Increased dietary sodium is independently associated with greater mortality among prevalent hemodialysis patients

Abstract

Dietary sodium is thought to play a major role in the pathogenesis of hypertension, hypervolemia and mortality in hemodialysis patients. Thus, restriction is almost universally recommended. However, the evidence on which these assumptions are based is limited.

We undertook a post-hoc analysis of the Hemodialysis Study with available dietary, clinical and laboratory information. Linear regression models were fit to estimate associations of dietary sodium with ultrafiltration requirement, blood pressure and nutritional indices. Cox regression models were fit to estimate the association of dietary sodium intake, sodium:calorie intake, sodium:potassium intake and prescribed sodium restriction with all-cause mortality.

Complete data were available in 1770 subjects, of whom 44% were male, 63% were black and 44% were diabetic. Mean age was 58 (±14) years; median dietary sodium intake was 2080 (IQR: 1490-2850) mg/day. After case-mix adjustment, higher reported dietary sodium was associated with greater ultrafiltration requirement, caloric and protein intake; sodium:calorie intake ratio associated with greater UF requirement; sodium:potassium ratio associated with higher serum sodium. None were associated with pre-dialysis systolic blood pressure. Higher baseline reported dietary sodium, sodium:calorie ratio and sodium:potassium ratio were independently associated with greater all-cause mortality. No associations between prescribed dietary sodium restriction and mortality were observed.

Higher reported dietary sodium intake is independently associated with greater mortality among prevalent hemodialysis subjects. Randomized trials are warranted to determine whether dietary sodium restriction improves survival.

Introduction

Sodium restriction has been a central tenet for management of hemodialysis patients since renal replacement therapy first became available. Total body sodium is a critical determinant of extracellular volume, plasma volume and blood pressure.¹ Under conditions of health, sodium balance is tightly regulated, principally through natriuresis. ² However, in oligo-anuric hemodialysis subjects, renal sodium excretion is severely impaired, and the hemodialysis procedure must provide for requisite sodium and water removal.

Previous studies in hemodialysis subjects have suggested that dietary salt restriction is associated with reduced inter-dialytic weight gain (IDWG), lesser requirement for anti-hypertensive medication, and ameliorative effects on left ventricular hypertrophy. ^3-5 It has therefore been recommended that hemodialysis patients carefully restrict sodium intake, ⁶ with the belief that this leads to improved volume and blood pressure control, and ultimately survival. However, in peritoneal dialysis, there is evidence that lower dietary sodium may actually be associated with higher mortality. ⁷ One explanation for this finding is that, because sodium and caloric intake are highly correlated,⁸ the fluid homeostatic benefits of salt restriction are outweighed by unintended concomitant nutritional compromise.

Overall, prior studies in hemodialysis patients have been limited by small size, lack of prospectively collected data, limited follow-up, and consideration of only surrogate endpoints. To our knowledge, there has been no directed study of the association between dietary sodium intake and mortality. In order to address these limitations, we conducted a post-hoc analysis of the Hemodialysis (HEMO) Study, a large-scale, prospective trial with detailed dietary assessments. We examined the associations of reported dietary sodium (diet-Na) intake with ultrafiltration (UF) requirement, systolic blood pressure (SBP), nutritional indices and all-cause mortality. In an attempt to further delineate the macronutrient-independent associations of dietary sodium intake with all-cause mortality, we performed analogous analyses considering sodium:calorie intake ratio (Na:Cal) and sodium:potassium intake ratio (Na:K) as the exposures of interest. Finally, we considered prescribed sodium restriction to explore the potential clinical effectiveness of advice regarding dietary sodium intake on outcomes.

Results

The primary cohort consisted of 1770 subjects; mean age was 58 (±14) years; 44% were male, 63% were black, and 44% were diabetic (Table 1). Mean reported diet-Na was 2240 (±1050) mg/day, median 2080 (IQR: 1490-2850) mg/day; mean Na:Cal was 1.46 ±0.53 mg/kcal; mean Na:K was 1.50 ±0.80 mg/mg (Figure 1). At baseline, higher diet-Na was associated with younger age, male sex, non-black race, older dialysis vintage, longer dialysis session length, and absence of diabetes, ischemic heart disease and oliguria. Higher Na:Cal was associated with non-black race, diabetes and longer dialysis session length. Higher Na:K was associated with younger age, older dialysis vintage, and absence of diabetes and ischemic heart disease. The intra-class correlation coefficients (95% CIs) within-subject over the course of follow up were 0.38 (95% CI: 0.34, 0.41) for diet-Na, 0.48 (95% CI: 0.45, 0.51) for caloric intake; and 0.34 (95% CI: 0.31, 0.37) for potassium intake.

Table 1.

Baseline characteristics of the total study cohort and comparisons of daily dietary sodium intake (mg/day), sodium:calorie intake ratio (mg/kcal/day) and sodium:potassium intake ratio (mg/mg/day)among sub-groups.^a

Mean Reported Dietary Intake
	N (%)	Diet-Na (mg/day)	p	Na:Cal (mg/kcal/day)	p	Na:K (mg/mg/day)	p
Age (years)			<0.001		0.10		<0.001
<50	489 (27.6)	2597 ±1155		1.53 ±0.50		1.70 ±0.91
50-65	613 (34.6)	2131 ±944		1.46 ±0.50		1.49 ±0.80
>65	668 (37.8)	2077 ±991		1.49 ±0.57		1.40 ±0.70
Sex			<0.001		0.59		0.05
Male	772 (43.6)	2546 ±1132		1.50 ±0.52		1.56 ±0.83
Female	998 (56.4)	2002±908		1.48 ±0.53		1.48 ±0.78
Race			<0.001		0.03		0.05
Black	1113 (62.9)	2170 ±1015		1.47 ±0.49		1.53 ±0.82
Non-black	657 (37.1)	2357 ±1089		1.52 ±0.58		1.47 ±0.77
Diabetes			<0.001		0.001		0.002
Present	786 (44.4)	2123 ±1028		1.54 ±0.52		1.45 ±0.74
Absent	984 (55.6)	2333 ±1053		1.45 ±0.53		1.57 ±0.84
IHD			0.005		0.88		0.01
Present	692 (39.1)	2153 ±998		1.49 ±0.53		1.45 ±0.77
Absent	1078 (60.9)	2295 ±1074		1.49 ±053		1.55 ±0.82
CHF			0.51		0.06		0.06
None	1077 (60.8)	2257 ±1052		1.47 ±0.50		1.55 ±0.85
Mild	492 (27.8)	2231 ±1072		1.53 ±0.57		1.48 ±0.70
Mod/Severe	201 (11.4)	2165 ±955		1.51 ±0.57		1.43 ±0.80
Vintage (months)			0.01		0.37		0.001
0-12	247 (13.9)	2156 ±1016		1.44 ±0.54		1.37 ±0.65
12-24	458 (25.9)	2174 ±1063		1.49 ±0.51		1.49 ±0.83
24-48	485 (27.4)	2207 ±996		1.51 ±0.56		1.49 ±0.69
>48	580 (32.8)	2353 ±1081		1.49 ±0.51		1.62 ±0.91
Session length (minutes)			<0.001		0.03		0.42
≤180	528 (29.8)	2079 ±1024		1.44 ±0.52		1.47 ±0.78
181-209	377 (21.3)	2245 ±1069		1.51 ±0.60		1.55 ±0.89
210-239	572 (32.3)	2260 ±1035		1.50 ±0.52		1.53 ±0.84
≥240	293 (16.6)	2483 ±1037		1.54 ±0.45		1.53 ±0.66
Access			0.002		0.04		0.45
AVG	1052 (59.4)	2181 ±1010		1.51 ±0.54		1.50 ±0.73
AVF	602 (34.0)	2360 ±1120		1.45 ±0.51		1.52 ±0.91
Catheter	116 (6.6)	2148 ±928		1.51 ±0.51		1.60 ±0.80
Urine outputb			0.003		0.39		0.48
Non-oliguric	236 (13.3)	2425 ±1173		1.52 ±0.49		1.48 ±0.64
Oliguric	1534 (86.7)	2211 ±1024		1.49 ±0.54		1.52 ±0.83
Flux group			0.99		0.16		0.22
High	882 (49.8)	2239 ±1038		1.47 ±0.51		1.49 ±0.76
Low	888 (50.2)	2240 ±1066		1.51 ±0.55		1.54 ±0.85
Kt/V group			0.59		0.26		0.42
High	884 (49.9)	2226 ±1054		1.50 ±0.51		1.53 ±0.80
Standard	886 (50.1)	2253 ±1040		1.48 ±0.55		1.50±0.81

^aP value for global difference; significance testing was by t-test for dichotomous variables and analysis of variance for multi-level categorical variables.

^bOliguria defined as residual urine output ≤200 ml/day.

Diet-Na, daily sodium intake; Na:Cal, daily dietary sodium:calorie intake ratio; Na:K, daily dietary sodium:potassium intake ratio; IHD, ischemic heart disease; CHF, congestive heart failure; AVG, arteriovenous graft; AVF, arteriovenous fistula.

Figure 1.

Distribution of daily dietary sodium intake (mg/day; Panel A), sodium:calorie intake ratio (mg/kcal/day; Panel B), sodium:potassium intake ratio (mg/mg/day; Panel C) and prescribed dietary sodium intake (mg/day; Panel D) at baseline.

Associations with nutritional indices, volume status and blood pressure

At baseline, higher diet-Na was modestly, but significantly correlated with height, post-dialysis weight, serum albumin and creatinine (Table 2). Upon multivariable adjustment, these correlations were attenuated and no longer statistically significant. Diet-Na was strongly correlated with daily caloric, protein and potassium intake. Again these associations were attenuated upon multivariable adjustment, after which only an association with daily caloric intake remained significant. Na:Cal was not associated with anthropometric measures, or serum nutritional markers. It was modestly inversely correlated with daily protein and potassium intake. Na:K was not associated with anthropometric measures, or serum nutritional markers (Supplementary Table A); it was modestly associated with daily caloric intake and inversely associated with protein intake. Longitudinal mixed linear models demonstrated no association between baseline diet-Na, Na:Cal and Na:K and subsequent changes in nutritional parameters over time (data not shown).

Table 2.

Distribution of anthropometric variables, nutritional indices and dietary intake parameters among the cohort and associations with daily dietary sodium intake (g/day) and sodium:calorie intake ratio (mg/kcal/day).

		Correlation with diet-Na intake			Correlation with Na:Cal intake ratio
	Mean (SD)	Unadjusteda	Model 1b	Model 2c	Unadjusteda	Model 1b	Model 2c
Height (cm)	165.8 (10.2)	0.22***	0.06**	−0.02	−0.004	−0.02	−0.01
Post-HD wt (kg)	69.6 (14.9)	0.05*	0.01	0.01	0.03	−0.004	−0.02
MAMC (cm)	24.3 (4.7)	0.02	−0.03	−0.04	−0.01	−0.02	−0.03
TSFT (mm)	18.6 (13.0)	−0.10***	−0.02	0.03	0.03	0.01	0.01
Alb (g/dL)	3.8 (0.4)	0.06**	−0.01	−0.02	−0.01	−0.01	−0.02
Cr (mg/dL)	10.3 (2.9)	0.12***	−0.01	0.001	−0.04	−0.01	−0.003
Caloric intake (kcal/d)	1523 (547)	0.68***	0.63***	0.40***	-	-	-
Protein intake (g/d)	62.7 (23.1)	0.50***	0.45***	−0.04	−0.10***	−0.11***	−0.12***
K intake (mg/d)	1612 (658)	0.46***	0.43***	−0.03	−0.10***	−0.11***	−0.04*

^aUnadjusted associations estimated by Pearson's correlation.

^bModel 1 estimates represent partial correlation coefficients based on linear regression models adjusted for age, sex, race (black vs non-black), HEMO Study Kt/V and flux group assignments, post-dialysis weight, sex-by-weight cross product terms, access (fistula, graft, catheter), congestive heart failure status (none, mild, moderate/severe), presence/ absence of diabetes and ischemic heart disease (post-dialysis weight and sex-by-weight cross product terms not included as predictors in models for which post-dialysis weight was the response variable).

^cModel 2 adjusted as per model 1 as well as for serum sodium, albumin (≤3.5, 3.5-4.0 and >4.0 g/dL), serum creatinine, ultrafiltration requirement, caloric intake and protein intake (individual covariates were not included as predictors in models for which they were the response parameter).

^*p ≤0.05

^**p ≤0.01

^***p ≤0.001

Diet-Na, daily dietary sodium intake; Na:Cal, daily dietary sodium:calorie intake ratio; HD, hemodialysis; MAMC, mid-arm muscle circumference; TSFT, triceps skin-fold thickness; Alb, serum albumin; Cr, pre-dialysis serum creatinine; K, potassium.

Higher diet-Na and Na:Cal were associated with increased UF requirement, though the magnitudes of these associations were small (Table 3). Diet-Na and Na:Cal demonstrated modest associations with pre-dialysis SBP, which variably achieved statistical significance depending on the modeling strategy employed. Neither diet-Na nor Na:Cal were associated with serum sodium. Na:K was modestly associated with increased serum sodium, but not with UF requirement or pre-dialysis SBP (Supplemental Table B).

Table 3.

Distribution of ultrafiltration requirement, systolic blood pressure and serum sodium and associations with daily dietary sodium intake (g/day) and sodium:calorie intake ratio (mg/kcal/day).^a

		Association with Diet-Na intake Per g/day			Association with Na:Cal intake ratio Per mg/kcal/day
	Mean (SD)	Unadjustedb	Model 1c	Model 2d	Unadjustedb	Model 1c	Model 2d
UF requirement (L)	2.9 (1.3)	0.14***	0.06*	0.09*	0.17**	0.10	0.13*
SBP (mmHg)	152 (26)	0.13	0.61	1.58*	2.48*	1.98	1.65
Serum Na (mEq/L)	138.3 (3.9)	0.16	0.15	0.22	0.05	0.23	0.29

^aAssociations estimated by linear regression models and expressed as adjusted mean difference in response parameter per 1000 mg/d increment in diet-Na or 1 mg/kcal/d increment in Na:Cal.

^bModel 1 adjusted for age, sex, race (black vs non-black), HEMO Study Kt/V and flux group assignments, post-dialysis weight, sex-by-weight cross product terms, access (fistula, graft, catheter), congestive heart failure status (none, mild, moderate/severe), presence/ absence of diabetes and ischemic heart disease.

^cModel 2 adjusted as per model 1 as well as for serum sodium, albumin (≤3.5, 3.5-4.0 and >4.0 g/dL), serum creatinine, ultrafiltration requirement, caloric intake and protein intake.

^*p ≤0.05

^**p ≤0.01

^***p ≤0.001

Diet-Na, daily dietary sodium intake; Na:Cal, daily dietary sodium:calorie intake ratio; UF, ultrafiltration; SBP, pre-dialysis systolic blood pressure; Na, sodium.

Associations with mortality

Overall, participants contributed 4378 patient-years of at-risk time, during which 750 deaths occurred; median follow-up time was 2.1 years.

Reported diet-Na was initially considered as a restricted cubic spline with knots at 1500, 2000 and 2500 mg/d; the observed association with log mortality hazard was essentially linear (Supplemental Figure A). Thus, reported diet-Na was considered as a linear term in all subsequent analyses. On unadjusted analysis, baseline diet-Na was not associated with all-cause mortality. However, diet-Na was associated with differences in important prognostic variables. When case-mix differences were accounted for through multivariable adjustment, a potent and significant association was observed; this was further accentuated upon additional adjustment for differences in nutritional parameters (Figure 2). Across quartiles of diet-Na, there was an incremental association with greater adjusted risk of mortality (Figure 3a). Considered in both linear and categorical forms, the association between diet-Na and mortality was similar regardless of body weight (p-interaction>0.10 for all models).

Figure 2.

Hazard ratios (95% CIs) between measures of daily sodium intake and all-cause mortality. Sodium was considered as total intake (mg/day), sodium:calorie intake ratio (Na:Cal; mg/kcal/day) and sodium:potassium intake ratio (Na:K; mg/mg/day). Unadjusted estimates are indicated by circles. Estimates from Model 1 (squares) were adjusted for age, sex, race (black vs non-black), HEMO Study Kt/V and flux group assignments, post-dialysis weight, sex-by post-weight cross product terms, access (fistula, graft, catheter), congestive heart failure status (none, mild, moderate/severe), presence/ absence of diabetes and ischemic heart disease, urine volume (≤200mL/day, >200mL/day) and dialysis session length (≤180, 181-209, 210-239, ≥240min). Estimates from Model 2 (triangles) were additionally adjusted for serum sodium, albumin (≤3.5, 3.5-4.0 and >4.0 g/dL), phosphorus, creatinine and ultrafiltration requirement. All models were stratified on clinical center.

Figure 3.

Hazard ratios (95% CIs) for all-cause mortality according to quartiles (Q; referent Q1) of reported daily dietary sodium intake (mg/day; Panel A), daily sodium:calorie intake ratio (mg/kcal/day; Panel B), and sodium:potassium intake ratio (mg/mg/day; Panel C). Estimates from Model 1 were adjusted for age, sex, race (black vs non-black), HEMO Study Kt/V and flux group assignments, post-dialysis weight, sex-by post-weight cross product terms, access (fistula, graft, catheter), congestive heart failure status (none, mild, moderate/severe), presence/ absence of diabetes and ischemic heart disease, urine volume (≤200mL/day, >200mL/day) and dialysis session length (≤180, 181-209, 210-239, ≥240min). Estimates from Model 2 were additionally adjusted for serum sodium, albumin (≤3.5, 3.5-4.0 and >4.0 g/dL), phosphorus, creatinine and ultrafiltration requirement. All models were stratified on clinical center.

Baseline Na:Cal was potently associated with all-cause mortality on unadjusted analysis. This association remained largely unchanged upon multivariable adjustment (Figure 2). Categorical analysis again revealed an incrementally greater adjusted mortality risk across quartiles of Na:Cal (Figure 3b). The pattern of association between Na:K and mortality was similar to that seen for diet-Na (Figures 2, 3c). For all metrics of sodium intake, observed estimates were essentially unchanged when reported caloric and protein intake were forced into Model 2 (only protein intake was forced into Na:Cal analyses to avoid collinearity).

Prescribed sodium restriction

Median prescribed sodium restriction was 2000 (IQR 2000-3000) mg/day (Figure 1d). At baseline, more liberal prescribed sodium restriction (>2g/day vs. ≤2g/day) was associated with male sex, non-black race, congestive heart failure, non-oliguria, longer dialysis session length, greater height, weight and mid-arm muscle circumference, lower triceps skin-fold thickness, and greater reported daily caloric, protein and potassium intake (Supplemental Table C).

At baseline participants with more liberal (vs. restrictive) prescribed sodium restriction had higher mean reported diet-Na (2370 ±1070 vs. 2170 ±1030 mg/day; p<0.001), but the difference was modest. More liberal prescribed sodium restriction was independently associated with lower UF requirement and serum sodium (adjusted mean differences per Model 1: −0.17 L (p=0.01) and −0.47 mEq/L (p=0.02), respectively), but not with pre-dialysis SBP.

On unadjusted analysis, intensity of baseline prescribed sodium restriction was not associated with all-cause mortality (Figure 4). Estimates were qualitatively unchanged upon multivariable adjustment. When marginal structural analysis was used to account for potential time-dependent confounding, no association between intensity of prescribed sodium restriction and mortality was observed: HR (95% CI) for liberal (vs. restrictive) 1.03 (0.86-1.23; p=0.74).

Figure 4.

Associations between liberal (>2g/day) versus restrictive (≤2g/day; referent group; HR=1, not shown) prescribed sodium restriction and all-cause mortality. Estimates from Model 1 were adjusted for age, sex, race (black vs non-black), HEMO Study Kt/V and flux group assignments, post-dialysis weight, sex-by post-weight cross product terms, access (fistula, graft, catheter), congestive heart failure status (none, mild, moderate/severe), presence/ absence of diabetes and ischemic heart disease, urine volume (≤200mL/day, >200mL/day) and dialysis session length (≤180, 181-209, 210-239, ≥240min). Estimates from Model 2 were additionally adjusted for serum sodium, albumin (≤3.5, 3.5-4.0 and >4.0 g/dL), phosphorus, creatinine and ultrafiltration requirement. All models were stratified on clinical center.

Discussion

To our knowledge, this is the first study to directly examine the association of indices of dietary sodium intake with mortality in the hemodialysis population. Moreover, it is the first to examine the association of diet-Na with IDWG and SBP in a large, prospective and well-characterized cohort. The primary findings of our study are that greater reported sodium intake is independently associated with an increased risk of death and modestly greater UF requirement, but bore no consistent independent association with pre-dialysis SBP. More restrictive (vs liberal) prescribed diet-Na is associated with modestly lower reported sodium intake, but not with mortality, UF requirement, macronutrient intake or SBP.

An association between diet-Na and hypertension in the general population has been well established,⁹ and reductions in diet-Na are commensurately associated with blood pressure reductions in hypertensive, pre-hypertensive and normotensive patients.¹⁰ Dietary sodium restriction (<2 g/day; <88mmol/day) has been long been championed as an important target in patients with ESRD, ⁶ with the belief that adherence would lead to meaningful reductions in IDWG, blood pressure and mortality. Prior studies in hemodialysis patients have demonstrated improvements in blood pressure control with dietary sodium restriction alone,¹¹ or combined with either reduced dialysate sodium^{12, 13} or intensive UF.³ However, small sample size, lack of dietary sodium measurements, and short follow up limit the utility of these findings. In this large, well-characterized cohort, we did not detect a consistent independent association of reported diet-Na, Na:Cal, Na:K, or prescribed sodium restriction with SBP. However, it should be noted that hemodynamic effects of sodium restriction may only become manifest in the setting of concurrent dry weight reductions, which were not prescriptively performed in the HEMO Study. ¹⁴

Prior work also suggests that greater sodium intake is associated with greater IDWG.^{3, 11} We found statistically significant associations between higher reported diet-Na, Na:Cal (but not Na:K) and greater UF requirement. However, the clinical implications of these findings may be limited considering the small magnitude of association (90 ml requisite ultrafiltrate per 1000mg increment in diet-Na intake; 130mL ultrafiltrate per 1mg/kcal increment in Na:Cal). It is possible that in HD patients other sources of sodium, (e.g. ‘loading’ from relatively hypernatremic dialysate), non-osmotically mediated thirst (e.g. from transient intravascular hypovolemia following dialysis) or habitual drinking, may play a greater role in determining IDWG and pre-dialysis SBP than previously recognized.

Of interest, unadjusted analyses failed to demonstrate an association between diet-Na (and dietary Na:K ratio) and mortality owning to extensive confounding (i.e. healthier patients tended to consume more sodium); however, when such differences were adjusted for (Model 1), a potent and independent association was observed. We estimated the diet-Na—mortality association without (Model 1) and with (Model 2) adjustment for nutritional indices. The former may more accurately represent the effects of diet-Na under prevailing clinical circumstances, where sodium intake is obligatorily linked to caloric intake and nutritional status. The latter may better estimate the therapeutic potential of sodium curtailment under a (currently hypothetical) paradigm where diet-Na can be manipulated without attendant nutritional compromise. That the diet-Na—mortality association was further potentiated upon adjustment for nutritional parameters (Model 2) suggests that, at present, the beneficial effects of diet-Na restriction are partially offset by unintended nutritional sequelae. Na:Cal analyses provide complementary information, essentially estimating the association between a 1 unit increment in diet-Na with caloric intake held constant. However, interpretation of Na:Cal estimates must consider that a higher ratio can also reflect lower caloric intake at constant sodium intake. Na:K estimates were chosen as another means to standardize diet-Na without incurring this interpretation difficulty. It is notable that for each of these three metrics we observed potent, significant and dose-dependent associations upon case-mix adjustment, which in turn provides reassurance that greater sodium intake is associated with greater mortality risk.

Assuming that the diet-Na—mortality association is causal, enthusiasm regarding clinical application must be tempered. Analyses of prescribed sodium restriction, which better assess clinical effectiveness (versus those of reported diet-Na, which point toward efficacy), did not find any association between the degree of prescribed sodium restriction with improvements in UF requirement, SBP or mortality. Likely, this stems from poor adherence to prescribed sodium restriction, as evidenced by the fact that mean diet-Na was only 200 mg/d lower among patients prescribed restrictive versus liberal sodium restrictions. It should be noted that our ability to detect associations between intensity of prescribed dietary sodium restriction and outcomes was limited by the fact that nearly all patients in the study were prescribed some dietary sodium restriction, and the range of prescribed limits was quite constrained (effectively, we studied threshold of 2 vs 3-4 g/d). In this light, it is important that findings not be extrapolated to suggest that liberal sodium restriction thresholds, beyond the range studied, are safe. Of note, subjects with significant residual renal function (urea clearance >1.5mL/min/35L urea) were excluded from participation in the HEMO Study. It is possible that those individuals may be better able to excrete excess sodium and thus be less susceptible to associated sequelae. Caution must therefore be applied before generalizing results from the present analyses to individuals with significant residual function.

Despite meticulous collection of prospective data in the setting of a randomized clinical trial, there are limitations of our study that deserve mention. The potential for exposure misclassification requires careful consideration. Several studies have questioned the utility of dietary recall in accurately assessing true dietary sodium intake (r= 0.30-0.61), ^15-18 compared with 24hr urine sodium excretion as a gold standard. For obvious reasons, urine collections would not provide accurate diet-Na assessments in oligo-anuric subjects. That the magnitude of the association between reported diet-Na intake and UF requirement we observed was consistent with those reported in the literature^{3, 11} offers a modicum of reassurance that exposure misclassification was not untoward. Moreover, in as much as random measurement error should bias toward the null, our estimates should be conservative and represent lower bounds on true biological associations. As with all observational studies, the possibility of residual confounding based on variables not considered (and for prescribed sodium restriction, confounding by indication), or due to incomplete adjustment of those that were considered, remains. Finally, known differences between clinical trial participants and the general hemodialysis population may limit the generalizability of our findings.

In conclusion, higher reported dietary sodium, whether expressed as raw intake, in proportion to caloric intake, or in proportion to potassium intake, was associated with greater mortality in prevalent hemodialysis subjects. Randomized trials are warranted to determine whether dietary sodium restriction, particularly if accomplished without decrement in overall caloric intake, improves survival.

Methods

Study Design and Population

The study was deemed exempt by the Partners Healthcare Institutional Review Board. All data were abstracted from the HEMO Study with the permission of the National Institute of Diabetic and Digestive and Kidney diseases. The design of the HEMO Study has been previously reported.^{19, 20} Briefly, HEMO was a prospective, multicenter, randomized clinical trial of standard/high dialysis dose and low/high flux membranes among prevalent adult subjects receiving thrice-weekly in-center hemodialysis. Exclusion criteria included a baseline serum albumin <2.6g/dL, residual urea clearance of ≥1.5ml/min/35L of urea distribution volume, inability to consistently achieve an equilibrated Kt/V of >1.3, or presence of end-stage co-morbid conditions. Of the 1846 HEMO Study participants, we excluded those who did not survive or were censored before the start of at-risk time (n=76); our final cohort consisted of 1770 individuals.

Exposures and Outcomes

The primary exposures of interest were reported and prescribed diet-Na intake, Na:Cal intake ratio and Na:K intake ratio. Reported dietary intake was assessed from 2-day dietary assisted recalls. Recall was performed on one dialysis and one non-dialysis day (most often on consecutive days) at annual intervals. A trained study dietician entered the data into customized versions of Nutritionist IV (version 4.0a; March 1995 – July 1999) or Nutritionist V (versions 2.0h to 2.1h; August 1999 – December 2001) software.

Dietary prescriptions were provided by the local dialysis center dietician, and were not influenced by HEMO Study dietitians, except in the following circumstances: normalized protein catabolic rate <1g/kg/d; caloric intake <28kcal/kg/d; persistent decline in serum albumin or unintentional weight loss. In these instances, dietary counseling was provided, and if insufficient, supplements were recommended. HEMO Study dieticians were not involved in diet-Na prescriptions.

The primary outcome was time to death from any cause. At-risk time began on post-randomization day 90, to enable capture of dietary recall data that lagged after randomization. Subjects were followed until censoring (transplant, transfer or change in modality; n=335), death (n=750), or study completion (December 31^st 2001; n=685). Secondary outcomes included the associations with UF requirement, pre-dialysis SBP and other biochemical, dietary and anthropometric parameters.

Study data

Per HEMO protocol, all study data were obtained via subject interview, chart review and self-reported questionnaires. Demographic data including sex, race, age and vintage were recorded at baseline and updated annually. Remaining variables of interest were recorded at baseline and at scheduled intervals during follow-up. These included diabetes mellitus, ischemic heart disease and congestive heart failure (annually); dialysis treatment parameters and hemodynamic parameters (weekly); mid-arm circumference and triceps skin fold thickness (annually; mid-arm muscle circumference was calculated as arm circumference – π*triceps skin fold thickness; both in cm); laboratory (6 weekly) and prescribed nutritional restrictions (biannually). Co-morbidities were graded on the Index of Co-existing Disease (ICED) scale; analytically these were dichotomized (0 if ICED score=0; 1 if ICED score≥1) except for congestive heart failure (0 if ICED score=0; 1 if ICED score=1; 2 if ICED score≥2).²¹

Statistical analysis

Continuous variables were examined graphically and recorded as means (± standard deviations) for normally distributed data, or medians (with inter-quartile ranges) for non-normally distributed data. Comparisons were made using t-tests or Wilcoxon rank sum tests, as appropriate. Categorical variables were examined by frequency distribution, recorded as proportions and comparisons made using the χ² test.

Exposure variables and covariates were considered as the most proximate value preceding at-risk time. The associations between dietary sodium indices and covariates were examined by Pearson's correlation and by linear regression models (with varying degrees of multivariable adjustment as described below). For nutritional response variables, estimates were presented as partial correlation coefficients (which are independent of the scale of the predictor or response parameters) to facilitate comparison of the associations across metrics of dietary sodium and across response parameters. For UF requirement, SBP and serum sodium, these were expressed as beta coefficients to facilitate clinical interpretation.

The associations between dietary sodium and mortality were examined by proportional hazards regression. All survival models were stratified on clinical center in order to minimize any potential center effect. Eligible covariates for model inclusion were those that demonstrated significant or near significant association with mortality (p<0.2) on bivariable analysis. HEMO Study flux and Kt/V assignment were forced into the model, as this was a post-hoc analysis of a randomized controlled trial. [Effect modification on the basis of HEMO Study flux and Kt/V assignments was examined for (separately) and excluded.] Two separate multivariable adjusted models were fit. Model 1 included covariate terms for eligible demographic, comorbidity, and dialysis-related variables (except UF requirement) that were unlikely to serve as intermediates in causal pathways between sodium intake and mortality. Model 2 additionally included eligible covariates related to dietary intake, nutritional adequacy and UF requirement that might serve as intermediates on some causal pathways.

Because the prognostic significance of body size may differ according to gender, two-way sex-by-dry weight cross product terms were included in all multivariable models. Effect modification on the basis of body weight was tested for (via likelihood ratio testing) by comparing nested models that did and did not contain two-way sodium-by-weight interaction terms.

The linearity assumption for continuous variables was tested by graphical examination of Martingale residual plots and by comparative model fit diagnostics using Akaike's Information Criterion (AIC). The proportionality assumption was tested by scaled Schoenfeld residual testing.

Time-updated marginal structural models were adjusted for the same variables as Model 2, through application of stabilized probability of exposure and censoring weights, as previously described.^22-26 Nominal two-sided p-values of <0.05 were considered statistically significant. Analyses were performed using STATA 10.0MP (College Station, TX).