The Effect of Increased Frequency of Hemodialysis on Volume-Related Outcomes: A Secondary Analysis of the Frequent Hemodialysis Network Trials

Fluid overload is common in patients receiving hemodialysis (HD); this leads to hypertension, structural and functional cardiac abnormalities and increased mortality [1,2,3,4]. The United States National Institutes of Health (NIH) sponsored the Frequent Hemodialysis Network (FHN) trials, which compared the effects of frequent (6 times per week) and conventional (3 times per week) HD on several intermediate outcomes in both in-center ‘daily' and home nocturnal settings. Study design [5], recruitment efficiency [6], baseline data [7,8], primary results for each trial [9,10] and the effects of frequent HD on selected secondary outcomes have been published previously [11,12,13,14,15,16,17,18].

Change in left ventricular mass (LVM) was one of the co-primary outcomes. Frequent HD significantly reduced LVM in the Daily Trial; the magnitude of that effect was similar in the Nocturnal Trial but did not reach statistical significance. The effects were more prominent in patients with low residual urine volume (UVol) and very large effects on LVM (i.e. ‘regression' of left ventricular hypertrophy) were seen exclusively in patients randomized to frequent HD [17]. The exact mechanism(s) responsible for the change in LVM is unknown, although reduction of volume load could be responsible, analogous to the effect of salt restriction [19]. In the Daily Trial, slightly longer weekly treatment times and shortened interdialytic intervals in the frequent HD group would be expected to result in reduction in volume load; the modeled volume load over the interdialytic period would also depend on intra- and interdialytic sodium balance, thirst, body fluid balance, and residual UVol. In this study, we introduce a novel model to quantify the interdialytic volume load, time-integrated estimate of fluid load (TIFL; units liter × day), described in detail in the Appendix. The model takes into account immediate postdialysis thirst-driven fluid intake as well as subsequent diet and sodium-driven fluid intake over the intradialytic period. We applied this model to analyze the treatment effect of frequent HD on estimated volume load over the entire HD cycle and to examine the relation between a new time-integrated estimate of fluid load TIFL and changes in LVM.

Patients for the Daily Trial (ClinicalTrials.gov number: NCT00264758) were recruited by 2 consortia of in-center HD units, one organized by the Renal Research Institute (RRI) in New York and the other by dialysis units at university and free-standing HD centers in California and Texas. Patients for the Nocturnal Trial (ClinicalTrials.gov number: NCT00271999) were recruited from a number of home-dialysis training centers in the United States and Canada. Both trials were approved by local Institutional Review Boards and conducted according to the Declaration of Helsinki.

The FHN Daily and Nocturnal Trials were controlled and randomized. The primary endpoints were assessed after a follow-up period of 12 months [5]. Inclusion and exclusion criteria were nearly identical for both trials, with the exception of residual kidney function at trial entry. In the Daily Trial, we excluded patients with residual urea clearance >3 ml/min per 35 liter of urea distribution volume, while in the Nocturnal Trial, where practical recruitment issues required a focus on incident patients, we set the residual kidney function exclusion considerably higher: at an average residual clearance of urea and creatinine >10 ml/min/1.73 m²[5].

We performed cardiac magnetic resonance imaging at baseline and during the 12^th month. Serum sodium concentration was measured monthly by potentiometric methods in certified local laboratories. Dialysate sodium concentration was not measured but retrieved from the dialysis provider medical record. The dialysate-to-serum sodium gradient was calculated as the prescribed dialysate sodium concentration minus the serum sodium value (in mmol/l). The interdialytic weight gain (IDWG) and the residual daily UVol were assessed according to the dialysis facility routine. To quantify the interdialytic volume overload, we computed a novel parameter designed to represent a time-integrated (‘area-under the curve') estimate of extracellular fluid load (TIFL). The TIFL curve was computed based on the IDWG, which is affected by the residual UVol and the dialysate-to-serum sodium gradient. We used the latter to calculate an estimated postdialysis serum sodium concentration, which in turn was used to estimate the volume of postdialysis fluid intake required to restore the predialysis serum sodium level to its usual value. The modeled TIFL also accounted for the sodium and fluid intake during the interdialytic period. In addition, we aimed at estimating the cardiac loading if no UVol had been present by calculating the TIFL corrected for urine volume (TIFL^UVol). See the Appendix for details on how TIFL was calculated.

Bioimpedance Measurements

Bioimpedance analysis (BIA) consisting of measurements of resistance and reactance were conducted by trained study coordinators using a single-frequency (50 kHz) wrist-to-ankle BIA method (RJL Systems, Clinton Township, N.J., USA). As per protocol, the bioimpedance measurements were conducted prior to a mid-week HD treatment; however, deviations in a minority of cases may be noted. Accuracy and precision of the BIA method that was used have been reported elsewhere [20,21]. Extracellular fluid (ECF) volume was calculated as the difference between BIA-estimates of total body water (TBW) [22] and intracellular fluid volume [23]. The ratio of ECF volume to TBW was used as an indicator of fluid overload.

All reported p-values are 2-sided without adjustments for multiple comparisons. P-values were considered significant when p < 0.05. We conducted all analyses using SAS version 9.2 (SAS Institute, Cary, N.C., USA).

Comparative Analysis

We evaluated the treatment effect of frequent versus conventional HD on the extracellular volume-related parameters IDWG and TIFL separately in both trials by development of linear mixed effects models adjusted for clinical center, age, diabetes mellitus, clinical center and baseline value. We explored the effect of modification by baseline residual UVol and the dialysate to the serum sodium gradient.

Correlational Analysis

We evaluated whether extracellular volume status mediated the effect of frequent HD on LVM by fitting and fitted a multivariable linear regression model, including baseline clinical characteristics, treatment assignment and (individually) IDWG, TIFL and sodium gradient. We have also fitted a companion multivariable model with systolic blood pressure (SBP) as the dependent variable.

We used data from all enrolled subjects in whom sufficient data were available in both trials to compute the TIFL. Subject characteristics in each trial (table 1) did not differ significantly between the frequent and conventional HD arms consistent with earlier analyses [8].

IDWG and TIFL changed significantly from baseline to the end of the study period in the intervention groups of both trials (table 2; fig. 1 and 2). The significant differences in changes in IDWG and TIFL between the treatment arms of the Daily Trial (fig. 1a and b) are in contrast to the findings in the Nocturnal Trial (fig. 2a and b), where the treatment effect on change in IDWG or TIFL was substantially less (table 2) and showed differences between the treatment arms only when TIFL^UVol including UVol was plotted (fig. 3). The treatment effect of more frequent dialysis on IDWG was significant in both trials, but the treatment effect on TIFL was significant only in the Daily Trial (table 2).

In the Daily Trial, there were significant interactions between the assigned modality and the residual UVol; patients with lower residual UVol exhibited more pronounced effects on IDWG and TIFL. The slope estimates for IDWG were 1.2 liter × days per liter/day residual UVol (p = 0.03) and 4.7 liter × days per liter/day residual UVol (p = 0.02) for TIFL. The interactions were not significant in the Nocturnal Trial. There was no significant interaction between the assigned modality and the dialysate to serum sodium gradient.

The multivariable generalized linear model based on data from all subjects (within each of the 2 trials) identified change in ECF as the only significant determinant of changes in LVM in the Daily Trial (table 3; slope estimate 4.1 g per liter ECF reduction; p = 0.01). When change in TIFL was replaced in the model with change in IDWG, neither change in IDWG nor change in ECF was a significant predictor of change in LVM (table 3; slope estimate 2.0 g per liter ECF reduction; p = 0.09). In the Nocturnal Trial, where the same multivariable models were constructed, only baseline ECF/TBW (slope estimate 1.9 g per % of ECF/TBW reduction; p = 0.04) was found to be a significant predictor of change in LVM. When change in TIFL was replaced in the model in the same population with change in IDWG, no significant predictors of change in LVM could be identified.

Using the same model in the Daily Trial, in subjects with baseline residual UVol <100 ml/day, change in TIFL (slope estimate 3.8 g per liter × day; p = 0.03) and change in ECF (slope estimate 6.1 g per l; p < 0.01) were each found to be significant determinants of change in LVM.

Using the same model, but focusing on predialysis SBP as the dependent variable, in both the Daily Trial and in the Nocturnal Trial, in subjects with residual UVol <100 ml/day, neither change in TIFL (slope estimate 1.1 mm Hg per liter × day; p = 0.11) nor change in ECF (slope estimate 0.4 mm Hg per liter; p = 0.75) was a significant predictor of change in predialysis SBP.

The FHN trials investigated two pre-specified co-primary intermediate outcomes: change in LVM and change in self-reported physical health. Our previous report suggested that blood pressure changes are the predominant influence on reduction of LVM [18]. However, a more detailed analysis of data presented here suggests that the effect of frequent dialysis on reduction in LVM, especially in those patients with minimal residual UVol, is to some extent also related to change in two measures of fluid overload, the TIFL, and the bioimpedance-estimated ECF volume.

Reduction of ECF is known to be followed by a reduction in blood pressure and consequently LVM [24,25]. In the Daily Trial, an association was found between change in predialysis SBP over the 12-month study period and change in LVM (slope estimate 6.6 g per 10 mm Hg in the group receiving 3/week HD and 10.1 g per 10 mm Hg in the group assigned to frequent HD) [17]. In the Daily Trial, a reduction in ECF or TIFL could affect LVM indirectly, via a reduction in blood pressure [26], or volume overload reduction per se might effect a reduction in LVM, since fluid overload during the interdialytic period increases cardiac strain [1,2,3] and subsequently connective tissue growth [27].

In our multivariable model examining factors associated with change in LVM during the 12-month trial period, change in ECF was indeed a significant predictor of LVM change (in the Daily Trial in the subset of patients with minimal residual UVol). However, neither change in TIFL nor changes in ECF were significant predictors of changes in predialysis SBP in either of the FHN trials. These findings suggest that changes in LVM associated with more frequent dialysis were likely caused by volume load and ECF expansion directly, and not via an effect of volume load reduction on blood pressure, although of course, both volume and blood pressure mechanisms may be contributory. It is possible that there may be nonlinear relationships between the reduction of volume-load measures and blood pressure, or other interaction between volume and blood pressure that was not captured by the statistical approaches used for the present analysis [15].

The length of the interdialytic period varied substantially between control and intervention arms in each trial. The effect of the interdialytic period on volume load is taken into account by the TIFL model but only partially by IDWG. In conventional (3 times per week) dialysis, the interdialytic period ranged from approximately 44 h for a 2-day interval to 68 h for a 3-day interval, and in the intervention arm, with more frequent dialysis, the interval ranged from 22 to 46 h in the Daily Trial and 17 to 41 h in the Nocturnal Trial. It is of interest to see that TIFL, as a concept, is in agreement with the gradual increase of TBW shown by Movilli et al. [28]. However, it needs to be noted that the gradient the authors report would not have altered postdialysis serum sodium concentration and thus would not have caused osmotically driven fluid intake.

In addition to the length of the interdialytic period, the TIFL also includes information about the timing at which fluid ingestion is hypothesized to occur. According to the current understanding of the consequences of intradialytic sodium loading, a positive dialysate to serum sodium gradient will raise the postdialysis serum sodium concentration and thereby affect serum osmolality, which stimulates thirst and consequently increases fluid intake in the early postdialysis period. A relationship between dialysate sodium prescription and thirst [29] and IDWG [30,31] is well established and corroborates the aforementioned dynamic. As a consequence of this postdialysis water intake, the ECF expands shortly after the dialysis session, an expansion that persists as a volume load with consequences, in particular on cardiac structure [27], until removal during the next HD (fig. 4). Such an early postdialysis volume increase due to the thirst-driven fluid ingestion is expected to result in a consequential early increase in ECF and in turn increased cardiac strain; the latter is then further augmented by continued ingestion of salt and water during the interdialytic period. The longer the interdialytic period and the exposure to the fluid load, the greater the overall increased cardiac strain and speculatively the more pronounced the increase in LVM [26].

Thus, a time-scaled parameter may more accurately reflect the effects of the actual volume load on the cardiovascular system. To support this notion, change in TIFL in addition to change in ECF were both significant predictors of change in LVM in those Daily Trial patients with minimal residual UVol (table 4).

The inclusion criteria in the Nocturnal trial for residual UVol were less stringent than those in the Daily Trial, and consequently, many patients had very substantial baseline residual urine output [17]. Most likely for this reason, the intervention in the Nocturnal Trial resulted in a smaller reduction of IDWG and TIFL than in the Daily Trial (table 2). This supposition is corroborated by a less pronounced separation of TIFL curves between treatment and control groups in the Nocturnal Trial compared to the groups in the Daily Trial (fig. 2). Furthermore, there was a significant interaction between residual daily UVol and treatment effect on IDWG in both trials (for TIFL it was significant in the Daily Trial only). In accordance with already published results [17,18,] the presence of high baseline urinary volume may have also reduced the effects of frequent HD on LVM change [24,25].

In conclusion, the current analysis presents additional evidence that a reduction in volume and sodium overload in HD may beneficially reduce LVM, which has been previously increased due to various degrees by fluid loading and hypertension. TIFL may be a useful tool to quantify the interdialytic volume load and associated cardiac stress, and demonstration of its reduction may be useful to more precisely ascertain the degree of interdialytic fluid load. In addition, our data also clearly shows that residual UVol has a substantial reducing effect on volume overload and positive consequences on cardiac structure. This emphasizes the great importance of measures to preserve residual renal function in dialysis patients. To what extent the use of increased HD frequency for reducing fluid overload should be preferred over stringent sodium restriction, possibly combined with more aggressive ultrafiltration and with lengthening of dialysis time, remains a subject for additional study.

The authors wish to acknowledge and thank the entire FHN Trials Group. Part of the data presented in this manuscript was presented at the Renal Week 2012 of the American Society of Nephrology in San Diego, Calif., USA; and at the World Congress of Nephrology 2013 of the International Society of Nephrology in Hong Kong, People's Republic of China.

Chair, Steering Committee: A. Kliger; National Institutes of Diabetes and Digestive and Kidney Diseases (NIDDK): P. Eggers, J. Briggs, T. Hostetter, A. Narva, R. Star; Centers for Medicare and Medical Services: B. Augustine, P. Mohr; Data Coordinating Center - Cleveland Clinic: G. Beck (PI), Z. Fu, J. Gassman, T. Greene, J. Daugirdas, L. Hunsicker, B. Larive, M. Li, J. MacKrell, K. Wiggins, S. Sherer, B. Weiss; MRI Core - Ohio State University and Mt. Sinai Medical Center: S. Rajagopalan, J. Sanz, S. Dellagrottaglie, M. Kariisa, T. Tran, J. West; Central Quality of Life Core - University of Pittsburgh: M. Unruh; R. Keene, J. Schlarb; Central Holter Core - Toronto General Hospital: C. Chan, M. McGrath-Chong; Biospecimen Repository - Fisher BioServices: R. Frome, H. Higgins, S. Ke, O. Mandaci, C. Owens, C. Snell; Data Safety and Monitoring Board: G. Eknoyan (Chair), L. Appel, A. Cheung, A. Derse, C. Kramer, N. Geller, R. Grimm, L. Henderson, S. Prichard, E. Roecker.

University of California San Francisco (UCSF)/Stanford Consortium: G. Chertow (PI); UCSF and San Francisco Bay Area: S. James, G. Chertow, M. Tamura, Y. Hall, C. McCulloch, P. Painter, I. Gorodetskaya, M. Tichy, M. Humphreys, J. Luan, R. Escalada, R. Rodriquez; UC Davis and Sacramento Area: T. Depner, G. Kaysen, M. Suter, J. Sonico, S. Anderson; El Camino Hospital and Satellite Health Care: G. Ting, B. Schiller, N. Coplon, S. Doss, J. Rogers, A. Dominguez, J. Atwal, D. Lemus; UCLA and Los Angeles Area: A. Rastogi, A. Nissenson, W. Goodman, I. Salusky, S. Schweitzer, M. Rivas, M. Smith, P. Gayda, A. Hernandez, M. Rashid; UCSD and San Diego Area: R. Mehta, J. Pepas, B. Bharti, A. Nabali, R. Manaster, R. Mathew, S. Shah, G. Sanz, J. Wei; University of Texas, San Antonio: J. Ayus, S. Achinger, M. Gutierrez; RRI New York Consortium: N. Levin (PI); W. Bay, M. Carter, R. Geronemus, M. Kuhlmann, G. Handelman, F. Gotch, F. Finkelstein, P. Kimmel, E. Lacson, D. Ornt, R. Greenwood, J. Vassalotti, J. Burrowes; RRI New York City: N. Levin, P. Kotanko, A. Kaufman, J. Winchester, I. Meisels, B. Radbill, J. Chang, Y. Fofie, R. Ramos, O. Sergeyeva, J. Callegari, B. Arthur, M. Tarallo, D. Ulloa, R. Apruzzese; University of Western Ontario: R. Lindsay, R. Suri, A. Garg, R. Bullas, A. Mazzorato; Wake Forest University School of Medicine: M. Rocco, J. Burkart, S. Moossavi, V. Mauck, T. Kaufman, A. Coppley; Vanderbilt University Medical Center: G. Schulman, S. McLeroy, M. Sika, E. Leavell; Barnes Jewish/Washington University: B. Miller, R. Schussler, J. Bardsley, R. Skelton.

Wake Forest University School of Medicine Consortium: M. Rocco (PI); Barnes-Jewish/Washington University: B. Miller, J. Riley, R. Schuessler; Lynchburg Nephrology: R. Lockridge, M. Pipkin, C. Peterson; Rubin Dialysis: C. Hoy, A. Fensterer, D. Steigerwald; University of Iowa: J. Stokes, D. Somers, A. Hilkin, K. Lilli, W. Wallace, B. Franzwa, E. Waterman; University of Toronto: C. Chan, M. McGrath-Chong; University of British Columbia: M. Copland, A. Levin, L. Sioson, E. Cabezon, S. Kwan, D. Roger; University of Western Ontario: R. Lindsay, R. Suri, J. Champagne, R. Bullas, A. Garg, A. Mazzorato, E. Spanner; Wake Forest University School of Medicine: M. Rocco, J. Burkart, S. Moossavi, V. Mauck, T. Kaufman; Humber River Regional Hospital: A. Pierratos, W. Chan, K. Regozo, S. Kwok.

Supported by the NIH, NIDDK, the Center for Medicare and Medical Services and the NIH Research Foundation. Contributors to the NIH Foundation in support of the FHN trials included Amgen, Baxter, and Dialysis Clinics. Additional support was provided by DaVita, Dialysis Clinics, Fresenius Medical Care, Renal Advantage, RRI, and Satellite Healthcare.

F.A.G. has a consultancy agreement with Fresenius Medical Care North America. P.K. holds stock in Fresenius Medical Care North America. N.W.L. is a scientific advisor to Aethlon and Affymax, received funding from the US National Institute of Health and owns stock in Fresenius Medical Care North America. R.L. has unrestricted research support from Fresenius Medical Care Canada. G.M.C. serves on the Board of Directors of Satellite Healthcare.

Concept

Despite well-documented associations between fluid overload as determined by bioimpedance and LVM in patients receiving hemodialysis [32], no correlation has been shown between IDWG and changes in LVM [33]. Development of an improved method to assess the degree of fluid loading in the interdialytic period (θ) in relation to intradialytic Na⁺ transfer, dietary Na⁺ intake and subsequent ECF expansion was stimulated by the need for better evaluation of cardiac stress. The model incorporates the length of exposure to fluid overload, not considered in previous approaches and allows quantification of the composite of the length of exposure and the magnitude of ECF expansion. By calculating the area under the volume curve over the entire interdialytic period, a TIFL can be estimated.

In line with existing literature, we propose that the parameter estimated with the aid of the proposed model represents an appropriate correlate to cardiac stress, reflecting the volume load as being a strain on the heart over the whole period of exposure reaching levels of right-ventricular pressure as seen in heart failure [34]. The TIFL incorporates information on the effects of the dialysate to serum sodium gradient (GNa⁺), IDWG and the length of the interdialytic interval (θ).

Model Outline

Based on the studies of Edelman et al. [35,] it is assumed for this model that the intracellular potassium and extracellular Na⁺ are in osmotic equilibrium and that extracellular serum sodium concentration (SNa⁺) and intracellular K⁺ are, for simplification, present in similar concentrations [35]:

This relationship can be rearranged to:

and assuming intracellular K⁺ to be equal to SNa⁺, to

Therefore, the change in total Na⁺ content (ΔNa⁺) in ECF during the interdialytic period can be evaluated as the product of IDWG and the difference between SNa⁺_pre and SNa⁺_post (after equilibrium with the interstitial fluid and ICF is reached)

ΔNa⁺ = IDWG × (SNa⁺_post - SNa⁺_pre),(4)

where SNa⁺_pre is the pre HD SNa⁺ at set-point SNa⁺, and SNa⁺_post the post HD SNa⁺ after equilibration. In accordance with this, the postdialysis, thirst-driven water intake ∆V_osm can be estimated based on the change in SNa⁺ during HD. The TBW (as the cation distribution volume), for the current analysis was assumed to be similar to the urea distribution volume derived from urea kinetic modeling:

Since in the current analysis SNa⁺_post is not known, it was estimated using linear regression analysis to develop an algorithm based on the pre HD SNa⁺ and GNa⁺ [31,36,37]

SNa⁺_post = 0.59 × SNa⁺_pre + 0.41 × DNa⁺ - 0.85.(6)

ΔV_osm and ΔV_iso have to be separated from IDWG during the first day of the interdialytic interval and then ΔV_iso calculated for the total interval in accordance with

ΔV_iso = IDWG - ΔV_osm,(7)

where ΔV_osm is estimated as per equation 5 and is applicable for both the situation of a positive GNa⁺ resulting in increased post-HD osmolality and subsequent compensatory fluid intake, and a negative GNa⁺ resulting in a lower SNa⁺_post and subsequent increased salt intake, possibly even ‘salt hunger,' without additional fluid intake. Figure 4 outlines the model and the respective areas under the curves as they are estimated in the model.

Subsequently, we calculated the areas under the curve of ΔV_osm for the entire interdialytic period (θ) as

TIFL_ΔVosm = ΔV_osm × θ.(8)

Consequently, the TIFL for the ΔV_iso over the entire interdialytic period (θ) is

TIFL_ΔViso = 0.5 × ΔV_iso × θ,(9)

with the unit liter × days.

The calculation of the total area under the curve TIFL, defined as the sum of TIFL_∆Viso and TIFL_∆Vosm, is simply the sum of the integrals.

TIFL = TIFL_∆Vosm + TIFL_∆Viso.(10)

The TIFL^UVol was calculated with the adjusted ΔIDWG^UVol for each patient as

ΔIDWG^UVol = IDWG + UVol × θ,(11)

and then all calculations repeated with the adjusted hypothetical IDWG^UVol following equations 7-11.