Original Investigation Serum Magnesium and Mortality in Hemodialysis Patients in the United States: A Cohort Study Eduardo Lacson Jr, MD, MPH,1 Weiling Wang, MS,2 Lin Ma, MS,1 and Jutta Passlick-Deetjen, MD3 Background: Low serum magnesium levels in patients with kidney disease have been linked to increased mortality. This study investigated whether similar associations existed in maintenance hemodialysis (HD) patients. Study Design: Cohort study. Setting & Participants: All Fresenius Medical Care North America in-center HD patients with available serum magnesium measurements were studied. The initial exploratory study in 21,534 HD patients evaluated associations among serum magnesium level, dialysate magnesium concentration, and mortality from April 2007 through June 2008. The follow-up study in 27,544 HD patients evaluated associations between serum magnesium levels and mortality over 1 year (January through December 2008). Predictors: The primary predictor was serum magnesium level, with adjustment for case-mix (age, sex, race, diabetes, and dialysis vintage and additionally for follow-up study: body surface area and vascular access) and laboratory variables (albumin, hemoglobin, phosphorus, equilibrated Kt/V, potassium, calcium, and intact parathyroid hormone values). Outcome: Primary outcome variable was 1-year mortality risk, evaluated using Cox proportional hazards models. Results: Among 21,534 HD patients in the exploratory study, there were 3,682 deaths. Higher dialysate magnesium level was associated with higher serum magnesium level (R 5 0.22; P , 0.001). Patients with the lowest serum magnesium levels (,1.30 mEq/L) were at highest risk for death (HR, 1.63; 95% CI, 1.30-1.96; reference serum magnesium, 1.60-,1.90 mEq/L). Among 27,544 HD patients in the follow-up study, there were 4,531 deaths. In Cox proportional hazards models, there was a linear decline in death risk from the lowest to the highest serum magnesium category, with the best survival at serum magnesium levels $ 2.50 mEq/L (HR, 0.68; 95% CI, 0.56-0.82). However, risk estimates were attenuated with case-mix and lab adjustment. This pattern was consistent within diabetes subgroups and for cardiovascular or noncardiovascular causes of death. Limitations: Observational study with cross-sectional serum magnesium measurements and no information for oral magnesium intake. Conclusions: Elevated serum magnesium levels . 2.10 mEq/L were associated with better survival than low serum magnesium levels , 1.30 mEq/L in HD patients. Prospective studies may determine whether manipulation of low serum magnesium levels affects survival. Am J Kidney Dis. -(-):---. ª 2015 by the National Kidney Foundation, Inc. INDEX WORDS: Serum magnesium concentration; dialysate magnesium prescription; hypomagnesemia; hypermagnesemia; hemodialysis (HD); end-stage renal disease (ESRD); mortality risk.

B

ased on observational studies in hemodialysis (HD) patients, it has been suggested that low serum magnesium levels are associated with cardiovascular (CV) morbidity (eg, mitral annular calcification, peripheral arterial calcification, and increased carotid intima-media thickness [cIMT]).1-3 Moreover, magnesium may have a myocardial protective role.4 In 1 Portuguese and 2 Japanese studies, higher mortality rates were observed in maintenance HD patients with low serum magnesium levels.5-7 The current study investigated whether similar associations existed in a large national US cohort. This observational study of patients from Fresenius Medical Care North America (FMCNA) outpatient dialysis facilities was conducted in 2 stages. The first stage was a retrospective exploratory evaluation into the distribution of serum magnesium levels and dialysate magnesium concentrations in a convenience sample of period-prevalent HD patients. We evaluated Am J Kidney Dis. 2015;-(-):---

both the correlation between serum magnesium level and dialysate magnesium prescription and the potential association with 1-year mortality risk, with further consideration for serum potassium, calcium, and intact parathyroid hormone (iPTH) levels. The second stage involved a follow-up study with a hypothesis that, unlike borderline hypermagnesemia, extremely high serum magnesium levels may associate From 1Fresenius Medical Care North America, Waltham, MA; Alcon Laboratories, Fort Worth, TX; and 3Department of Nephrology, University of Duesseldorf, Duesseldorf, Germany. Received March 6, 2015. Accepted in revised form June 4, 2015. Address correspondence to Eduardo Lacson, Jr, MD, MPH, c/o Lin Ma, Fresenius Medical Care North America, 920 Winter Street, Waltham, MA 02451. E-mail: [email protected]  2015 by the National Kidney Foundation, Inc. 0272-6386 http://dx.doi.org/10.1053/j.ajkd.2015.06.014

2

1

Lacson et al

with higher death risk. This study was nested within a prospective observational evaluation of potentially actionable factors associated with mortality risk in a larger prevalent cohort of FMCNA HD patients.8 This stage investigated a wider range of serum magnesium levels, with greater focus on serum magnesium levels above the normal range because elevated levels are not uncommon in HD patients. Furthermore, we determined whether serum magnesium levels have similar associations in patients with diabetes and those without and evaluated non-CV causes of death and CV death, as well as the subset of sudden cardiac death within all CV deaths, each investigated as distinct outcomes among patients who had a documented cause of death.

METHODS Initial Exploratory Study The source population (n 5 83,744) included all FMCNA incenter HD patients who were actively treated as of July 1, 2007, and with serum magnesium results (n 5 21,534) for the prior 3month baseline period (April 1 to June 30, 2007) from all facilities reporting data to the Knowledge Center data warehouse.9 Patients were followed up for 1 year (up to June 30, 2008) with the combination of death or withdrawal from dialysis therapy leading to death as the primary outcome. Almost all (95%) patients were consistently treated by thriceweekly HD, with the rest having additional fourth treatments intermittently for fluid overload and other indications and very few patients treated 2 times weekly, with proportions comparable within serum magnesium subgroups. Patients were treated almost exclusively using Fresenius polysulfone dialyzers. Median treatment time was 225 (interquartile range, 210-240) minutes and was similarly distributed within serum magnesium subgroups. No hemodiafiltration was performed. Case-mix variables at baseline included patient age, sex, race, with or without diabetes mellitus, and duration of dialysis therapy prior to study entry (dialysis vintage). Prescribed dialysate magnesium concentration was obtained from physician orders. If patients had their dialysate magnesium concentration changed during the baseline period, a weighted mean of different dialysate magnesium prescriptions was calculated in proportion to the time that each order was in effect. Prescribed dialysate magnesium concentration was stratified into 5 categories: ,0.75, 0.75 to 0.99, exactly 1.0 (reference), 1.01 to 1.49, and $1.50 mEq/L. In general, dialysate magnesium prescriptions were not based on an individual level, but determined at a facility level. For models involving dialysate magnesium concentration, an additional model was constructed that included an adjustment for serum magnesium level only, with serum magnesium input as a continuous variable. Dialysate prescriptions for calcium, sodium, and bicarbonate/acetate concentrations were not systematically linked to dialysate magnesium concentrations. All laboratory examinations were performed by a single laboratory (Spectra Laboratories). A photometric color test for quantitative magnesium determination (Beckman Coulter; Olympus America, Inc) was used, which has a normal range of 1.30 to 2.10 mEq/L and coefficient of variation of about 1.3% to 1.6% (though the latter has been observed to be as low as w1.2%). The test is linear within a concentration range of 0.5 to 8.0 mEq/L for serum and plasma samples. Change in serum magnesium concentrations over about 6 months was analyzed and stability was confirmed in a subset of more than 16,768 study patients (w78%) with available repeat test results (Table S1, available as online supplementary material). In addition to 2

recording the mean value of all available serum magnesium levels and equilibrated Kt/V for patients during the baseline period, mean predialysis values were obtained for albumin, hemoglobin, phosphorus, potassium, calcium, and iPTH. Correlation coefficients were obtained between serum magnesium and dialysate magnesium values, as well as to each of 3 laboratory variables selected a priori: potassium, calcium, and iPTH. Cox proportional hazard models produced hazard ratios (HRs) for mortality from unadjusted, case-mix–adjusted, and casemix plus laboratory variable–adjusted models (ie, albumin, hemoglobin, phosphorus, equilibrated Kt/V, potassium, calcium, and iPTH), with higher serum magnesium level associating with lower mortality risk. The proportionality assumption was tested using visual inspection of survival curves and the SAS Proc PHREG Proportionality Test; all variables satisfied proportionality assumptions with the exception of albumin and hemoglobin, but this was mitigated by the large sample size. With .80% of patients in the study cohort having baseline serum magnesium levels within the normal range (1.30-2.10 mEq/L),10 categorization into equal patient numbers in the form of quartiles and sextiles would have not allowed for differentiating clinically meaningful serum magnesium levels. Thus, the categories chosen were based on clinical criteria: hypomagnesemia: magnesium , 1.30 mEq/L; low, mid, and high-normal magnesium levels: 1.30 to ,1.60, 1.60 to ,1.90, and 1.90 to 2.10 mEq/L, respectively; and hypermagnesemia: .2.10 mEq/L. The mid-normal range (1.60,1.90 mEq/L) was used as the reference group.

Follow-up Analyses Follow-up analyses were conducted according to the same methodology, with the following exceptions. The baseline period required serum magnesium levels obtained October 1 through December 31, 2007. Similarly, mortality was tracked for 1 year (up to December 31, 2008). Both CV disease (CVD)-related death (International Classification of Diseases, Ninth Revision [ICD-9] codes: 390.xx-447.xx) and the subset with sudden death (ICD-9 codes: 427, 427.4x-427.5x, 427.8x, and 427.9) were identified. Case-mix variables included all prior exploratory variables with the addition of body surface area and vascular access type. Laboratory data were the same except for the omission of potassium level. This analysis of higher serum magnesium level categories was made possible by the larger sample size of the analytical file, which was primarily intended for another study designed to determine the top 5 potentially actionable variables associated with hospitalization and death.8 Survival analyses were conducted as described in a previous section. A stratified analysis was also performed to elicit potential effect modification by diabetes. Cox models were similarly constructed using the subgroup of patients with cause-of-death data to determine the association between serum magnesium level and non-CV deaths, serum magnesium level and CV deaths, and (separately) serum magnesium level and the smaller group with sudden cardiac deaths. The follow-up cohort maintained the first 4 categories and the reference category, but further broke down the hypermagnesemic categories into 3 more: .2.10 to ,2.30, 2.30 to ,2.50, and $2.50 mEq/L. Therefore, the exploratory and follow-up cohorts had differing categories of serum magnesium. Post hoc, follow-up for all-cause mortality was extended to 3 years to determine whether the findings were maintained (Fig S1). All statistical analyses were performed using SAS, version 9.2/9.3 (SAS Institute Inc).

RESULTS Baseline Demographic Characteristics of Both Studies The initial exploratory study cohort represented 21,534 of 83,744 (26%) of all FMCNA maintenance Am J Kidney Dis. 2015;-(-):---

Serum Magnesium and Mortality in HD

HD patients during the period, whereas the follow-up analysis was 27,544 of 110,271 (25%) of a larger study cohort of patients available for analysis.8 Table 1 shows baseline characteristics of both the exploratory and follow-up cohorts, together with their respective source populations (ie, all eligible HD patients during each respective baseline period). Both study cohorts show a similar distribution of baseline characteristics compared with their source populations, with the exception of a slight oversampling of black relative to white patients. Initial Exploratory Study Distributions of serum magnesium and dialysate magnesium concentrations are shown in Fig 1A. For each category of prescribed dialysate magnesium, mean serum magnesium level was higher overall than the prescribed dialysate magnesium concentration (only 0.2% of patients had serum magnesium levels less than dialysate magnesium levels), with a positive correlation (Pearson coefficient: R 5 0.22; P , 0.001). Dialysate

magnesium concentrations showed minimal or no correlation with serum potassium (R 5 0.09), calcium (R 5 0.13), or iPTH (R 5 20.02) levels. However, serum magnesium level correlated positively with baseline serum potassium, albumin, phosphorus, and calcium levels (R 5 0.30, 0.27, 0.21, and 0.20, respectively; all P , 0.001), but was not correlated with iPTH level (R 5 0.016). There were 3,682 deaths (17.1%), and as shown in Fig 1B, increasing serum magnesium levels were associated with decreasing 1-year mortality risk (reference serum magnesium category, 1.60-1.89 mEq/L). The HRs indicated an almost linear trend, with the highest risk associated with hypomagnesemia (serum magnesium , 1.30 mEq/L) at unadjusted HR of 1.6 (95% confidence interval [CI], 1.30-1.96), although few patients (n 5 377) were within this category. However, patients with low-normal serum magnesium levels (1.30-,1.60 mEq/L) also exhibited a higher risk for death (unadjusted HR, 1.33; 95% CI, 1.21-1.45), whereas above-normal serum magnesium

Table 1. Patients’ Characteristics for 2 Baseline Periods for Both the Exploratory and Follow-up Analyses April 1, 2007-June 30, 2007

October 1, 2007-December 31, 2007

Exploratory Analysis (n 5 21,534)

All FMCNA HD (n 5 83,744)

Follow-up Analysis (n 5 27,544)

All FMCNA HD (n 5 110,271)

Age, y Male sex Race White Black Other

61.7 6 14.8 11,650 (54.1)

61.7 6 14.9 45,611 (54.5)

61.8 6 14.8 14,799 (53.7)

61.9 6 15.0 60,067 (54.5)

10,463 (48.6) 9,590 (44.5) 1,481 (6.9)

43,204 (51.6) 33,956 (40.6) 6,584 (7.9)

13,394 (48.6) 12,070 (43.8) 2,080 (7.6)

56,648 (51.4) 44,662 (40.5) 8,961 (8.1)

Diabetes Dialysis vintage, y Body surface area, m2 Dialysate Mg, mEq/L Laboratory values Mg,a mEq/L Potassium,a mEq/L Calcium,a mg/L iPTH,a pg/mL Albumin,a g/dL Hemoglobin,a g/dL Phosphorus,a mmEq/L eKt/V

11,444 (53.1) 2.4 [1.0-4.7]

44,616 (53.6) 2.5 [1.0-4.8]

0.94 6 0.22

0.96 6 0.22

—

14,756 (53.6) 2.5 [1.0-4.8] 1.84 6 0.28

59,277 (53.8) 2.5 [1.1-4.9] 1.84 6 0.28

1.84 6 0.32 4.74 6 0.60 9.16 6 0.68 279 [180-462] 3.79 6 0.41 11.95 6 1.14 5.42 6 1.42 1.48 6 0.30

1.84 6 0.32b 4.77 6 0.60 9.16 6 0.68 279 [181-458] 3.80 6 0.40 11.96 6 1.13 5.42 6 1.40 1.48 6 0.30

1.86 6 0.32 4.79 6 0.60 9.08 6 0.68 272 [178-436] 3.80 6 0.40 11.88 6 1.14 5.39 6 1.40 1.48 6 0.29

1.86 6 0.32b 4.78 6 0.60 9,08 6 0.68 273 [180-438] 3.81 6 0.39 11.88 6 1.12 5.39 6 1.40 1.48 6 0.29

— — — —

— — — —

12,158 (44.1) 6,699 (24.3) 8,524 (30.9) 163 (0.6)

49,914 (45.3) 26,723 (24.2) 33,058 (30.0) 576 (0.5)

Characteristics

Vascular access Fistula Graft Catheter Other

—

—

—

Note: Values for categorical variables are given as number (percentage); values for continuous variables, as mean 6 standard deviation or median [interquartile range]. Conversion factors for units: Mg in mEq/L to mmol/L, 30.5; serum calcium in mg/dL to mmol/ L, 30.2495; serum phosphorus in mg/dL to mmol/L, 30.3229. Abbreviations: eKt/V, equilibrated Kt/V; FMCNA, Fresenius Medical Center North America; HD, hemodialysis; iPTH, intact parathyroid hormone, Mg, magnesium. a Serum values. b Value same as in study cohort, by definition. Am J Kidney Dis. 2015;-(-):---

3

Lacson et al

Figure 1. Exploratory study hemodialysis (HD) patient cohort (n 5 21,534). (A) The distribution of HD patients according to the weighted mean of their prescribed dialysate magnesium concentration (bars) during the 3-month baseline period. Mean serum magnesium level with 95% confidence interval is shown for the same period (solid line). (B) Hazard ratio (HR) of death according to baseline serum magnesium levels. The horizontal dotted line tracks the HR of 1 (same as the reference group) across serum magnesium categories to facilitate interpretation. The 2 vertical dashed lines represent the boundaries of common lower and upper limits of the laboratory serum magnesium assay’s normal range. P , 0.05; *P , 0.001. Abbreviations: CM, case-mix; lab, laboratory. ˇ

levels (.2.10 mEq/L) exhibited the lowest unadjusted HR of 0.67 (95% CI, 0.60-0.75). With subsequent case-mix and case-mix plus laboratory variable adjustments, the trend was attenuated, although a significant survival advantage remained for patients with high-normal serum magnesium levels (1.90-2.10 mEq/L: HR, 0.89; 95% CI, 0.80-0.95) and patients with hypermagnesemia (.2.10 mEq/L: HR, 0.79; 95% CI, 0.67-0.85). There was no independent association between dialysate magnesium concentration 4

and mortality risk, and a center effect was not detected (data not shown). Follow-up Study The follow-up study was performed in order to further explore survival at serum magnesium categories greater than the laboratory upper limit in a larger cohort. Kaplan-Meier survival curves confirmed that patients with serum magnesium levels above this limit (.2.10 mEq/L) had a survival advantage. There was Am J Kidney Dis. 2015;-(-):---

Serum Magnesium and Mortality in HD

no significant difference between patients with serum magnesium levels of .2.10 to ,2.30 mEq/L (n 5 3,593) or 2.30 to ,2.50 mEq/L (n 5 1,564) versus patients with serum magnesium levels $ 2.50 mEq/L, although there were fewer patients within this highest serum magnesium category (n 5 976; Fig 2). There were 4,531 (16.5%) deaths during the year, and as shown in Fig 3A, unadjusted HRs for death (reference serum magnesium category, 1.60-,1.90 mEq/L) decreased significantly from the lowest to the highest serum magnesium category ($2.50 mEq/L: HR, 0.68; 95% CI, 0.56-0.82). While differences in case-mix and laboratory variable results were observed between serum magnesium categories (Table 2), adjusting for these differences only attenuated the risk estimates but maintained similar results. For the highest serum magnesium category ($2.50 mEq/L), the case-mix–adjusted HR remained significantly lower (HR, 0.74; 95% CI, 0.61-0.88) than the reference category (serum magnesium, 1.60-,1.90 mEq/L), but lost statistical significance after further adjustment for laboratory variables (although the HR of 0.89 [95% CI, 0.74-1.09] remained below that for the reference range and did not indicate higher death risk). Subgroup analyses included stratification by the presence of diabetes (n 5 14,756) versus none (n 5 12,788; Figs 3B and C, respectively) and limited outcomes analyses for non-CVD deaths (n 5 2,077), CVD deaths (n 5 1,711), and the subgroup of sudden cardiac deaths within CVD deaths (n 5 1,174), shown in Fig 4A, B, and C, respectively. The association of

Figure 2. Unadjusted Kaplan-Meier survival analysis for all follow-up study patients (n 5 27,544) according to different serum magnesium level categories (unadjusted log-rank P , 0.001). Unadjusted Cox models used serum magnesium level $ 2.50 mEq/L as reference. Compared with the reference category, serum magnesium levels .2.10 to ,2.30 and 2.30 to ,2.50 mEq/L were not significantly different (P 5 0.4 and 0.9, respectively). However, serum magnesium categories of 1.30 to 2.10 and ,1.3 mEq/L were significantly different from the reference group (P , 0.001 for both) and from each other (P 5 0.001). Am J Kidney Dis. 2015;-(-):---

high serum magnesium level with lower mortality risk was consistent in diabetic and nondiabetic patients, but lost significance after adjustment. The association of high serum magnesium level with better survival trended for both non-CVD and CVD deaths, as well as in the subset of CVD deaths due to sudden cardiac death, but was not significant for adjusted models. Similarly, low and low-normal serum magnesium levels only had higher mortality risk prior to laboratory variable adjustment, primarily driven by non-CV deaths (Fig 4A).

DISCUSSION This study demonstrated that higher serum magnesium levels were associated with lower mortality risk in HD patients. To our knowledge, this is only the second large nationally representative study investigating the relationship between serum magnesium levels and mortality risk in HD patients, allowing for categories that represent the broad range of observed serum magnesium levels in this population. Considering the complexity of determinants of death risk, this observational study was designed to detect signals that could be tested in prospective clinical trials and was not intended to prove a causal relationship between serum magnesium level and mortality. Initially, we investigated the relationship between dialysate magnesium and serum magnesium concentrations and other serum parameters, along with an overview of their association with overall survival. After the exploratory study indicated a positive association between increasing serum magnesium level and improving survival that appeared to extend to patients with hypermagnesemia, the follow-up study evaluated the association of survival with categories of higher-than-normal serum magnesium levels. Elevated serum magnesium levels occurred frequently in HD patients and in this study indicated no increase in mortality rate even with hypermagnesemia with magnesium levels $ 0.40 mEq/L higher than the normal reference range. These results are potentially generalizable because the baseline composition of both study cohorts was similar to their respective source populations (all period-prevalent FMCNA HD patients). Outcome studies in the general population have indicated potential associations between low serum magnesium levels and atherosclerosis, hypertension, diabetes, and left ventricular hypertrophy, as well as both CVD mortality and all-cause mortality.11-13 In Japanese HD patients, an initial small study (n 5 514) reported increased all-cause and non-CVD mortality, but not CVD mortality, associated with lower serum magnesium levels (,1.14 mmol/L [,2.28 mEq/L]).5 This study was later confirmed in a national Japanese cohort, extending the results to include a significant 5

Lacson et al

Figure 3. (A) Hazard ratio (HR) of death during the 1-year follow-up study according to baseline serum magnesium levels in hemodialysis patients (total group, n 5 27,544). (B, C) HR of death during the 1-year follow-up study according to baseline serum magnesium levels in hemodialysis patients (B) with and (C) without diabetes mellitus (n 5 14,756 and 12,788, respectively). The horizontal dotted line tracks the HR of 1 (same as the reference group) across serum magnesium categories to facilitate interpretation. The 2 vertical dashed lines represent the boundaries of common lower and upper limits of the laboratory serum magnesium assay’s normal range. P , 0.05; *P , 0.001. Abbreviations: CM, case-mix; lab, laboratory. ˇ

6

Am J Kidney Dis. 2015;-(-):---

Serum Mg Category, mEq/L

Characteristics

Age, y Male sex Race, White Black Other Diabetes Dialysis vintage, y Body surface area, m2 Laboratory results Mg,a mEq/L Potassium,a mEq/L Calcium,a mg/dL iPTH,a pg/mL Albumin,a g/dL Hemoglobin,a g/dL Phosphorus,a mg/dL eKt/V Vascular access Fistula Graft Catheter Other

,1.30 (n 5 454)

1.30-,1.60 (n 5 4,275)

1.60-,1.90 (n 5 10,466)

1.90-2.10 (n 5 6,216)

.2.10-,2.30 (n 5 3,593)

2.30-,2.50 (n 5 1,564)

$2.50 (n 5 976)

64.8 6 13.5 220 (48.5)

64.2 6 14.5 2,290 (53.6)

62.6 6 14.6 5,745 (54.9)

61.0 6 14.8 3,333 (53.6)

60.0 6 14.9 1,863 (51.9)

58.6 6 14.9 834 (53.3)

59.4 6 14.8 514 (52.7)

302 (66.5) 134 (29.5) 18 (4.0)

2,324 (54.4) 1,718 (40.2) 233 (5.5)

5,162 (49.3) 4,612 (44.1) 692 (6.6)

2,809 (45.2) 2,891 (46.5) 516 (8.3)

1,641 (45.7) 1,615 (44.9) 337 (9.4)

696 (44.5) 692 (44.2) 176 (8.5)

460 (47.1) 408 (41.8) 108 (11.1)

231 (50.9) 1.3 [0.5-3.0] 1.85 6 0.28

2,332 (54.5) 1.8 [0.6-3.9] 1.86 6 0.28

5,742 (54.9) 2.3 [0.9-4.5] 1.86 6 0.28

3,296 (53.0) 2.8 [1.2-5.1] 1.84 6 0.27

1,865 (51.9) 2.9 [1.5-5.3] 1.81 6 0.27

815 (52.1) 3.2 [1.6-5.7] 1.80 6 0.27

475 (48.7) 3.3 [1.6-5.7] 1.79 6 0.27

1.16 6 0.08 4.38 6 0.61 8.69 6 0.68 250 [151-4,031] 3.50 6 0.49 11.42 6 1.27 4.68 6 1.35 1.52 6 0.31

1.46 6 0.08 4.52 6 0.58 8.88 6 0.69 262 [172-6,506] 3.63 6 0.46 11.64 6 1.19 4.89 6 1.27 1.49 6 0.31

1.72 6 0.08 4.67 6 0.58 9.07 6 0.65 270 [176-7,736] 3.77 6 0.39 11.87 6 1.13 5.29 6 1.37 1.49 6 0.30

1.86 6 0.06 4.82 6 0.56 9.19 6 0.64 281 [184-6,138] 3.87 6 0.36 11.96 6 1.10 5.55 6 1.41 1.48 6 0.29

2.14 6 0.06 4.95 6 0.56 9.24 6 0.63 282 [187-5,193] 3.91 6 0.34 12.03 6 1.08 5.7 6 1.43 1.48 6 0.27

2.24 6 0.06 5.03 6 0.58 9.25 6 0.66 273 [181-5,100] 3.94 6 0.36 11.98 6 1.12 5.74 6 1.40 1.47 6 0.26

2.68 6 0.26 5.07 6 0.61 9.31 6 0.62 264 [164-3,512] 3.98 6 0.34 12.12 6 1.13 5.73 6 1.45 1.50 6 0.27

146 (32.2) 92 (20.3) 211 (46.5) 5 (1.1)

1,538 (36.0) 955 (22.3) 1,746 (40.8) 36 (0.8)

4,468 (42.7) 2,549 (24.4) 3,373 (32.2) 76 (0.7)

2,936 (47.2) 1,590 (25.6) 1,661 (26.7) 29 (0.5)

1,758 (48.9) 898 (25.0) 927 (25.8) 10 (0.3)

798 (51.0) 369 (23.6) 392 (25.1) 5 (0.3)

514 (52.7) 246 (25.2) 214 (21.9) 2 (0.2)

Note: Follow-up analysis (n 5 27,544) October 1, 2007, to December 31, 2007. Values for categorical variables are given as number (percentage); values for continuous variables, as mean 6 standard deviation or median [interquartile range]. All P values for trend were significant at P , 0.001 except for male sex (P 5 0.01) and eKt/V (P 5 0.8). Conversion factors for units: Mg in mEq/L to mmol/L, 30.5; serum calcium in mg/dL to mmol/L, 30.2495; serum phosphorus in mg/dL to mmol/L, 30.3229. Abbreviations: eKt/V, equilibrated Kt/V; iPTH, intact parathyroid hormone, Mg, magnesium. a Serum values.

Serum Magnesium and Mortality in HD

Am J Kidney Dis. 2015;-(-):---

Table 2. Patients’ Baseline Characteristics for the Follow-up Analysis (October 1, 2007-December 31, 2007).

7

Lacson et al

Figure 4. Hazard ratio (HR) of death during the 1-year follow-up study according to baseline serum magnesium levels in hemodialysis patients with (A) cause of death other than cardiovascular disease (CVD; n 5 2,077 deaths), (B) CVD as the cause of death (n 5 1,711 deaths), and (C) the subgroup of patients with CVD categorized as having had sudden cardiac death (n 5 1,174 deaths). The horizontal dotted line tracks the HR of 1 (same as the reference group) across serum magnesium categories to facilitate interpretation. The 2 vertical dashed lines represent the boundaries of common lower and upper limits of the laboratory serum magnesium assay’s normal range. P , 0.05; *P , 0.001. Abbreviations: CM, case-mix; lab, laboratory. ˇ

8

Am J Kidney Dis. 2015;-(-):---

Serum Magnesium and Mortality in HD

association between serum magnesium level and CVD mortality risk.6 Thus, the current study fills a knowledge gap, not only by confirming the association between serum magnesium level and survival, but doing so in a different large nationally distributed population. It demonstrated that the relationship was consistent in subgroups that might be vulnerable (eg, diabetic patients), as well as independent of categorical causes of death (eg, non-CVD, CVD, and sudden death). Furthermore, the study was able to evaluate categories of hypermagnesemia and, contrary to expectation,6 did not detect a higher death risk with levels $ 0.40 mEq/L above the laboratory upper limit of normal values. In the exploratory study, not surprisingly, serum magnesium levels were higher than prescribed dialysate magnesium concentrations and a correlation between serum magnesium and prescribed dialysate magnesium concentrations existed, but other factors such as unaccounted comorbid illness, use of diuretics, and/or nutritional intake of magnesiumcontaining compounds, including food items, dietary supplements, phosphate binders, or antacids, either singly or in combination, may have likely influenced serum magnesium levels. More importantly, a clear, almost linear positive relationship existed between risk for death and serum magnesium level (Fig 1B). Thus, in follow-up, a more refined analysis was performed to evaluate outcomes above the normal serum magnesium range (.2.10 mEq/L). This group was further divided into several elevated serum magnesium level categories, providing greater detail than results from Ishimura et al.5 Post hoc, we found a similar risk pattern when follow-up was extended to 3 years (Fig S1). Although Sakaguchi et al6 also evaluated higher serum magnesium levels, it is important to note that they used a hypermagnesemic reference group (group 5: serum magnesium of 2.8-,3.1 mg/dL [2.3,2.6 mEq/L]); therefore, findings of a higher death risk relative to this reference group (corresponding to the current study’s group with the best survival) should not immediately cause concern. Sakaguchi et al’s6 group 6 ($3.1 mg/dL [$2.6 mEq/L]) has the same or lower relative death risk compared with patients with high-normal serum magnesium levels (group 2: 2.3,2.5 mg/dL [1.9-2.1 mEq/L]). Therefore, despite using different serum magnesium categories, we found a clear increase in mortality rates in the 2 lower serum magnesium groups (,1.30 and 1.30-,1.60 mEq/L) relative to the highest serum magnesium category ($2.50 mmol/L; both P , 0.001), consistent with the findings of Sakaguchi et al.6 In the study from Ishimura et al,5 lower serum magnesium levels also associated with lower serum phosphorus, calcium, and albumin levels, possibly indicating malnutrition. A similar pattern was also Am J Kidney Dis. 2015;-(-):---

found in our study. In our multivariate analysis, an increased mortality risk for lower serum magnesium categories (,1.60 mEq/L) remained even after adjustment for case-mix, but was not present after adjusting for multiple laboratory parameters, including albumin, phosphorus, and calcium, among others. Recently, elevated levels of adiponectin, which has a role in energy homeostasis and lipid/ glucose metabolism, have been associated with increased mortality risk in end-stage renal disease in the presence of low serum magnesium levels.14 These findings may link serum magnesium level, nutritional status, and mortality in this population. High serum magnesium levels may be a consequence of better nutrition, consistent with the Honolulu Heart study.15 In the present study, adjustment for body surface area and serum albumin and phosphorus levels did not abrogate the lower death rates with higher serum magnesium levels. However, we could not adjust for other markers, such as C-reactive protein and alkaline phosphatase, which have previously been associated with worse outcome.16-18 Our findings remained consistent in all subsequent subgroup analyses, whether stratified by diabetic status or changing the outcome variable by categorical cause of death. The risk for both CVD- and non– CVD-related death was reduced with increasing serum magnesium levels. This is in contrast to Ishimura et al5 not finding a reduced HR for death from CVD (HR, 0.98; P 5 0.9).5 This difference may be due to a smaller sample size (515 vs 27,544) plus a different study design and study population. Compared to the current study, they had a longer follow-up period (4.25 vs 1 year) but a lower mortality rate (21% in 4.25 years vs 16.6% in 1 year), proportionally fewer cases of sudden death (8.7% vs 31%), and a smaller proportion of patients with diabetes (24% vs 54%). Sakaguchi et al,6 using a similar design, likewise showed that low serum magnesium levels were associated with increased risk for both CVD and nonCVD mortality. However, as mentioned, we did not detect a J-shaped association but instead a continuous almost linear relationship up to serum magnesium levels $ 2.50 mEq/L.6 In addition to choice of reference group, contributing factors for differing results could include the following: (1) differences in laboratory standards and variability (we used only a single reference laboratory) and (2) a different study population that, in addition to race, varied by dialysis vintage (3.5 vs 7 years) and percentage of diabetic patients (w50% vs 35%). In the general population, Reffelmann et al12,13 found lower CV mortality in patients with serum magnesium levels $ 0.74 mmol/L ($1.48 mEq/L) after many adjustments, including for left ventricular 9

Lacson et al

hypertrophy, whereas in patients with non–dialysisdependent chronic kidney disease, Kanbay et al19 reported a lower CV event rate at serum magnesium levels . 0.84 mmol/L (.1.68 mEq/L). Furthermore, higher serum magnesium levels correlate with reduced vascular calcification in dialysis patients,1-3,20-22 a potential mechanism to reduce CV death. Additionally, sudden (CV) death is not uncommon in HD patients (w30% identified in this study), along with arrhythmias such as atrial fibrillation,23 and electrocardiogram changes have been associated with abnormal serum magnesium levels.9 The reduced HR in our study for sudden cardiac death in higher serum magnesium level groups may relate to prevention of arrhythmia.24 Three small published studies have indicated that interventions involving magnesium may have beneficial effects.21,22,25 In one of these studies, 47 longterm dialysis patients were randomly assigned, and those receiving magnesium supplementation had reduced cIMT after 2 months of treatment.22 A second study randomly assigned 29 of 54 HD patients to magnesium oxide 3 times daily for 6 months with a significant reduction in cIMT, whereas the control group had significantly increased cIMT.25 The third report was a small uncontrolled pilot study showing the prevention of progressive coronary artery calcification in HD patients given magnesium carbonate as a phosphate binder.21 In vivo/ex vivo studies in Wistar rats indicated that the presence of magnesium ions reduced aortic ring calcifications despite the presence of high phosphorus levels.26 In addition, a magnesiumbased phosphate binder (magnesium carbonate/calcium acetate) reduced aortic calcification in rates with adenine-induced chronic kidney failure.27 However encouraging these small experimental studies are, even in combination with our epidemiologic study, we caution that it would not be justified to infer a treatment recommendation; prospective, adequately powered, randomized clinical trials are needed. The current observational study, although robust, has limitations inherent to its design, particularly with susceptibility to selection biases, absence of information for nutrition and oral magnesium-containing medications, and residual confounding (eg, comorbid illnesses). First, because serum magnesium is not routinely measured in all patients at FMCNA, we used convenience sampling; only patients whose physicians ordered serum magnesium tested during the entry period. Testing frequency appears to be practice related and we are not aware of a differential bias for obtaining serum magnesium levels selectively in particular patient subgroups. Nevertheless, we cannot exclude such bias. Second, both hypo- and hypermagnesemic groups contained fewer patients than the other groups and thus may sometimes not achieve statistical significance. However, despite abrogating significance with laboratory adjustment, the former 10

tended to be associated with increased risk for death, while the latter groups tended toward lower risk than the reference category. Finally, the current study only provided cross-sectional representations of serum magnesium levels, and longitudinal stability for the patients in our study was inferred from a subset of patients but not tested in all. However, these stability results were consistent with those from Ishimura et al,5 in which serum magnesium levels at baseline correlated strongly (R 5 0.835) with repeat measurement after 1 year. In conclusion, the study findings suggest that whereas hypomagnesemia could be associated with increased risk for death, hypermagnesemia was associated with either lower or no additional risk. However, serum magnesium level is only one of many factors that may influence risk of death. Thus, although it is prudent for physicians to evaluate for treatable causes of hypomagnesemia, only adequately powered long-term prospective interventional studies will be able to determine whether therapeutic adjustment of serum magnesium levels can be beneficial for HD patients.

ACKNOWLEDGEMENTS Some of the analyses presented here have been previously reported in the form of an oral presentation (Passlick-Deetjen J, et al. Magnesium and mortality risk in hemodialysis patients. European Renal Association (ERA) and European Dialysis and Transplant Association (EDTA), XLVII Congress, June 25-28, 2010, Munich, Germany) and conference poster (Lacson E, et al. Magnesium and mortality risk in hemodialysis patients. American Society of Nephrology 42nd Annual Meeting & Scientific Exposition [Renal Week], October 27-November 1, 2009, San Diego, CA. Poster F-PO1488). Support: Dr Richard Clark, Dunchurch, Warwickshire, United Kingdom, provided writing and editorial assistance on behalf of Fresenius Medical Care, AG. Dr Franklin Maddux, Chief Medical Officer of FMCNA, and Dr Robert Hootkins (retired clinician, consulting for FMCNA) reviewed the manuscript draft. Two authors (Drs Lacson Jr and Passlick-Deetjen) collaboratively made the final decision regarding the main points to be communicated in the manuscript, with review by Dr Maddux. Financial Disclosure: Dr Passlick-Deetjen was an employee of Fresenius Medical Care, AG, and Dr Lacson Jr and Ms Wang were employed by FMCNA when the study was conducted. Contributions: Research idea and study design: EL, WW, JP-D; data acquisition: WW, LM; data analysis/interpretation: EL, WW, JP-D; statistical analysis: WW, LM. Each author contributed important intellectual content during manuscript drafting or revision and accepts accountability for the overall work by ensuring that questions pertaining to the accuracy or integrity of any portion of the work are appropriately investigated and resolved. EL, WW, and JP-D take responsibility that this study has been reported honestly, accurately, and transparently; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.

SUPPLEMENTARY MATERIAL Table S1: Stability study of serum Mg on subset of patients who overlapped between both studies. Figure S1: Kaplan-Meier survival analysis for all patients in main cohort during 3-year follow-up. Am J Kidney Dis. 2015;-(-):---

Serum Magnesium and Mortality in HD Note: The supplementary material accompanying this article (http://dx.doi.org/10.1053/j.ajkd.2015.06.014) is available at www.ajkd.org

REFERENCES 1. Ishimura E, Okuno S, Kitatani K, et al. Significant association between the presence of peripheral vascular calcification and lower serum magnesium in hemodialysis patients. Clin Nephrol. 2007;68(4):222-227. 2. Tzanakis I, Pras A, Kounali D, et al. Mitral annular calcifications in hemodialysis patients: a possible protective role of magnesium. Nephrol Dial Transplant. 1997;12:2036-2037. 3. Tzanakis I, Virvidakis K, Tsomi A, et al. Intra- and extracellular magnesium levels and atheromatosis in hemodialysis patients. Magnes Res. 2004;17:102-108. 4. Chakraborti S, Chakraborti T, Mandal M, Mandal A, Das S, Ghosh S. Protective role of magnesium in cardiovascular diseases: a review. Mol Cell Biochem. 2002;238(1-2):163-179. 5. Ishimura E, Okuno S, Yamakama T, et al. Serum magnesium concentration is a significant predictor of mortality in maintenance hemodialysis patients. Magnes Res. 2007;20:237-244. 6. Sakaguchi Y, Fujii N, Shoji T, et al. Hypomagnesemia is a significant predictor of cardiovascular and non-cardiovascular mortality in patients undergoing hemodialysis. Kidney Int. 2014;85:174-181. 7. Matias PJ, Azevdo A, Laranjinha I, et al. Lower serum magnesium is associated with cardiovascular risk factors and mortality in haemodialysis patients. Blood Purif. 2014;38:244-252. 8. Lacson E Jr, Wang W, Hakim RM, et al. Associates of mortality and hospitalization in hemodialysis: potentially actionable laboratory variables and vascular access. Am J Kidney Dis. 2009;53(1):79-90. 9. Krishnan M, Wilfehrt HM, Lacson E Jr. In data we trust: the role and utility of dialysis provider databases in the policy process. Clin J Am Soc Nephrol. 2012;7(11):1891-1896. 10. Jahnen-Dechent W, Ketteler M. Magnesium basics. Clin Kidney J. 2012;5(suppl 1):i3-i14. 11. Ma J, Folsom AR, Melnick SL, et al. Associations of serum and dietary magnesium with cardiovascular disease, hypertension, diabetes, insulin, and carotid arterial wall thickness: the ARIC Study. J Clin Epidemiol. 1995;48:927-940. 12. Reffelmann T, Dorr M, Ittermann T, et al. Low serum magnesium concentrations predict increase in left ventricular mass over 5 years independently of common cardiovascular risk factors. Atherosclerosis. 2010;213:563-569. 13. Reffelmann T, Ittermann T, Dorr M, et al. Low serum magnesium concentrations predict cardiovascular and all-cause mortality. Atherosclerosis. 2011;219:280-284.

Am J Kidney Dis. 2015;-(-):---

14. Markaki A, Kyriazis J, Stylianou K, et al. The role of serum magnesium and calcium on the association between adiponectin levels and all-cause mortality in end-stage renal disease patients. PLoS One. 2012;7(12):e52350. 15. Joffres MR, Reed DM, Yano K. Relationship of magnesium intake and other dietary factors to blood pressure. The Honolulu Heart Study. Am J Clin Nutr. 1987;45:469-475. 16. Bazeley J, Bieber B, Li Y, et al. C-Reactive protein and prediction of 1-year mortality in prevalent hemodialysis patients. Clin J Am Soc Nephrol. 2011;6:2452-2461. 17. Blayney MJ, Pisoni RL, Bragg-Gresham JL, et al. High alkaline phosphatase levels in hemodialysis patients are associated with higher risk of hospitalization and death. Kidney Int. 2008;74: 655-663. 18. Regidor DL, Kovesdy CP, Mehrotra R, et al. Serum alkaline phosphatase predicts mortality among maintenance hemodialysis patients. J Am Soc Nephrol. 2008;19:2193-2203. 19. Kanbay M, Yilmaz MI, Apetrii M, et al. Relationship between serum magnesium levels and cardiovascular events in chronic kidney disease patients. Am J Nephrol. 2012;36: 228-237. 20. Meema HE, Oreopoulos DG, Rapoport A. Serum magnesium level and arterial calcification in end-stage renal disease. Kidney Int. 1987;32:388-394. 21. Spiegel DM, Farmer B. Long-term effects of magnesium carbonate on coronary artery calcification and bone mineral density in hemodialysis patients: a pilot study. Hemodial Int. 2009;13: 453-459. 22. Turgut F, Kanbay M, Metin MR, et al. Magnesium supplementation helps to improve carotid intima media thickness in patients on hemodialysis. Int Urol Nephrol. 2008;40: 1075-1082. 23. Khan AM, Lubitz SA, Sullivan LM, et al. Low serum magnesium and the development of atrial fibrillation in the community: the Framingham Heart Study. Circulation. 2013;127: 33-38. 24. Geiger H, Wanner C. Magnesium in disease. Clin Kidney J. 2012;5(suppl 1):i25-i38. 25. Mortazavi M, Moeinzadeh F, Saadatnia M, et al. Effect of magnesium supplementation on carotid intima-media thickness and flow-mediated dilatation among hemodialysis patients: a double-blind, randomized, placebo-controlled trial. Eur Neurol. 2013;69(5):309-316. 26. Salem S, Bruck H, Bahlmann FH, et al. Relationship between magnesium and clinical biomarkers on inhibition of vascular calcification. Am J Nephrol. 2012;35:31-39. 27. De Schutter TM, Behets GJ, Geryl H, et al. Effect of a magnesium-based phosphate binder on medial calcification in a rat model of uremia. Kidney Int. 2013;83:1109-1117.

11