Serum vitamin D and fracture risk in children and adolescents: A systematic review and meta-analysis highlighting age and altitude as key moderators

Search strategies

According to PRISMA statement (13), two review authors independently conducted a systematic electronic and manual search in PubMed (https://pubmed.ncbi.nlm.nih.gov/), Cochrane (https://www.cochranelibrary.com/) and Embase (https://www.embase.com/) databases to identify observational studies that had compared the serum vitamin D levels in a population of children and adolescents with and without fracture. All the available literature without language restrictions was retrieved, with the last search made on October 1, 2025. Search terms included ‘vitamin D’, ‘25-hydroxyvitamin D2’, ‘children’ and ‘fractures’. A combination search using the subject terms (Medical Subject Heading, MeSH) and free terms was carried out. The references of these full-text assessed articles were also reviewed.

A combination search in PubMed revealed the following: (‘vitamin D’ [Title/Abstract] OR ‘vitamin D’ [MeSH Terms] OR ‘cholecalciferol’ [Title/Abstract] OR ‘hydroxycholecalciferols’ [Title/Abstract] OR ‘25 hydroxyvitamin D2’ [Title/Abstract] OR ‘dihydrotachysterol’ [Title/Abstract]) AND (‘fractures, bone’ [MeSH Terms] OR ‘fractures bone’ [Title/Abstract]) AND (‘adolescent’ [MeSH Terms] OR ‘adolescents’ [Title/Abstract] OR ‘adolescence’ [Title/Abstract] OR ‘youth’ [Title/Abstract] OR ‘youths’ [Title/Abstract] OR ‘teens’ [Title/Abstract] OR ‘teen’ [Title/Abstract] OR ‘teenagers’ [Title/Abstract] OR ‘teenager’ [Title/Abstract OR ‘pediatric’ [Title/Abstract] OR ‘child’ [MeSH Terms] OR ‘child’ [Title/Abstract] OR ‘children’ [Title/Abstract]). Corresponding search strategies adapted for Embase (Emtree thesaurus, :ti,ab field labels) and Cochrane Library (MeSH terminology, :ti,ab field labels) were developed with identical core search concepts, free-text synonyms and Boolean logic to the PubMed search string. The full search formulas for Embase and Cochrane Library are presented in Table SI.

Grey literature including clinical trial registry records, conference abstracts, dissertations and unpublished research reports was not retrieved in this meta-analysis. Most grey literature lacked complete standardized data of serum 25(OH)D and fracture outcomes and their inconsistent methodological quality would introduce substantial heterogeneity to pooled analyses and reduce the reliability of synthetic results.

The eligibility criteria

Studies were included if they met the following strict criteria: i) Compared the vitamin D level between fracture and non-fracture groups; ii) involved only children and adolescents population.

Study design

Prospective or retrospective case-control studies, cross-sectional studies.

Data extraction

For each study, two independent reviewers (BT and JM) extracted the following information: Surname of the first author, publication year, country where the study was conducted, design of the study, sample size, latitude, altitude, season, ethnicity, fracture site, trauma energy and Newcastle-Ottawa Scale (NOS).

Statistical analysis

Review Manager statistical software (version RevMan 5.4; https://revman.cochrane.org/) and R software (version 4.4.3; https://www.r-project.org/) were used to analyze all the datasets. The mean difference (MD) was used to compare measurement data. The odds ratio (OR) was calculated for counting data. Cochran's Q χ² test was employed to assess statistical heterogeneity across studies. In accordance with the Cochrane Handbook, the selection between fixed-effect and random-effects models should not be guided by the results of statistical heterogeneity tests. The present study therefore employed the random-effects model consistently across all analyses. The heterogeneous studies were analyzed, univariate and further multivariate meta-regression analysis was performed to isolate the possible factors causing heterogeneity. For the subgroup analyses, if a study exhibited marked clinical heterogeneity (such as markedly different baseline characteristics) compared with the other studies within the same stratum, it was manually excluded from the pooled subgroup estimates to avoid misleading results. These excluded studies are clearly marked as having zero weight or labelled as ‘Not estimable’ in the corresponding forest plots, ensuring the clinical homogeneity of the specific subgroup analyses. If heterogeneity between the 2 groups was still significant, descriptive statistics were employed. All the statistical tests were two-sided. P<0.05 was considered to indicate a statistically significant difference.

Publications assessment

The methodological quality of the included studies was assessed using the NOS (14). This tool employs a ‘star system’ to appraise studies across three domains: i) The selection of study groups, ii) the comparability of groups and iii) the ascertainment of exposure (for case-control studies) or outcome (for cohort studies). Based on the total star score, studies were categorized as high quality (≥7 stars), moderate quality (5-6 stars), or low quality (≤4 stars). Two review authors independently performed the assessments, with any discrepancies resolved through discussion until consensus was reached.

Selection of publications

In total, 4,987 literature sources were retrieved from PubMed, Cochrane and Embase Databases. A total of 3,046 literature articles were left for further evaluation after the removal of duplicates. 124 literatures were screened out through title and abstract evaluation, and 28 studies were finally included according to the inclusion criteria (7-9,12,15-38) (Fig. 1).

This systematic review and meta-analysis included a total of 28 observational studies, comprising five cross-sectional and 23 case-control designs. The pooled analysis encompassed 4,135 pediatric and adolescent patients with fractures and 6,344 age- and sex-matched controls. The included studies were conducted across diverse geographical regions: Nine from the United States, three from Turkiye, two from Lebanon, two from Poland and one each from Romania, Ireland, New Zealand, China, Iran, Canada, Egypt, Jordan, Greece, Finland, Switzerland and Saudi Arabia (Table I).

Quality assessment of included studies

Based on the NOS ‘star system’, the 28 included studies were predominantly moderate in methodological quality. Specifically, two studies achieved a rating of seven stars, six received six stars, nine were assessed as five stars, another nine as four stars, while one study was rated three stars and one two stars. Notably, a primary factor contributing to the relatively lower NOS scores in certain studies was insufficient documentation of seasonal factors, particularly the unclear comparability of vitamin D blood sampling seasons between the fracture and non-fracture groups. This methodological limitation consequently diminished the overall methodological rigor and the reliability of their respective findings.

Egger test and funnel plot for publication bias

The results of Egger's linear regression test indicated that the intercept term was not statistically significant (P=0.53; 95% CI: -2.16 to 1.08), suggesting no significant asymmetry in the funnel plot and thus no evidence of substantial publication bias. Through a combination of funnel plot visualization and Egger's test, the present study did not detect significant publication bias, indicating a low risk of bias among the included studies and supporting the reliability of the pooled effect estimate (Fig. 2).

Sensitivity analysis

The sensitivity analysis results of the present study demonstrated that upon the sequential removal of any single study, the direction and statistical significance of the pooled effect size remained unchanged, with minimal fluctuation in heterogeneity (<5%; Fig. 3). Leave-one-out sensitivity analysis included low-quality studies rated two or three stars; sequential individual omission of each low-quality study did not materially change the pooled effect estimate or its statistical significance. This indicates that the core conclusion of the present study, that adolescents in the fracture group have significantly lower vitamin D levels than those in the control group (pooled MD~ -1.95), exhibits good robustness and is not unduly influenced by individual studies. Even after excluding the studies with the largest sample sizes [such as Herdea 2023(12), Matcovic 2023(8)], the pooled effect size remained stable (MD=-1.78 to -1.94), indicating that large-sample studies did not disproportionately influence the overall results. This minimized the risk of ‘weight bias’ interfering with the conclusions. In summary, the findings of this meta-analysis were robust and reliable. The core conclusions possess strong external validity and can provide stable evidence to support clinical practice and public health decision-making.

Comparison of serum vitamin D levels between the fracture and the non-fracture groups

A total of 27 studies involving 10,433 adolescent participants were included, comprising 4,112 in the fracture group and 6,321 in the non-fracture group. The mean vitamin D levels ranged from 12.28-35.4 ng/ml in the fracture group and from 15.08-36.64 ng/ml in the non-fracture group. Considerable variation in sample sizes was observed across studies, with the smallest study including 34 participants [Chan 1984(15)] and the largest including 1,577 [Herdea 2023(12)]. Study weights were evenly distributed (range: 3.1-5.0%), indicating that no single study unduly influenced the pooled results. Three studies [Contreras 2014(21), Younes 2019(33), Segal 2023(9)] reported slightly higher vitamin D levels in the fracture group compared to the non-fracture group. However, the effect sizes were small (MD=0.10-2.72), and only the result from Younes 2019 reached statistical significance (MD=1.29; 95% CI: 0.15-2.43). The overall effect of these studies on the pooled results was limited. Meta-analysis revealed that the mean vitamin D level was significantly lower in the fracture group than in the non-fracture group (MD=-1.94 [-2.89-0.99]; P<0.00001; I²=80%). However, this finding requires cautious interpretation due to significant heterogeneity (Fig. 4).

Comparison of risk of serum vitamin D deficiency between the fracture and the non-fracture groups

A total of 17 studies, comprising 5,473 adolescent participants (2,488 in the fracture group and 2,985 in the non-fracture group), were included in this analysis on the correlation of vitamin D deficiency (vitamin D level ≤20 ng/ml) with fracture (Fig. 5). Within these studies, vitamin D deficiency was reported in 1,620 participants from the fracture group and 1,894 from the non-fracture group. The distribution of study weights varied, with the highest weight assigned to Herdea 2023(12) (34.9%) and the lowest to Saglam 2017(30) (0.5%). Importantly, no single study contributed >50% of the total weight, thus mitigating the risk of any individual study exerting undue influence on the pooled result. Of the 17 studies, 13 reported a positive association between vitamin D deficiency and fracture risk in adolescents (OR>1), with eight reaching statistical significance, the strongest effect was observed in the study by Saglam et al (30) (OR=13.14; 95% CI: 3.84-45.02), indicating a 13.14-fold higher fracture risk among adolescents with vitamin D deficiency in that study. The studies with the highest statistical weights, Herdea 2023(12) (OR=2.42; 95% CI: 1.94-3.01) and Liu 2019(34) (OR=1.78; 95% CI: 1.30-2.45); both supported the conclusion that vitamin D deficiency increases fracture risk, further validating the robustness of the pooled result.

The pooled analysis revealed that adolescents with vitamin D deficiency (vitamin D level≤20 ng/ml) had a significantly higher risk of fracture compared to those with sufficient vitamin D levels (OR=1.71 [1.29-2.27]; P<0.0001; I²=65%). This result indicates that vitamin D deficiency is associated with a 71% increased risk of fracture in adolescents, confirming a significant positive association between vitamin D deficiency and fracture risk in this population. Similarly, this result should be interpreted with caution owing to substantial heterogeneity (Fig. 5).

Building upon the previous unadjusted analyses, the present study further incorporated studies that explicitly reported adjusted ORs to focus on the independent association between vitamin D deficiency and fracture risk in adolescents. Statistical analysis was performed using the metafor package in R software version 4.4.3 (https://www.r-project.org/). Adjusted ORs with their corresponding 95% CI served as the core effect measure to evaluate the strength of the independent association. The pooled analysis demonstrated that, even after adjusting for potential confounders, vitamin D deficiency remained significantly associated with an increased risk of fracture [OR=2.41 (95% CI: 1.74-3.34); P<0.0001; I²=11.11%; P=0.3373; Fig. 6]. This result suggested that, independent of confounding factors such as age and altitude, adolescents with vitamin D deficiency face a 2.41 times higher risk of fracture compared to those with sufficient levels, further confirming the independence and robustness of this association.

Univariate and multi-variate meta-regression analysis on the heterogeneity

Given the considerable number of original studies, the present study further conducted meta-regression to explore sources of heterogeneity. Metafor package in R software was mainly employed to perform the univariate regression analysis, the results demonstrated that age, altitude, latitude, and season could markedly explain the heterogeneity (Table II). Subsequent multivariate meta-regression indicated that a model combining age and altitude sufficiently accounted for the heterogeneity, while latitude and season also contributed partially to explaining its sources (Table III). The two variables with the strongest explanatory power in the univariate analysis (age and altitude, both accounting for 100% of the heterogeneity) remained capable of fully explaining the heterogeneity within the multivariate core model, confirming their central moderating roles. Variables with weaker explanatory power in the univariate analysis (such as Study design and Trauma energy) were not included in the multivariate model. Since the core model already explained 100% of the heterogeneity, the addition of Latitude and Season did not alter the model's explanatory power or the AIC value, indicating that these variables are either collinear with the core variables (age/altitude) or that their moderating effects are encompassed by the core variables.

The multivariate regression model further delineated the specific moderating effects of age and altitude, the age groups of 8.3 and 9.5 years were identified as high-risk periods for fracture (β=-6.1; P=0.0048; and β=-5.81, P=0.0191, respectively), indicating that the association between vitamin D deficiency and fracture risk is significantly stronger in these age groups compared to others. For ≥600 m altitude regions, the regression coefficient was β=-2.28 (P=0.0002), indicating a stronger correlational pattern linking vitamin D deficiency and fracture risk among adolescents living aforementioned 600 m, alongside observations of higher fracture prevalence in this subgroup.

Subgroup analyses

To explore potential sources of heterogeneity, the present study conducted prespecified subgroup analyses using random-effects models, stratified by latitude, season/month, trauma energy and study design (Figs. S1, S2, S3 and S4). All forest plots present subgroup analyses performed using a random-effects model. Studies exhibiting substantial clinical heterogeneity within the same stratum were manually excluded from the pooled estimates; these excluded studies are displayed with zero weight or labelled as ‘not estimable’ in the respective plots.

In the latitude-based analysis, studies at ≤40 ˚N showed no significant fracture benefit from vitamin D after excluding Al-Daghri 2016(22) (unique Saudi ethnicity) and Matcovic 2023(8) (distal forearm fractures only), with heterogeneity dropping from 74-44% (P=0.73). Conversely, among studies at >40˚N, excluding Herdea 2023(12) (pandemic period), Mäyränpää 2012(19) (60˚N; low winter sunlight) and Talmac 2021(36) (February-May) reduced heterogeneity from 78-47%, and revealed a significant protective effect of vitamin D on fractures (P<0.00001).

In the seasonal subgroup analysis (18 of 27 studies with clear seasonal data), studies conducted from May to October showed a significant fracture benefit after excluding Delshad 2020 (Southern Hemisphere) and Younes 2019 (wide altitude range), with heterogeneity falling from 87-0% (P<0.00001). In the December-April subgroup, after removing Talmac 2021(36) (late winter/early spring, lowest solar radiation) and extreme exposure conditions, heterogeneity decreased from 93-0%, with no significant fracture benefit (P=0.41). In the full-year subgroup, excluding Herdea 2023(12) (pandemic) and Segal 2023(9) (age up to 15.9 years) reduced heterogeneity from 61-46% and showed a significant effect (P<0.00001).

In the trauma energy stratification (24 studies with clear energy data), the low-energy subgroup showed reduced heterogeneity from 79-38% after excluding seasonally or temporally conflicting studies (P=0.01). The low-or-moderate subgroup exhibited a marked fracture benefit, with heterogeneity dropping from 72-0% (P<0.00001). The mixed-energy subgroup (three studies) showed 0% heterogeneity and no significant benefit (P=0.57), and the unintentional injury subgroup (two studies) also showed 0% heterogeneity and no significant benefit (P=0.35).

In the study design analysis, cross-sectional studies demonstrated significant fracture risk reduction after excluding the two largest studies, which were the primary sources of heterogeneity; heterogeneity decreased from 87-0% (P=0.03).