Hyponatraemia and fluid overload are associated with higher risk of mortality in dialysis patients

ABSTRACT

Background

The 5-year mortality rate for haemodialysis patients is over 50%. Acute and chronic disturbances in salt and fluid homeostasis contribute to poor survival and are established as individual mortality risk factors. However, their interaction in relation to mortality is unclear.

Methods

We used the European Clinical Database 5 to investigate in a retrospective cohort analysis the relationship between transient hypo- and hypernatremia, fluid status and mortality risk of 72 163 haemodialysis patients from 25 countries. Incident haemodialysis patients with at least one valid measurement of bioimpedance spectroscopy were followed until death or administrative censoring from 1 January 2010 to 4 December 2019. Fluid overload and depletion were defined as >2.5 L above, and −1.1 L below normal fluid status, respectively. N = 2 272 041 recorded plasma sodium and fluid status measurements were available over a monthly time grid and analysed in a Cox regression model for time-to-death.

Results

Mortality risk of hyponatremia (plasma sodium <135 mmol/L) was slightly increased when fluid status was normal [hazard ratio (HR) 1.26, 95% confidence interval (CI) 1.18–1.35], increased by half when patients were fluid depleted (HR 1.56, 95% CI 1.27–1.93) and accelerated during fluid overload (HR 1.97, 95% CI 1.82–2.12).

Conclusions

Plasma sodium and fluid status act independently as risk factors on mortality. Patient surveillance of fluid status is especially important in the high-risk subpopulation of patients with hyponatremia. Prospective patient-level studies should examine the effects of chronic hypo- and hypernatremia, risk determinants, and their outcome risk.

Graphical Abstract

haemodialysis, hyponatraemia, mortality, observational study, serum sodium

KEY LEARNING POINTS

What is already known about this topic?

• Dysnatremia and fluid overload contribute to increased mortality in haemodialysis patients.

• Individually, these have been established as risk factors, but their mortality effect in interaction is unclear.

What this study adds?

• We investigated the risk association between plasma sodium, fluid status and risk of mortality in a historical cohort study including more than 72 000 dialysis patients from 25 countries with over 2 million plasma sodium measurements. Our study showed that mortality risk doubled when patients with transient hyponatremia were fluid overloaded.

• Thus, in this high-risk subpopulation, adequate correction of fluid overload is especially important.

What impact this may have on practice or policy?

• Our study suggests that in fluid-overloaded hyponatremia, dry weight status should be checked and possible causes identified to counteract the negative mortality effects.

INTRODUCTION

In oligoanuric kidney failure, the kidneys can no longer concentrate or dilute urine [1]. Excess of salt and water inevitably occur unless corrected by dialysis techniques [2]. Hyponatraemia (hypotonic hyponatraemia, usually defined as values <135 mmol/L) [3] occurs in up to one-fifth of patients and is related to increased mortality [4, 5]. Fluid overload and depletion can be measured and classified by bioimpedance spectroscopy when the threshold of normal fluid status is exceeded by 2.5 L or undercut by 1.1 L [6–10]. Both hyponatraemia and fluid overload are established as individual mortality risk factors, but it is unclear how the interplay affects mortality [7, 9, 11].

Although various scenarios have been discussed of the relationship between plasma sodium, fluid overload and mortality in dialysis patients [12], uncertainty persists about the cause, correlations and implications [13]. Most studies in this area have focused only on plasma sodium, without considering fluid status, or vice versa [4]. No study has investigated whether the effect of plasma sodium on mortality might depend on fluid status. The aim of this retrospective cohort analysis was to investigate the longitudinal association between time-varying plasma sodium concentrations, fluid status and mortality risk in many patients on incident haemodialysis.

MATERIALS AND METHODS

We analysed data from a clinical database of an international dialysis network (Fresenius Medical Care NephroCare, FMC). This network includes centres in 25 countries in Europe, Africa, the Middle East and Latin America. Patients’ electronic medical records are centralized in the European Clinical Database 5 (EuCliD5) of FMC clinics. The EuCliD5 database is a real-time electronic health record for routine clinical care in dialysis centres. The database contains patient characteristics, daily treatment data, laboratory parameters and medications. Our longitudinal analysis (EuCliD5 download, 4 December 2019) included more than 2 million measurements (n = 2 272 041) of plasma sodium and fluid status (based on bioimpedance spectroscopy) in 72 163 patients.

The patients/participants provided their written informed consent to utilize their medical data for clinical studies. This study was approved by the ethics committee at the University of Wuerzburg.

Patient population

We included incident haemodialysis patients into the study if they had available repeated plasma sodium measurements and at least one valid body composition monitor (BCM) measurement within the first 3 months after commencement of renal replacement therapy. Patients were followed until death or administrative censoring.

Plasma sodium concentration and fluid status

The primary exposure measure was plasma sodium concentration. The plasma sodium level was assessed as a time-varying variable as plasma sodium was routinely measured in EuCliD. The secondary exposure variable was fluid status. We used bioimpedance measurements as they provide more objective assessment than clinical evaluation alone [7, 14]. The bioimpedance measurements available for analysis were taken with the BCM (FMC, Bad Homburg, Germany). This device is regularly used on more than 50 000 patients/month in FMC centres and more than 5 million measurements have been recorded to date. The BCM measurements were taken in the supine position before the dialysis machine was connected, so that the measurements reflected the fluid status before dialysis. The BCM measures whole-body impedance at 50 frequencies between 5 kHz and 1 MHz. It determines extra- and intracellular resistance based on Cole modelling, and a special fluid volume model estimates extra- and intracellular water [6]. These variables are used in a three-compartment body composition model that divides body weight into normally hydrated lean tissue mass, normally hydrated adipose tissue mass and excess fluid [15]. The BCM algorithm and formula have been validated using gold standard references (bromide and deuterium dilutions) in a variety of healthy volunteers and patients aged 2–95 years [7, 8, 16]. BCM uses −1.1 to 1.1 L as normal values for fluid status. An increase of +2.5 L above normal fluid status (or in relative terms fluid overload/extracellular water ≥15%) represents severe fluid overload.

Outcomes

The primary endpoint of our study was all-cause death. Documentation of death was reliable and time-accurate for each patient registered in the database.

Statistical analysis

All study variables that were obtained repeatedly during the follow-up, i.e. plasma sodium, fluid status or other markers obtained at each haemodialysis session, were evaluated as time-varying variables on a monthly grid as over 90% of measurements were available within the same month. If there were two or more measurements during a month, we calculated the monthly average of all measurements. Fluid status was evaluated as the relative weekly averaged fluid overload pre-dialysis (FO pre rel.), calculated as the ratio of weekly averaged absolute fluid overload (in L) and extracellular water. Time-varying measurements of plasma sodium und fluid overload have been classified in three categories according to clinical cut-off values that were established by experts based on their clinical relevance. Fluid status was categorized as normal (FO pre rel.: −7%), depletion (FO pre rel. ≤−7%) and overload (FO pre rel. ≥13% for females and FO pre rel. ≥15% for males) [17]. Plasma sodium was categorized as normonatremia (135–142 mmol/L), hyponatremia (<135 mmol/L) and hypernatremia (≥142 mmol/L) [18–20].

We used the upper high value of ≥142 mmol/L to define hypernatremia because the prevalence of haemodialysis patients over 145 mmol/L is marginal and reflects severe central nervous system disorders or errors in haemodialysis proportioning system [21]. The term ‘transient’ was used if the condition, for example dysnatremia, was recorded between two laboratory tests, the gap between which was usually 1 month.

Missing values in repeated measurements were replaced by the ‘last observation carried forward’ method. If the first value was missing cross-sectional mean imputation was performed. Descriptive statistics of study variables were calculated stratified by categories of plasma sodium and fluid overload including means and standard deviations (SDs) for continuous variables and frequencies for categorical variables.

Multivariate Cox proportional hazard models were used to quantify the effect of exposure on the risk of mortality adjusted for confounders. Clustering per country and medical centre was addressed by including a frailty term for country and using a robust sandwich estimator for the between-centre variance, which is a random effects model. Covariates were selected based on a conceptual model that classified factors that could potentially influence mortality as confounders, mediators or effect modifiers [see Supplementary data, Fig. S1 A (plasma sodium) and B (fluid overload)].

The aim was to obtain a consistent multivariate estimate of the effect of the two exposure variables, adjusted for all confounding variables. Any potentially mediating factors, i.e. variables that might be on the pathway between exposure and mortality, have not been adjusted. First, we fitted for each exposure a multivariate model including the exposure variables coded in deciles to examine the dose–response association pattern. The pattern of association was visualized with bar charts showing the multivariate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) of each decile compared with the reference decile with the lowest risk. If the association pattern was shown to be continuous (linear or curvilinear), we added a polynomial fitting curve to the decile plots. Secondly, because our conceptual model assumes interaction between the two exposure factors, we additionally fitted a multivariate model examining the effect of joint exposure of fluid status and plasma sodium. This model used a nine-level categorical variable, cross-classifying fluid status and plasma sodium according to the three categories described above. This model was adjusted for all confounding variables defined in the two conceptual models above. All statistical analyses were performed using STATA (StataCorp 2019; Stata Statistical Software: Release 16; StataCorp LLC, College Station, TX, USA).

RESULTS

Patient characteristics

We included incident patients receiving dialysis treatment between 1 January 2010 and 4 December 2019. Of 154 512 patients identified in EuCliD5 of FMC clinics, plasma sodium measurements and at least one BCM measurement within the first 3 months of renal replacement therapy were available for 72 163 patients (Supplementary data, Fig. S2). The average follow-up of patients was 40 months. After aggregating the measurements on a monthly time grid, 2 272 041 valid measurements were available for analysis.

Demographics, clinical variables and comorbidity were compared in the three groups defined by plasma sodium levels (Table 1). Most measurements (75%) were in patients with normonatremia, 14% in patients with hyponatremia and 11% in patients with hypernatremia. Across the three plasma sodium strata, the mean age of patients was 63 years and two-thirds were male. Most patients were overweight [body mass index (BMI) >25], the average duration of dialysis treatment was 1 month, and the three groups had similar dialysis efficiency (Kt/V). In the hyponatremic state, systolic blood pressure (139 mmHg) was slightly lower than in the normonatremic and hypernatremic states (140 mmHg for both). The mean serum albumin concentration was lower and the mean ferritin concentration was higher in hyponatremia than in normonatremia or hypernatremia. Of the hyponatremic measurements, 30% were associated with diabetes, 9% with chronic heart failure and 22% with cancer. The HbA1c values did not differ between groups. The mean HbA1c for diabetic patients was 6.6% (SD 0.96).

Table 1:

Characteristics of patients classified by plasma sodium level.

Variable	Hyponatremia	Normonatremia	Hypernatremia
Monthly measurements	257 064	1 810 709	204 268
Plasma sodium (mmol/L)a	132.2 (3.1)	138.1 (1.4)	143.7 (0.0)
Age at start of studies (years)	63.1 (14.7)	62.7 (14.0)	63.4 (14.7)
Men (%)	57	60	63
Ethnicity, white (%)	53	55	45
BMI (kg/m²)	26.9 (5.9)	27.5 (6.0)	27.2 (5.9)
Time of haemodialysis (months)	1.0 (0.8)	0.9 (0.8)	1.0 (0.8)
Systolic blood pressure (mmHg)	139 (21)	140 (20)	140 (20)
Creatinine (mg/dL)	6.9 (2.5)	7.4 (2.6)	7.3 (2.8)
Haemoglobin (g/L)	108.5 (16.3)	109.2 (15.6)	110.3 (15.3)
CRP (mg/L), median (IQR)	9.9 (14.7)	9.5 (13.7)	11.0 (12.6)
HbA1c	6.5 (0.9)	6.4 (0.7)	6.3 (0.7)
Leukocytes (×109 cells/L)	7.0 (1.4)	7.0 (1.5)	6.9 (1.7)
Albumin (g/L)	37.8 (5.4)	38.9 (4.7)	39.1 (4.6)
Ferritin (ng/mL)	610 (462)	552 (431)	561 (435)
Phosphate (mg/dL)	4.5 (1.4)	4.6 (1.4)	4.6 (1.4)
Relative average weekly fluid overload before dialysis (%)	11% (10)	10% (10)	10% (9)
Kt/V	1.44 (0.32)	1.46 (0.30)	1.42 (0.31)
Vascular access
Fistula (%)	63	63	57
Catheter (%)	30	29	31
Graft (%)	3	3	2
Other (%)	4	5	7
Medical History
Diabetes mellitus (%)	30	25	22
Liver disease (%)	6	7	4
Cardiovascular diseases (%)	31	36	33
Peripheral vascular disease (%)	2	2	1
Chronic heart failure (%)	9	12	12
Malignancy (%)	22	33	26
Dementia (%)	1	1	1
Connective tissue disease (%)	8	6	4
Chronic lung disease (%)	3	4	4

^aWeekly plasma sodium measurements were combined into a monthly grid. Values are number of patients or mean (SD) unless otherwise stated.

CRP: C-reactive protein; IQR: interquartile range; Relative average weekly fluid overload before dialysis is a % value derived from recorded body composition measurements that objectively assesses fluid status: fluid overload to extra cellular water ratio ≥−7% and <13% in women and <15% in men fluid overload: fluid overload to extra cellular water ratio in women ≥13% and men ≥15%; fluid depletion: fluid overload to extra cellular water ratio <−7% [21]; Kt/V: solvent clearance per body fluid volume (dialysis efficiency); hyponatremia: sodium <135 mmol/L; normonatremia: sodium 135–142 mmol/L; hypernatremia: sodium ≥142 mmol/L [18–20].

Table 2:

Characteristics of patients with hyponatremia compared with patients with hypernatremia, categorized by fluid status.

Plasma sodium group characteristics	Fluid depletion	Normal fluid status	Fluid overload
Hyponatremia
Measurement count	10 188	160 966	85 910
Age (years)	59.4 (15.5)	63.8 (14.6)	62.2 (14.6)
Males (%)	44.2	58.4	56.7
Ethnicity (white) (%)	46.9	55.2	50.2
BMI (kg/m2)	30.2 (7.4)	27.7 (5.8)	24.9 (5.2)
Chronic heart failure (%)*	6.3	9.0	10.5
Liver disease (%)*	2.4	5.8	6.2
Malignancy (%)*	15.0	23.2	21.0
CRP	10.0 (8.6)	8.2 (8.6)	11.0 (7.7)
Leukocyte count	7.1 (1.6)	7.0 (1.3)	7.1 (1.6)
Ferritin	606.5 (471.8)	586.5 (445.5)	655.0 (488.8)
Normonatremia
Measurement count	68 996	1 240 253	501 460
Age (years)	58.6 (15.4)	62.9 (14.9)	62.9 (14.9)
Males (%)	45.2	61.1	60.2
Ethnicity (white) (%)	46.5	57.1	52.5
BMI (kg/m2)	31.1 (7.9)	28.2 (5.9)	25.2 (5.2)
Chronic heart failure (%)*	8.6	12.0	13.1
Liver disease (%)*	4.4	7.7	7.4
Malignancy (%)*	22.4	33.8	31.2
CRP	10.7 (11.9)	10.4 (13.0)	12.7 (15.4)
Leukocyte count	7.1 (1.6)	7.0 (1.5)	7.0 (1.8)
Ferritin	535.4 (421.3)	539.4 (424.3)	585.9 (447.2)
Hypernatremia
Measurement count	5510	101 630	44 609
Age (years)	60.4 (15.8)	63.6 (14.6)	62.0 (14.4)
Males (%)	48.1	64.2	63.5
Ethnicity (white) (%)	38.3	47.4	41.7
BMI (kg/m2)	30.0 (7.7)	27.8 (5.9)	25.1 (5.2)
Chronic heart failure (%)*	8.8	12.3	12.0
Liver disease (%)*	3.0	5.2	7.2
Malignancy (%)*	18.0	26.5	27.9
CRP	6.0 (11.0)	8.0 (10.6)	7.1 (11.0)
Leukocyte count	6.9 (1.5)	6.9 (1.6)	7.1 (1.8)
Ferritin	557.7 (430.1)	434.0 (466.5)	438.0 (496.3)

*history of

Values are mean (SD) or percentage as indicated.

CRP, C-reactive protein; hyponatremia, sodium <135 mmol/L; normonatremia, sodium 135–142 mmol/L; hypernatremia, sodium ≥142 mmol/L [18–20]; normal fluid status, fluid overload to extra cellular water ratio ≥−7% and <13% in women and <15% in men); fluid overload (fluid overload to extra cellular water ratio in women ≥13% and men ≥ 15%); fluid depletion (fluid overload to extra cellular water ratio <−7%) [21].

Effects of plasma sodium and fluid overload

In the decile analysis, we observed that the mortality risk of plasma sodium changed substantially at 134 and 138 mmol/L. The relative risk increased by 80% for <134 mmol/L and by 14% between 134–138 mmol/L as compared with plasma sodium levels >138 mmol/(Fig. 1). The mortality risk associated with the relative average weekly fluid overload before dialysis (FO pre rel. avw, %) increased exponentially when the ratio between fluid overload and extra cellular water was >15% as compared with 0% (Fig. 2).

Figure 1:

HRs are presented with 95% CIs (whiskers). Estimates are adjusted for age, sex, BMI, ethnicity, comorbidities [diabetes mellitus, liver disease, chronic heart failure, cardiovascular disease, cancer, dementia, connective tissue disease, chronic lung disease, serum creatinine, leukocytes, ferritin, haemoglobin, duration of dialysis treatment, vascular access and Kt/V (dialysis efficacy)]; N = 72 163 patients.

Figure 2:

HRs are presented with 95% CIs (whiskers). Relative average weekly fluid overload before dialysis is a % value derived from recorded body composition measurements that objectively assesses fluid status. Normal fluid status: fluid overload to extra cellular water ratio ≥−7% and <13% in women and <15% in men. Fluid overload: fluid overload to extra cellular water ratio in women ≥13% and men ≥15%. Fluid depletion: fluid overload to extra cellular water ratio <−7% [21]. Estimates are adjusted for age, sex, BMI, ethnicity, comorbidities [diabetes mellitus, liver disease, chronic heart failure, cardiovascular disease, cancer, dementia, connective tissue disease, chronic lung disease, serum creatinine, leukocytes, ferritin, haemoglobin, duration of dialysis treatment, vascular access and Kt/V (dialysis efficacy)]; N = 72 163 patients.

Combined effects of sodium and fluid excess

Figure 3 visualizes the effect of bivariate exposure on mortality showing the HRs of the eight categories that classified the joint exposure of plasma sodium and fluid status in relation to the reference category (normonatremia/normal fluid status). The risk of mortality from transient dysnatremia was significantly increased when patients were fluid overloaded. The mortality risk doubled when hyponatremic patients were fluid overloaded (HR 1.97, 95% CI 1.82–2.12) (Table 2, 33% of measurements), and a smaller but significant increase (HR 1.56, 95% CI 1.27–1.93) was seen in fluid depleted patients (Table 2, 4% of measurements). Fluid-depleted hypernatremia was not associated with an increased risk of death (HR 1.15, 95% CI 0.83–1.60). The mortality risk pattern associated with transient dysnatremia (hypo- and hypernatremia) was consistent irrespective of fluid status categorization. On the other hand, the risk of fluid overload was consistently higher compared with normal and fluid depletion status among all categories of plasma sodium.

Figure 3:

HRs are shown with 95% CIs (whiskers) in relation to the reference category (normonatremia/normal fluid status). Fluid status is objectively assessed using recorded body composition measurements taken before dialysis. Weekly average indices in % values define fluid overload and depletion. N refers to the number of patients that were categorized in the indicated fluid status stratum at least once. Red: fluid overload (fluid overload to extra cellular water ratio in women ≥13% and men ≥15%) [21]; N = 61 883 patients. Green: normal fluid status (fluid overload to extra cellular water ratio ≥−7% and <13% in women and <15% in men) [21]; N = 55 262. Blue: fluid depletion (<−7%) [21]; N = 71 354. Hyponatremia is defined as plasma sodium <135 mmol/L; normonatremia is defined as plasma sodium 135–142 mmol/l; hypernatremia is defined by the clinical cut-off of ≥142 mmol/L [18–20]. Estimates are adjusted for age, sex, BMI, ethnicity, concomitant diseases [diabetes mellitus, liver disease, chronic heart failure, cardiovascular disease, cancer, dementia, connective tissue disease, chronic lung disease, serum creatinine, leukocytes, ferritin, haemoglobin, dialysis time, vascular access and Kt/V (dialysis efficacy)].

DISCUSSION

In this large cohort of 72 163 haemodialysis patients, we investigated the association between time-varying plasma sodium concentrations, fluid status and risk of death over a period of 10 years. It is well known that the risk of death increases linearly when sodium concentration decreases or fluid overload increases [1, 4, 22]. While we could confirm that both fluid overload and hyponatremia independently act as risk factors for mortality, we also identified that the signal of hazard was strongest when joint sodium and water excess were present.

The finding that hyponatremia (reflecting hypotonicity) robustly increased the risk of mortality confirms the finding of a recent study of 27 180 American haemodialysis patients, in which sodium levels <138 and ≥144 mmol/L were associated with a higher risk of mortality [23]. However, previous observational data showed mixed mortality association curves [4]. Conventional Cox regression models showed an inverse linear relationship [22, 24], while analyses of time-varying sodium concentrations showed a U-shaped association with all-cause mortality [4, 23]. As did recently a study of 178 114 haemodialysis patients from the Japanese national registry [19]. Our finding correlates with their classification that even transient hyponatremia significantly affects survival [19]. However, without randomized evidence, our understanding of whether this is causal to death or a marker of severe underlying disease remains limited [25].

Although we did not see a direct interaction between fluid status and plasma sodium in this retrospective study, our analysis showed that fluid overload is most critical in the most vulnerable subpopulation of patients with hyponatremia. It is known that plasma sodium concentration serves as a surrogate for effective tonicity and thus determines the distribution of water between extra- and intracellular compartments [26]. However, if the kidneys fail to regulate water excretion, other mechanisms for salt and water regulation might prevail [12, 13]. One possible mechanism could be dysregulation of thirst in dialysis patients due to either osmotic threshold changes (e.g. setpoint) [27] or non-osmotic factors such as neuroendocrine mediators (e.g. arginine vasopressin, angiotensin) [25] acting on various organs (e.g. gut, lung, oxidative mechanisms) outside the kidney. Another possible mechanism could be that various underlying diseases (e.g. inflammation, protein energy wasting, hypoalbuminemia, chronic heart failure, liver disease, sick cell-like syndrome) induce severe hypercatabolism and tend to increase production of endogenous water through enhanced oxidative mechanisms (muscle proteolysis) [28]. This may reprioritize cellular energy flows to conserve water and sodium, which then leads to the formation of a third compartment with significant compartment water imbalance [28, 29].

We were able to show that hyponatremia is more frequently associated with fluid overload and normal fluid status, than fluid depletion. In the logistic observational regression model based on 8883 haemodialysis patients from the international MONitoring Dialysis Outcomes (MONDO) initiative, hyponatraemia was predicted by the presence of fluid overload [11]. Our observations are consistent with the pathophysiological assumption that hypotonic hyponatremia in dialysis patients is a marker of free water excess, more generally associated with a mass imbalance between sodium and potassium [5, 12, 25].

The findings that mortality risk was associated with indicators of malnutrition, diabetes, chronic heart failure and liver disease were consistent with current knowledge [4, 25]. These associations suggest that in dialysis patients who are in a fluid-overloaded, hyponatremic state, multiple disease processes interact to contribute to the escalation of mortality risk [5]. Hyponatremia must be taken seriously in dialysis patients as it may not just be a fluid disorder, but be indicative of an underlying active disease, inflammation and protein energy wasting processes [30]. Future randomized trials should determine if systematic assessment and management of sodium and fluid has the potential to restore sodium and water homeostasis more effectively and improve outcomes in dialysis patients.

In the recent retrospective cross-sectional analysis in the Japanese registry, simple correction of hyponatraemia during dialysis treatment (based on a fixed sodium concentration in the dialysate) was not sufficient to have a protective effect on survival [19]. Instead, the presence of hyponatraemia may alert clinicians to assess and manage more adequately extracellular fluid volume status, i.e. reduce dry weight, and, more importantly, initiate detailed causal work-up rather than solely acting on dialysate sodium prescription [5]. A potential cause may be related to an intercurrent illness, an acute event or to a superimposed chronic disease.

The risk of death may be related to the sodium gradient in the dialysate rather than the pre-dialysis sodium concentration per se [19, 24, 31]. Further prospective studies in large cohorts are needed to identify factors related to the optimal management of dialysis prescriptions (i.e. extended treatment time or additional isolated ultrafiltration) as well as the role of dialysate sodium and/or dialysate plasma gradient prescription to mitigate plasma tonicity changes (i.e. isonatremic dialysis). To better understand such pathophysiological relationships, it seems necessary to build a patient-based model that examines the effects of the dialysate plasma sodium gradient prescription (neutral, positive and negative gradient) on fluid status, haemodynamic tolerance, blood pressure control, nutritional status, tissue sodium content and other important predictors of outcomes.

The strength of our study is that the data came from a large international cohort. By analysing more than 2 million plasma sodium and body composition measurements over a 10-year period, we were able to show that even a temporary excess of salt and water correlates with a significant increase in mortality risk. However, the risk assessment of transient effects did not allow us to investigate the risk in the context of chronic exposure, which is particularly relevant for dialysis patients who usually suffer from chronic fluid overload.

Despite extensive adjustments, we could not exclude the possibility that residual confounding may have influenced the results and causality cannot be investigated due to potential unknown risk factors. We did not correct hyponatremia for glucose levels. Another potential limitation was that we analysed fewer measurements in the hypo- and hypernatremic state than in the normonatremic state. Nevertheless, the 2 000 000 measurements available should be sufficient for a robust association analysis. Finally, fluid depletion was rarely observed before dialysis; consequently, the risk ratios calculated for hypo- and hypernatremia during fluid depletion may not have been as robust as those for the other associations.

Using this large contemporary cohort, we showed that normal to relatively higher pre-dialysis plasma sodium concentrations are associated with a lower risk of mortality in normohydrated and fluid-overloaded haemodialysis patients. In contrast, lower plasma sodium concentrations in fluid overloaded patients are associated with worst outcomes. In brief, improvement of dialysis patients outcomes requires both strict control of fluid status, and identification and correction of causes of hyponatremia to counteract the deleterious effects on mortality of such combined disorders.

ACKNOWLEDGEMENTS

The Country Medical Directors of the NephroCare–Fresenius Medical Care oversaw data collection and were responsible for checking data quality at the country level. These endeavors were performed by the following individuals: Marija Bojic (Bosnia and Herzegovina), Vlasta Kupres (Croatia), Daniela Voiculescu (Romania), Konstantin Gurevich (Russia), Reina Dovc-Dimec (Slovenia), Jelena Maslovaric (Serbia), Charles Swanepoel (South Africa), Serkan Kubilay KOC (Turkey), Tomas Jirka, Michaela Sagova (Czech Republic), Martin Lepiksoo (Estonia), Erzsebet Ladanyi (Hungary), Wojciech Marcinkowski, Adam Riemel (Poland), Jaroslav Rosenberger (Slovakia), Kira Enden (Sweden), Christophe Ridel (France), Kolitha Basnayake (Ireland), Mario Cioffi (Italy), Pedro Ponce (Portugal), Maria Eva Baro Salvador (Spain), Kolitha Basnayake (UK), Volodymyr Novakivskyy (Ukraine), Yerkebulan Karibayev (Kazkhastan), Alejandro Kohn (Argentina), Ana Beatriz Barra (Brazil), Eduardo Machuca (Chile), Jesús Muñoz (Colombia), Leonor Briones (Ecuador), Maria Teresa Lopera (Perú).

FUNDING

The statistical analysis was funded by Fresenius Medical Care. Jule Pinter is funded and supported by the German Research Foundation (DFG, Projektnr. 413657723 (Clinician Scientist-Programm UNION-CVD)).

DATA AVAILABILITY STATEMENT

The data underlying this article are available in the article and in its online supplementary material.

CONFLICT OF INTEREST STATEMENT

The results presented in this paper have not been published previously in whole or part, except in abstract format. J.P. received a grant from Fresenius Medical Care for medical education and publication related expenses. C.W. received honoraria from Fresenius Medical Care for lectures given outside the scope of this study, as well as honoraria for membership of the study steering committee, participation in the advisory board and for lectures. B.G. received a consulting fee from Fresenius Medical Care for conducting the statistical analysis. J.P., C.W. and B.G. declare that they have no conflict of interest with respect to the contents of this manuscript. The other authors are all employees of Fresenius Medical Care.