in Table 5) and double-blind trials (Q test: P = .07), suggesting that trial quality introduces heterogeneity.

In the 8 high-quality trials (40 886 individuals), the pooled RR for any dose of vitamin D preventing hip fractures was 0.91 (95% CI, 0.78-1.05). However, heterogeneity in results was seen among studies (Q test: P = .08). After stratifying trials by received dose, heterogeneity was resolved. For the 3 high-quality trials (9014 individuals) on a received low dose of less than 400 IU/d of vitamin D (340-380 IU; all using cholecalciferol), the pooled RR was 1.09 (95% CI, 0.90-1.32; Q test: P = .81) (Table 6).

For the 5 trials with a higher received dose of more than 400 IU/d (482-770 IU; 31 872 individuals) (Table 7), the pooled RR was 0.82 (95% CI, 0.69-0.97; Q test: P = .18). Thus, the higher dose of vitamin D reduced hip fracture risk by 18% (Figure 2B). The pooled risk difference for the higher received dose was 0.60% (0.23%-0.96%; P = .02), so the NNT was 168 (95% CI, 104-440) for 12 to 84 months.

In metaregression analyses, a greater reduction in hip fractures was seen both with higher received dose (P = .07; Figure 4A) and higher achieved 25-hydroxyvitamin D levels in the treatment group (P = .01; Figure 4B). Owing to the smaller number of trials with a hip fracture end point, subgroup analyses were limited because there was only 1 trial using ergocalciferol at a higher dose,⁴ 1 trial among individuals 65 to 74 years of age,³ 1 trial among men,¹⁷ and 2 trials using vitamin D alone^4,17 (Table 7).

For any received dose, after adding 2 open study design trials to the 8 double-blinded trials (47 917 individuals), the pooled RR was 0.92 (95% CI, 0.80-1.06), but variation among studies remained significant (Q test: P = .10).

For the higher received dose, after adding 2 open study design trials to the 5 double-blinded trials (38 903 individuals), the pooled RR was 0.84 (95% CI, 0.71-0.99; Q test: P = .17; Table 7; higher-dose open study design trials are summarized in Table 8). For hip fractures, there were no lower-dose trials with an open study design.

Table 9 shows characteristics of 7 RCTs that were included in the analysis for 1-α-hydroxylated vitamin D.^32-38 None of the trials reported separate data for hip fractures. The 7 trials included 1484 individuals, all 65 to 74 years of age and 99.7% of whom were women.

The pooled RR for any type of 1-α-hydroxylated vitamin D preventing nonvertebral fractures compared with placebo or calcium was 0.58 (95% CI, 0.37-0.92), similar to the RR for a higher dose of supplemental vitamin D in the same age group (Table 4 and Table 10; the ratio of the 2 effect sizes: pooled RR of supplemental vitamin D to pooled RR of 1-α-hydroxylated vitamin D = 0.67/0.58 = 1.16; 95% CI, 0.44-3.03).

In this meta-analyses of 12 double-blinded trials among individuals aged 65 years or older, the antifracture efficacy of supplemental vitamin D increased significantly with higher received dose or higher achieved 25-hydroxyvitamin D levels for any nonvertebral fractures and for hip fractures. No fracture reduction was observed for a received dose of 400 IU/d or less, whereas a higher received dose of 482 to 770 IU/d of supplemental vitamin D reduced nonvertebral fractures by 20% and hip fractures by 18%. Subgroup analyses for the prevention of nonvertebral fractures with the higher received dose suggested possibly better fracture reduction with cholecalciferol compared with ergocalciferol, whereas additional calcium did not further improve antifracture efficacy. Nonvertebral fracture reduction with the higher received dose was significant among all subgroups by age and dwelling, including younger individuals aged 65 to 74 years and those living in the community (see Table 4 for P values).

In August 2007, a review and meta-analysis⁵ commissioned by the US Department of Health and Human Services (DHHS) addressed the effect of vitamin D supplementation on all fractures in postmenopausal women and men aged 50 years or older. The pooled results for all fractures included 10 double-blinded and 3 open study design trials (n = 58 712) and did not support a significant reduction of fractures with vitamin D (pooled odds ratio, 0.90; 95% CI, 0.81-1.02). The report⁵ suggested that the benefit of vitamin D may depend on additional calcium and may be primarily seen in institutionalized individuals, which is consistent with the meta-analysis of Boonen et al.⁶ However, in both reports,^5,6 heterogeneity by dose may have been missed owing to the inclusion of open study design trials plus a dose evaluation that did not incorporate adherence. Biologically, the exclusion of heterogeneity by dose seems implausible even if a formal test of heterogeneity is not statistically significant.

In our meta-analysis, the dose of vitamin D and achieved 25-hydroxyvitamin D levels were identified as important sources of variation in the antifracture efficacy of supplemental vitamin D. Our findings confirm the findings of an earlier 2005 primary prevention meta-analysis¹² after including 5 additional high-quality, double-blinded trials (2005: total n = 9820; 2008: total n = 42 279). New to these analyses is the primary use of received dose (dose × adherence) as opposed to treatment dose. The received dose allows assessment of antifracture efficacy by a dose that accounts for the low adherence in several recent large trials.^1-3 The consistency of our results for both received dose and achieved 25-hydroxyvitamin D levels in the treatment group across all 12 masked trials lends support to the presence of a dose-response relationship between supplemental vitamin D and fracture reduction. Despite the well-documented interlaboratory and interassay variation for 25-hydroxyvitamin D,^39,40 the consistency in the dose-response analyses for both received dose and achieved 25-hydroxyvitamin D level also lends support to our use of 25-hydroxyvitamin D levels from different trials.

Confirming our findings with some limitations, Tang et al⁷ suggested in their meta-analysis that, together with calcium supplementation, a daily intake of at least 800 IU of vitamin D increases total fracture reduction by 3% compared with daily doses of vitamin D of less than 800 IU. However, with their focus on calcium, Tang et al⁷ excluded 4 high-quality trials of vitamin D alone compared with placebo.^4,17,19,22

The pooled RR reduction was 21% with or without additional calcium for the higher dose of vitamin D. Previous meta-analyses may have missed this finding owing to their analyses including all doses of vitamin D. Physiologically, the calcium-sparing effect of vitamin D may explain why we did not see an additional benefit of calcium supplementation at a higher dose of vitamin D.^39,40 Similarly, our findings suggest that, at a sufficiently high dose, vitamin D benefits are not limited to institutionalized and frail individuals, as suggested by the DHHS report.⁵

To our knowledge, the type of supplemental vitamin D was not addressed previously. With a higher received dose, the pooled effect of cholecalciferol was significant with 23% fracture reduction, whereas the pooled effect with ergocalciferol was not significant with 10% fracture reduction. One explanation may be that ergocalciferol is less potent than cholecalciferol in maintaining 25-hydroxyvitamin D levels, as suggested by 2 direct comparison trials,^13,41 although this was challenged by a recent trial⁴² showing similar potency of daily ergocalciferol and daily cholecalciferol. Another explanation may be dosing frequency in that 1 trial⁴ of 2 ergocalciferol trials^4,15 dosed intermittently, which may have decreased efficiency. However, the higher-dose cholecalciferol supplement given either daily or intermittently did reduce fractures. Thus, future research efforts may wish to simply focus on higher doses of cholecalciferol.

We performed sensitivity analyses, including 4 open study design trials.^1,24-26 This increased the number of trials to 16 and the number of individuals to 56 459 for nonvertebral fractures. The pooled RR risk from these 16 trials was 0.88 (95% CI, 0.80-0.97), suggesting that with all evidence considered, supplemental vitamin D should reduce nonvertebral fracture risk by 12% among individuals 65 years or older. However, the study variation was larger than expected for the pooled result from all 16 trials. Even within the higher received dose, adding 3 open study design trials to the 9 double-blinded trials, variation was larger than expected (pooled RR, 0.83; 95% CI, 0.74-0.95) supporting our predefined strategy of focusing on fracture efficacy from double-blinded trials.

Based on the pooled results, 1-α-hydroxylated vitamin D reduced nonvertebral fractures by 42% and among individuals of comparable age a higher dose of supplemental cholecalciferol reduced these fractures by 33%. Thus, although the number of studies for 1-α-hydroxylated vitamin D was small and differences in efficacy could not be excluded, our analyses do not support prevention of nonvertebral fractures with 1-α-hydroxylated vitamin D owing to its higher cost and higher risk profile compared with an adequate dose of supplemental vitamin D. Importantly, the efficacy of 1-α-hydroxylated vitamin D adds to the evidence that improved vitamin D status will reduce fracture risk.

In conclusion, a higher received dose of supplemental vitamin D (482-770 IU/d) should reduce nonvertebral fractures by at least 20% and hip fractures by at least 18%. The greater fracture reduction with a higher received dose or higher achieved 25-hydroxyvitamin D levels for both any nonvertebral fractures and hip fractures suggests that higher doses of vitamin D should be explored in future research to optimize antifracture efficacy. Also, it is possible that greater benefits may be achieved with earlier initiation of vitamin D supplementation and longer duration of use. Our results do not support use of low-dose vitamin D with or without calcium in the prevention of fractures among older individuals.

Correspondence: Heike A. Bischoff-Ferrari, DrPH, Centre on Aging and Mobility, University of Zurich, University Hospital, Gloriastrasse 25, Zurich 8091, Switzerland (heike.bischoff@usz.ch).

Accepted for Publication: October 1, 2008.

Author Contributions:Study concept and design: Bischoff-Ferrari, Stuck, and Staehelin. Acquisition of data: Bischoff-Ferrari and Henschkowski. Analysis and interpretation of data: Bischoff-Ferrari, Wong, Orav, Thoma, Kiel, and Henschkowski. Drafting of the manuscript: Bischoff-Ferrari, Thoma, and Henschkowski. Critical revision of the manuscript for important intellectual content: Willett, Wong, Stuck, Staehelin, Orav, and Kiel. Statistical analysis: Bischoff-Ferrari, Willett, Wong, Orav, and Henschkowski. Obtained funding: Bischoff-Ferrari and Stuck. Administrative, technical, and material support: Thoma. Study supervision: Willett, Staehelin, and Kiel.

Financial Disclosure: None reported.

Funding/Support: This study was supported by a Swiss National Foundation Professorship grant and a fellowship grant by the Robert Bosch Foundation.