Introduction

Fracture is a major cause of disability, morbidity, and mortality, thus creating a considerable burden on the healthcare system annually [1–3]. Prevention of fractures by identifying and confirming the modifiable risk factors is important. Some factors such as age, smoking, bone mineral density (BMD), physical activity, and body mass index (BMI) are known to be involved in the risk of fracture [4–7].

Serum 25-hydroxyvitamin D (25(OH) vitamin D) level is associated with BMD, bone size (relative to body size), and bone strength [8, 9]. Several studies reported that low serum 25(OH) vitamin D level is a potentially modifiable risk factor for fractures. However, the findings were controversial [10–28]. On the basis of a literature search up to April 2009, only one meta-analysis—performed by Lai et al. [29], which included 17 case–control studies with 1903 fractures—has examined the association between serum 25(OH) vitamin D level and the risk of hip fracture. The results of that study showed a 33% lower serum 25(OH) vitamin D level in cases compared with controls, with significant heterogeneity. The authors did not include cohort studies in their meta-analysis, which may have reduced the strength of their conclusions and the level of evidence. After 2009, 15 cohort studies with a focus on this topic have been published. Moreover, Lai and colleagues only examined the hip fracture risk but did not assess the risk of total fractures.

We therefore performed a meta-analysis to evaluate the association between serum 25(OH) vitamin D level and the risk of total fractures and hip fractures, by using data from the published prospective cohort or nested case–control studies.

Materials and methods

We performed a meta-analysis of the available literature according to the PRISMA statement and the MOOSE guidelines [30, 31].

Search strategy and data sources

Two reviewers searched the PubMed, EMBASE, and OVID databases for published prospective cohort or nested case–control studies that investigated the association between serum 25(OH) vitamin D level and the risk of total fractures and hip fractures, from their inception to June 1, 2016, without restrictions. Our searches combined MeSH (medical subject) headings and free text, and the following search keywords were used: “vitamin D level” or “25-OH-D” or “25-hydroxyvitamin D” or “25(OH) vitamin D” and “fracture”. In addition, manual searches for the references of all relevant studies and the abstracts of meetings related to osteoporosis were performed to identify additional studies.

Study selection

The two reviewers evaluated the articles independently. Discrepancies were resolved by arbitration, and a consensus was reached on study inclusion and interpretation of data after a discussion. Studies were included in the present meta-analysis if they met the following inclusion criteria: (1) a prospective cohort or nested case–control design, (2) reported adult population, (3) reported serum 25(OH) vitamin D level as a risk factor and total fractures or hip fracture as the outcome, or (4) reported risk estimates such as relative risks (RRs), odds ratios (ORs), or hazard ratios (HRs) with 95% confidence intervals (CIs). Studies that did not meet the inclusion criteria were excluded. If different articles investigated the same cohort, we selected the most detailed study.

Data extraction

Two reviewers (Y.F. and G.C.) extracted the data independently by using a standardized data collection form for analysis. A third reviewer (B.C.) checked the reliability. The standard data extraction form included the first author’s last name, publication year, name of cohort, country where the study was performed, sex and age of the participants, recruitment time of the participants, years of follow-up, sample size, number of fractures, ascertainment of fracture, cutoff of 25(OH)vitamin D level, variables adjusted for analysis, and RR estimates with corresponding 95% CIs. We extracted the RRs and 95% CIs that reflected the greatest degree of control for potential confounders. The nine-star Newcastle–Ottawa Scale (NOS) [32] was used to assess study quality. The third reviewer (B.C.) was involved to resolve any disagreement concerning the abstracted data.

Statistical analyses

In the present analysis, we used RRs as the means of measuring the association across studies. Multivariable-adjusted HRs or ORs were transformed into RRs [33, 34]. One study [14] reported stratified risk estimates according to race, and we combined these estimates by using a random-effects model and then used pooled estimates for the meta-analysis [35–37]. For two studies [13, 28] that reported more than one cutoff of serum 25(OH) vitamin D level, we used 20 ng/mL (the definition of vitamin D deficiency in most studies [38, 39]) or the closest value to 20 ng/mL as the assigned cutoff [40, 41]. For studies that presented graded associations, we only used the estimates for the highest category [35, 37].

We used the Cochran Q test and I 2 statistics to estimate heterogeneity across studies [42]. If the p value for heterogeneity was <0.1, we considered I 2 values of <30% as low heterogeneity, 30–50% as moderate heterogeneity, and >75% as high heterogeneity, according to Higgins et al. [43]. We performed subgroup analyses for relevant study characteristics (i.e., geographical location, number of participants, length of follow-up, NOS scores, and adjustments). Publication bias was assessed by using the Begg and Egger regression asymmetry tests [44, 45]. To assess the possible effect of publication bias in our meta-analysis, we also performed the “trim and fill” process. This process is used to evaluate the possibility of hypothetical “missing” studies, imputing their RRs, and obtaining a pooled RR that includes the hypothetical missing studies as though they had actually been performed [44, 46]. Sensitivity analysis involved removing any one study and assessing whether the results would be markedly affected. All statistical tests were performed by using STATA software (version 12.0; StataCorp, College Station, TX, USA), and p <0.05 was considered statistically significant.

Results

The procedure of the study selection is presented in Fig. 1. A total of 2679 studies were included from the initial database search. After title and abstract assessment, we excluded the duplicated studies and those that did not satisfy inclusion criteria, and 42 articles remained. Thereafter, we excluded some studies because of having a case–control design, a cross-sectional design, duplicate cohorts, or missing data. Finally, 19 studies were included in this meta-analysis [10–28].

Fig. 1
figure 1

Flowchart of the study selection

Full size image

Characteristics of the included studies

Table 1 shows the characteristics of the included prospective cohort (n = 15) or nested case–control (n = 4) studies. This meta-analysis included 47,341 participants, with 4762 total fractures and 3091 hip fractures. Eight studies included men and women, eight studies included only women, and three studies included only men. The included studies were performed in ten different countries (nine studies from the United States; two studies from Japan; and one study each from Sweden, Iceland, Australia, Norway, China, Netherlands, France, and Finland). All participants were ≥19 years old. The follow-up time of the included studies ranged from 4 to 16.9 years. The recruitment time of the included studies ranged from 1986 to 2007. The fractures were identified by using verified self-reports of fracture experience, radiological diagnosis, and medical records. The covariates most commonly taken into account were age, BMI, sex, race, smoking, alcohol use, and physical activity. Only six studies adjusted for BMD [16, 18, 22, 25–27].

Table 1 Characteristics of the included prospective cohort studies
Full size table

Serum 25(OH) vitamin D level and total fracture risk

Eleven studies were concerned with the association between serum 25(OH) vitamin D level and the risk of total fractures. The results of the meta-analysis are shown in Fig. 2a. The meta-analysis with a random-effects model showed a significant increase in the risk of total fractures in patients with low serum 25(OH) vitamin D level (RR 1.25, 95% CI 1.06–1.43), with moderate heterogeneity across studies (p = 0.149, I 2 = 31.3%). The Egger test revealed a publication bias (p < 0.01), and the Begg test showed no evidence of publication bias (p = 0.10). The trim-and-fill method confirmed that the six possibly missing studies could alter the pooled estimation of RR to 1.13 (95% CI 1.01–1.24). The sensitivity analysis showed that excluding any one study from the meta-analysis did not change the results substantially (Fig. 3a).

Fig. 2
figure 2

Adjusted relative risk (RR) of 25-hydroxyvitamin D (serum 25(OH) vitamin D) level and total fracture (a) and hip fracture (b) risk by using the random-effects model

Full size image
Fig. 3
figure 3

Sensitivity analysis of serum 25-hydroxyvitamin D (25(OH) vitamin D) level and total fracture (a) and hip fracture (b) risk by using the random-effects model

Full size image

The subgroup analyses for the association between serum 25(OH) vitamin D level and the risk of total fractures are shown in Table 2. We assessed the geographical location; number of participants; NOS scores; and adjustments for BMI, physical activity, race, smoking status, alcohol intake, diabetes mellitus, and history of fracture in the subgroup analyses. The RRs were 1.15 (95% CI 1.03–1.28; p for heterogeneity = 0.09, I 2 = 47.2%) for studies conducted in the United States, 1.50 (95% CI 1.04–1.96; p for heterogeneity = 0.70, I 2 = 0%) for studies conducted in Asia, and 1.17 (95% CI 0.73–1.61; p for heterogeneity = 0.13, I 2 = 57.5%) for studies conducted in other geographical locations (p = 0.35). For the length of follow-up, the RRs were 1.02 (95% CI 0.83–1.20; p for heterogeneity = 0.32, I 2 = 15.1%) for <6 years and 1.28 (95% CI 1.13–1.43; p for heterogeneity = 0.40, I 2 = 2.8%) for >6 years (p = 0.02). For NOS scores, the RRs were 1.37 (95% CI 1.06–1.68; p for heterogeneity = 0.31, I 2 = 16.3%) for scores >7 and 1.18 (95% CI 0.95–1.41; p for heterogeneity = 0.17, I 2 = 33.8%) for scores ≤7 (p = 0.17). Concerning adjustment for BMI, the RRs were 1.24 (95% CI 1.09–1.39; p for heterogeneity = 0.53, I 2 = 0%) with adjustment for BMI and 1.09 (95% CI 0.90–1.27; p for heterogeneity = 0.05, I 2 = 62.1%) without adjustment for BMI (p = 0.22). Concerning adjustment for history of fracture, the RRs were 1.53 (95% CI 1.15–1.92, p for heterogeneity = 0.54, I 2 = 0%) with adjustment for history of fracture and 1.14 (95% CI 1.02–1.26; p for heterogeneity = 0.19, I 2 = 31.8%) without adjustment for history of fracture (p = 0.06).

Table 2 Subgroup analysis for serum 25-hydroxyvitamin D and the risk of fracture
Full size table

Serum 25(OH) vitamin D level and hip fracture risk

Eleven studies were concerned about the association between serum 25(OH) vitamin D level and the risk of hip fractures. The results of the meta-analysis are shown in Fig. 2b. The meta-analysis with a random-effects model showed a significant increase in the risk of hip fractures among patients with low serum 25(OH) vitamin D level (RR 1.48, 95% CI 1.29–1.68), with no heterogeneity across studies (p = 0.514, I 2 = 0%). The Egger test revealed a publication bias (p = 0.03), and the Begg test showed no evidence of publication bias (p = 0.31). The trim-and-fill method confirmed that the seven possibly missing studies could alter the pooled estimation of RR to 1.36 (95% CI 1.18–1.54). The sensitivity analysis showed that excluding any one study from the meta-analysis did not change the results substantially (Fig. 3b).

The subgroup analyses for the association between serum 25(OH) vitamin D level and the risk of hip fractures are shown in Table 3. We assessed the geographical location; number of participants; NOS scores; and adjustments for BMI, physical activity, race, smoking status, alcohol intake, diabetes mellitus, and history of fracture in the subgroup analyses. The RRs were 1.47 (95% CI 1.21–1.73; p for heterogeneity = 0.57, I 2 = 0%) for studies conducted in the United States, 2.12 (95% CI 0.57–3.67; p for heterogeneity = 0.87, I 2 = 0%) for studies conducted in Asia, and 1.48 (95% CI 1.18–1.77; p for heterogeneity = 0.05, I 2 = 73.0%) for studies conducted in Europe (p = 0.72). For the length of follow-up, the RRs were 1.61 (95% CI 1.18–2.04; p for heterogeneity = 0.29, I 2 = 20.2%) for <6 years and 1.45 (95% CI 1.24–1.67; p for heterogeneity = 0.54, I 2 = 0%) for >6 years (p = 0.53). For NOS scores, the RRs were 1.90 (95% CI 1.50–2.30; p for heterogeneity = 0.66; I 2 = 0%) for scores >7 and 11.36 (95% CI 1.14–1.58; p for heterogeneity = 0.93; I 2 = 0%) for scores ≤7 (p = 0.02). Concerning adjustment for BMI, the RRs were 1.48 (95% CI 1.27–1.68; p for heterogeneity = 0.58, I 2 = 0%) with adjustment for BMI and 1.54 (95% CI 1.04–2.04; p for heterogeneity = 0.22, I 2 = 32.1%) without adjustment for BMI (p = 0.82). Concerning adjustment for history of fracture, the RRs were 1.18 (95% CI 1.03–2.58; p for heterogeneity = 0.63, I 2 = 0%) with adjustment for history of fracture and 1.46 (95% CI 1.27–1.66; p for heterogeneity = 0.41, I 2 = 3.1%) without adjustment for history of fracture (p = 0.40).

Table 3 Subgroup analysis for serum 25-hydroxyvitamin D and the risk of hip fracture
Full size table

Per SD decrease of serum 25(OH) vitamin D level and the risk of total fractures and hip fractures

Four studies [11, 21, 26, 27] were concerned with the association between the per SD decrease of serum 25(OH) vitamin D level and the risk of total fractures, and three studies [12, 20, 27] were concerned with the association between the per SD decrease of serum 25(OH) vitamin D level and the risk of hip fractures. The covariates most commonly taken into account were age, BMI, and physical activity. Only two studies adjusted for BMD [26, 27]. The hip fracture risk was increased by 40% for each SD decrease in serum 25(OH) vitamin D level (RR 1.40, 95% CI 1.20–1.61; I 2 = 0%, p for heterogeneity = 0.51). The per SD decrease in serum 25(OH) vitamin D level was not associated with the risk of total fractures (RR 1.14, 95% CI 0.93–1.35; I 2 = 63.2%, p for heterogeneity = 0.04).

Discussion

The present meta-analysis indicated that there was an increased risk of total fractures and hip fractures in patients with lower serum 25(OH) vitamin D levels. The hip fracture risk was increased by 40% for each SD decrease in serum 25(OH) vitamin D level. However, the per SD decrease in serum 25(OH) vitamin D level was not associated with the risk of total fractures.

Several plausible mechanisms have been proposed for the associations between serum 25(OH) vitamin D level and the risk of fractures. A low serum 25(OH) vitamin D level usually indicates vitamin D deficiency. First, vitamin D increases the serum calcium concentrations and stimulates osteoblasts to produce RANKL, a protein that stimulates osteoclastogenesis [8]. Low serum 25(OH) vitamin D level may induce an increase in parathyroid hormone (PTH) level, which may result in bone loss [47]. Calcium mobilization from the bone is affected by both vitamin D and PTH. Second, vitamin D deficiency has also been shown to be related to low muscle mass and muscle weakness [9, 48]. Low serum 25(OH) vitamin D level may also be a cofounder in idiopathic inflammatory myopathies [49], alcoholic skeletal muscle myopathy [50], and diffuse musculoskeletal pain [9]. Koeckhoven et al. used the data of the Amsterdam osteoarthritis cohort and found that serum 25(OH) vitamin D level was significantly associated with muscle strength [51]. Moreover, another study performed by Orces et al. [52] showed coincident results: compared with subjects with normal muscle strength, the prevalence rates of 25(OH) vitamin D deficiency were 31 and 43% higher among men and women with muscle weakness. Third, numerous studies suggested an association between vitamin D insufficiency and falls. Snijder et al. reported that low serum 25(OH) vitamin D level was significantly associated with increased falls in elderly persons [53]. Rothenbacher et al. [54] performed a prospective population-based cohort study and showed an association between serum 25(OH) vitamin D level and the risk of first fall [HRR = 1.93 (95% CI 1.10–3.37) for serum 25(OH) vitamin D level < 20 mg/mL]. Fourth, the association between serum 25(OH) vitamin D level and BMD was investigated in several studies [24, 26, 55]. The cohort study performed by Swanson et al. [27] showed that higher levels of serum 25(OH) vitamin D were associated with higher baseline BMD and slower bone loss at the hip. Steingrimsdottir et al. [24] performed a prospective study of 5764 men and women, and showed that compared with reference values (50–75 nmol/l), values <30 nmol/l were associated with significantly lower BMD of the femoral neck.

Furthermore, many previous meta-analyses investigated the associations between oral vitamin D supplementation and the risk of fractures. The meta-analysis conducted by Bischoff-Ferrari et al. found that, in elderly persons, oral vitamin D supplementation of between 700 and 800 IU/day may reduce the risk of non-vertebral fractures and hip fractures [56]. Furthermore, the same team reported that prevention of non-vertebral fractures with vitamin D supplementation was dose dependent [57]. Moreover, another meta-analysis showed that oral calcium plus vitamin D supplements may reduce the fracture risk in both community-dwelling and institutionalized elderly adults [58].

Only one meta-analysis [29] examined the association between serum 25(OH) vitamin D level and the risk of hip fractures (from the included publications up to April 2009—a total 17 case–control studies), and the results showed 33% lower serum 25(OH) vitamin D level in cases compared with controls, based on 1903 cases, with significant heterogeneity existing among studies. The authors, however, did not include cohort studies in their meta-analysis, which may have reduced the strength of their conclusions. Moreover, Lai and colleagues only examined hip fracture risk but did not assess the risk of total fractures. In comparison with this previous meta-analysis, our meta-analysis included 15 prospective cohort studies and 4 nested case–control studies with a total of 47,341 participants, and focused on total fractures and hip fractures. The present study is the most comprehensive research in terms of the amount of data contributing to the summary estimates.

Our study has some advantages. First, through a wide-ranging search of the literature, we included a substantial number of participants and fracture cases (up to 47,341 participants with 4762 total fractures cases and 3091 hip fractures cases), which significantly increased the statistical power of our analysis. Second, the quantitative assessment of the analysis was based on prospective cohort studies, which may have minimized the selection or recall bias. Third, all the included studies in the present meta-analysis had a long duration of follow-up, which may have increased the statistical power of the results. Fourth, the fractures in most of the included studies were verified.

Despite these strengths, our meta-analysis has some limitations. First, as a meta-analysis, there could be inherent confounding factors in the included studies, which may have underestimated or exaggerated the risk estimates. Nevertheless, most of the prospective studies adjusted for major potential confounders, including age, BMI, sex, race, smoking, alcohol use, and physical activity. Moreover, the moderate heterogeneity in the present study may have been due to the methodological differences among the studies. Although many of the I 2 values we estimated were assessed as moderate to high in the subgroup analysis, we investigated the length of follow-up for potential sources of heterogeneity in the analysis of the total fracture risk; only NOS scores were considered for the analysis of hip fracture risk. The heterogeneity in the present study may have reduced the power of our conclusions.

Second, another concern could be the potential publication bias because smaller studies or studies reporting null results are difficult to publish, as we have established in the Egger test in this meta-analysis. However, the Begg test showed no evidence of publication bias, and further trim-and-fill analysis showed that after adding possibly missing studies, the pooled estimation RRs were 1.13 (95% CI 1.01–1.24) for serum 25(OH) vitamin D level and total fracture risk and 1.36 (95% CI 1.18–1.54) for serum 25(OH) vitamin D level and hip fracture risk. Moreover, our sensitivity test showed that the findings were robust.

Third, only four studies on the association between the per SD decrease of serum 25(OH) vitamin D level and the risk of total fractures and three studies on the association between the per SD decrease of serum 25(OH) vitamin D level and the risk of hip fractures were included in our meta-analysis, which made the results less meaningful.

Fourth, the included studies used different serum 25(OH) vitamin D assays, and the 25(OH) vitamin D levels have not been standardized. Therefore, the results need to be interpreted with caution. However, we performed further meta-analyses by using different cutoffs of serum 25(OH) vitamin D and recalculated the pooled RR. The results showed that the pooled RR for the association between serum 25(OH) vitamin D level and the risk of total fractures did not change substantially [the pooled RR decreased from 1.34 (95% CI 1.10–1.58) to 1.23 (95% CI 1.03–1.43)]. In addition, in this meta-analysis, three included studies [15, 18, 27] reported non-vertebral fracture risk, and we considered these non-vertebral fractures as total fractures according to previous meta-analyses [2, 59]. However, we performed further meta-analysis after excluding these studies, and the results showed that the pooled RR for the association between serum 25(OH) vitamin D level and the risk of total fractures did not change substantially (RR 1.33, 95% CI 1.11–1.55; I 2 = 17.7%, p for heterogeneity = 0.29). Moreover, the definition of fractures differed across studies (osteoporotic fractures, low-trauma fractures, low-energy fractures, or exclusion of pathological fractures, fractures of unknown etiology, traumatic fractures, etc.). Nevertheless, it is difficult to determine whether the differences of definition of fractures were responsible for the observed outcomes. Further individual patient data meta-analysis may offer an additional insight into this issue.

Fifth, owing to a lack of related studies, we did not conduct a non-linear dose–response analysis for the association between serum 25(OH) vitamin D level and the risk of total fractures. Especially, our meta-analysis showed a significant increase in the risk of total fractures in patients with low serum 25(OH) vitamin D level. However, the per SD decrease in serum 25(OH) vitamin D level was not associated with an increased risk of total fractures. Therefore, a non-linear dose–response analysis may offer potential insights to this issue. Bleicher et al. [22] reported that the association between serum 25(OH) vitamin D level and the risk of total fractures was U-shaped, and an increased risk of total fractures was observed with either high or low serum 25(OH) vitamin D levels. Further studies should be conducted to quantitatively assess the dose–response association between serum 25(OH) vitamin D level and the risk of total fractures.

Conclusion

This study found an increased risk of total fractures and hip fractures in patients with low serum 25(OH) vitamin D levels. The hip fracture risk was increased by 40% for each SD decrease in serum 25(OH) vitamin D level. However, the per SD decrease in serum 25(OH) vitamin D level was not associated with the risk of total fractures. Further well-designed and stratified cohort studies should be conducted to quantitatively assess the non-linear dose–response association between serum 25(OH) vitamin D level and the risk of total fractures.