Circulating magnesium levels and incidence of coronary heart diseases, hypertension, and type 2 diabetes mellitus: a meta-analysis of prospective cohort studies

We conducted a systematic literature review to identify all the studies on the associations between circulating Mg levels and incidence of CHD, hypertension, or T2DM through May 2017. We first searched the electronic database of PubMed (https://​www.​ncbi.​nlm.​nih.​gov/​pubmed) using the following MeSH terms: (“micronutrients” OR “magnesium” OR “magnesium deficiency”) AND (“cardiovascular disease” OR “myocardial infarction” OR “ischemic heart disease” OR “hypertension” OR “blood pressure” OR “coronary heart disease” OR “diabetes mellitus” OR “hyperglycemia” OR “insulin resistance”) AND (“cohort studies” or “follow-up studies” OR “longitudinal studies” OR “prospective studies”). We further reviewed EmBase (http://​www.​elsevier.​com/​online-tools/​embase), SCOPUS (https://​www.​scopus.​com/) and Google Scholar (http://​scholar.​google.​com/​). In addition, we identified relevant articles by manually searching the references of the retrieved studies and review articles.

Articles were selected if they met the following criteria: published in English; had a prospective design; evaluated the association between baseline circulating Mg levels and CHD, hypertension, or T2DM; and reported estimated relative risks (RRs), hazard ratios (HRs), or odds ratios (ORs) with 95% confidence intervals (CIs) or these data could be calculated from the available information. In the included studies, CHD was defined as any coronary heart disease, including ischemic heart disease (IHD) incidence or death, angina, myocardial infarction and sudden cardiac death. Hypertension was defined as a systolic blood pressure ≥ 140 mmHg, or a diastolic blood pressure ≥ 90 mmHg, or the use of anti-hypertensive drugs. T2DM was defined based on the blood glucose levels (fasting plasma glucose ≥7.0 mmol/l, or, non-fasting or 2 h post-load glucose ≥11.1 mmol/l) or the use of anti-diabetic drugs. New-onset CHD, hypertension and T2DM was determined by medical records, self-reports or death certificates. All the articles were identified by an initial screen of abstracts, followed by a full-text review, which was independently conducted by two investigators (JW and PX).

The relevant data were independently extracted by two investigators (JW and PX) using a standardized form, and disagreements were resolved by consensus after discussion with the third investigator (KH). The following information was extracted: study characteristics (study name, publication year, authors, country where the study was conducted, study design, sample size, and follow-up time), participants’ characteristics (age at baseline, race, proportion of male gender, major covariates), exposure (method of assessment, classification), main outcomes, and estimated RRs with 95% CIs for corresponding categories and/or continuous exposure. If more than one multivariate model was reported, the estimates with full adjustment for potential confounders were extracted.

The quality of included studies was ascertained with the Newcastle-Ottawa quality assessment scale (NOS) [36] by 2 investigators (JW and PX) independently. Any disagreement was solved by group discussion with the third investigator (KH). This assessment allowed a total score of up to 9 points. The NOS for cohort studies was divided into three groups: selection of cohort (4 points), comparability of cohort (2 points), and assessment of outcome (3 points). The quality of study was considered high or moderate if the sum score was ≥8 points or between 5 and 7 points, respectively.

In this meta-analysis, we used RRs and 95% CIs as a measure of the effect size for all studies. Hazards ratio was considered RR directly, and OR was considered RR in the main analysis and was transformed into RR in a sensitivity analysis by using the formula: RR = OR/[(1-P ₀) + OR*P ₀)], where P ₀ indicated incidence of the outcome of interest in the reference group [37]. RRs and 95% CIs transformed to their natural logarithms (ln) were used to compute the corresponding standard errors. RRs and 95% CIs were converted to per 0.1 mmol/L increment consistently to describe the linear associations of interest regardless of the original unit (mEq/L, mmol/L or mg/dL) for Mg levels. If a study did not provide the linear association of circulating Mg concentration with the outcomes of interest, we estimated it based on the categorical associations by using Greenland and Longnecker’s method, if the person-time of participants as well as cases were reported for each subgroup of Mg levels [38], or using variance-weighted least squares linear regression if they were not reported. If the highest or the lowest group of Mg levels was an open range, then its upper or lower limit was estimated by assuming its range as the same width as the adjacent category.

We pooled RR estimates separately for each outcome using a random-effects model. We evaluated the statistical heterogeneity of the RRs using Cochran’s Q test with a significance level of 0.10, and quantified the heterogeneity using the I ² statistic with a value of 0–25%, 26–50%, 51–75%, or >75% denoting very low, low, moderate, and high degrees of heterogeneity, respectively. Publication bias was visually examined by funnel plots, and statistically assessed by using Egger’s regression asymmetry test and Begg’s rank correlation test, with a significance level of 0.10. If publication bias existed, the Duval and Tweedie nonparametric “trim and fill” method was used to adjust for the pooled results [39].

The average of follow-up year was calculated as the sum of person-year divided by total number of participants. Sensitivity analyses were conducted to evaluate the influence of replacing the random-effects model with the fixed-effects model and the influence of a single study on the overall association by calculating the pooled estimates while omitting one primary study at each time. In addition, OR was transformed to RR in a sensitivity analysis.

As shown in Fig. 1, we retrieved 805 relevant articles from PubMed. Of them, 795 articles were excluded for one of the following reasons: 1) not a human study; 2) a review/meta-analysis, editorial, or abstract; 3) not published in English; 4) not conducted in the general population, but in patients (e.g., diabetic patients); 5) not relating circulating Mg levels to an outcome of interest; 6) not a prospective cohort design; or 7) did not exclude prevalent cases at baseline. In addition, we found 1 additional articles by searching Google Scholar. Thus, 11 studies were identified and included in this meta-analysis.

Table 1 summarizes the characteristics of the included studies, all of which had a prospective cohort design and had participants without a prior diagnosed outcome of interest at baseline. Some studies estimated RRs/HRs/ORs with 95% CIs for men and women separately [24, 28], or for blacks and whites independently [11], and these were counted as separate cohorts in the meta-analysis. Of these 11 prospective studies, 5 studies (6 cohorts) reported results on CHD with 45,808 individuals and 4437 cases during an average of 10.5 years of follow-up [23‐27], 3 studies (4 cohorts) on hypertension with 14,876 participants and 3149 cases during an average of 6.7 years of follow-up [26, 28, 29], and 4 studies (5 cohorts) on T2DM with 31,284 participants and 2680 cases during an average of 8.8 years of follow-up [11, 30‐32].

The average age at baseline was approximately 52 years old, and about 43.4% participants were male. All the included studies adjusted for age and sex, except for the studies that only included one gender. Also, the primary studies controlled for various potential confounders, including body mass index (n = 9) [11, 23, 25‐30, 32] and/or other body composition variables such as waist/hip ratio (n = 3) [11, 24, 28] and waist circumference (n = 1) [31], smoking (n = 7) [23‐27, 29, 32], alcohol consumption (n = 7) [11, 23‐25, 27, 29, 32], physical activity (n = 6) [11, 23‐25, 28, 30], and education(n = 5) [11, 24, 25, 28, 30]. Few studies adjusted for circulating levels of other nutrients such as calcium and potassium (n = 4) [11, 27, 29, 32] or estimated glomerular filtration rate (n = 2) [26, 32]. Ten out of the 11 included studies were assessed as high quality and only 1 yielded moderate quality (see Additional file 1: Table S2 in supplemental materials).

Five studies (6 cohorts) reported data for CHD (Fig. 2). The summary estimate was 0.86 (95% CI: 0.74, 0.996; P = 0.04). There was no significant heterogeneity among the studies (I ² = 39.6%, P = 0.14) and no significant publication bias (Egger’s test: P = 0.48; Begg’s test: P = 0.57).

Dose-response analysis did not reveal a linear association between circulating Mg levels and incidence of CHD (RR, 0.89; 95% CI: 0.77, 1.03; P _trend = 0.10 with a 0.1 mmol/L increment in Mg levels). High heterogeneity was observed among the 6 included cohorts (I ² = 61.9%, P = 0.02). There was no evidence indicating significant publication bias (Begg’s test, P = 0.14; Egger’s test, P = 0.57).

Three studies (4 cohorts) reported data on incidence of hypertension (Fig. 3). The pooled RR of incident hypertension was 0.91 (95% CI: 0.80, 1.02; P = 0.10) comparing the highest to the lowest circulating Mg levels. No significant heterogeneity among studies was found (I ² = 0.0%, P = 0.48). Neither Egger’s (P = 0.86) nor Begg’s (P = 0.50) test indicated significant publication bias.

A significant inverse linear association was observed between circulating Mg levels and incidence of hypertension (RR, 0.96; 95% CI: 0.93, 0.99; P _trend = 0.02 with a 0.1 mmol/L increment in Mg levels). No significant heterogeneity across studies (I ² = 0.0%, P = 0.69) and no evidence of publication bias were found (Egger’s test: P = 0.62; Begg’s test: P = 0.50).

Four studies (5 cohorts) represented results on T2DM (Fig. 4). The pooled RR for incidence of T2DM comparing the highest to the lowest category of Mg levels was 0.64 (95% CI: 0.50, 0.81; P = 0.01). There was no significant heterogeneity among the studies (I ² = 27.3%, P = 0.24), and no significant publication bias was found (Egger’s test: P = 0.94; Begg’s test: P = 0.46).

According to the available data from 3 cohorts, a non-significant linear association was observed (RR, 0.90; 95% CI: 0.81, 1.002; P _trend = 0.054 with a 0.1 mmol/L increment in circulating Mg levels). But, a high heterogeneity across studies was detected (I ² = 76.4%, P = 0.01). No evidence of publication bias was found (Egger’s test: P = 0.26; Begg’s test: P = 0.31).

The funnel plots did not indicate publication bias for any pooling in this meta-analysis (See Additional file 1: Figure S1 in supplemental materials).

When replacing a random-effects model with a fixed-effects model, the results were generally consistent, except that the linear association between circulating Mg levels and incidence of T2DM became statistically significant (0.94; 95% CI: 0.90, 0.98; P < 0.01) (see Additional file 1: Table S3 in supplemental materials).

Additional file 1: Table S4 (see supplemental materials) presents the influence of a single study on the overall associations of interest by omitting one study at each time from the pooled analysis. The categorical association between circulating Mg levels (highest vs. lowest) and incidence of CHD was attenuated to some extent by omitting Gartside et al. [23] (RR, 0.92; 95% CI: 0.82, 1.03; P = 0.16), Liao et al. [24] (female cohort: RR, 0.88; 95% CI: 0.76, 1.01; P = 0.07; male cohort: RR, 0.86; 95% CI: 0.73, 1.02; P = 0.08), Ford et al. [25] (RR, 0.83; 95% CI: 0.68, 1.02; P = 0.08), or Khan et al. [26] (RR, 0.84; 95% CI: 0.70, 1.02; P = 0.08). The linear association between circulating Mg levels and incidence of CHD was strengthened and became statistically significant (RR, 0.84; 95% CI: 0.71, 0.99; P = 0.04) when Ford et al. [25] was omitted.

No single study substantially changed the categorical association between circulating Mg levels (highest vs. lowest) and incidence of hypertension. However, when the female cohort from Peacock et al. [28] was omitted, the overall linear association between circulating Mg levels and incidence of hypertension was attenuated and became statistically non-significant (RR, 0.97; 95% CI: 0.93, 1.01; P = 0.20).

The overall categorical association between circulating Mg levels (highest vs. lowest) and incidence of T2DM persisted with excluding any single study each time in the meta-analysis, while the linear association between circulating Mg levels and incidence of T2DM was strengthened and became statistically significant when Kao et al. [11] (RR, 0.86; 95% CI: 0.75, 0.97; P = 0.02) was omitted in the meta-analysis.

In addition, when we transformed OR to RR in one study for hypertension [28], and the other one for diabetes [11], the related results were not materially changed (data not shown).

Although no previous meta-analyses specifically focused on CHD, several meta-analyses have studied the association of circulating Mg levels and CVD risk [13, 34]. One meta-analysis of prospective studies published in 2013, which combined CVD incidence and mortality, reported that every 0.2 mmol/L increment in circulating Mg levels was associated with a 30% lower risk of CVD and a 17% lower risk of ischemic heart disease [13]. Another meta-analysis also published in 2013 found that the pooled RR of total CVD events (including CVD incidence and mortality) was 23% lower comparing the highest to the lowest serum Mg levels [34]. Also, a recent meta-analysis observed a significant but heterogeneous inverse association between serum Mg and metabolic syndrome [40]. Another recent meta-analysis investigated dietary Mg intake, but not circulating Mg level, in relation to type 2 diabetes and other CVD endpoints [41]. By contrast, we focused on research about CHD events and our findings contribute additional information to the literature.

To the best of our knowledge, this is the first systematic review to summarize studies on circulating Mg levels and incidence of hypertension. Our findings are generally consistent with previous reviews and meta-analyses of interventional studies on Mg supplementation [42‐44]. For example, one meta-analysis of clinical trials found that Mg supplementation decreased systolic blood pressure (SBP) (3–4 mmHg) and diastolic blood pressure (DBP) (2–3 mmHg) in both normotensive and hypertensive individuals. [43] Another meta-analysis found that Mg supplementation significantly reduced SBP (18.7 mmHg) and DBP (10.9 mmHg) in hypertensive patients with SBP > 155 mmHg [44].

Several meta-analyses examined the association of dietary Mg intake and diabetes risk [9, 45‐47]. For example, an updated meta-analysis of prospective cohort studies published in 2015 found that Mg intake was significantly and inversely associated with risk of T2DM in a nonlinear dose-response manner (P _nonlinearity = 0.003) [45]. In addition, a most-recent published systematic review and meta-analysis of randomized clinical trials reported that oral Mg supplementation improved insulin-sensitivity parameters in those who were at high risk for diabetes [48]. Similarly, findings from our meta-analysis of circulating Mg levels and risk of T2DM are supplementary to previous systematic reviews on dietary Mg intake or Mg supplementation.

Our meta-analysis has several strengths. First, this is the first up-to-date meta-analysis examining the associations between circulating Mg levels and incidence of CHD, hypertension, and T2DM together in the general population. Second, this meta-analysis was based on prospective cohort studies of high or moderate quality from various populations, which reduced the likelihood that our findings were substantially biased by the inherent limitations in the primary studies. Third, the combined sample size was relatively large and the long duration of follow-up enabled us to examine the long-term association. Considering that randomized controlled trials are relatively small and include short follow-up periods, our data provide important supplementary information to the literature. Finally, our conclusions are strengthened by generally consistent findings from both categorical and linear analyses, as well as the robust findings from the sensitivity analyses.

Some limitations also need to be acknowledged. Although we found significant associations, the present meta-analysis was based on 10 published cohort studies. The limited data sources not only restricted us from doing stratified or subgroup analyses, but also may potentially lead to over- or under-estimation of the true associations. Second, similar to other meta-analyses of observational studies, the inherent limitations in primary studies may bias the pooled results, though various potential confounders were considered in the original studies. Of note, factors such as family history of chronic diseases and medications related to Mg metabolism were seldom mentioned in the included studies. Third, the possibility of misclassification from both exposure and outcome could not be completely excluded. However, objective biomarkers were used for Mg assessment, and objective evidence (e.g., medical records and death certificates) was used for outcome assessment in most of the included studies. Fourth, although we did not observe evidence of publication bias, the likelihood could not be completely excluded due to publications in other languages. Finally, we realize that circulating Mg may not be the best biomarker, thus alternative biomarkers (e.g., red blood cell Mg and ionized Mg) are definitely needed in future studies depending on the scale of the study and other considerations such as budget and ethics.

Mg is a co-factor of more than 350 essential metabolic reactions. Most importantly, as a component of the Mg-adenosine triphosphate complex, Mg is involved with all phosphate transfer reactions [6, 7, 49]. Experimental studies show that Mg could: 1) regulate vascular smooth muscle tone through modulating calcium entry and intracellular signal pathways [50, 51]; and 2) regulate endothelial function through adjusting the synthesis and release of vasodilatory prostacyclin and nitric oxide [52, 53]. These vascular effects of Mg form the link between its deficiency and the pathogenesis of CHD and hypertension. Studies also found that Mg inhibited experimental arterial thrombus formation by inhibition of platelet aggregation [54], and Mg deficiency resulted in inflammation in various parts of the heart [55] and acceleration of the atherosclerotic process [56], which were related to the development of CHD. In recent meta-analyses and reviews, researchers found that low diet Mg intake and hypomagnesemia were correlated with low-grade inflammation and oxidative stress [57‐59], both of which are known to be part of the pathogenesis of chronic diseases such as CHD and T2DM.

In addition to Mg involvement in glucose metabolism, including the glycolytic pathway and the Krebs cycle, studies have shown that Mg is essential in insulin signal transduction through the activation of the β-subunit of the tyrosine kinase domain of the insulin receptor, which is a critical step in the trans-membrane signaling cascade of the insulin reaction [6, 7, 60]. Animal experiments found that Mg deficiency was associated with reduced glucose uptake and utilization in insulin-sensitive tissues, thus promoting insulin resistance and the development of diabetes [25, 61]. Furthermore, randomized controlled trials including non-diabetic individuals found improvement in insulin sensitivity together with increased levels of serum Mg after Mg supplementation [62‐64].