This project was registered in the PROSPERO International prospective register of systematic reviews (registration number: CRD42019138283).

A systematic search was executed in PubMed, Scopus, Web of Science, The Cochrane Library, and PROSPERO databases to identify relevant studies published in English and Spanish restricted to human with no restriction in publication date. The query and keywords of the search are presented in Supplementary S1A. additional manual search was conducted to identify additional relevant publications. The first search was run on July 20, 2018. Afterward, the update was extended until March 20, 2022. This review was carried out adhering to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines (20), and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews and meta-analysis (21, 22). Two independent evaluators evaluated the abstracts (GCC & BJRN), blinded to authorship, authors' institutional affiliation, and study results. If the abstract fulfilled the inclusion criteria, full-text articles were then reviewed. The third and fourth investigators (JRVB & RJMP) independently resolved any disagreement between evaluators. Corresponding authors were reached by e-mail to request such data in the case of relevant studies with missing information.

We include in this systematic review and meta-analysis only observational cross-sectional studies that included pediatric patients (less than 18 years old) with T1D, T2D, or nondiabetic controls reporting mean serum levels of osteocalcin (cOC, ucOC or tOC) and/or coefficient correlations between any type of osteocalcin (cOC, ucOC or tOC) and fasting glucose, fasting insulin, HOMA-IR, HabA1c, HDL-c, LDL-c, body mass index (BMI), waist circumference (WC), height, weight, and age. We excluded studies with no information on mean levels of osteocalcin or the previously mentioned correlations. Nondiabetic controls were those groups of individuals reported as healthy by the authors of the primary studies, namely subjects without type 1, type 2 diabetes, or carbohydrate intolerance.

The following information was extracted using a datasheet based on Cochrane Consumers and Communication Review Group's data extraction template (23): author, year of publication, the country where the study was conducted, type of study, original inclusion and exclusion criteria, diabetes type, type of osteocalcin, the method for measurement of osteocalcin, intra and inter-assay coefficient of variation, the total number of patients included in the study, the total number of participants, mean BMI and age, gender, method of quantification and brand of the kit used for the measurement. Additionally, information about the mean UV index and altitude (as surrogates of sun exposure) of the city where the study was conducted was obtained from Weather Government Agencies of each Country.

Two reviewers (FEPM & RDC) independently assessed the quality of the selected studies. In case of disagreement in the evaluation, a third researcher resolved it (JRVB or RJMP). Quality assessment of observational studies was carried out using the Newcastle–Ottawa Scale for cross-sectional studies. Each study was judged on three dimensions: the study groups' selection, the groups' comparability, and the ascertainment of exposure. One star was given for each signaling question among each dimension. The total number of possible stars were nine, and studies with six or more stars were considered high-quality, while studies with less than six were considered low-quality (24).

Before analysis, all osteocalcin values were converted to the same unit of measurement. Then extracted quantitative data were pooled in the metanalysis. For data analysis, the recommendations of Cochrane Handbook were followed as described: ucOC and tOC were compared between T1D or T2D and healthy controls. The effect size was expressed as standardized mean difference (SMD) by random-effects model (REM) weighted by the inverse of the variance since all studies used randomly sampled (25). Results are presented using forest plots of SMDs and Pearson correlations for the main groups (T1D or T2D and controls). Inter-study variability was assessed using the τ², Cochran's Q, and I² statistics (26). The contribution of individual study heterogeneity was visually assessed by Baujat plots (25).

Additionally, the effect size was determined for correlations (Pearson or Spearman) of tOC, ucOC, or cOC with fasting glucose, fasting insulin, HOMA-IR, HabA1c, HDL-c, LDL-c, body mass index (BMI), waist circumference (WC), height, weight, and age.

Since enough information was available only for tOC, univariate and multiple meta-regressions were performed to add another approach for unexplained heterogeneity and to determine which variables influenced tOC serum levels. The following covariates individually or combined were used: patient age, altitude, UV index, HbA1c, and method for tOC quantification. I² and R² values were reported to present residual heterogeneity and the amount of heterogeneity explained by each variable or by the multiple meta-regression model. Residual analysis was performed to test the validity of the multiple meta-regression model.

Publication bias was visually assessed by contour-enhanced funnel plots and quantified by Egger method. Moreover, a cumulative analysis was performed and presented as a forest plot to assess an “small study effect” defined as the chance of finding a trend towards a larger effect due to the higher probability of a small study of being published when a more “significant” result is found (27, 28). Statistical analysis was conducted using R studio v1.1.463 (R Foundation for Statistical Computing) [package “meta v4.2”].

A total of 557 studies were identified by database searching, and 45 were eligible for full-text review. After review, 20 studies were retained for the systematic review and 17 for meta-analysis. Specific reasons for excluding 25 full-text studies are presented in Supplementary S1B, including no existence of a control group, the inclusion of adults and children (mixed population) in the same analysis, or with diabetic ketoacidosis. Furthermore, studies with no observational design or failure to report osteocalcin serum levels were excluded (Figure 1). Some authors were reached by email to obtain missed data about their article, but no answer was achieved. The general characteristics of the included articles are described in Table 1 (10, 11, 15–17, 29–45).

The Newcastle–Ottawa scale was used for the quality assessment of observational studies. All studies were considered as high quality (had 6 or more points) (10, 11, 15–17, 30–35, 38, 39, 41–44) (Table 2).

From the 20 included studies, a total of 3,000 pediatric patients underwent tOC, cOC, or ucOC assessment, including 983 with T1D, 30 with T2D, and 1987 healthy individuals. A total of 14 studies were performed in T1D (10, 15, 16, 30, 32, 34, 35, 39–45), two in T2D (11, 45), two in prediabetes, and three in HS. The mean patient age at inclusion was 10.6 years, and the male-to-female ratio was 1 : 0.9. The methods used for osteocalcin measurement were EIA in 5 studies (30, 35, 38, 45, 46), RIA in 4 studies (36, 37, 40, 41), ELISA in 6 studies (11, 29, 31, 43), ECLIA in 4 studies (16, 34, 42) and in one study the technique for quantification of osteocalcin was not reported. The intra-and interassay coefficients of variation were from 1.1% to 10.8%.

A total 14 studies compared the total OC serum concentrations among T1D and HS (15–17, 30, 33–35, 39–43, 47). There was lower SMD for serum levels of total OC in T1D than in HS (SMD: −2.2, CI 95%: −3.75 to −0.66). A Q value of 706.39 with 13 degrees of freedom and p < 0.001 provides evidence that the effect size varies across studies, and I² indicates that 98.0% of the variation can be attributed to a true effect rather than a random error (Figure 2).

Baujat plot showed that the studies from Faienza influenced the most to the overall results, whereas Huneif contributed the most to the overall heterogeneity (Figure 3) (15–17, 30, 35, 39, 41). The funnel plot suggested publication bias (Supplementary S2). The cumulative analysis showed no “small-study effect” on effect size (Supplementary S3).

In total, three studies compared the ucOC serum levels among individuals with abnormal glycemic status (AGS = T2D or prediabetes) and normal glycemic status. There was lower SMD for serum levels of ucOC in AGS than in HS (−1.34, CI95%: −2.42 to −0.25). A Q value of 51.95 with 2 degrees of freedom and p < 0.001 provides evidence that the effect size varies across studies, and I² indicates that 96% of the variation can be attributed to a true effect rather than a random error (Figure 4) (11, 36, 37).

Only Huneiff measured ucOC in T1D and found lower serum levels in these patients (1.08 ± 0.56 ng/ml) than in HS (2.58 ± 0.55 ng/ml) (17).

One study measured ucOC serum concentrations in pediatric patients with T2D, reporting lower ucOC serum levels among T2D (0.65 ± 0.46 ng/ml) than in controls (1.25 ± 0.49 ng/ml; p < 0.01) (11).

A total of two studies reported correlations of ucOC with metabolic and anthropometric parameters in HS (29, 31) and three in diabetic patients (11, 17, 45). In healthy subjects, significant positive correlations of ucOC with WC and weight were found but not with other parameters (Supplementary S4). In diabetic children, significant negative correlations of ucOC with HbA1c and glycemia were found in T1D and T2D (Figure 5).

Two studies reported correlations of tOC with metabolic and anthropometric parameters in HS (29, 31) and two in diabetic patients (30, 44). In HS, a significant positive correlation was identified for FPG, LDL-c, WC, height, and weight (Supplementary S5A–J). Among T1D children, a significant negative correlation was also found between tOC and HbA1c, but no data regarding T2D were identified in publications (Supplementary S5K).

One report compared the serum levels of cOC between HS and T2D and found no significant differences (11). No differences in T1D or T2D. Only one study performed in healthy children found that high serum levels of cOC were related to lower HMW-adiponectin in leaner children (a hormone released by osteocalcin in animals that increases insulin sensitivity). Moreover, they found that a higher ucOC-to-cOC ratio was associated with higher HOMA-β in leaner children and related with higher HMW-adiponectin in heavier children, concluding that carboxylation of osteocalcin relates to metabolic and anthropometric parameters in a weight-dependent manner (38).

Since variability was found to be high for tOC estimates, univariate and multivariate meta-regressions were performed to determine which variables influence the serum levels of tOC. In the individual meta-regressions, covariates significantly explaining the heterogeneity were patient age (R²= 25.53%), altitude (R²= 34.67%) and HbA1c (R²= 26.43%) (Table 3; Figure 6).

The multivariate meta-regression indicates that 62.8% of the total heterogeneity was explained by the covariate age, altitude, and HbA1c. In the moderator test, the estimated model was highly significant with QM (df = 4) = 41.21 and p < 0.0001. The patient age (p = 0.007), altitude (p = 0.008), and HbA1c (p = 0.012) have a significant influence on the levels of tOC serum, while the UV index was not significant (p = 0.64) (Table 4).

The estimated model indicates that every one-year increase in the patient's age corresponds to an increment of 0.7557 units in terms of the SMD, while keeping the other covariates constant. Similarly, for every unit of altitude increasing, the true effect size decreases at a rate of 0.0028. Finally, for every one-unit change in HbA1c, the SMD of levels of tOC serum decreases at a rate of 1.01.

Residual analysis was performed (see Supplementary S6), in which it can be noted that the assumptions of homoscedasticity, independence, and normality are satisfied—such analysis allowed the validation of the meta-regression model.

This meta-analysis revealed the following findings: (i)

Among healthy children there was a:   (a)

Positive correlation of ucOC with WC and weight

Since obesity, insulin resistance, glycemic, and lipid abnormalities predispose to children and adolescents to the development of chronic diseases in adulthood such as diabetes, hypertension, and cardiovascular diseases, it is quite important to identify serum markers related to abnormal metabolic phenotypes. Because osteocalcin was consistently related to FPG, HbA1c, LDL-c, WC, height, and weight in infants and adolescents, its routine clinical measurement as a marker children/adolescent metabolic status is plausible. Especially because ucOC has several metabolic effects discovered in preclinical studies, such as increased insulin synthesis, beta-cell proliferation, and insulin sensitivity throughout an action on the pancreas, adipose, and muscle tissue (48). Furthermore, a decrease in ucOC is related to fat accumulation in adipocytes and hepatocytes (49). In contrast, the administration of this hormone decreases the content of triglycerides at these cells (50), improves glucose metabolism, and prevents T2D in mice. Thus, a plausible biological explanation exists for our findings about the correlation of ucOC with glucometabolic and anthropometric parameters.

Our study has several strengths. Firstly, we conducted a rigorous systematic review by independent reviewers and a third one to evaluate and assess bias. Database searching was also done by independent investigators, all blinded to authorship and the hospital where the study was conducted, allowing us to minimize bias when selecting publications for inclusion. Secondly, the standardized mean difference was used to avoid discrepancies in unit measurements along with an exhaustive assessment of heterogeneity by 10⁶ simulations, subgroup analysis, and meta-regression. In addition, multiple methods were used to measure bias in each group, from advanced techniques such as the Copas model to cumulative analysis, aiming to determine the trend toward higher effects in smaller studies. Finally, multiple intrinsic and extrinsic factors were analyzed by single and multivariate meta-regression to address the influence of all these factors on the variability of tOC. It allows us to understand that tOC levels were mainly influenced by age, glycemic control, and altitude but not by latitude, BMI, insulin, or other covariates. This complete analysis is a strength because it should be considered when designing clinical research involving osteocalcin or stablishing reference values. On another way, we must mention some weaknesses, including a relatively small number of studies performed in children with T2D, the use of different methods (RIA, EIA, ELISA) to measure the serum levels of osteocalcin and the non-reporting of the inter-assay coefficients of variation for all the studies. In addition, we cannot assure that the primary studies considered all factors that modify OC serum levels such as the treatment, exercise, and the season when osteocalcin was measured.

Osteocalcin is involved in energy metabolism in pediatric subjects because it is consistently related to FPG, HbA1c, LDL-c, and anthropometric parameters such as WC, height, and weight. Although serum levels of OC are highly variable among populations, such variability is explained by diverse intrinsic and extrinsic factors such as age, altitude, and glycated hemoglobin.

Moreover, its utility as a predictor of metabolic disease risk should be explored in the future, together with potential strategies to increase uCOC serum levels to improve metabolic status.