Association between hypomagnesemia and mortality among dialysis patients: a systematic review and meta-analysis

Search strategy and selection criteria

This meta-analysis was conducted according to the Preferred Reporting Items of Systematic Reviews and Meta-Analyses (PRISMA) statement (Higgins et al., 2011) and following the Cochrane methods (Salguero et al., 2008). The protocol was prospectively submitted to PROSPERO [CRD42021256187].

Two investigators (CY Huang; CC Yang) searched published studies in PubMed, Embase, Cochrane, Collaboration Central Register of Controlled Clinical Trials, and Cochrane Systematic Reviews until April 2022 without any language limitation (Data S1). Using Medical Subject Headings (MeSH) terms and the PICO (population, intervention, comparison, outcome), we selected key words such as “Renal Dialysis,” “Hemodialysis,” “Peritoneal dialysis,” “Magnesium,” “Hypomagnesemia,” “Hypermagnesemia,” “Mortality,” and “Death” as search terms for our literature review. Prospective and retrospective cohort studies and observational studies were included, but case reports and case series were excluded. Two investigators (CY Huang, CC Yang) independently performed searches and checked all articles for inclusion. If they disagreed on the inclusion of an article, a third author (MY, Jiang) resolved the dispute.

Inclusion and exclusion criteria

Published studies were included if they: (1) reported the impact of serum magnesium on all-cause or CV mortality for KRT patients; (2) used a cohort study design, including retrospective (e.g., cohort and case control study) and prospective cohort studies; and (3) reported the hazard ratio (HR) with 95% confidence intervals (CIs) or provided sufficient data to investigate the outcomes. The following studies were excluded: (1) studies that did not include the outcomes of CV mortality or all-cause mortality, or (2) had follow-up times less than 12 months.

Data extraction

The following data were extracted from the articles: first author, year of publication, patient characteristics (sample size, age, and sex), follow-up duration, and clinical outcomes. Both abstracts and full papers were selected for quality assessment and data synthesis. If the data were incomplete in the text but extractable from the figures (ex. Kaplan-Meier (KM) survival curve), we used WebPlotDigitizer Version 4.1 to extrapolate the data (Drevon, Fursa & Malcolm, 2017). Another author (YT, Huang) then verified these data. We did our best to contact the corresponding authors for more information about the selected studies.

Quality assessment

The risk of bias of the included studies was assessed by two authors (CC, Hsieh; MH, Chuang), independently, using the Newcastle-Ottawa Scale (NOS). The NOS contains three domains which consisted of eight items totally, which represent three quality assessment parameters: selection, comparability, and outcome. One point was given for a low risk of bias, and 0 points were given for a high or unclear risk of bias. A score of 0–3 points was considered to be a low-quality study, a score of 4–6 points equated to a moderate-quality study, with a score of 7–9 points indicating a study was high quality.

Outcomes and definition of hypomagnesemia or non-hypomagnesemia

The primary outcome of this meta-analysis was all-cause mortality, with CV mortality as a secondary outcome. Individual trials had different serum magnesium cut-off values for defining hypomagnesemia and non-hypomagnesemia in KRT patients. Among these studies, the HRs of serum magnesium as dichotomous variables were regarded as “lower magnesium vs. non-hypomagnesium,” according to their respective categories.

Data synthesis and statistical analysis

The random-effects model was employed to analyze the selected outcomes (all-cause mortality and CV mortality) between the two groups. The effect size was expressed as the pool HRs with 95% CIs.

We used funnel plots to demonstrate potential publication bias. Statistical heterogeneity was calculated using I², with an I² more than 50% indicating the studies were not homogeneous. P-values < 0.05 were considered statistically significant. We used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach to assess the certainty of the evidence for each outcome. We used the Comprehensive Meta-Analysis software (Version 3.3.070, November 20, 2014) for all statistical analyses.

Subgroup analysis

We performed subgroup analyses of adjusted/unadjusted outcomes, dialysis modality (HD, PD), serum magnesium cut-off value for defining the hypomagnesemia and non-hypomagnesemia groups (1.58–2.5 mg/dL vs. 2.5–2.79 mg/dL), follow-up period (≥5 years vs. <5 years), study design (retrospective vs. prospective), and if the data were extracted from the KM survival curve.

Search results and included studies

Our initial screening of cohort studies focused on published studies that were available on May 20^th, 2021. We then submitted our study to PROSPERO and completed our comprehensive database collection on April 4^th, 2022.

A total of 3,416 studies were identified from the databases and 224 studies were identified from registers as original research. We removed 810 studies due to duplication. Of the remaining 2,830 articles, a further 2,806 articles were excluded after reviewing the title and abstract (Fig. 1). The remaining 24 studies were then assessed for eligibility. After excluding three more studies that lacked usable data for an analysis, a total of 21 observational studies of 55,232 patients undergoing dialysis were selected for inclusion in the final meta-analysis.

Study characteristics

As shown in Table 1, the included studies were published from 2007 to 2020, including 12 studies from Asia (Ishimura et al., 2007; Kurita et al., 2015; Ago et al., 2016; Cai et al., 2016; Yang et al., 2016; Ye et al., 2018; Mizuiri et al., 2019; Shimohata et al., 2019; Tamura et al., 2019; Wu et al., 2019; Lu et al., 2020; Mizuiri et al., 2020), seven from Europe (Markaki et al., 2012; Vervloet et al., 2013; João Matias et al., 2014; Lacson et al., 2015; de Roij van Zuijdewijn et al., 2015; Garagarza et al., 2015; Selim et al., 2017), and two from North America (Fein et al., 2014; Li et al., 2015). Thirteen studies were retrospective studies (Ishimura et al., 2007; Fein et al., 2014; de Roij van Zuijdewijn et al., 2015; Lacson et al., 2015; Ago et al., 2016; Cai et al., 2016; Li et al., 2016; Yang et al., 2016; Mizuiri et al., 2019; Shimohata et al., 2019; Wu et al., 2019; Lu et al., 2020; Mizuiri et al., 2020), and the other eight studies were prospective studies (Markaki et al., 2012; Vervloet et al., 2013; João Matias et al., 2014; Garagarza et al., 2015; Kurita et al., 2015; Selim et al., 2017; Ye et al., 2018; Tamura et al., 2019).

Hemodialysis (HD) patients were the study population in 15 studies, (Ishimura et al., 2007; João Matias et al., 2014; de Roij van Zuijdewijn et al., 2015; Garagarza et al., 2015; Kurita et al., 2015; Lacson et al., 2015; Li et al., 2015; Ago et al., 2016; Selim et al., 2017; Mizuiri et al., 2019; Shimohata et al., 2019; Tamura et al., 2019; Wu et al., 2019; Lu et al., 2020; Mizuiri et al., 2020), peritoneal dialysis (PD) patients were the study population in four studies (Fein et al., 2014; Cai et al., 2016; Yang et al., 2016; Ye et al., 2018), and two studies had a study population that combined HD and PD patients (Markaki et al., 2012; Vervloet et al., 2013). The mean ages of the HD and PD populations were 62 ± 6 and 55 ± 6 years, respectively. The follow-up duration of the 21 included studies ranged from 12 to 129.6 months, with one study not reporting the follow-up duration (Garagarza et al., 2015). Twenty studies reported the association of serum magnesium and all-cause mortality (Ishimura et al., 2007; Markaki et al., 2012; Fein et al., 2014; João Matias et al., 2014; de Roij van Zuijdewijn et al., 2015; Garagarza et al., 2015; Kurita et al., 2015; Lacson et al., 2015; Li et al., 2015; Ago et al., 2016; Cai et al., 2016; Yang et al., 2016; Selim et al., 2017; Ye et al., 2018; Mizuiri et al., 2019; Shimohata et al., 2019; Tamura et al., 2019; Wu et al., 2019; Lu et al., 2020; Mizuiri et al., 2020) and nine studies reported the association of serum magnesium and CV mortality (Ishimura et al., 2007; Vervloet et al., 2013; João Matias et al., 2014; de Roij van Zuijdewijn et al., 2015; Cai et al., 2016; Ye et al., 2018; Mizuiri et al., 2019; Tamura et al., 2019; Lu et al., 2020). The cut-off value of serum magnesium used for defining hypomagnesemia ranged from 1.7 to 2.77 mg/dl.

Heterogeneity and publication bias

The heterogeneity was 82.77% for all-cause mortality and 71.46% for CV mortality according to the I² test. Publication bias, as evaluated using funnel plots and Egger’s test, was significant for all-cause mortality (Fig. S1), but not significant for CV mortality (Fig. S2). We used subgroup analyses and meta-regression to investigate the observed heterogeneity.

Quality of enrolled trials

The NOS score of the included studies was 6–9 (Table S1), with 20 of the 21 studies reaching the high-quality NOS score range of 7–9. The one study that did not reach a score between 7–9 (Garagarza et al., 2015) received a score of six points. This study was considered to be of moderate quality because it did not include the follow-up duration of the study. There was a high inter-observer reliability in article selection (Kappa coefficient = 0.90).

Assessment of evidence quality and summary of findings

An evidence quality assessment was performed using the GRADE system for the outcomes of all-cause mortality and CV mortality (Table S3).

Serum magnesium and all-cause mortality

Twenty studies of 54,471 KRT patients reported the association between serum magnesium levels and all-cause mortality. Ten studies presented both unadjusted and adjusted HR (de Roij van Zuijdewijn et al., 2015; Kurita et al., 2015; Li et al., 2015; Ago et al., 2016; Cai et al., 2016; Yang et al., 2016; Selim et al., 2017; Shimohata et al., 2019; Wu et al., 2019; Mizuiri et al., 2020), seven studies included only adjusted HR (Ishimura et al., 2007; Markaki et al., 2012; Fein et al., 2014; João Matias et al., 2014; Garagarza et al., 2015; Ye et al., 2018; Lu et al., 2020), and three studies included only unadjusted HR (Lacson et al., 2015; Mizuiri et al., 2019; Tamura et al., 2019). Four studies extracted data for the HR calculation from the KM survival curve (Table 2; Fein et al., 2014; Mizuiri et al., 2019; Tamura et al., 2019; Lu et al., 2020; Mizuiri et al., 2020). The results of our meta-analysis showed that hypomagnesemia was associated with increased risk of all-cause mortality compared with non-hypomagnesemia among KRT patients (random effect, HR:1.67, 95% CI [1.41–2.00], p < 0.001, (Fig. 2A)). However, we found high heterogeneity among the included studies (random effect model, I² value of 82.77%, Fig. 2A).

Serum magnesium and cardiovascular mortality

Nine out of the 21 studies reported the relationship between serum magnesium level and CV mortality among dialysis patients (Ishimura et al., 2007; Vervloet et al., 2013; João Matias et al., 2014; de Roij van Zuijdewijn et al., 2015; Cai et al., 2016; Ye et al., 2018; Mizuiri et al., 2019; Tamura et al., 2019; Lu et al., 2020). Four of these studies presented both unadjusted and adjusted HR (João Matias et al., 2014; de Roij van Zuijdewijn et al., 2015; Cai et al., 2016; Ye et al., 2018), three included only unadjusted HR (Tamura et al., 2019; Lu et al., 2020; Mizuiri et al., 2020), and the remaining two studies only included adjusted HR (Ishimura et al., 2007; Vervloet et al., 2013). One of these studies extracted data for the HR calculation from the KM survival curve. The results of this meta-analysis showed an increased risk of CV mortality for dialysis patients with hypomagnesemia compared to those with non-hypomagnesemia. (random effect, HR:1.56, 95% CI [1.08–2.25], p < 0.001, (Fig. 2B)). However, we found high heterogeneity among the included studies (random effect model, I² value of 71.46%, Fig. 2B).

Subgroup analysis

A subgroup analysis showed that the association between hypomagnesemia and increased all-cause mortality (Fig. 3A) and CV mortality (Fig. 3B) among dialysis patients was consistently significant when stratified by follow-up duration and data extraction. A subgroup analysis by dialysis modality revealed that hypomagnesemia was associated with increased all-cause and CV mortality among HD patients, but this association was not present among PD patients. In addition, while the associations between hypomagnesemia and all-cause mortality were significant for serum magnesium cut-off values both above and below 2.5 mg/dL, the association between hypomagnesemia and CV mortality was only significant for a cut-off value less than 2.5 mg/dL.

Meta-regression analysis

The quantitative measures of serum magnesium levels, older age, prevalence of diabetes, or male proportion were not associated with all-cause mortality (Age, Z = 0.12, p = 0.90, Fig. S3A; DM, Z = −1.27; p = 0.205, Fig. S3B; Male, Z = 0.35, p = 0.72, Fig. S3C; the quantitative measures of serum levels of Mg, Z = 0.95, p = 0.34) or CV mortality (Age, Z = 0.29, p = 0.7685, Fig. S4A; DM, Z = 0.75; p = 0.4548, Fig. S4B; Male, Z = 1.69, p = 0.0.0907, Fig. S4C; the quantitative measures of serum levels of Mg, Z = 0.91, p = 0.3636) between the hypomagnesemia and non-hypomagnesemia groups (Table S4).

Strengths and limitations

Our study has some significant strengths. This is the first meta-analysis comprehensively evaluating the association of serum magnesium concentration on all-cause mortality and CV mortality in dialysis patients, specifically. Our findings represent the current evidence supporting the potential influence of hypomagnesemia on relevant clinical outcomes. We also extracted data from the KM survival curve using the WebPlotDigitizer, making our data more complete.

However, our study results should be interpreted cautiously because of its limitations. First, a high heterogeneity was observed among the included studies, which could be related to differences among the studies in serum magnesium cut-off values and follow-up period lengths. Two included studies (Vervloet et al., 2013; Garagarza et al., 2015) were only available as abstracts and were thus lacking complete, available data. The inclusion criteria for patients also differed among the included studies. For example, the proportion of diabetes was vastly different among the studies, with one study even excluding DM patients (Shimohata et al., 2019). These inclusion differences mean the patients included in this meta-analysis may not match the characteristics of real-world patients on dialysis. Two of the included studies (Markaki et al., 2012; Vervloet et al., 2013) combined the data for both HD and PD patients, which may interfere with the analyses of HD and PD patients, separately. More studies investigating the relationship between serum magnesium and clinical outcomes in PD patients, specifically, are needed. In three of the included studies (Kurita et al., 2015; Lacson et al., 2015; Yang et al., 2016), participants were separated into multiple serum magnesium subgroups (ex. low, middle, higher groups). For these studies, we compared the outcomes of the lowest and the highest magnesium groups, which may have induced relevant bias in the effect estimates.