Dietary calcium intake in relation to type-2 diabetes and hyperglycemia in adults: A systematic review and dose–response meta-analysis of epidemiologic studies

Several epidemiological studies investigated the relation of Ca intake with type 2 diabetes mellitus (T2DM), but there were inconsistencies in their findings. So, we conducted a systematic review and dose–response meta-analysis to quantify the relation of dietary Ca intake with the risk of T2DM/hyperglycemia in adults. A systematic search was conducted up to May 2021, in MEDLINE (Pubmed), Web of Science (WOS), Scopus electronic databases and Google Scholar, for epidemiological studies that investigated the relation of dietary Ca intake (as the exposure) and T2DM/hyperglycemia (as the outcome) in adults, without restriction in publication date and language. Finally, 8 cohort and 9 cross-sectional studies were included in the analysis. The body of evidence was assessed by the GRADE approach. Combining effect sizes from prospective cohort studies included 255,744 general adult population illustrated that highest level of dietary Ca intake, compared to lowest category, was related to an 18% reduced risk of T2DM (RR: 0.82; 95% CI 0.74–0.92). Based on linear dose–response analysis (including 255,744 healthy individuals and 13,531 patients with T2DM), each 300, 600 and 1000 mg/day increment in dietary Ca intake was respectively associated to 7, 14 and 23% reduced risk of T2DM. There was a steeper reduction in risk of T2DM when dietary Ca intake increased from low levels to 750 mg/day. Nevertheless, meta-analysis of cross-sectional studies revealed an inverse significant association between dietary Ca intake and T2DM/hyperglycemia only in the female population (OR: 0.66; 95% CI 0.50–0.88). This meta-analysis illustrated an inverse association between dietary Ca intake and risk of T2DM in general adult populations in prospective cohort studies, in a dose–response manner. It seems that increasing dietary Ca intake from low levels to around 750 mg/day was inversely related to risk of T2DM. In cross-sectional studies, an inverse relation between dietary Ca intake and T2DM/hyperglycemia was found only in females.

ence (WOS) and Scopus electronic databases as well as Google Scholar, up to May 2021. There was no restriction in the time of publication or language. Details of the applied MeSH and non-MeSH keywords in the systematic search are presented in Supplemental Table 1. Furthermore, we performed a manual search in bibliographies of the relevant investigations to identify additional studies. Grey literature including conference proceedings, unpublished articles and theses were not included in the present review. Two researchers (Z.H and P.R) independently performed all processes of the search strategy and the third investigator (P.S) supervised them.
The current analysis was performed according to the PRISMA checklist that is presented in Supplemental Table 2. We registered the study protocol at PROSPERO (CRD42021244394) (http:// www. crd. york. ac. uk/ Prosp ero).
Inclusion criteria. All published papers were included in the current analysis if they: (1) had cohort, crosssectional or case-control design; (2) investigated adult population (≥ 18 years), regardless of their health status; (3) considered dietary Ca intake as the exposure and reported the risk for abnormal glucose homeostasis including, T2DM, prediabetes or hyperglycemia as the outcomes of interest; (4) reported relative risks (RRs), hazard ratios (HR), or odds ratios (ORs), with 95% confidence intervals (CIs) for the association of dietary Ca intake and abnormal glucose homeostasis. The details of PICOS criteria (population, intervention/exposure, comparison/control, outcome, and study design) are presented in Table 1.
In this systematic review and meta-analysis, we considered abnormal blood glucose homeostasis as the primary outcome. Included studies have used various validated methods and cut-off-points for their populations to define abnormal blood glucose homeostasis. They categorized abnormal blood glucose homeostasis as T2DM, prediabetes or hyperglycemia. Therefore, we considered T2DM, prediabetes or hyperglycemia as the main outcomes of interest.  Table 3 contains details of more relevant studies that were not included in the current study. Investigations were excluded if they: (1) standard or un-standard regression coefficient (β or B), reported correlation coefficient, mean ± SE or mean ± SD or median (Inter Quartile Range) for blood glucose across categories of dietary Ca intake; (2) considered change in blood glucose concentrations as the outcome of interest; (3) considered type 1 diabetes (T1D) or gestational diabetes mellitus (GDM) as the outcome; (4) considered hyperglycemia or diabetes as the exposure and dietary Ca intake as the outcome. In addition, studies with overlapping study population were excluded. We found 2 reports from the following investigations: National health and nutrition examination surveys (NHANES 2001(NHANES -2014 19,31 and Multi-Rural Communities Cohort (MRCohort 2005-2011) 18,28 . Therefore, only the publication with higher sample size from each pair of these investigations (NHANES 2007(NHANES -2014 , MRCohort 2005-2011 18 ) was included in the analysis.
Data extraction. The  Quality assessment of studies. We used the Newcastle-Ottawa Scale (NOS) 32 to determine the quality score of studies. Based on NOS a maximum 9 score was given to each cohort investigation, including 5 for individual selection (demonstration that T2DM or hyperglycemia was not present at the start of the study, selection of the non-exposed cohort, ascertainment of dietary Ca intake as the exposure, representativeness of the exposed cohort), 2 scores for comparability (adjustment for confounders including, age and sex), and 3 scores for assessment of the outcome (validated assessment of T2DM or hyperglycemia, adequacy of follow-up of cohorts and enough duration of follow-up for incidence of T2DM or hyperglycemia). Moreover, a maximum of 10 score was assigned for each cross-sectional study. Five scores for participant selection (representativeness of study population, satisfaction of sample size and ascertainment of dietary Ca intake as the exposure, description of non-respondents), 2 scores for comparability (controlling for confounders including, age and sex), and 3 scores for the outcome (validated assessment of T2DM or hyperglycemia and using an appropriate statistical test for the analysis). Details of quality assessment of eligible studies are presented in Supplemental Table 4. In the current meta-analysis, studies with a quality score of 8 or more were determined as "high quality" and other studies were deemed to be "low quality". Moreover, Grading of Recommendations, Assessment, Development and Evaluations (GRADE) was used to determine the quality of evidence 33 through GRADEpro (GRADEproGDT, www. grade pro. org 34 . According to this approach, we evaluated the main factors that could downgrade the study quality including indirectness of evidence, risk of bias, inconsistency of findings, imprecision of findings, and publication bias. The factors upgrading quality were also considered through the evaluation of dose-response analysis, large effect and plausible confounding. Based on GRADE approach, the certainty of the body of evidence could be rated in one of four categories: high, moderate, low and very low. Results of GRADE assessment of this meta-analysis are presented in Supplemental Table 5. Statistical analysis. We used the reported RR/HR/OR and 95% CI for dietary Ca intake in relation to T2DM or hyperglycemia to compute the log RR/HR/OR and its standard error. In order to calculate the overall effect size, the random-effects model that takes between-study variation into account was applied. In order to indicate heterogeneity, we calculated Cochran's Q test as well as I 2 . When heterogeneity was significant, in order to examine its source, we performed subgroup analyses based on several confounders including sex, age (< 50 vs. ≥ 50 years), Ca intake assessment tools, quality of studies, study location (Asian vs. Non-Asian countries), developmental status of the countries, methods of blood glucose measurement, representativeness of study population, and health status of participants. We performed a fixed-effect model to determine between subgroups heterogeneity. Utilizing meta-regression, we also tested the influence of continuous variables (including latitude, quality of studies, longitude and age) on the overall estimate. To assess whether each study could influence the results, sensitivity analysis was applied. Publication bias was examined visually by inspecting funnel plot asymmetry, as well as using statistical tests of Egger's and Begg's. According to the methods previously described by  36 , dose-response analysis was applied by using RRs/HRs/ORs and their 95% CIs, number of cases with T2DM or hyperglycemia, total number of individuals and median or mean level of dietary Ca intake in each category. When an investigation reported the range of Ca intake, we computed the midpoint by averaging the lower and the upper bounds. In studies that reported an open-ended highest category, we considered the length of interval as the same as the adjacent category. In case of the open-ended lowest category, the lower boundary was considered zero and the mean level of dietary Ca intake was calculated. The specific slopes (linear trends) and 95% CIs for 300, 600 and 1000 mg/day dietary Ca intake were computed through the use of natural logs of the RRs/HRs/ORs and their 95% CIs across dietary Ca levels. Non-linear dose-response analysis required studies with at least 3 different categories of dietary Ca intake. Non-linear relation between dietary Ca intake and T2DM or hyperglycemia was assessed using the restricted cubic splines (considering three knots at fixed percentiles of 10%, 50% and 90% of the distribution). We applied statistical analyses through the use of statistical software package Stata version14.0. P values less than 0.05 were considered statistically significant.

Results
Overall, 3924 publications were obtained in our comprehensive systematic search. Additionally, 2 papers were found through the manual search of reference lists. Then, 861 duplicate studies were excluded. In the primary round, the title and abstract of 3065 reports were screened and 2965 investigations were excluded. In the secondary round, the full-text of 100 reports was carefully investigated and after excluding 82 irrelevant investigations, 17 eligible studies were included in the systematic review and meta-analysis, as shown in Fig. 1 17,23 and the remaining in Turkey 20 and Japan 39 (eight studies in Asian countries and 9 others in non-Asian regions). The assessment of dietary Ca intake was performed through the use of food frequency questionnaires (FFQs) in 12 reports, food recall in 4 investigations and food record in the last study. Nine of included studies considered T2DM as the outcome and the others reported high blood glucose (HBG) or hyperglycemia (fasting blood  www.nature.com/scientificreports/ glucose (FBG) ≥ 100 or 110 mg/dL) as the outcome of interest. Fifteen studies investigated generally healthy populations and just 2 cross-sectional investigations were conducted on patients with hypertension 24 and renal transplant recipients 25 . Thirteen investigations have used random sampling method to select their participants, so their subjects were representative of general adult populations; while 4 others did not randomly select their individuals. Fifteen of 17 studies were included in the meta-analysis of highest vs. lowest dietary Ca intake in relation to T2DM/hyperglycemia, while the two remaining investigations 21,23 (one cohort and one cross-sectional study) that reported linear relationships, were only included in the linear dose-response analysis. Among the included studies, 12 studies received a score of 8 or more; therefore, they were classified as high-quality studies, while the remaining studies were classified as low-quality.  (Fig. 2). Nevertheless, the heterogeneity between studies was moderate (I 2 = 53.6, P Q-test = 0.02). Therefore, subgroup analysis was conducted based on study location (Asian vs. non-Asian countries); between-study heterogeneity was removed in both Asian (I 2 = 0.0, P Q-test = 0.57) and non-Asian (I 2 = 0.0, P Q-test = 0.63) subgroups. However, the Ca intake-T2DM relation was significant only in Asian countries (RR: 0.75; 95% CI: 0.69, 0.82) and there was no significant relation in non-Asian regions (RR: 0.98; 95% CI: 0.87, 1.11) (Fig. 2). Further subgroup analyses were also conducted based on other confounders, and the results are shown in Table 3. Higher dietary Ca intake was protectively associated with lower risk of T2DM in subgroups of females, high-quality studies, investigations with adjustment for Mg intake, among participants with mean age ≥ 50 and in developing countries. However, heterogeneity was still significant in some of these subgroups. Additionally, meta-regression was applies and showed that mean age of participants (β = − 0.014, P = 0.02, I 2 residual = 0.0%) and the latitude of study location (β = 0.024, P = 0.03, I 2 residual = 17.92%) would contribute to the overall estimate; nevertheless, longitude (β = −0.002, P = 0.40, I 2 residual = 51.51%) and quality of included studies (β = − 0.019, P = 0.85, I 2 residual = 58.01%) did not have significant effect on the pooled RR. Furthermore, sensitivity analysis showed that none of the eligible studies influenced the    (Fig. 4). The threshold effect of dietary Ca intake was about 750 mg/day; such that, there was a steeper reduction in risk of T2DM when dietary Ca intake increased from low levels to 750 mg/ day. Nevertheless, at Ca intakes above 750 mg/day, the risk was not reduced anymore.
Quality of the evidence. GRADE evidence profile for dietary Ca intake in relation to T2D and hyperglycemia is presented in Supplemental Table 5. The certainty of evidence was rated as "high quality" for cohort stud- www.nature.com/scientificreports/ ies and "moderate quality" for cross-sectional investigations. The endpoint for both cohort and cross-sectional studies was upgraded for "risk of bias", "inconsistency" and "indirectness". For both cohort and cross-sectional studies, 95% CI of overall effect contained a minimal value of 0.75; so, the certainty of evidence was downgraded for "imprecision". Cohort studies had also reported essential data for dose-response analysis; so, the endpoint of these investigations was upgraded for "other considerations". Cross-sectional studies did not provide enough data for dose-response analysis and the endpoint for these studies was downgraded for "other considerations".

Discussion
This meta-analysis documented that in prospective cohort studies highest category of Ca intake, compared to lowest one, was significantly related to decreased risk of T2DM in adults. This inverse significant relation between dietary Ca intake and T2DM was also confirmed in almost all subgroups. Moreover, based on linear dose-response analysis, each 300, 600 and 1000 (mg/day) increment in dietary Ca intake was respectively  www.nature.com/scientificreports/ associated with 7, 14 and 23% reduced risk of T2DM. It seems that the threshold effect of dietary Ca intake was around 750 mg/day; a steeper reduction in risk of T2DM was observed when dietary Ca intake increased from low levels to 750 mg/day. There was no significant relation between dietary Ca intake above 750 mg/day and risk of T2DM. Nevertheless, meta-analysis on cross-sectional studies revealed an inverse significant association between dietary Ca intake and T2DM/hyperglycemia only in the female population and there was no significant linear relationship in cross-sectional studies. Previous investigations documented straight associations between T2DM and risk of chronic conditions including hypertension, dyslipidemia, obesity and degenerative diseases such as Parkinson's disease 40,41 . In addition, T2DM is correlated to double risk of mortality 42,43 ; so, management the increasing rate of T2DM prevalence is a major issue. In the current analysis, we illustrated that subjects with lower dietary Ca intake have higher risk of T2DM; this point could be clinically recommended to individuals to increase their Ca intake via food in a hope to decrease the risk of hyperglycemia and subsequently T2DM.
In line with our study, previous systematic reviews and meta-analyses of epidemiologic studies investigated the relation of Ca intake with the risk of non-communicable disease (NCDs). These researches revealed inverse significant relations between Ca intake and colorectal cancer 44 , breast cancer 45 , ovarian cancer 46 and stroke 47 . Nevertheless, prior investigations could not confirm significant relation between Ca intake and CVD 48 and depression 49 . Yang et al. 8 conducted a meta-analysis on prospective cohort studies and randomized clinical trials (RCTs). They found that dietary Ca intake ranging from 200 to 1500 mg/day did not affect the risk of CVD, coronary heart disease (CHD), or stroke. However, Ca supplements could raise the risk of CHD, and especially myocardial infarction (MI). In 2012, Dong et al. 26 have conducted a meta-analysis on prospective studies and documented an inverse significant relation between Ca intake and T2DM; however, they reported that this relation was dependent on magnesium intake and the overall estimate was not significant in studies that controlled the effect of magnesium intake. In contrast, our analysis illustrated an inverse significant relationship between dietary Ca intake and T2DM in the subgroup of studies that made adjustment for magnesium intake.
In the current analysis, an inverse significant relationship between dietary Ca intake and T2DM was observed in cohort studies. In addition, a significant linear relation was documented. Notably, the threshold for the beneficial effect of dietary Ca intake was around 750 mg/day. Similar to our study, Wang et al. 50 conducted a metaanalysis on prospective cohort studies and documented a U-shaped relation between Ca intake and CVD mortality; such that, there was significant relation between dietary Ca intake lower or higher than around 800 mg/ day and increased risk of mortality from CVD. For all-cause mortality, they found a threshold effect at Ca Table 4. Results of subgroup-analyses of dietary Ca intake in relation to T2DM/hyperglycemia in crosssectional studies. 1 P for heterogeneity, within subgroup. 2 P for heterogeneity, between subgroups. 3 Quality Scores were according to Newcastle-Ottawa Scale. www.nature.com/scientificreports/ intakes around 900 mg/day and the risk did not decrease further, at intakes more than 900 mg/day. They have also found a threshold effect of 900 mg/day for all-cause mortality; the risk did not decline at Ca intakes more than 900 mg/day. Although we documented that Ca intake was related to a lower risk of T2DM in prospective cohort studies, no significant relationship was found in the whole population in cross-sectional studies. In this case, some points might affect the results and should be considered. First, in cross-sectional studies, the number of effect sizes and the sample size of studies were lower than those in cohort studies (n = 40,614 vs. 255,744). Second, all cohort studies were conducted on apparently healthy individuals who were representative of the whole adult populations; so, the results of these cohorts could be generalizable to the whole population. However, both representative and non-representative populations were included in cross-sectional studies. A small number of eligible crosssectional studies (n = 2) were also conducted on patients. Third, most cohort studies controlled the effect of Mg intake; while none of the cross-sectional studies made this adjustment. It is also should be considered that all eligible cohorts had used FFQs to measure dietary Ca intake; whereas cross-sectional studies have applied FFQs, food record and food recall questionnaires. Several replications of 24-h dietary recalls or records (at least 3 to 5) are needed to estimate the usual dietary intakes.
Our stratified analysis by gender among cohort studies revealed an inverse association between highest vs. lowest Ca intake and risk of T2DM in females and both genders together, but the relation was not significant in males. Similarly, in cross-sectional studies, higher Ca intake was related to the reduced odds of T2DM/hyperglycemia in females. Increasing Ca absorption and its bioavailability in premenopausal women by estrogen hormone might be the reason for different findings in men and women 51 .
Although previous studies have documented significant relation between dietary Ca intake and T2DM, they could not exactly clarify underlying mechanisms. Some pathways were suggested to explain this relationship. Earlier investigations documented that hypocalcemia due to vitamin D deficiency could lead to insulin resistance and impaired insulin secretion 52 . An experimental study showed that Ca intake could increase the extracellular Ca, affect the beta cells of pancreas, and improve insulin secretion and insulin resistance in rats with vitamin D deficiency, while vitamin D supplementation had no significant effect 53 . In human studies, hypocalcemia in non-diabetic individuals could also impair insulin release 54,55 , whereas vitamin D supplementation had no beneficial effect on glucose/insulin homeostasis 56,57 and the impact of Ca intake was not tested in this situation. Furthermore, oral calcium load could lead to increased glucose-induced insulin secretion in patients with diabetes 58 . In addition, dietary Ca intake is related to reduced lipogenesis, less inflammation and increased lipolysis. Therefore, it seems that higher dietary Ca intake would decrease obesity, inflammation and consequently insulin resistance and T2DM 22,59 . Furthermore, there are mechanisms for explanation the cellular role of Ca including, first, insulin secretion is calcium-dependent, any change in Ca flux might have an adverse effect on this secretion 60 ; second, Ca plays an important role in insulin action. The phosphorylation of insulin receptors depends on Ca 61 . Hypocalcemia could impair insulin signal transduction 62,63 and decrease the activity of glucose transporter-4 63,64 . Therefore, in insulin-target tissues (such as skeletal muscle and adipose tissue) Ca is necessary for insulin-mediated intracellular processes 62,65 .
As far as we know, this is the first meta-analysis that examined a dose-response relationship between dietary Ca intake and the risk of T2DM/hyperglycemia in epidemiologic studies. In the current analysis, a large population of adults was included and we found significant relation in cohort studies, in a dose-response fashion. Considering the representative population of cohort studies, the results could be generalized to the whole adult population. Additionally, we performed subgroup analyses based on several confounders. GRADE approach provided the certainty that dietary Ca intake is related to reduced odds of T2DM and may have a role in decreasing hyperglycemia. However, some limitations should be kept in mind. First, only a few included studies have separately reported the sources of Ca intake (dietary, supplemental, plant-based and animal-based); therefore, we could only consider dietary Ca intake as the exposure. Second, none of the included cross-sectional studies controlled the impact of Mg intake. Third, cross-sectional studies have used different questionnaires to examine dietary Ca intake including FFQ, food record and food recall. Fourth, included studies had a wide range of followup, centers of data collection and populations. Additionally, participants had a wide range of dietary calcium intake due to the various dietary intakes in each region. Fifth, some included studies had significant sources of bias because of having low sample size, non-representative study population, and high rate of dropout or using invalid food questionnaire. Finally, gray literatures that have not usually passed peer review process and might be low quality or significant sources of bias, were not included in the current analysis. The above-mentioned restrictions might lead to insignificant results and moderate between-study heterogeneity in cross-sectional studies. Furthermore, only a few studies investigated the relation of dietary Ca intake with abnormal glucose homeostasis and their results were inconsistent. Therefore, there was low evidence to support the finding in this field.
In conclusion, this meta-analysis illustrated an inverse association between dietary Ca intake and risk of T2DM in general adult populations in prospective cohort studies, in a dose-response manner. It seems that increasing dietary Ca intake from low levels to around 750 mg/day was related to a lower risk of T2DM. Metaanalysis on cross-sectional studies showed an inverse relation between dietary Ca intake and T2DM/hyperglycemia only in females.