Associations of Abnormal Fluid Status, Plasma Sodium Disorders, and Low Dialysate Sodium with Mortality in Patients on Hemodialysis

Visual Abstract

Abstract

Key Points

• This multicenter observational study followed 68,196 patients on hemodialysis over a decade and investigated the effect of cumulative exposure burden of abnormal fluid and sodium concentrations and dialysate sodium on mortality.

• Of >2 million patient-months, 61% were spent in any degree of fluid overload, 28% in severe fluid overload, and 4% in fluid depletion; 11% were spent in hyponatremia and 1% in hypernatremia; and 64% were spent on low dialysate sodium.

• Cumulative exposure to even mild fluid overload was associated with higher risk of death, and this association was independent of abnormal plasma sodium and low dialysate sodium.

Background

Abnormal fluid and plasma sodium concentrations are established prognostic factors for patients on hemodialysis. However, the cumulative effects of abnormal salt and water and potential effect modifications and the effect of dialysate sodium remain incompletely understood.

Methods

The study followed 68,196 patients on incident hemodialysis from 875 dialysis clinics in 25 countries over 10 years (2010–2020) investigating dose-response patterns between cumulative exposure time of fluid overload/depletion (measured by bioimpedance spectroscopy using the Fresenius Body Composition Monitor), abnormal plasma sodium levels, low dialysate sodium, and all-cause mortality. We calculated time-varying cumulative exposure (in months) of relative fluid overload (any degree; >7% or severe; >13 or >15% in women or men, respectively) and fluid depletion (<−7%), hypo- or hypernatremia (sodium <135 or >145 mmol/L, respectively), low dialysate sodium (≤138 mmol/L), and estimated hazard ratios for all-cause mortality using a multivariable Cox model.

Results

Of 2,123,957 patient-months, 61% were spent in any degree of fluid overload, 4% in fluid depletion, 11% in hyponatremia, and 1% in hypernatremia. Any degree of fluid overload was associated with higher all-cause mortality (hazard ratio peak at 3.42 [95% confidence intervals, 3.12 to 3.75] relative to no exposure), and this association with all-cause mortality seemed to be stronger with severe fluid overload. The risk pattern associated with hyponatremia was approximately linear in the first four patient-months and then plateaued after the fourth patient-month. We did not observe effect modification between fluid overload and hyponatremia.

Conclusions

Even mild fluid overload was associated with higher mortality in patients on hemodialysis. Whether more stringent fluid management results in clinical improvement requires further investigation.

Introduction

Patients on hemodialysis frequently experience deranged fluid and salt levels, leading to cardiocirculatory stress, ventricular remodeling, vascular stiffening, arrhythmias, and heart failure, all increasing mortality.^1–4 Multifrequency bioimpedance devices measure volume status, reducing clinical assessment uncertainty, but randomized studies on routine bioimpedance use have shown inconsistent results.^5,6 Improving bioimpedance-guided decisions requires better understanding clinically significant fluid overload. Typically, fluid overload is defined as an extracellular volume increase >15% (approximately 2.5 L in a 70-kg person),^3,6–10 but a threshold of >7% before dialysis (approximately 1.1 L) might better identify at-risk individuals.^6,8–12 Although the hazard associated with acute or time-varying fluid overload and plasma sodium has been identified, previous studies were limited in follow-up or considered risk markers separately.^3,13,14

The associations of abnormal fluid and sodium status with long-term mortality in patients on hemodialysis remain incompletely understood, especially the interplay of chronic exposure to abnormal fluid status, plasma sodium concentrations, and dialysate sodium with mortality. This international multicenter historical cohort study followed patients on incident hemodialysis for 10 years to assess the cumulative associations of abnormal fluid and plasma sodium status and dialysate sodium with mortality. We evaluated the association of fluid overload on mortality using two thresholds and explored potential interactions between fluid overload and hyponatremia, aiming to refine the understanding of these associations in patients on hemodialysis.

Methods

We analyzed data from an international dialysis network (NephroCare, Fresenius Medical Care [FMC]) with dialysis centers in 25 countries in Europe, Africa, Middle East, and Latin America (supplemental research protocol). Patients' records were stored in the European Clinical Database 5, a real-time electronic health record system. Ethical approval was obtained from the University Hospital at Würzburg, Germany (reference: 255/22).

Study Population

Patients on incident hemodialysis were included from January 1, 2010, to December 4, 2019, if they had one valid plasma and dialysate sodium and bioimpedance measurement within 90 days of starting hemodialysis. Patients prescribed ultrafiltration and dialysate sodium profiles (1–6) were excluded to avoid bias by indication.^15,16 Follow-up began after the first bioimpedance measurement and continued until death, transplantation, modality change, site change, or administrative censoring (December 4, 2019).

Monthly reviews of key performance indicators within the FMC network enable continuous monitoring of clinical and operational performances across all countries and clinics, ensuring clinical care improvements are driven by data and not by regional differences.¹⁷ Clinical governance guides nephrologists in Nephrocare clinics to conduct at least three clinical assessments per week, monthly blood draws, and one Body Composition Monitor (BCM) measurement every 13 weeks. Dialysis prescriptions, including dry weight, are updated monthly. Specific interventions, such as plasma volume–adjusted ultrafiltration, or extra sessions are performed at the physician's discretion.

Exposures

Fluid status was assessed by bioimpedance spectroscopy using the BCM in the supine position before dialysis. This provided a more objective assessment than clinical evaluation.¹⁸ Fluid status was based on average weekly relative fluid status before dialysis in liters (L) related to extracellular water in L. Nephrocare advises midweek BCM measurements, but European Clinical Database (EuClid) records varied. We used average weekly predialysis fluid status to standardize comparisons. This measure adjusts for weekday variations: subtracting 0.4 L on the first dialysis day and adding 0.2 L on the other days. Relative overhydration is standardized as percentage. Percentage values better reflect individual variance than absolute values. The equation we use was validated and corrected for body mass index to reduce systematic errors associated with extremes in body composition, thus enhancing the reliability of fluid volume measurements.¹⁹

Fluid status was classified using clinical cutoffs: any degree of fluid overload (>7%, approximately equivalent to 1.1 L), severe fluid overload (>13% or 15% for women and men, respectively, equivalent to approximately 2.5 L), and fluid depletion (<−7%, equivalent to approximately <−1.1 L). This approximation of +1.1 or 2.5 L excess for a typical adult with an average body weight of 70 kg aids in translating a percentage-based threshold to a tangible volume for easier clinical interpretation.

The second exposure variable was plasma sodium concentration, categorized as hyponatremia (<135 mmol/L), isonatremia (≥135 to ≤145 mmol/L), or hypernatremia (>145 mmol/L), on the basis of established clinical thresholds.^17,20 The third exposure variable was dialysate sodium concentration considering ≤138 mmol/l as low dialysate sodium compared with >138 mmol/L.²¹ Low dialysate sodium was considered exposed because of higher risk, i.e., a hazard ratio (HR) >1.

BCM-measured fluid status (mean: 0.9, SD: 1.5) and plasma sodium measurements (mean: 0.7, SD: 1.5) were mostly available on a monthly basis and dialysate sodium for every session.

Covariates were selected on the basis of previous work^13,21 classifying models for plasma sodium, dialysate sodium, and chronic fluid overload, considering factors influencing mortality as risk factors, mediators, or effect modifiers (see Supplemental Figure 1, A–C and Supplemental Table 9). Our primary end point was all-cause death. Deaths are reliably recorded in EuClid, as the absence of a patient from the clinic triggers a follow-up to determine why the patient is missing. In the event of death, the cause of death is recorded according to 10th revision of the International Classification of Diseases codes and verified at the clinic level. For each clinic, the distribution of deaths is monitored monthly for outliers.

Statistical Analyses

Descriptive statistics of study variables in total and stratified by exposure were calculated, including means and SDs or median and interquartile range for non-normal distributed data for continuous variables and frequencies for categorical variables. We considered five distinct longitudinal exposure processes that were presumed to associate with higher risk of mortality, namely (1) fluid overload, using two different definitions: severe overload only defined as >13% (women) or 15% (men), or any overload defined as >7%; (2) fluid depletion <−7%; (3) hyponatremia (plasma sodium <135 mmol/L); (4) hypernatremia (plasma sodium >145 mmol/L); or (5) low dialysate sodium ≤138 mmol/L. The exposure was first coded using time-varying binary variables classifying each patient-month into the risk conditions. We then counted the cumulative number of months spent in in each exposure status as time-varying variables. Given that a patient could have both, in the past, history of fluid overload/depletion or hyponatremia/hypernatremia, the reference category of no exposure was tailored specifically for each exposure state and thus not directly comparable between processes (e.g., the patient was either nonoverloaded or nondepleted or nonhyponatremic or nonhypernatremic in each month). Cox regression models, including time-varying variables, stratified by country and adjusted for clustering by medical center using a robust sandwich estimator, were used to estimate the HRs of exposure. All time-varying variables were evaluated on a monthly grid, and covariate time series were smoothed calculating moving averages over 3 months. If for a particular month there were two or more measurements available, we calculated monthly averages; if there was no measurement, we used the last observation carried forward imputation from the previous month. For missing covariates in the first month, we used cross-sectional mean imputation; a multiple imputation was not feasible in this large dataset because of the computational complexity.

We fitted the following models in the total study population: first, three univariate models including fluid overload/depletion, hyponatremia/hypernatremia, or low dialysate sodium. Because a patient can have both, past periods of overload and depletion (and hypo- and hypernatremia, respectively), we adjusted even in the univariate model the complementary process. Second, three multivariable core models including additional covariates according to our conceptual model (Supplemental Figure 1, A–C) aimed to estimate the overall strength of the association of the exposure adjusted for potential confounding variables. Third, to explore effect modification between fluid overload and plasma sodium, we fitted a model with a combination variable reflecting the joint bivariate exposure of fluid overload and hyponatremia (both categorized in quartiles). Finally, we fitted a full multivariable model including all five exposure processes with the pooled set of confounding variables. Then, we conducted sensitivity analyses (1) to explore effect mediation refitting the models including mediation factors and (2) to explore potential survivorship bias, i.e., surviving longer increases opportunity to accrue exposure time, refitting all models only in patients surviving at least 48 months.

Association patterns were visualized by decile plots, both in the total population and in patients surviving longer than 48 months. As high-risk patients dropped out earlier, the hazards of the total population were expected to be biased downward along with exposure time. The medians of the decile groups of cumulative exposure time are shown on the x axes and the HRs of all-cause death with respect to the reference category no exposure (=0) on the y axes. In each figure, the first panel shows the hazard associated with cumulative exposure time in the total population. The second panel represents the sensitivity analysis conducted in the subpopulation surviving for at least 48 months to explore the potential effect of survivorship bias. All statistical analyses were conducted using STATA (StataCorp. 2023. Stata Statistical Software: Release 18. College Station, TX: StataCorp LLC).

Results

Patient Characteristics

We identified 72,163 patients on incident hemodialysis in the EuClid disease register, with at least one BCM, plasma sodium, and dialysate sodium measurement within 90 days within 90 days of starting hemodialysis between January 1, 2010, and December 4, 2019, from 875 hemodialysis clinics in 25 countries. Of these, we selected 68,196 patients without dialysate profiles (Supplemental Figure 2). from 25 countries (Supplemental Table 4). The minimum follow-up period was 30 days and median (interquartile range) survival time was 71 (95% confidence interval [CI], 70 to 73) months, and we analyzed a total of 2,123,957 patient-months of exposure. The baseline mean age was 63 (SD=15) years, and 39% of the patients were female. Patients were frequently overweight and hypertensive, and 35% and 12% of patients had a history of cardiovascular disease and chronic heart failure, respectively (see Table 1 for further baseline descriptive statistics). At the end of the study period, 21,644 patients (32%) died, 6298 (9%) received a kidney transplant, and 25,636 (38%) were alive and on hemodialysis. The remaining 14,618 (21%) were censored during follow-up because of other reasons. The most common reason for censoring was transfer to another site (see Supplemental Table 5 for the full list of reasons for censoring).

Table 1.

Descriptive statistics of study variables in N=68,196 patients on hemodialysis

Patient-months of exposure	2,123,957 (100.0%)
Age (yr)a	62.9 (14.8)
Sex (male)a	1,281,336 (60.3%)
Dialysis vintage at first BCM (d)a	28.6 (23.1)
Ethnicity (%)a
Asian	15,726 (0.7%)
Black	28,614 (1.3%)
Unknown	923,620 (43.5%)
White	1,155,997 (54.4%)
BMI (kg/m2)	27.4 (6.0)
Systolic BP before dialysis (mm Hg)b	140.2 (20.3)
Creatinine (mg/dl)	7.3 (2.6)
Hemoglobin (g/L)	109.3 (15.7)
CRP (mg/L), median (IQR)	9.7 (8.1)
Leukocytes (×109 cells/L)	7.0 (1.6)
Albumin (g/L)	38.8 (4.8)
Ferritin (ng/ml)	558.5 (432.2)
Phosphate (mg/dl)	4.6 (1.4)
Single-pool Kt/V	1.4 (0.3)
Ultrafiltration (ml)	1951.5 (1370.9)
Vascular access
Fistula	1,332,875 (62.8%)
Catheter	641,454 (30.2%)
Graft	55,008 (2.6%)
Unknown/other	94,620 (4.5%)
Hypertension medication	2008 (2.9%)
Diabetes	541,213 (25.5%)
Liver disease	149,297 (7.0%)
Cardiovascular disease	751,175 (35.4%)
Peripheral vascular disease	42,250 (2.0%)
Chronic heart failure	252,891 (11.9%)
Malignancy	652,075 (30.7%)
Dementia	22,500 (1.1%)
Connective tissue disease	128,315 (6.0%)
Chronic lung disease	86,874 (4.1%)

Data are mean (SD) or patient-months (%) unless otherwise stated. BCM, Body Composition Monitor; BMI, body mass index; CRP, C-reactive protein; IQR, interquartile range; Kt/V, dialysis efficacy; %, percentages.

^aFixed variables evaluated at the beginning of follow-up. All other variables were time varying evaluated in a monthly grid. Comorbidities were considered as chronic diseases and coded as time varying with beginning of exposure at the first diagnosis of the disease.

^bBP measurements displayed in the table were automated measurements using a cuff before each hemodialysis session that were automatically transferred to the European Clinical Database 5.

Descriptive Statistics

Fluid Overload/Depletion

Of 2,123,957 patient-months of exposure, 61% were spent in any degree of fluid overload and 4% in fluid depletion. Patients who spent more time in fluid overload had more comorbidities, such as diabetes mellitus, cancer, or connective tissue disorders, and most notably a higher prevalence of cardiovascular disease (Supplemental Table 1A). By contrast, only 16% of patient-months were spent in a fluid-depleted state. Patients who spent more time in fluid depletion were younger, had a higher body mass index, had higher hemoglobin, had lower systolic BP before dialysis, and had fewer comorbidities (Supplemental Table 1, A and B).

Hypo-/Hypernatremia

Eleven percent of patient-months were spent in hyponatremia. Patients who spent more time in hyponatremia tended to be younger (mean: 62, SD: 14 years) in the highest quartile (versus mean: 64, SD: 15 years in the lowest quartile). Compared with no exposure, patients most exposed to hyponatremia had higher C-reactive protein and ferritin. A longer duration spent in hyponatremia was associated with higher prevalence of diabetes mellitus and connective tissue disorders and lower prevalence of cardiovascular disease or malignancy. Patients who spent more time in hyponatremia showed similar characteristics, regardless of the chronic exposure time (Supplemental Table 1C). Hypernatremia was the least commonly encountered condition with only 1% of patient-months spent in this state. A longer cumulative time spent in hypernatremia was associated with chronic heart failure, higher ferritin, higher prevalence of central venous catheter use, and lower ultrafiltration volume (Supplemental Table 1D). There was no association between exposure of hyponatremia and severity of fluid overload: The mean relative fluid overload before dialysis was similar in months with hyponatremia (mean: 10%, SD: 10%) and months with hyper- or normonatremia (mean: 10%, SD: 10%).

Dialysate Sodium Concentration

Dialysate sodium ≤138 mmol/L was used in 64% of patient-months. Longer exposure to low dialysate sodium correlated with cardiovascular disease and chronic heart failure. Older individuals had less cumulative exposure to lower dialysate sodium, likely because of their shorter lifespan (Supplemental Table 1E).

Cumulative Exposure Time–Risk Pattern (Decile Plots)

Figures 1–4 show the decile plots visualizing the association patterns between cumulative past exposure burden and mortality. Because of tied observations, we were not able to calculate decile groups (i.e., ten groups with equal observation counts) for all exposures. Therefore, for some exposures, less than ten quantile groups are shown.

Figure 1.

Effect of cumulative exposure time of any degree of fluid overload on mortality. (A) Estimates of the total population. (B) Estimates of a subpopulation of patients who survived 48 months or longer. The x axis shows medians of decile groups of cumulative exposure time (months) of mild and severe relative fluid overload. A threshold of 7% defines any fluid overload of >1.1 L above normal fluid status. The y axis shows estimates of HRs and 95% CIs on a log scale as compared with the reference category (=0 months of fluid overload). Estimates are adjusted for the complementary exposure process fluid depletion and 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, hemoglobin, vascular access, and single-pool Kt/V). BMI, body mass index; CI, confidence interval; HR, hazard ratio; Kt/V, dialysis efficacy.

Figure 4.

Exposure time–risk association pattern of any degree of fluid overload, stratified by burden of hyponatremia. (A) Estimates of the total population. (B) Estimates of a subpopulation of patients who survived 48 months or longer. The x axis shows five groups of error bars, each group reflecting a different exposure burden of hyponatremia (plasma sodium <135 mmol/L): 0: no exposure (70% of patient-months, 0 months cumulative exposure), 1: first quarter of exposure time=1 month of exposure, 2: second quarter=2–4 months, 3: third quarter=5–11 months, 4: fourth quarter=12–118 months. The y axis shows HRs and 95% CIs as compared with the reference category (no hyponatremia/no fluid overload). The bars in different colors within each group of hyponatremia exposure represents relative risk estimates of subgroups with different exposure burden of any degree of fluid overload. Green: no fluid overload (11.5% of patient-months). Blue: first quarter: 1–4 months. Yellow: second quarter: 5–11 months. Brown: third quarter=12–24 months. Red: fourth quarter=25–120 months. 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, hemoglobin, vascular access, and single-pool Kt/V).

Risk Pattern for Fluid Overload/Depletion

Using the threshold classifying severe fluid overload (>13% or >15% for women or men, respectively), 28% of patient-months were spent in severe fluid overload as compared with 61% of patient-months with any fluid overload (>7%). There were significant associations between mortality for any degree of and severe fluid overload (Figure 1). Using the lower more sensitive BCM-measured cutoff, we could identify seven more months on average of mild fluid overload that added to the cumulative exposure time (Figure 2). The pattern of this cumulative exposure distribution, including mild fluid overload periods, was associated with a linear and steeply higher risk compared with quantifying only severe fluid overload (Supplemental Table 2): Whereas the hazard peaked in exposure months 24–31 in any degree of fluid overload (HR peak, 3.42; 95% CIs], 3.12 to 3.75), the hazard peaked in months 9–12 of exposure in severe fluid overload (HR peak, 2.57; 95% CI, 2.42 to 2.73). The risk of death in patients who survived longer than 48 months was higher by almost seven-fold (HR, 6.83; 95% CI, 5.17 to 9.02). Fluid depletion before dialysis was less prevalent and less strongly associated with the risk of death compared with fluid overload (Supplemental Figure 3). Nevertheless, the duration of time spent in a state of fluid depletion was also associated with a gradual and stepwise higher risk of mortality (HR peak, 1.97; 95% CI, 1.74 to 2.23).

Figure 2.

Effect of cumulative exposure time of severe fluid overload. (A) Estimates of the total population. (B) Estimates of a subpopulation of patients who survived 48 months or longer. The x axis shows medians of decile groups of cumulative exposure time (months) of mild and severe relative fluid overload. A threshold of 13% or 15% for women and men, respectively, defines any fluid overload of more than 2.5 L above normal fluid status. The y axis shows estimates of HRs and 95% CIs on a log scale as compared with the reference category (=0 months of severe fluid overload). Estimates are adjusted for the complementary process fluid depletion and covariates 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, hemoglobin, vascular access, and single-pool Kt/V).

Risk Pattern for Hypo-/Hypernatremia

With hyponatremia, the risk pattern was approximately linearly increasing and then remained significantly higher if hyponatremia persisted for more than four patient-months leading to a doubling in all-cause death (HR peak at 1.83; 95% CI, 1.69 to 1.99 as compared with 2.14, 95% CI, 1.84 to 2.49 among survivors >24 months; Figure 3 and Supplemental Table 3).

Figure 3.

Effect of cumulative exposure time of hyponatremia on mortality. (A) Estimates of the total population. (B) Estimates of a subpopulation of patients who survived 48 months or longer. The x axis shows medians of eight quantile groups of the cumulative exposure time (months) of hyponatremia (plasma sodium <135). The y axis shows HRs and 95% CIs as compared with the reference category (no exposure to hyponatremia). Estimates are adjusted for the complementary exposure process (hypernatremia) and covariates 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, hemoglobin, vascular access, and single-pool Kt/V).

The association between hypernatremia and all-cause death was modest, and prevalence of exposure to hypernatremia was lower than for hyponatremia. The hazard associated with prolonged hypernatremia was higher by more than 40% (HR peak, 1.43; 95% CI, 1.27 to 1.61) (Supplemental Figure 4 and Supplemental Table 2).

Risk Pattern for Low Dialysate Sodium

Although the hazards of all-cause death for patients exposed to dialysate sodium ≤138 mmol/L were markedly higher compared with no exposure (HR, 1.58; 95% CI, 1.20 to 2.10), the CIs were wide precluding definitive conclusions (Supplemental Figure 5 and Supplemental Table 2).

Risk Pattern of Stratified Analysis (Joint Exposure Fluid Overload/Hyponatremia)

Analyzing the effect of fluid overload in four strata (quartiles) with different cumulative exposure burden of hyponatremia (Figure 4), we observed a homogenous risk association pattern among strata. This indicates that there is no evidence for effect modification between fluid overload and hyponatremia. The sensitivity analysis in patients who survived longer than 48 months confirmed the mortality risk pattern of fluid overload that we observed (HR peak, 14.46; 95% CI, 8.67 to 24.12).

Full Model

A full multivariable model including all five exposure processes (fluid overload/depletion, hypo-/hypernatremia, and low dialysate sodium) showed similar HR estimates to the models shown in Figures 1–4 and Supplemental Figures 3–5 (results shown in Supplemental Table 7). This indicated no strong evidence of confounding or effect modification among the exposure processes.

Sensitivity Analysis

Sensitivity analysis including/excluding mediators according to the conceptual model (Supplemental Figure 1) did not indicate any evidence for effect mediation Supplemental Table 6. The HRs were similar after adjusting our models for potential mediating factors (Supplemental Table 8).

Discussion

Our study showed that cumulative exposure to abnormal fluid and plasma sodium concentrations as well as low dialysate sodium was associated with higher mortality. Any degree of fluid overload (>7%) and severe fluid overload (>13/15%) each exhibited an increasingly higher risk of association in a near-linear pattern, even when adjusted for confounding factors. Any degree of fluid overload was a better predictor of mortality than severe fluid overload alone. In chronic hyponatremia, we observed a continuously higher risk. Fluid overload and hyponatremia seemed to act independently with similar risk patterns across different hyponatremia exposure levels.

Our finding that fluid overload is a strong prognostic risk factor resonates with previous research.^3,7,13,14 The increasingly higher risk pattern with exposure time indicates that each month spent in fluid overload may influence survival. No gold standard for fluid overload assessment exists.^6,8,18 A recent systematic review found multiple definitions for relative fluid overload using bioimpedance spectroscopy,⁶ with one fifth of studies defining it as overhydration to extracellular water >15%. Few cohort studies report granular mortality risk stratification for relative fluid overload before dialysis.^14,22 Our data suggest that a 7% threshold (approximately 1.1 L) accurately identifies and quantifies lifetime exposure risk in patients on dialysis.

Chronic hyponatremia showed an increasingly higher mortality risk, suggesting it is not solely a fluid-related disorder. Recognized as an independent mortality risk factor,^23,24 hyponatremia should be seen as a proxy marker, prompting clinicians to investigate underlying disease processes, such as inflammation, organ failure, and protein-energy wasting.¹³

To our knowledge, no study has quantified the cumulative exposure burden of abnormal fluid status with plasma and dialysate sodium. Our stratified analysis showed high relative risks of fluid overload in patients with high hyponatremia exposure. However, the consistent risk pattern across hyponatremia exposure levels indicated no effect modification. Fluid overload and hyponatremia seem to act independently with distinct pathways. Because patients with hyponatremia have a high mortality risk, fluid management is crucial, not just correcting plasma sodium with isotonic or hypertonic dialysate.

Our study highlighted that the cumulative exposure time of even mild fluid overload was associated with higher risk. The severe fluid overload threshold may overlook important patient-months contributing to dialysis lifetime risk. The adoption of the European Clinical Database 5 as the real-time electronic health record enabled continuous quality improvement program across FMC Nephrocare clinics. Fluid status, a key performance indicator, is regularly monitored with standardized BCM measurements.¹⁷ Our analysis was robust because observed mortality rates aligned with those from routine clinical governance processes at FMC.

In FMC NephroCare clinics, a cutoff of 13% for women and 15% for men has been the target for relative fluid overload for 7 years, balancing clinical risk and intradialytic hypotension. We found that a 7% cutoff for predialytic fluid overload may be more sensitive to mortality risks, but this epidemiological finding may not be clinically applicable. The experience in FMC clinics (unpublished data) indicates that <25%–40% of the dialysis population would be able to reach a level below 7% without an increase in intradialytic hypotensive episodes. Future trials should investigate the characteristics of patients who can achieve a relative fluid overload level of 7% before dialysis and compare fluid management protocols and optimal targets for long-term fluid control. These trials should compare conventional target assessment against more intensive fluid control while assessing the risk of intradialytic hypotension in anuric patients to generate evidence on which targets improve outcomes^25,26 and are safely achievable, such as using a cluster randomized stepped-wedge design with clinics step by step entering the intervention group. Subgroup analyses would be useful to test and implement sex-specific thresholds.²⁷ For dialysate sodium, we observed an inconclusive exposure–risk pattern. Trials are underway to define optimal prescribing practices (Randomised Evaluation of SOdium Dialysate Levels on Vascular Events Study: NCT02823821).

Future longitudinal modeling should explore hyponatremia evolution to better understand pathophysiological correlations and inform intervention strategies. Analyzing cause-specific mortality may provide further insights.

Our findings should be considered within the context of study limitations. First, as an observational study, residual confounding may remain. Although we adjusted for many clinical markers, some important data, such as ethnicity, N-terminal pro-brain natriuretic peptide, or urine output, were unavailable. Second, BCM measurements have limitations in assessing fluid status in patients on hemodialysis. Factors such as electrode placement, patient movement, and body temperature can affect readings. These measurements do not consider nutritional status, inflammation, or comorbidities. Accurate assessments require combining BCM measurements with comprehensive clinical evaluations. Third, our study was affected by survivor bias; longer survival associates with higher cumulative exposure time. By re-estimating risks in long-time survivors, we showed that the risk pattern in the full population is likely underestimated. We did not have access to intradialytic BP changes, which should be analyzed in future studies.

In conclusion, our study showed a strong cumulative exposure time–mortality risk association for fluid overload. The lower threshold (any fluid overload) showed a striking cumulative risk pattern, suggesting that each episode of even mild fluid overload may cumulatively introduce higher risk to patients on hemodialysis. We found no evidence for correlation, confounding, or effect modification between fluid overload and hyponatremia, suggesting that they likely act independently as prognostic factors with distinct pathways. Future investigations are needed to determine clinical applicability. Efforts should focus on stringent fluid management, especially in high-risk patients.

Supplementary Material

cjasn-19-1444-s001.pdf ^{(1.4MB, pdf)}

cjasn-19-1444-s002.pdf ^{(1.1MB, pdf)}

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/CJN/C16.

Funding

J. Pinter: Deutsche Forschungsgemeinschaft (Projektnr. 413657723). B. Smyth: Royal Australasian College of Physicians. This work was supported by Flinders Medical Centre.

Author Contributions

Conceptualization: Bernard Canaud, Bernd Genser, Jule Pinter.

Data curation: Bernd Genser.

Investigation: Bernard Canaud, Bernd Genser, Jule Pinter.

Methodology: Bernard Canaud, Bernd Genser, Jule Pinter.

Project administration: Jule Pinter.

Software: Bernd Genser.

Visualization: Bernd Genser.

Writing–original draft: Bernd Genser, Jule Pinter.

Writing–review & editing: Bernard Canaud, Nicholas C. Chesnaye, Bernd Genser, Kitty J. Jager, Jeroen Kooman, Kaitlin J. Mayne, Ulrich Moissl, Jule Pinter, Brendan Smyth, Stefano Stuart.

Data Sharing Statement

Due to data protection issues, the full study dataset is not available on a public repository because it is an extraction of a real-time electronic health record system, namely the European Clinical Database 5 (EuCliD 5). An anonymized version of the study dataset can be requested upon request.

Supplemental Material

This article contains the following supplemental material online at http://links.lww.com/CJN/C15.

Supplemental Figure 1. Conceptual framework for the epidemiological effect of fluid overload, plasma, and dialysate sodium exposure.

Supplemental Figure 2. Patient selection and analysis procedures.

Supplemental Figure 3. Exposure time–risk association pattern of fluid depletion.

Supplemental Figure 4. Exposure time–risk association pattern of hypernatremia.

Supplemental Figure 5. Exposure time–risk association pattern of low dialysate sodium.

Supplemental Table 1. Distribution of study variables according to cumulative exposure split in quartiles of (A) mild relative fluid overload, (B) fluid depletion, (C) hyponatremia, (D) hypernatremia, and (E) dialysate sodium ≤138 mmol/L.

Supplemental Table 2. Cumulative exposure burden of abnormal fluid status: distribution and multivariable exposure–risk association pattern with all-cause mortality.

Supplemental Table 3. Cumulative exposure burden of abnormal plasma sodium status (hypo-/hypernatremia) and low dialysate sodium: distribution and multivariable exposure–risk association pattern with all-cause mortality.

Supplemental Table 4. Distribution of patients among the countries participating in the European clinical database.

Supplemental Table 5. Reasons for censoring.

Supplemental Table 6. Survivor bias sensitivity analysis refitting the models for patients who at least survived 48 months (N=16,058; 2813 events).

Supplemental Table 7. Hazard ratios of the full multivariate model.

Supplemental Table 8. Sensitivity analysis: hazard ratios of models with additional adjustment for mediating variables.

Supplemental Table 9. Life history covariable risk stratification research protocol.