Predialysis Serum Sodium Level, Dialysate Sodium, and Mortality in Maintenance Hemodialysis Patients: The Dialysis Outcomes and Practice Patterns Study (DOPPS)

Original Investigation Predialysis Serum Sodium Level, Dialysate Sodium, and Mortality in Maintenance Hemodialysis Patients: The Dialysis Outcomes and Practice Patterns Study (DOPPS) Manfred Hecking, MD,1,2 Angelo Karaboyas, MS,1 Rajiv Saran, MD, MRCP, MS,3,4 Ananda Sen, PhD,5,6 Walter H. Hörl, MD, PhD,2 Ronald L. Pisoni, MS, PhD,1 Bruce M. Robinson, MD, MS,1 Gere Sunder-Plassmann, MD,2 and Friedrich K. Port, MD, MS1 Background: Predialysis serum sodium concentrations recently have been linked to patient characteristics and outcomes in hemodialysis patients and may have implications for the dialysate sodium prescription. Study Design: Prospective cohort study. Participants: 11,555 patients from 12 countries in the Dialysis Outcomes and Practice Patterns Study (DOPPS), phases I (1996-2001) and III (2005-2008). Predictors: Demographics, comorbid conditions, laboratory measurements (model 1); mean serum sodium level, dialysate sodium concentration (model 2). Outcomes: Serum sodium level, using adjusted linear mixed models (model 1); all-cause mortality, using Cox proportional hazards models (model 2). Results: Median follow-up was 12 months, with 1,727 deaths (15%) occurring during the study period (12,274 patient-years). Mean serum sodium level in the DOPPS countries was 138.5 ⫾ 2.8 mEq/L. Japan had the highest (139.1 ⫾ 2.6 mEq/L) and Australia/New Zealand had the lowest mean serum sodium level (137.4 ⫾ 2.8 mEq/L). Serum sodium level was associated positively with male sex, black race, body mass index, serum albumin level, and creatinine level and negatively with neurologic and psychiatric disease, white blood cell count, and intradialytic weight loss (0.16 mEq/L lower per 1% loss). Higher serum sodium level was associated with lower adjusted all-cause mortality in a continuous model (HR, 0.95 per 1 mEq/L higher; 95% CI, 0.93-0.97). Dialysate sodium prescription was not associated with serum sodium level. Mortality analyses restricted to the serum sodium tertile with the highest mortality (serum sodium ⬍137 mEq/L) showed lower mortality risk in patients with dialysate sodium prescriptions ⬎140 mEq/L. Limitations: Causality cannot be established in this observational study, which does not consider potential effects of dialysate sodium level on postdialysis thirst, dietary salt and water intake, interdialytic weight gain, and cardiovascular stability. Conclusions: Lower serum sodium levels are associated with certain hemodialysis patient characteristics and higher adjusted risk of death. The lower mortality observed in our adjusted analyses in patients with serum sodium levels ⬍137 mEq/L dialyzed against dialysate sodium prescriptions ⬎140 mEq/L is intriguing, may be related to intradialytic cardiovascular stability, and deserves further study. Am J Kidney Dis. 59(2):238-248. © 2012 by the National Kidney Foundation, Inc. Published by Elsevier Inc. All rights reserved. INDEX WORDS: End-stage renal disease; predictive model; predialysis serum sodium; dialysate sodium prescription; Dialysis Outcomes and Practice Patterns Study.

S

odium, the main extracellular cation in humans,1 is held relatively constant around a predetermined level by a cybernetic or feedbackcontrolled system.2-4 Hemodialysis (HD) patients also may have a fixed “setpoint” for sodium5-7 or osmolarity. Based on data from a clinical trial,8 it has been suggested that a patient who is dialyzed against a high dialysate sodium concentration and

develops relative hypernatremia after HD likely will drink enough free water to drive plasma osmolality back to the setpoint with little variability.9 In line with this assumption, recent results have indicated that interdialytic weight gain correlates with the dialysate sodium to serum sodium gradient6,10 and more strongly with pre- to postdialysis serum sodium level change.11

1 From the Arbor Research Collaborative for Health, Ann Arbor, MI; 2Department of Internal Medicine III–Nephrology, Medical University of Vienna, Vienna, Austria; 3Department of Internal Medicine–Nephrology, 4University of Michigan–Kidney Epidemiology and Cost Center, and Departments of 5Family Medicine and 6 Statistics, University of Michigan, Ann Arbor, MI. Received March 21, 2011. Accepted in revised form July 8, 2011. Originally published online September 26, 2011.

Address correspondence to Friedrich K. Port, MD, MS, Arbor Research Collaborative for Health, 340 E Huron, Suite 300, Ann Arbor, MI 48104. E-mail: [email protected] © 2012 by the National Kidney Foundation, Inc. Published by Elsevier Inc. All rights reserved. 0272-6386/$36.00 doi:10.1053/j.ajkd.2011.07.013

238

Am J Kidney Dis. 2012;59(2):238-248

Serum Sodium and Mortality

An earlier analysis of the Dialysis Outcomes and Practice Patterns Study (DOPPS) showed that interdialytic weight gain is associated with mortality12 and was confirmed by another report of a very large data set.13 Previous guidelines and reviews have stressed the importance of reducing the sodium overload caused by relatively high sodium dialysate concentration, as a contributor to hypertension14-16 and interdialytic weight gain,17 with recent publications showing increased attention for sodium individualization as a beneficial dialysis practice pattern.10,18,19 However, lower dialysate sodium prescriptions potentially can be hazardous by causing osmotic disequilibrium,20 a more common phenomenon in earlier years when dialysate sodium concentration typically was ⬃132 mEq/L.21 To our knowledge, the benefit or harm of the dialysate sodium prescription has not yet been analyzed in the context of a patient’s predialysis serum sodium concentration. As a prerequisite to this study objective, we sought to confirm previously reported associations between predialysis serum sodium concentration and HD patient characteristics, dialysisassociated variables, and mortality risk22 in the large international data set of the DOPPS. We also aimed to explore variations in serum and dialysate sodium concentrations across countries and investigate whether a relationship existed between serum sodium concentration, dialysate sodium concentration, and mortality

because this could have implications for dialysate sodium prescription practices.

METHODS Data Source and Determination of Predialysis Mean Serum Sodium Level The DOPPS design has been published previously23,24 and recently was reviewed.25 Quarterly determined predialysis serum sodium measurements were available from DOPPS phases I (19962001) and III (2005-2008). The total number of serum sodium measurements available per patient varied from 1 (n ⫽ 2,053 patients) to 12 (n ⫽ 1 patient). It was important to decide on a set number of measurements to compute the mean patient serum sodium level because the number of serum sodium measurements drives the variability of the sampling distribution. We calculated various serum sodium averages with increasing numbers of serum sodium measurements per patient to determine how many serum sodium measurements would best approximate a patient’s mean serum sodium level. As more measurements were used, less variation in each patient’s mean serum sodium level was observed, and the standard deviation of the global mean decreased. However, because fewer patients were available with increasing numbers of serum sodium measurements, the standard error of the global mean began to increase with more than 3 measurements. Furthermore, in patients who had 4-9 serum sodium measurements reported, the mean serum sodium level calculated with all available measurements differed by no more than 0.1 mEq/L from the mean serum sodium level calculated with the first 3 serum sodium measurements. Therefore, we used the first 3 serum sodium measurements to compute each patient’s mean serum sodium level and included, in consequence, 11,555 patients in the present cohort, based on the algorithm described in Fig 1.

Figure 1. Flow chart of the cohort creation. Abbreviations: DNa, dialysate sodium; DOPPS, Dialysis Outcomes and Practice Patterns Study; HD, hemodialysis; SNa, serum sodium. Am J Kidney Dis. 2012;59(2):238-248

239

Hecking et al

Sodium Measurement and Glucose Concentration A clinical rule of thumb states that the serum sodium level decreases by 1.6 mEq/L for every 100 mg/dL increase in glucose level because water transfers from the intracellular to extracellular compartment.26 The resulting correction formula for sodium in patients with abnormal glucose concentrations recently has been validated in HD patients.27 In the present study, we reported main findings with and without exclusion of patients who had at least one glucose measurement ⬎140 mg/dL or all 3 glucose measurements missing, but also confirmed our results after applying the correction formula. Glucose values ⬍50 and ⬎500 mg/dL are rare in an outpatient setting and therefore were excluded to prevent the use of potentially erroneous data.

Statistical Analysis Descriptive statistics (mean ⫾ standard deviation or absolute and relative frequency) were used to examine the prevalent crosssection of DOPPS I and III patients at baseline. Linear mixed models were used to analyze the association of mean serum sodium level and patient characteristics. Models were adjusted for country, age, sex, black race, vintage, body mass index, intradialytic weight loss, dialysate sodium prescription, residual kidney function, predialysis systolic blood pressure, vascular access, albumin level, hemoglobin level, serum creatinine level, ferritin level, white blood cell count, and 14 comorbid conditions. Clustering within facilities was accounted for by random facility effect. Proportional hazards (time-to-event) models (Cox) were used to determine the association between mortality and serum sodium group (tertile) starting 30 days after the third serum sodium measurement. Participants were followed until the earliest of death or up to 7 days after departure from the facility for kidney transplant, change of treatment modality, withdrawal from dialysis therapy, return of kidney function, or transfer to another facility. Follow-up time was censored at study end for patients who did not depart from the facility. Cox models for all-cause, cardiovascular, and noncardiovascular mortality were adjusted for variables as listed, stratified by study phase and country, and accounted for facility clustering effects by using the sandwich estimator for standard errors. Similarly adjusted Cox models were used to examine the relationship between all-cause mortality and dialysate sodium concentration, both stratified by serum sodium tertiles and also with the variable of interest as 9 combinations of dialysate sodium concentration and serum sodium tertiles. Patients with missing data for age and vintage (n ⫽ 215) were excluded from models, and missingness indicators were used to account for missing data for other adjustment covariates. Of these 215 patients, 34 died, which explains the discrepancy between the total number of deaths and deaths reported in the third table. Sensitivity analyses excluding Japan were conducted on Cox models to ensure that this subset of patients was not driving the mortality rate. Additional sensitivity analyses were conducted restricting to patients with Kt/V ⬎1.2 and vintage longer than 1 year. The proportional hazards assumption was tested using Martingale residuals28 and examining log-log survival plots. For calculations, we used MS Excel 2003 and SAS, version 9.2, for Windows (SAS Institute Inc, www.sas.com).

RESULTS Distribution of Mean Serum Sodium Levels Across DOPPS Countries Of the sample of 11,555 patients analyzed, mean serum sodium level was 138.5 ⫾ 2.8 (standard deviation) mEq/L and therefore lower in comparison to a 240

cross-section of 2,172 nonhypertensive, nondialysis Framingham Offspring Study participants at baseline (139.3 ⫾ 2.2 mEq/L).29 Interestingly, mean serum sodium levels differed considerably among the various DOPPS countries (overall P ⬍ 0.001). Upon categorization in approximate tertiles of mean serum sodium levels (⬍137, 137-139.9, and ⱖ140 mEq/L), Japan had the highest and Australia/New Zealand the lowest relative proportion of HD patients with mean serum sodium levels ⱖ140 mEq/L (40.7% and 19.1%, respectively; Table 1). Mean serum sodium level also was the highest in Japan (139.1 ⫾ 2.6 mEq/L) and the lowest in Australia/New Zealand (137.4 ⫾ 2.8 mEq/L; Table 1). Patient Characteristics Associated With Mean Serum Sodium Level Several demographic factors and laboratory values were associated significantly with higher mean serum sodium levels (Table 2). Because hyperglycemia is associated with decreased serum sodium readings,26 Table 2 lists results from the fully adjusted linear mixed model (1) for all patients and (2) for a subgroup of patients for whom the simultaneous blood glucose level was never ⬎140 mg/dL. Not surprisingly, diabetes was associated with lower mean serum sodium levels. However, this association was not observed in the subgroup without hyperglycemia. Significant predictors for higher serum sodium levels in the overall and subgroup analyses included male sex, black race, higher body mass index, higher serum albumin level, and higher serum creatinine level (all P ⱕ 0.03). In the same models (1 and 2), several comorbid conditions, patient characteristics, and laboratory values were associated with lower mean serum sodium levels, including neurologic disease, psychiatric disease, recurrent cellulitis, greater intradialytic weight loss, and white blood cell count. All corresponding P values were ⱕ0.005 in the overall model (1) and ⱕ0.1 in the smaller subgroup analysis (2). Mortality Risk Associated With Mean Serum Sodium Level Median follow-up was 12 months, with a maximum of 39 months. During the study period, 1,727 patients died (15%), and total patient-years for all patients were 12,274. Higher mean serum sodium level was associated with lower all-cause mortality (Table 3). Models with various levels of adjustment showed that this association was significant and consistent. Even after controlling for 10 patient characteristics, 14 comorbid conditions, 5 laboratory parameters, and facility clustering effects and stratifying by DOPPS phase and country, patients with serum sodium levels ⬍137 mEq/L had a 45% higher risk of death Am J Kidney Dis. 2012;59(2):238-248

Serum Sodium and Mortality Table 1. Demographics and Patient Characteristics of the Study Cohort Predialysis Mean Serum Sodium Tertilea Variable

No. of patientsb Men Residual kidney function present Diabetes mellitusc Black race Comorbid conditions Hypertension Coronary heart disease Cardiovascular disease Congestive heart failure Peripheral vascular disease Cerebrovascular disease Psychiatric disorder Cancer (nonskin) Lung disease Neurologic disease Recurrent cellulitis GI bleeding HIV/AIDS Age (y) Dialysis vintage (y) BMI (kg/m2) spKt/V IDWL (% of body weight) Predialysis SBP (mm Hg) nPCR (g/kg/d) Laboratory measurements Predialysis mean serum sodium (mEq/L)d Japan (n ⫽ 3,583) Sweden (n ⫽ 445) Germany (n ⫽ 1,018) UK (n ⫽ 681) Canada (n ⫽ 189) Italy (n ⫽ 1,017) Spain (n ⫽ 1,062) Belgium (n ⫽ 354) US (n ⫽ 1,613) France (n ⫽ 1,054) ANZ (n ⫽ 539) Serum albumin (g/dL) Hemoglobin (g/dL) Serum creatinine (mg/dL) Serum ferritin (ng/mL) WBC count (⫻ 103/␮L)

All Patients

<137 mEq/L

137-139.9 mEq/L

>140 mEq/L

11,555 (100) 59 42 34 6

3,097 (27) 56 43 45 6

4,824 (42) 59 41 33 6

3,634 (31) 64 41 28 5

76 40 33 29 22 15 13 11 9 8 6 5 1

80 44 36 36 27 18 18 11 12 11 8 6 1

76 39 32 28 21 14 12 10 9 8 5 4 1

73 38 31 26 18 13 10 11 8 6 4 4 1

62 ⫾ 14 4.7 ⫾ 6.0

62 ⫾ 15 3.8 ⫾ 5.3

61 ⫾ 14 4.8 ⫾ 6.0

62 ⫾ 14 5.3 ⫾ 6.6

23.9 ⫾ 5.3 1.39 ⫾ 0.32 3.3 ⫾ 4.6 147 ⫾ 24 1.03 ⫾ 0.24

24.1 ⫾ 5.4 1.41 ⫾ 0.34 3.5 ⫾ 4.6 147 ⫾ 26 1.02 ⫾ 0.26

24.0 ⫾ 5.3 1.39 ⫾ 0.32 3.4 ⫾ 4.9 146 ⫾ 24 1.03 ⫾ 0.24

23.5 ⫾ 5.0 1.37 ⫾ 0.31 3.1 ⫾ 4.1 148 ⫾ 24 1.04 ⫾ 0.23

138.5 ⫾ 2.8 139.1 ⫾ 2.6 138.7 ⫾ 2.7 138.6 ⫾ 2.8 138.5 ⫾ 2.8 138.4 ⫾ 2.9 138.4 ⫾ 2.8 138.3 ⫾ 2.7 138.2 ⫾ 3.0 137.9 ⫾ 2.9 137.7 ⫾ 2.8 137.4 ⫾ 2.8 3.8 ⫾ 0.5 10.8 ⫾ 1.7 9.2 ⫾ 3.1 352 ⫾ 392 6.9 ⫾ 2.4

135.0 ⫾ 1.6 (27) 135.2 ⫾ 1.5 (18) 135.2 ⫾ 1.4 (24) 135.1 ⫾ 1.5 (26) 134.7 ⫾ 1.7 (25) 134.8 ⫾ 1.7 (26) 135.1 ⫾ 1.5 (29) 135.0 ⫾ 1.4 (28) 134.8 ⫾ 1.7 (33) 134.8 ⫾ 1.8 (35) 134.8 ⫾ 1.8 (36) 134.7 ⫾ 1.7 (40) 3.7 ⫾ 0.5 10.8 ⫾ 1.7 8.5 ⫾ 3.0 385 ⫾ 397 7.3 ⫾ 2.6

138.4 ⫾ 0.8 (42) 138.5 ⫾ 0.8 (42) 138.4 ⫾ 0.8 (42) 138.4 ⫾ 0.8 (41) 138.4 ⫾ 0.9 (44) 138.4 ⫾ 0.8 (41) 138.4 ⫾ 0.8 (43) 138.4 ⫾ 0.9 (45) 138.4 ⫾ 0.8 (38) 138.4 ⫾ 0.9 (39) 138.4 ⫾ 0.8 (43) 138.2 ⫾ 0.9 (41) 3.8 ⫾ 0.5 10.8 ⫾ 1.7 9.3 ⫾ 3.1 349 ⫾ 409 6.8 ⫾ 2.4

141.5 ⫾ 1.3 (31) 141.5 ⫾ 1.3 (41) 141.5 ⫾ 1.4 (35) 141.5 ⫾ 1.3 (33) 141.5 ⫾ 1.2 (32) 141.5 ⫾ 1.4 (33) 141.8 ⫾ 1.5 (28) 141.5 ⫾ 1.3 (27) 141.7 ⫾ 1.4 (29) 141.4 ⫾ 1.3 (25) 141.4 ⫾ 1.3 (21) 141.2 ⫾ 1.1 (19) 3.8 ⫾ 0.5 10.7 ⫾ 1.7 9.7 ⫾ 3.2 328 ⫾ 362 6.5 ⫾ 2.2

Note: Except where otherwise indicated, categorical variables given as percentage, and continuous variables, as mean ⫾ standard deviation. Conversion factors for units: serum albumin and hemoglobin in g/dL to g/L, ⫻10; serum creatinine in mg/dL to ␮mol/L, ⫻88.4; no conversion necessary for sodium in mEq/L and mmol/L, serum ferritin in ng/mL and ␮g/L, and WBC count in 103/␮L and 109/L. Abbreviations: ANZ, Australia/New Zealand; BMI, body mass index; GI, gastrointestinal; HIV, human immunodeficiency virus; IDWL, intradialytic weight loss; nPCR, normalized protein catabolic rate; SBP, systolic blood pressure; spKt/V, single-pool Kt/V; UK, United Kingdom; US, United States; WBC, white blood cell. a Approximate tertiles. b Percentage given in parentheses. c Comorbid condition or cause of end-stage renal disease. d Values given as mean ⫾ standard deviation; when given, number in parentheses is percentage.

Am J Kidney Dis. 2012;59(2):238-248

241

Hecking et al Table 2. Demographics and Patient Characteristics Associated With Predialysis Mean Serum Sodium Patients With All Glucose Measurements <140 mg/dLa (n ⴝ 4,188)

All Patients (n ⴝ 11,340) Estimate

Age (/5 y) Men (vs women) Black race (vs nonblack) Vintage (/y) BMI (/5 kg/m2) IDWL (/% lost) Albumin (/1 g/dL) Creatinine (/1 mg/dL) WBC count (/103/␮L) Residual kidney function Comorbid conditions Cardiovascular Diabetes Neurologic disease Psychiatric disorder Recurrent cellulitis

P

Estimate

P

0.08 0.33 0.31 0.03

⬍0.001 ⬍0.001 0.03 ⬍0.001

0.06 0.21 0.47 0.03

⬍0.001 0.01 0.02 ⬍0.001

0.14 ⫺0.16 0.31 0.08 ⫺0.07 0.14

⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 ⬍0.001 0.02

0.19 ⫺0.18 0.35 0.05 ⫺0.03 0.01

⬍0.001 ⬍0.001 0.001 0.002 0.09 0.9

⫺0.12 ⫺0.64 ⫺0.41 ⫺0.38 ⫺0.31

0.03 ⬍0.001 ⬍0.001 ⬍0.001 0.005

⫺0.13 ⫺0.10 ⫺0.68 ⫺0.25 ⫺0.33

0.1 0.4 ⬍0.001 0.06 0.1

Note: Linear mixed model simultaneously adjusted for all covariates in the table plus dialysate sodium prescription, predialysis systolic blood pressure, vascular access, hemoglobin level, serum ferritin level, 9 other comorbid conditions, and Dialysis Outcomes and Practice Patterns Study phase and country and accounted for facility clustering effects. Abbreviations: BMI, body mass index; IDWL, intradialytic weight loss; WBC, white blood cell. a During 12 months; corresponds to 37% of patients. All glucose measurements missing in 4,275 patients. At least one glucose measurement ⱖ140 mg/dL in 2,877 patients.

Table 3. Mortality Risk Associated With Predialysis Mean Serum Sodium With 5 Levels of Adjustment Model 1

Model 2

Model 3

Model 4

Model 5

All Patientsa Continuous model: per 1 mEq/L higher SNa Categorical model: SNa ⬍137 mEq/L SNa ⫽ 137-139.9 mEq/L SNa ⱖ140 mEq/L

0.92 (0.91-0.94)

0.93 (0.92-0.95)

0.94 (0.92-0.96)

0.94 (0.92-0.96)

0.95 (0.93-0.97)

1.76 (1.53-2.02) 1.28 (1.12-1.46) 1.00 (reference)

1.64 (1.42-1.88) 1.26 (1.10-1.44) 1.00 (reference)

1.55 (1.35-1.79) 1.25 (1.09-1.44) 1.00 (reference)

1.54 (1.34-1.78) 1.25 (1.09-1.43) 1.00 (reference)

1.45 (1.26-1.67) 1.22 (1.06-1.39) 1.00 (reference)

Patients With All Glucose Measurements <140 mg/dLb Continuous model: per 1 mEq/L higher SNa Categorical model: SNa ⬍137 mEq/L SNa ⫽ 137-139.9 mEq/L SNa ⱖ140 mEq/L

0.94 (0.91-0.98)

0.95 (0.92-0.98)

0.96 (0.93-0.99)

0.95 (0.92-0.99)

0.96 (0.93-0.99)

1.50 (1.17-1.91) 1.23 (1.00-1.52) 1.00 (reference)

1.49 (1.17-1.90) 1.26 (1.02-1.55) 1.00 (reference)

1.37 (1.08-1.74) 1.23 (1.00-1.52) 1.00 (reference)

1.40 (1.10-1.78) 1.25 (1.01-1.55) 1.00 (reference)

1.39 (1.10-1.78) 1.28 (1.04-1.57) 1.00 (reference)

Note: Cox model; values shown are HR (95% confidence interval). Adjustments in model 1: age, race, sex, vintage, body mass index, Dialysis Outcomes and Practice Patterns Study phase, country, and facility clustering; model 2: model 1 plus diabetes (comorbid condition or cause of end-stage renal disease); model 3: model 2 plus neurologic and psychiatric disorder and 11 other comorbid conditions; model 4: model 3 plus intradialytic weight loss, dialysate sodium prescription, residual kidney function, and predialysis systolic blood pressure; and model 5: model 4 plus vascular access, albumin level, hemoglobin level, creatinine level, ferritin level, and white blood cell count. Abbreviations: HR, hazard ratio; SNa, serum sodium. a N ⫽ 11,340 patients, 1,695 deaths. b n ⫽ 4,188 patients, 591 deaths. All glucose measurements missing in 4,275 patients, at least one glucose measurement ⱖ140 mg/dL in 2,877 patients. 242

Am J Kidney Dis. 2012;59(2):238-248

Serum Sodium and Mortality

Figure 2. Adjusted mortality risk associated with predialysis mean serum sodium (SNa) level. ⫹Range, 127.0-134.9 mEq/L; ⫹⫹range, 142.0148.0 mEq/L. *P ⬍ 0.05; **P ⬍ 0.001; *** P ⬍ 0.0001. Cox model stratified by phase and country and adjusted for age, sex, black race, vintage, body mass index, intradialytic weight loss, dialysate sodium prescription, residual kidney function, predialysis systolic blood pressure, vascular access, albumin level, hemoglobin level, serum creatinine level, serum ferritin level, white blood cell count, 14 comorbid conditions, and facility clustering effects. Abbreviations: CI, confidence interval; HR, hazard ratio; ref, reference.

compared with patients with serum sodium levels ⱖ140 mEq/L (hazard ratio [HR], 1.45; 95% confidence interval [CI], 1.26-1.67; model 5). The Martingale residual test showed no violation of the proportional hazards assumption (P ⫽ 0.1), and this result was supported by examination of the log-log survival plot. Using serum sodium level as a continuous variable, the mortality HR was 0.95 per 1-mEq/L higher serum sodium level (95% CI, 0.93-0.97), indicating that higher serum sodium level was associated significantly with lower risk of mortality. A sensitivity analysis excluding Japan, the DOPPS country with the highest mean serum sodium level and typically the lowest mortality rates, showed consistent results (HR, 0.95 per 1 mEq/L higher serum sodium level; 95% CI, 0.94-0.97). Sensitivity analyses restricting to Kt/V ⬎1.2 (HR, 0.94 per 1 mEq/L higher serum sodium level; 95% CI, 0.92-0.97) and vintage longer than 1 year (HR, 0.95 per 1 mEq/L higher serum sodium level; 95% CI, 0.92-0.97) also showed consistent results. The mortality risk associated with lower serum sodium levels was similar for cardiovascular (HR, 0.95 per 1 mEq/L higher serum sodium level; 95% CI, 0.92-0.98) and noncardiovascular deaths (HR, 0.94 per 1 mEq/L higher serum sodium level; 95% CI, 0.92-0.97). Categorization of all 11,555 patients into 9 serum sodium groups of approximately similar size indicated that mortality risk was almost continuously lower in subgroups with higher mean serum sodium concentrations, in steps of 1 mEq/L, through the 2 tails of ⬍135 and ⱖ142 mEq/L (Fig 2). Mean serum sodium level in the group ⱖ142 mEq/L was 143.0 mEq/L. Lumping all patients ⬍142 mEq/L (n ⫽ 10,329) as an alternative reference group, the HR for mortality was 0.76 (95% CI, 0.59-0.99) for patients with a mean serum sodium level of 142Am J Kidney Dis. 2012;59(2):238-248

142.9 mEq/L (n ⫽ 657), 0.73 (95% CI, 0.52-1.03) for patients with a mean serum sodium level of 143-143.9 mEq/L (n ⫽ 338), and 0.45 (95% CI, 0.25-0.84) for patients with a mean serum sodium level of 144-144.9 mEq/L (n ⫽ 160). Only 71 patients (0.6%) had a mean serum sodium level ⱖ145 mEq/L and mortality HR not significantly different from the reference group (0.89; 95% CI, 0.51-1.56). These results show that in general, the risk of death was lower at higher mean serum sodium levels and better approximated by a straight line than a U-shaped relationship. Of 4,261 patients with all reported glucose measurements ⬍140 mg/dL, 841 had a diagnosis of diabetes as either a comorbid condition or cause of native kidney disease. However, in those patients, diabetes was not associated with a lower serum sodium concentration (Table 2), and adding diabetes into the Cox model only marginally decreased the mortality risk of patients who were in the lowest serum sodium group (Table 3). Application of the correction formula for glucose level27 yielded results similar to the sensitivity analysis of patients with all reported glucose measurements ⬍140 mg/dL (Item S1, available as online supplementary material). Together, these findings indicate that the association between lower serum sodium level and mortality is not influenced per se by the presence of diabetes mellitus. Dialysate Sodium Prescription by Serum Sodium Categories and Mortality A dialysate sodium prescription of 140 mEq/L was used in 6,412 patients (55%), whereas 2,295 (20%) patients were dialyzed with dialysate sodium concentrations ⬎140 mEq/L and 2,848 (25%) patients were dialyzed with dialysate sodium concentrations ⬍140 243

Hecking et al

Figure 3. Practice pattern of the dialysate sodium (DNa) prescription. Distribution of DNa by predialysis mean serum sodium (SNa). No correlation of DNa and SNa by MantelHaenszel ␹2 (P ⫽ 0.4) or Pearson correlation (r ⫽ 0.002; P ⫽ 0.8).

mEq/L. The dialysate sodium prescription was distributed nearly identically throughout all mean serum sodium categories, indicated by very similar percentages of patients with a dialysate sodium prescription ⬍140 mEq/L, exactly 140 mEq/L, and ⬎140 mEq/L throughout the 9 groups of mean serum sodium concentrations (Fig 3). A scatterplot confirmed the lack of correlation between dialysate sodium concentration and mean serum sodium level (Pearson correlation coefficient r ⫽ 0.002, P ⫽ 0.8, scatterplot not shown). With a mean of 141.0 ⫾ 2.5 (standard deviation) mEq/L, the highest dialysate sodium prescriptions were used in Italy, whereas the United Kingdom and Germany had the lowest dialysate sodium prescriptions (138.4 ⫾ 1.6 mEq/L) and, correspondingly, a smaller percentage of patients with dialysate sodium concentration ⱖ142 mEq/L (Fig 4). Interestingly, the

greatest discrepancy in practice patterns concerning dialysate sodium prescription was identified upon analyzing the number of dialysis facilities predominately using a single level of dialysate sodium prescription (Fig 4). There were several countries, such as Australia/New Zealand and Japan, in which the majority of facilities used only one level of dialysate sodium concentration (mostly 140 mEq/L), whereas in countries such as Italy and France, facilities using only a single level of dialysate sodium were the minority. Compared with a dialysate sodium prescription of 140 mEq/L (reference), a dialysate sodium prescription ⬎140 mEq/L was not associated with higher mean serum sodium concentrations (mean serum sodium, 0.02 mEq/L higher; P ⫽ 0.8; same model as in Table 2), and neither was a dialysate sodium prescription ⬍140 mEq/L associated with significantly lower mean serum sodium concentrations (mean serum sodium, 0.11 mEq/L lower; P ⫽ 0.3). These findings suggest that a patient’s mean serum sodium level was not modified through higher or lower dialysate sodium prescriptions in the observed ranges of dialysate sodium concentrations. To assess the potential effect of dialysate sodium prescription on mortality, we applied the fully adjusted Cox model from our previous analysis (model 5, Table 3) stratified by serum sodium tertile (Fig 5) and with the variable of interest as a 9-category combination of dialysate sodium and serum sodium tertiles (Table 4). Compared with HD patients with mean serum sodium levels ⱖ140 mEq/L and a dialysate sodium prescription of 140 mEq/L (reference), mortality risk was higher in HD patients with mean serum sodium levels ⬍140 mEq/L and dialysate so-

Figure 4. Dialysate sodium (DNa) prescription by country. * Percentage of facilities in which ⱖ90% of patients use a single DNa concentration. Abbreviations: ANZ, Australia/New Zealand; US, United States; UK, United Kingdom. 244

Am J Kidney Dis. 2012;59(2):238-248

Serum Sodium and Mortality

Figure 5. Adjusted mortality risk associated with dialysate sodium (DNa) prescription and predialysis mean serum sodium (SNa) level. Mortality hazard ratio (HR) and 95% confidence interval (CI) based on approximate tertiles of serum sodium (3 Cox models), stratified by phase and country and adjusted for age, sex, black race, vintage, body mass index, intradialytic weight loss, DNa prescription, residual kidney function, predialysis systolic blood pressure, vascular access, albumin level, hemoglobin level, serum creatinine level, serum ferritin level, white blood cell count, 14 comorbid conditions, and facility clustering effects (n ⫽ 11,392).

dium prescriptions of all 3 categories (P ⱕ 0.05), consistent with our noted findings. In analyses stratified by serum sodium tertiles, we observed that within the higher mean serum sodium tertiles (137-137.9 and ⱖ140 mEq/L), there were no significant mortality risks or benefits associated with dialysate sodium prescriptions ⬎140 or ⬍140 mEq/L compared with a prescription of exactly 140 mEq/L (reference). However, in the lower mean serum sodium tertile group (⬍137 mEq/L), the mortality HR was relatively lower for dialysate sodium prescriptions ⬎140 mEq/L (HR, 0.77; 95% CI, 0.60-0.98) versus dialysate sodium prescriptions of 140 mEq/L (reference). These findings were consistent when dialysate sodium prescriptions were divided into 5 categories of ⬍138, 138-139, exactly 140, 141-142, and ⬎142 mEq/L: In patients with mean serum sodium levels ⬍137 mEq/L, those with dialysate sodium prescriptions ⬎142 mEq/L had the lowest mortality risk (HR, 0.65; 95% CI, 0.47-0.91) in comparison to patients with dialysate sodium prescriptions of exactly 140 mEq/L (reference). Similar results were obtained in sensitivity analyses excluding Japan. In the lower mean serum sodium tertile group (⬍137 mEq/L), cardiovascular deaths (HR, 0.53; 95% CI, 0.33-0.86) were decreased to a greater extent than noncardiovascular deaths (HR, 0.87; 95%

CI, 0.63-1.20) with the use of higher dialysate sodium prescriptions (⬎140 mEq/L).

DISCUSSION In the present analysis, we have investigated the association of predialysis serum sodium concentration with patient characteristics and mortality and the influence of dialysate sodium prescription on this relationship using a large subset of the international DOPPS. We identified several associations between serum sodium level and patient characteristics, such as race, body mass index, and comorbid conditions, as well as a strong inverse relationship between serum sodium level and mortality risk, assessed with various levels of detailed adjustments. We observed that as practiced in the DOPPS, dialysate sodium prescription was independent of a patient’s serum sodium level, although the typical dialysate sodium prescription varied by country. Because serum sodium is measured routinely but rarely interpreted in HD patients, these observations are clinically meaningful and set the stage for further study. The apparent benefit of higher mean serum sodium values, with the lowest mortality risk observed at mean serum sodium concentrations ⱖ142 mEq/L, was unexpected. In the Cox proportional hazards model, most factors associated with mean serum so-

Table 4. Adjusted Mortality Risk Associated With Dialysate Sodium Prescription by Predialysis Mean Serum Sodium Serum Sodium < 137

Serum Sodium ⴝ 137-139.9

Serum Sodium >140

Dialysate sodium ⬍140 No. of patients

1.55 (1.26-1.91) 799

1.46 (1.17-1.82) 1,172

1.11 (0.87-1.43) 877

Dialysate sodium ⫽ 140 No. of patients

1.66 (1.37-2.00) 1,687

1.21 (1.00-1.45) 2,690

1.00 (reference) 2,035

Dialysate sodium ⬎140 No. of patients

1.32 (1.00-1.75) 611

1.39 (1.10-1.76) 962

1.21 (0.92-1.60) 722

Note: Single Cox model with adjustments as in Fig 5; risk values shown are hazard ratio (95% confidence interval). Serum sodium and dialysate sodium are given in mEq/L. Am J Kidney Dis. 2012;59(2):238-248

245

Hecking et al

dium level (other than diabetes) did not affect the mortality risk associated with lower mean serum sodium level (Table 3). Specifically, intradialytic weight loss, as one of the main variables of interest, seemed clinically relevant in predicting lower mean serum sodium levels (mean serum sodium, 0.16 mEq lower per 1% weight lost), but this variable did not appear to have a substantial confounding effect on mortality, judged by the consistent findings after additional adjustment for intradialytic weight change in model 4 (Table 3). Higher dialysate sodium to serum sodium gradients have been suggested to be associated with mortality in incident HD patients.30 In the present study, we addressed whether this elevated mortality risk is related to higher dialysate sodium prescriptions in patients at the same mean serum sodium level or the consequence of lower mean serum sodium levels in patients at the same dialysate sodium prescription. Importantly, our analyses showed a potential benefit of dialysate sodium concentration ⬎140 mEq/L in the subgroup of patients with low mean serum sodium levels (⬍137 mEq/L) and no association of dialysate sodium prescriptions ⬎140 mEq/L with mortality in patients with higher serum sodium levels (Fig 5). These findings indicate that the higher mortality HR found in patients with higher dialysate sodium to serum sodium gradients in recent reports are a consequence of lower mean serum sodium levels as opposed to higher dialysate sodium prescriptions per se, which if anything appeared protective in our analyses in the lower mean serum sodium tertile. Extracting implications for treatment from results of the present study by recommending higher dialysate sodium concentrations in patients with lower mean serum sodium levels may be premature. However, we emphasize that the recommended individualization of dialysate sodium concentration, “aligned to serum sodium,”19 is based on sodium gradient analyses, which did not take into account the mortality risk associated with low serum sodium levels, whereas our study of the components of the gradient has dissected the role of both predialysis serum sodium level and dialysate sodium prescription. Most recently, low serum sodium concentrations were found to be associated with mortality in a data set of HD patients from the HEMO Study.22 Although this analysis included fewer (n ⫽ 1,549) and highly selected patients in the setting of a randomized trial, the reported all-cause mortality HR, adjusted to very similar variables as in the present study, was similar to the HR observed here (0.89 per 4 mEq/L higher serum sodium level, equivalent to 0.890.25 ⫽ 0.97, per 1 mEq/L higher serum sodium level). In contrast to our 246

3-measurement approximation of mean serum sodium level and lag of 30 days before starting the mortality analysis to account for preterminal patient conditions with potentially low serum sodium levels, Waikar et al22 used the baseline predialysis serum sodium level, as well as time-updated analyses with repeated measures of serum sodium, but mortality results from the 2 models differed only slightly. Our present study not only confirms the HEMO Study findings and extends them to a larger and international data set, but also provides observational information for the consequences of dialysate sodium prescription. Regarding the surprising benefit of higher serum sodium levels in HD patients observed in both studies, we refer to the well-known risks associated with hypernatremia in the nondialysis population and note that, interestingly, the risk with higher predialysis mean serum sodium levels identified in the present study did not begin until relatively high values (ⱖ145 mEq/L) were reached, which affected only a very small minority of HD patients. The DOPPS questionnaires did not require facilities to report on measurements of sodium in delivered dialysate baths (quality control) or the method of their sodium assays, which represents potential study limitations. Sodium measurement by indirect potentiometry typically is performed in centralized laboratories and compares with flame photometry.31 Direct potentiometry may be used at some facilities and implies that measured sodium activity is converted to the readout concentration by a fixed (ion-specific) multiplier31 to avoid variable electrolyte levels resulting from different measurement techniques.31 Therefore, although most facilities in DOPPS were likely to measure serum sodium by indirect potentiometry, even when direct potentiometry is used, the recorded sodium results were anticipated to compare closely with indirect potentiometry. The remaining uncertainty, also with the sodium concentration in the dialysate bath, likely is overcome by the large number of data points analyzed in the present study. In conclusion, lower mean serum sodium levels in HD patients were associated continuously with higher risk of death. Although our mortality analyses were adjusted for patient characteristics and comorbid conditions, serum sodium level was associated with certain indicators of general health, which might point to an apparently “frail phenotype” of patients with lower serum sodium levels. Despite adjusting extensively for case-mix, the possibility of residual confounding remains and causality cannot be established in the setting of an observational study. Although diabetes mellitus was associated with lower serum sodium levels, this finding was explained largely by concurAm J Kidney Dis. 2012;59(2):238-248

Serum Sodium and Mortality

rent hyperglycemia. The survival advantage associated with higher dialysate sodium prescriptions (⬎140 mEq/L), observed in patients in the tertile with low serum sodium levels (⬍137 mEq/L), is intriguing. This benefit could be related to improved intradialytic cardiovascular stability with higher dialysate sodium concentration, as might be deduced from decreased death rates due to cardiovascular compared with noncardiovascular causes. A prospective evaluation of this treatment regimen could become a potentially rewarding new research focus, particularly if total volume status is assessed simultaneously and ultrafiltration is regulated to avoid the risk associated with higher ultrafiltration rates.32,33 Our otherwise negative finding of survival benefits with different dialysate sodium prescriptions does not consider potential effects of dialysate sodium concentration on postdialysis thirst, dietary salt and water intake, interdialytic weight gain, and intradialytic changes in blood pressure. These factors also should be considered in future studies.

ACKNOWLEDGEMENTS Jennifer McCready-Maynes, employee of Arbor Research Collaborative for Health, provided editorial assistance for this manuscript. Support: The DOPPS is administered by Arbor Research Collaborative for Health and supported by scientific research grants from Amgen (since 1996), Kyowa Hakko Kirin (since 1999, in Japan), Genzyme (since 2009), and Abbott (since 2009) without restrictions on publications. For the present analysis, Dr Hecking received a research grant from the Else Kröner Fresenius-Foundation without restrictions on publication. Dr Hecking is a visiting scientist at and Mr Karaboyas, and Drs Pisoni, Robinson, and Port are employees of Arbor Research Collaborative for Health. Financial Disclosure: The authors declare that they have no other relevant financial interests.

SUPPLEMENTARY MATERIAL Item S1: Sensitivity analysis: associations of glucose-corrected mean serum sodium. Note: The supplementary material accompanying this article (doi:10.1053/j.ajkd.2011.07.013) is available at www.ajkd.org.

REFERENCES 1. Adrogue HJ, Madias NE. Sodium and potassium in the pathogenesis of hypertension. N Engl J Med. 2007;356:19661978. 2. Braunwald E, Plauth WH Jr, Morrow AG. A method for the detection and quantification of impaired sodium excretion. Results of an oral sodium tolerance test in normal subjects and in patients with heart disease. Circulation. 1965;32:223-231. 3. Hollenberg NK. Set point for sodium homeostasis: surfeit, deficit, and their implications. Kidney Int. 1980;17:423-429. 4. Strauss MB, Lamdin E, Smith WP, Bleifer DJ. Surfeit and deficit of sodium; a kinetic concept of sodium excretion. AMA Arch Intern Med. 1958;102:527-536. 5. Flanigan MJ. Role of sodium in hemodialysis. Kidney Int Suppl. 2000;76:S72-S78. Am J Kidney Dis. 2012;59(2):238-248

6. Keen ML, Gotch FA. The association of the sodium “setpoint” to interdialytic weight gain and blood pressure in hemodialysis patients. Int J Artif Organs. 2007;30:971-979. 7. Levin NW, Zhu F, Keen M. Interdialytic weight gain and dry weight. Blood Purif. 2001;19:217-221. 8. de Paula FM, Peixoto AJ, Pinto LV, Dorigo D, Patricio PJ, Santos SF. Clinical consequences of an individualized dialysate sodium prescription in hemodialysis patients. Kidney Int. 2004;66: 1232-1238. 9. Santos SF, Peixoto AJ. Revisiting the dialysate sodium prescription as a tool for better blood pressure and interdialytic weight gain management in hemodialysis patients. Clin J Am Soc Nephrol. 2008;3:522-530. 10. Munoz Mendoza J, Sun S, Chertow GM, Moran J, Doss S, Schiller B. Dialysate sodium and sodium gradient in maintenance hemodialysis: a neglected sodium restriction approach? Nephrol Dial Transplant. 2011;26:1281-1287. 11. Hecking M, Kainz A, Horl WH, Herkner H, SunderPlassmann G. Sodium setpoint and sodium gradient: influence on plasma sodium change and weight gain. Am J Nephrol. 2011;3:39-48. 12. Saran R, Bragg-Gresham JL, Rayner HC, et al. Nonadherence in hemodialysis: associations with mortality, hospitalization, and practice patterns in the DOPPS. Kidney Int. 2003;64:254-262. 13. Kalantar-Zadeh K, Regidor DL, Kovesdy CP, et al. Fluid retention is associated with cardiovascular mortality in patients undergoing long-term hemodialysis. Circulation. 2009;119:671679. 14. Agarwal R. Management of hypertension in hemodialysis patients. Hemodial Int. 2006;10:241-248. 15. Charra B. Fluid balance, dry weight, and blood pressure in dialysis. Hemodial Int. 2007;11:21-31. 16. Jindal K, Chan CT, Deziel C, et al. Hemodialysis clinical practice guidelines for the Canadian Society of Nephrology. J Am Soc Nephrol. 2006;17(3 suppl 1):S1-S27. 17. Wystrychowski G, Levin NW. Dry weight: sine qua non of adequate dialysis. Adv Chronic Kidney Dis. 2007;14:e10-e16. 18. Lomonte C, Basile C. Do not forget to individualize dialysate sodium prescription. Nephrol Dial Transplant. 2011;26:11261128. 19. Penne EL, Sergeyeva O. Sodium gradient: a tool to individualize dialysate sodium prescription in chronic hemodialysis patients? Blood Purif. 2011;31:86-91. 20. Lindley EJ. Reducing sodium intake in hemodialysis patients. Semin Dial. 2009;22:260-263. 21. Port FK, Johnson WJ, Klass DW. Prevention of dialysis disequilibrium syndrome by use of high sodium concentration in the dialysate. Kidney Int. 1973;3:327-333. 22. Waikar SS, Curhan GC, Brunelli SM. Mortality associated with low serum sodium concentration in maintenance hemodialysis. Am J Med. 2011;124:77-84. 23. Pisoni RL, Gillespie BW, Dickinson DM, Chen K, Kutner MH, Wolfe RA. The Dialysis Outcomes and Practice Patterns Study (DOPPS): design, data elements, and methodology. Am J Kidney Dis. 2004;44:7-15. 24. Young E, Goodkin DA, Mapes DL, et al. The Dialysis Outcomes and Practice Patterns Study (DOPPS): an international hemodialysis study. Kidney Int Suppl. 2000;57:S74-S81. 25. Robinson BM, Port FK. Caring for dialysis patients: international insights from the Dialysis Outcomes and Practice Patterns Study (DOPPS). Identifying best practices and outcomes in the DOPPS. Semin Dial. 2010;23:4-6. 26. Katz MA. Hyperglycemia-induced hyponatremia—calculation of expected serum sodium depression. N Engl J Med. 1973;289: 843-844. 247

Hecking et al 27. Penne EL, Thijssen S, Raimann JG, Levin NW, Kotanko P. Correction of serum sodium for glucose concentration in hemodialysis patients with poor glucose control. Diabetes Care. 2010;33: e91. 28. Lin D, Wei L, Ying Z. Checking the Cox model with cumulative sums of Martingale-based residuals. Biometrika. 1993; 80:557-572. 29. Lago RM, Pencina MJ, Wang TJ, et al. Interindividual variation in serum sodium and longitudinal blood pressure tracking in the Framingham Heart Study. J Hypertens. 2008;26:2121-2125. 30. Sergeyeva O, Usvyat L, Kotanko P, Levin N. Positive intradialytic sodium gradients relate to increasing intradialytic

248

weight gain, hospitalization rate, and mortality in chronic hemodialysis patients. WCN Poster NDT Plus. 2009;2(suppl 2):iii650. 31. Holbek CC. Understanding the different values in electrolyte measurements. http://acutecaretestingorg. 2002. Accessed August 20, 2010. 32. Flythe JE, Kimmel SE, Brunelli SM. Rapid fluid removal during dialysis is associated with cardiovascular morbidity and mortality. Kidney Int. 2011;79:250-257. 33. Saran R, Bragg-Gresham JL, Levin NW, et al. Longer treatment time and slower ultrafiltration in hemodialysis: associations with reduced mortality in the DOPPS. Kidney Int. 2006;69: 1222-1228.

Am J Kidney Dis. 2012;59(2):238-248