Abstract
Background and objectives: Theoretical advantages exist of online hemodiafiltration (HDF) over high-flux hemodialysis (HD), but outcome data are scarce. Our objective was to compare outcomes between these modalities.
Design, setting, participants, & measurements: We studied 858 incident patients in our incremental high-flux HD and online HDF program during an 18-yr period. We compared outcomes, including survival, in those who were treated predominantly with HDF (>50% sessions) and those with high-flux HD. Survival comparisons used a Cox model taking into account the time-varying proportion of time spent on HDF. All data were prospectively collected.
Results: A total of 152,043 sessions were delivered as HDF and 291,222 as high-flux HD. A total of 232 (27%) patients were treated predominantly with HDF and 626 (73%) with high-flux HD. Total Kt/V, serum albumin, erythropoietin resistance index, and BP were similar in both groups up to 5 yr after HD initiation. Intradialytic hypotension was less frequent in the predominant HDF group. Predominant HDF treatment was associated with a reduced risk for death after correction for confounding variables. In a second Cox model, proportion of time spent on HDF predicted survival, such that patients who were treated solely by HDF would have a hazard for death of 0.66 compared with those who solely used high-flux HD.
Conclusions: We found no benefits of HDF over high-flux HD with respect to anemia management, nutrition, mineral metabolism, and BP control. The mortality benefit associated with HDF requires confirmation in large randomized, controlled trials. These data may contribute to their design.
Hemodiafiltration (HDF) was first described in the 1970s (1) as a treatment that combines the benefits of diffusion and convection but remains little used. Standard hemodialysis (HD) provides excellent small molecule clearance, but mortality remains high in the dialysis population. Increasing recognition that “middle molecules” contribute to morbidity has fueled the development of techniques to enhance their clearance. Unmodified cellulosic membranes allow ultrafiltration (UF) rates of up to 5 to 6 ml/h/mmHg per m2. Development of high-flux membranes has allowed higher rates over 20 ml/h/mmHg per m2, favoring convection, while being thin enough to permit diffusion. HDF combines diffusion and convection. The large convective volumes that are generated using high-flux membranes are replaced with substitution fluid, from either commercially available bags or using ultrapure dialysis fluid in a technique known as online HDF (2). Development of the online HDF technique has overcome many of the practical and economic difficulties in delivering HDF to large numbers of patients.
Despite these developments, there is a paucity of outcome data comparing HDF with conventional HD. Furthermore, no large published outcome studies have directly compared survival between online HDF and high-flux HD. HDF has been demonstrated to be a superior provider of middle molecule clearance (3–5). Observational studies have suggested that increased β2 microglobulin clearance by high-flux HD is associated with reduced incidence of dialysis-related amyloidosis compared with low-flux HD (6,7). Compelling data from the Membrane Permeability Outcome Study (8) suggest that high-flux HD provides a survival benefit compared with low-flux HD, at least in patients with low serum albumin.
HDF outcome data range from moderate-sized safety and efficacy studies (4,9–11) to large registry studies (12,13). Recent data from the Cardiovascular Risk in Dialysis (RISCAVID) study suggest a survival benefit for both online HDF and sterile bag HDF over standard low-flux HD (14); however, data comparing outcomes between high-flux HD and HDF are limited to a very small (n = 76), single-center, randomized, crossover study (15), which demonstrated improved β2 microglobulin clearance with online HDF, but the study design did not allow conclusions to be drawn about mortality.
The aim of our study was to compare outcomes, including survival, between online HDF and high-flux HD at our renal unit, where these techniques have been used routinely for almost two decades. Incremental dialysis has been integral to our HD program throughout this period. Our total Kt/V target for both high-flux HD and HDF is the sum of dialysis and residual renal components. This approach has required monthly monitoring of residual renal function. For much of the study period, selection of patients for treatment with HDF favored those with less residual function.
Comparing outcomes between HDF and HD in retrospective studies in which treatment modality was not randomly selected may introduce bias, as described by Fisher and Lin (16). In our study, comparison of survival between patients who were treated predominantly with HDF (>50% treatments) and predominantly with high-flux HD could create bias. Patients with little residual renal function, who had survived longer, may have been crossed over from HD to HDF, distorting differences between groups. We addressed this by comparing survival on the basis of proportion of time that each patient had spent on HDF during his or her dialysis career as a time-dependent variable. This statistical technique has been recommended to limit bias when treatment crossover may have occurred (16).
Materials and Methods
Patients
We retrospectively studied all incident HD patients at our renal unit who underwent dialysis for ≥3 mo between 1989 and 2007 (n = 858). Patients were excluded for the following reasons: Previous renal replacement therapy of any type including transplantation, transfer of care from a different renal unit, or recovery of renal function beyond 3 mo.
HD Program
Patients were treated exclusively using high-flux membranes, predominantly polysulphone. Polyacrylonitrile and other biocompatible membranes were used in a small minority. During the earlier part of the study, dialyzers were reused for patients who did not have hepatitis, but reuse was zero after 2003. Bicarbonate was used exclusively as the buffer, and ultrapure water was used for HD, online HDF, and the reuse procedure. The HDF fraction was 0.35 of the blood flow rate, and the target HDF volume was 40% of the Watson volume. Water quality was regularly monitored to ensure tight bacteriologic standards (total viable count <0.1 cfl/ml and endotoxin level <0.03 EU/ml). Selection of patients for online HDF favored those with large body size and little residual renal urea clearance (KRU <1 ml/min). Dialysate and replacement fluid ion concentrations were as follows: Sodium 138 mmol/L, chloride 108.5 mmol/L, bicarbonate 32 mmol/L, acetate 3 mmol/L, calcium 1.25 mmol/L, magnesium 0.5 mmol/L, glucose 5.5 mmol/L, and potassium 2 mmol/L.
Dialysis Adequacy and Residual Renal Function
Target total two-pool Kt/V urea (Kt/VTotal) was 1.2 per session for thrice-weekly dialysis for both patients who were on HD and patients who were on HDF. Twice-weekly dialysis was used in a small proportion of patients usually for a few months after HD initiation (target Kt/VTotal 2.0). Delivered session Kt/VTotal comprised the sum of the dialysis component (Kt/VDialysis) and the intermittent equivalent of continuous residual renal urea clearance (Kt/VRenal), calculated using the method described by Gotch (17,18). This takes into account the higher efficiency of urea removal by residual renal function compared with that of a dialyzer. The following formulas were used:
Kt/VTotal = Kt/VRenal + Kt/VDialysis
Kt/VRenal = (KRU*f)/V
where KRU is residual renal urea clearance (ml/min), f = 9500 for twice-weekly HD or f = 5500 for thrice-weekly HD, and V is Watson volume (ml).
KRU was calculated from a 48-h interdialytic urine collection performed monthly. Collections were performed during the interdialytic period from Monday to Wednesday or Tuesday to Thursday depending on HD schedule. KRU was calculated as follows:
KRU (ml/min) = 2(UID × VID)/tID (Cpost + Cpre)
where UID is urinary urea concentration (mmol/L) in the interdialytic urine collection, VI is urine collection volume (ml), tID is duration of interdialytic urine collection (min), Cpost is postdialysis serum urea (mmol/L), and Cpre is predialysis serum urea (mmol/L).
Data Gathering
For all patient demographic, body mass index (BMI) and comorbidity data were collected at dialysis initiation. The following data were collected at 3, 6, 12, 24, 36, 48, and 60 mo after HD initiation:
Treatment modality (HDF or high-flux HD) for each individual session, predominant treatment modality from dialysis initiation to each time point, and overall predominant modality throughout dialysis career. The term “predominant” describes the treatment modality that provided >50% of sessions for each patient.
Session data: Treatment duration, UF volume, and predialysis BP.
Anemia, erythropoietin, nutrition, and inflammation outcome data: Serum hemoglobin (g/dl), weekly erythropoietin dose (IU/wk), erythropoietin resistance index (19) (U/wk per kg per g/dl), serum albumin (g/L) and C-reactive protein (CRP; mg/L).
HD adequacy parameters: Kt/VTotal, Kt/VDialysis, Kt/VRenal, KRU, serum urea (μmol/L), serum creatinine (mmol/L), sodium (mmol/L), potassium (mmol/L), and bicarbonate (mmol/L).
Mineral metabolism parameters: Serum phosphate (mmol/L), corrected calcium (mmol/L), and parathyroid hormone level (PTH; pmol/L).
Intradialytic complication data (cramps, hypotensive episodes, and intravenous saline boluses) were prospectively routinely collected at each HD session. Survival outcome data collected were date of HD initiation and date of leaving program by transplantation, death, or transfer.
Statistical Analysis
Definition of Groups for Outcome Comparisons.
Survival was compared using two methods. In Kaplan-Meier and multivariate analysis (survival model 1), patients whose predominant treatment modality during their whole HD career was HDF were compared with those for whom it was high-flux HD. Baseline characteristics and intradialytic complications were compared in the same manner. In multivariate survival model 2, we calculated proportion of time spent on each treatment modality up to each time point and examined the relationship of this with survival (see Statistical Techniques).
Other outcomes were compared at different time points with patients classified as predominant-high flux HD or predominant HDF according to their predominant treatment modality up to that time point. These included the data relating to HD adequacy, nutrition, mineral metabolism, UF requirements, BP, and anemia management.
Statistical Techniques.
Statistical techniques were conducted using SPSS 16.0.2.
Comparison of groups.
T tests were used for normally distributed variables, Mann-Whitney U Tests were used for nonparametric data, and χ2 tests were used for frequency data.
Survival.
Kaplan-Meier analysis compared those whose predominant treatment modality had been high-flux HD throughout their HD career and those whose had been HDF using log-rank tests. Patients were censored for transplantation, conversion to peritoneal dialysis, or loss to follow-up (transfer elsewhere). A Cox proportional hazards model (survival model 1) was used to compare survival in those who were treated with predominant HDF and predominant high-flux HD, correcting for covariates. In this model, survival and risk were calculated with respect to one of the factors, assigned a unitary risk of 1. For example, patients with diabetes were compared to patients without diabetes, the latter not being listed and having a unitary risk of 1. The proportional hazards assumption was tested when appropriate for each covariate using two methods: Inclusion of interacting terms in the model (product of covariate and natural logarithm of survival time) and the Schoenfeld residuals method.
A second Cox regression model (survival model 2) was used to correct for the confounding effects of switching patients between treatment modalities, dialysis dosage, and residual renal clearance. Missing KRU data equating to 11% of the sample were replaced using linear regression to estimate values on the basis of other measurements for the respective patient. Proportion of time spent by each patient on HDF was calculated at each time point and entered as a time-dependent covariate (HDFTime). HDFTime varied from 0 to 1 for each patient at different time points. Other covariates entered were Kt/VDialysis (time-dependent covariate; Kt/V DialysisTime), KRU (time-dependent covariate; KRUTime), age, and comorbidities as in model 1.
Results
Baseline Characteristics
Of the 858 patients studied, 626 were treated predominantly with high-flux HD and 232 with HDF throughout their dialysis careers. A total of 50.7% of patients had no dialysis sessions delivered as HDF. Patients who were treated predominantly with high-flux HD had a median percentage of sessions delivered as HDF of 0% (interquartile range [IQR] 14%). Those who were treated predominantly with HDF had median of 79% of sessions delivered as HDF (IQR 27%). The patients in the predominant HDF group were more likely to be male (P < 0.001; Table 1) and younger (P < 0.001) and to have a higher body weight (P < 0.001) and BMI (P < 0.001). Peripheral vascular disease and malignancy also were less prevalent in this group (P = 0.002 in both cases). Mean UF volume was 14.9 ± 3.9 L for patients on HDF with a range of 5.8 to 33.2 L.
Table 1.
Descriptive analysis
| Parameter | All Patients | Predominantly High-Flux HD Group | Predominantly HDF Group | HD versus HDF Group (P) |
|---|---|---|---|---|
| N | 858 | 626 | 232 | |
| Age (yr; mean ± SD) | 61.2 ± 15.2 | 64.2 ± 13.9 | 53.1 ± 15.7 | <0.001a |
| Gender (%; M:F) | 68.2:31.8 | 62.6:37.4 | 83.2:16.8 | <0.001b |
| Weight (kg; mean ± SD) | 74.7 ± 17.7 | 71.7 ± 15.9 | 82.9 ± 19.8 | <0.001a |
| BMI (mean ± SD) | 25.5 ± 5.5 | 25.0 ± 5.3 | 26.9 ± 5.8 | <0.001a |
| Comorbidity at dialysis initiation (%) | ||||
| diabetes status | 26.3 | 26.5 | 25.9 | 0.846b |
| ischemic heart disease | 22.1 | 23.8 | 17.7 | 0.055b |
| peripheral vascular disease | 17.2 | 19.6 | 10.8 | 0.002b |
| malignancy | 13.1 | 15.2 | 7.3 | 0.002b |
| Underlying renal disease (%) | ||||
| diabetic nephropathy | 16.8 | 17.1 | 15.9 | 0.69b |
| chronic glomerulonephritis | 17.9 | 16.3 | 22.4 | 0.04b |
| obstructive uropathy | 8.6 | 8.6 | 8.6 | 1.00b |
| congenital cystic renal disease | 5.9 | 4.6 | 9.5 | 0.008b |
| renovascular disease | 7.7 | 8.9 | 4.3 | 0.02b |
| chronic renal failure of uncertain cause | 31.2 | 32.9 | 26.7 | 0.08b |
| other causes | 11.8 | 11.5 | 12.5 | 0.16b |
Independent t test.
χ2 test.
Dialysis Dosage Differences between High-Flux HD and HDF
Figure 1 demonstrates that mean Kt/VTotal delivered to patients in the predominant HDF and high-flux HD groups did not differ except that at 24 mo Kt/VTotal was lower in the HDF group (P = 0.049). Kt/V as a result of residual renal function (Kt/VRenal) was significantly lower in the patients who were treated predominantly with HDF at all time points. Because Kt/VRenal was lower in patients who were on HDF, the higher Kt/VDialysis necessary to achieve equality of Kt/VTotal was achieved through a combination of longer treatment duration in the HDF group (P < 0.001 all time points), higher dialysis fluid flow in the first 2 yr of dialysis (P < 0.006), and higher blood flow (significant at the 48- and 60-mo time points). There was no difference in dialysis session frequency between patients who were on HDF and high-flux HD at any time point (P varying from 0.461 to 1.000).
Figure 1.
Total Kt/Vurea in patients who were on predominantly high-flux HD and HDF. The incremental dialysis algorithm adjusted dialysis dosage to a total Kt/V of 1.2 regardless of dialysis type. Kt/VTotal comprised the sum of Kt/VDialysis and Kt/VRenal. There was no significant difference in Kt/VTotal between patients who were on predominantly high-flux HD or HDF at all time points except at 24 mo.
Dialysis Adequacy Parameters
There was no difference in blood urea (Table 2) between patients who were treated predominantly with high-flux HD and HDF at any time point, but serum creatinine was higher in those who were on HDF throughout (P varying from <0.001 to 0.004). There were no differences in serum sodium, potassium, and bicarbonate between the groups at any time point up to 60 mo except for a higher mean potassium at 6 mo in HDF patients.
Table 2.
Dialysis adequacy parameters in patients who were treated by predominantly HDF and predominantly high-flux HD
| Parameter | Group | 3 Mo | 6 Mo | 12 Mo | 24 Mo | 36 Mo | 48 Mo | 60 Mo |
|---|---|---|---|---|---|---|---|---|
| Sodium (mmol/L) | HD mean | 138.1 | 138.1 | 138.0 | 138.0 | 137.9 | 138.3 | 137.5 |
| HDF mean | 137.7 | 138.4 | 138.7 | 138.3 | 137.5 | 138.1 | 138.4 | |
| df | 711 | 657 | 559 | 384 | 273 | 195 | 127 | |
| P | 0.413 | 0.362 | 0.051 | 0.580 | 0.421 | 0.720 | 0.169 | |
| Potassium (mmol/L) | HD mean | 4.8 | 4.9 | 4.9 | 5.1 | 5.2 | 5.1 | 5.3 |
| HDF mean | 4.9 | 5.1 | 5.1 | 5.1 | 5.2 | 5.2 | 5.3 | |
| df | 721 | 672 | 558 | 393 | 281 | 203 | 132 | |
| P | 0.320 | 0.006 | 0.125 | 0.820 | 0.579 | 0.494 | 0.859 | |
| Urea (mmol/L) | HD mean | 21.0 | 21.6 | 21.9 | 21.7 | 21.8 | 20.4 | 20.5 |
| HDF mean | 21.9 | 22.8 | 22.4 | 22.3 | 23.1 | 21.1 | 21.0 | |
| df | 787 | 713 | 593 | 407 | 294 | 206 | 130 | |
| P | 0.28 | 0.09 | 0.38 | 0.42 | 0.09 | 0.37 | 0.64 | |
| Creatinine (μmol/L) | HD mean | 654 | 698 | 723 | 749 | 769 | 746 | 766 |
| HDF mean | 812 | 841 | 872 | 919 | 988 | 889 | 893 | |
| df | 725 | 677 | 569 | 391 | 281 | 202 | 130 | |
| P | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | <0.001 | 0.004 | |
| Bicarbonate (mmol/L) | HD mean | 23.5 | 23.2 | 23.2 | 23.0 | 23.0 | 23.0 | 23.6 |
| HDF mean | 23.6 | 22.6 | 23.0 | 23.0 | 22.5 | 22.9 | 23.1 | |
| df | 818 | 748 | 635 | 449 | 326 | 234 | 149 | |
| P | 0.718 | 0.142 | 0.658 | 0.981 | 0.171 | 0.825 | 0.424 |
Fluid Balance Parameters
Systolic and diastolic BP did not differ between the predominant high-flux HD and HDF groups at any time point up to 60 mo (Figure 2), except for a higher mean systolic BP at 36 mo in the high-flux HD group (P = 0.02). Mean UF rate was significantly higher in the predominant HDF group at all time point up to 60 mo (P varying from 0.003 to <0.001; Figure 3).
Figure 2.
BP in patients who were treated with predominantly high-flux HD and HDF at different time points. There was no significant difference in BP between groups at any time point except that predialysis systolic BP was significantly higher in the high-flux HD group at 36 mo (P = 0.02).
Figure 3.
Ultrafiltration rate in patients who were treated with predominantly high-flux HD and HDF at different time points. UF rate was significantly higher in the group that was treated with predominantly HDF at each time point up to 60 mo.
Anemia Parameters
Mean serum hemoglobin levels were higher in patients who were treated predominantly with HDF at 24 and 36 mo (Table 3). There were no significant differences at other time points. Erythropoietin dosage was higher in the predominant HDF group at 3 and 36 mo (P < 0.05 in both cases), but erythropoietin resistance index did not differ significantly between groups at any time point.
Table 3.
Anemia and erythropoietin treatment parameters in patients who were treated by predominantly HDF and high-flux HD
| Parameter | Group | 3 Mo | 6 Mo | 12 Mo | 24 Mo | 36 Mo | 48 Mo | 60 Mo |
|---|---|---|---|---|---|---|---|---|
| Hemoglobin (g/dl) | HD mean | 9.93 | 10.69 | 10.90 | 11.00 | 11.12 | 11.00 | 11.24 |
| HDF mean | 9.89 | 10.77 | 11.17 | 11.46 | 11.63 | 11.33 | 11.41 | |
| df | 770 | 712 | 609 | 461 | 335 | 240 | 162 | |
| P | 0.794 | 0.704 | 0.079 | 0.003 | 0.004 | 0.089 | 0.445 | |
| Erythropoietin dosage (U/wk) | HD mean | 7660 | 8061 | 8762 | 8818 | 9090 | 10109 | 10053 |
| HDF mean | 8810 | 8691 | 9397 | 9963 | 10553 | 11523 | 12184 | |
| df | 646 | 632 | 573 | 416 | 304 | 221 | 142 | |
| P | 0.041 | 0.283 | 0.252 | 0.065 | 0.045 | 0.100 | 0.063 | |
| Erythropoietin resistance index (U/wk per kg per g/dl) | HD mean | 11.13 | 11.15 | 12.15 | 12.19 | 12.63 | 14.68 | 14.80 |
| HDF mean | 10.83 | 10.82 | 11.24 | 11.93 | 12.68 | 13.70 | 15.26 | |
| df | 580 | 579 | 522 | 389 | 291 | 208 | 133 | |
| P | 0.760 | 0.755 | 0.334 | 0.785 | 0.967 | 0.443 | 0.805 |
Nutrition/Inflammation Parameters
Mean serum albumin (Table 4) was not significantly different between patients who were treated predominantly with HDF or high-flux HD at any time point. BMI was higher in patients who were treated with predominantly HDF at all time points from 6 to 60 mo (Table 4). Although higher at baseline, serum CRP levels were lower in patients who were treated predominantly with HDF at all subsequent time points, significantly so at 12 and 24 mo (P = 0.031 and 0.015, respectively).
Table 4.
Nutrition, inflammation, and bone function parameters in patients who were treated by predominantly HDF and high-flux HD
| Parameter | Group | 3 Mo | 6 Mo | 12 Mo | 24 Mo | 36 Mo | 48 Mo | 60 Mo |
|---|---|---|---|---|---|---|---|---|
| Albumin (g/L) | HD mean | 34.2 | 35.3 | 35.6 | 35.7 | 35.2 | 34.7 | 34.0 |
| HDF mean | 34.5 | 35.9 | 36.1 | 35.9 | 34.5 | 34.8 | 34.5 | |
| df | 823 | 744 | 631 | 445 | 324 | 232 | 149 | |
| P | 0.612 | 0.276 | 0.230 | 0.681 | 0.223 | 0.877 | 0.570 | |
| BMI | HD mean | 25.39 | 25.30 | 25.35 | 24.87 | 25.20 | 24.84 | 24.28 |
| HDF mean | 26.39 | 26.93 | 26.41 | 26.61 | 26.76 | 27.25 | 26.63 | |
| df | 828 | 752 | 636 | 450 | 327 | 234 | 149 | |
| P | 0.123 | 0.004 | 0.034 | 0.001 | 0.015 | 0.001 | 0.007 | |
| CRP (mg/L) | HD median (IQR) | 10.0 (19.0) | 9.0 (16.1) | 10.0 (16.2) | 9.4 (17.8) | 10.0 (18.6) | 9.9 (27.0) | 14.0 (21.0) |
| HDF median (IQR) | 11.0 (25.5) | 7.0 (9.0) | 7.0 (12.5) | 7.0 (12.0) | 7.2 (10.5) | 7.7 (11.7) | 9.0 (22.1) | |
| Total (n) with data | 443 | 371 | 350 | 283 | 215 | 149 | 110 | |
| P | 0.136 | 0.054 | 0.031 | 0.015 | 0.083 | 0.078 | 0.862 | |
| Corrected calcium (mmol/L) | HD mean | 2.32 | 2.34 | 2.35 | 2.36 | 2.38 | 2.39 | 2.39 |
| HDF mean | 2.31 | 2.33 | 2.36 | 2.35 | 2.39 | 2.39 | 2.38 | |
| df | 754 | 677 | 572 | 399 | 294 | 214 | 135 | |
| P | 0.512 | 0.551 | 0.466 | 0.661 | 0.807 | 0.957 | 0.685 | |
| Phosphate (mmol/L) | HD mean | 1.68 | 1.79 | 1.72 | 1.78 | 1.81 | 1.71 | 1.64 |
| HDF mean | 1.75 | 1.80 | 1.82 | 1.92 | 1.75 | 1.86 | 1.79 | |
| df | 747 | 665 | 565 | 393 | 289 | 211 | 131 | |
| P | 0.273 | 0.873 | 0.060 | 0.019 | 0.370 | 0.039 | 0.121 | |
| PTH (pmol/L) | HD median (IQR) | 30.3 (36.0) | 28.0 (36.1) | 25.4 (38.0) | 27.2 (38.8) | 26.0 (44.0) | 30.0 (42.7) | 37.0 (57.0) |
| HDF median (IQR) | 23.0 (38.0) | 26.0 (34.7) | 25.1 (38.6) | 32.3 (39.0) | 35.2 (49.4) | 37.5 (47.4) | 45.0 (52.3) | |
| Total (n) with data | 443 | 371 | 350 | 283 | 215 | 149 | 110 | |
| P | 0.489 | 0.450 | 0.488 | 0.045 | 0.141 | 0.289 | 0.197 |
Bone Function Parameters
Corrected serum calcium, phosphate, and PTH were similar in the predominant HDF and high-flux HD groups at all time points (Table 4) except that at 24 mo, phosphate and PTH levels were higher in patients who were on HDF (P = 0.019 and 0.045, respectively).
Intradialytic Complications
Hypotensive episodes were more frequent in patients who were treated predominantly with high-flux HD compared with those who were treated predominantly with HDF (median 0.05/session [IQR 0.08] versus 0.03/session [IQR 0.05]; P < 0.001). Intradialytic saline boluses were administered more frequently in patients who were on high-flux HD (median 0.03/session [IQR 0.04] versus 0.02 [IQR 0.03]; P < 0.001). There was no difference in frequency of reported intradialytic cramps (median 0.01/session [IQR 0.03] versus 0.01 [IQR 0.03]; P = 0.89).
Survival Analysis
In a Kaplan-Meier survival analysis patients in the predominant HDF group had significantly better survival compared with those in the predominant high-flux HD group (P < 0.001). Median survival was 3.4 yr (95% confidence interval [CI] 3.0 to 3.8) for those who were on predominantly high-flux HD and 7.2 yr (95% CI 6.1 to 8.3) for those who were on predominantly HDF.
Survival model 1 compared survival in the predominant high-flux HD group with that in the predominant HDF group, after correcting for covariates (Table 5, Figure 4). The hazard ratio (HR) for death was significantly lower in those who were treated with predominantly HDF compared with those who were on high-flux HD (HR 0.45; 95% CI 0.35 to 0.59; P < 0.001). Age, malignancy, and diabetes status were significant independent predictors of increased risk for death with HRs shown in Table 5.
Table 5.
Survival model 1: Comparison of survival between patients who were treated with HDF and high-flux HD
| Parameter | HR | P | 95% CI for HR |
|---|---|---|---|
| Predominant HDF treatment | 0.454 | 0.000 | 0.350 to 0.590 |
| Age | 1.030 | 0.000 | 1.021 to 1.039 |
| Gender | 1.060 | 0.570 | 0.868 to 1.295 |
| BMI | 0.983 | 0.101 | 0.964 to 1.003 |
| Ischemic heart disease | 1.144 | 0.225 | 0.921 to 1.422 |
| Peripheral vascular disease | 1.103 | 0.411 | 0.874 to 1.392 |
| Malignancy | 1.848 | 0.000 | 1.452 to 2.353 |
| Diabetes | 1.339 | 0.006 | 1.088 to 1.648 |
Treatment with predominantly HDF (>50% sessions) was a significant independent predictor of reduced risk for death compared with treatment with predominantly high-flux HD in this multivariate analysis. Age, malignancy, and diabetes status were significant independent predictors of increased risk for death.
Figure 4.
Cox proportional hazards model demonstrating survival differences between patients in whom the predominant treatment modality was HDF and high-flux HD. Treatment with HDF was an independent risk factor predicting a lower hazard for death (HR 0.46; P < 0.001). Other factors included in the model are shown in Table 5.
Survival model 2 (Table 6) was designed to study the interaction between proportion of dialysis career time spent on HDF and survival. For each patient, the time-dependent covariate HDFTime varied in a binary way between 0 and 1 with time, depending on whether treatment type was HD (0) or HDF (1). Age, BMI, and comorbidities were entered into the model in addition to KRU and Kt/VDialysis, the last two being included as time-dependent covariates. Proportion of time spent on HDF (HDFTime) was a significant independent predictor of reduced risk for death (HR 0.66; 95% CI 0.47 to 0.92; P = 0.014). The interpretation of this is that a patient who was treated with HDF for 100% of their dialysis career would have a hazard of death of 0.66 compared with a patient who was treated for 100% of the time with high-flux HD. In this model, KRU was a significant independent predictor of survival after taking into account its time-dependent variation with a unitary 1-ml/min increase in KRU conferring a reduced hazard for death of 0.80 (95% CI 0.75 to 0.86). Kt/V from dialysis as a time-dependent covariate was also a significant independent predictor for reduced risk for death (HR 0.58; 95% CI 0.41 to 0.83). A unitary increase in BMI of 1 was associated with a reduced hazard for death of 0.98 (95% CI 0.960 to 0.999). Malignancy and diabetes status were associated with a significantly increased hazard for death.
Table 6.
Survival model 2: Effect of the proportion of dialysis career spent on HDF and high-flux HD on survival
| Parameter | HR | P | 95 CI for HR |
|---|---|---|---|
| Age | 1.035 | 0.000 | 1.026 to 1.044 |
| BMI | 0.979 | 0.040 | 0.960 to 0.999 |
| HDFTime | 0.655 | 0.014 | 0.468 to 0.918 |
| KRUTime | 0.803 | 0.000 | 0.747 to 0.862 |
| Kt/V DialysisTime | 0.582 | 0.003 | 0.408 to 0.831 |
| Diabetes | 1.395 | 0.002 | 1.134 to 1.716 |
| Ischemic heart disease | 1.076 | 0.508 | 0.867 to 1.335 |
| Peripheral vascular disease | 1.257 | 0.051 | 0.999 to 1.580 |
| Malignancy | 1.992 | 0.000 | 1.564 to 2.538 |
Proportion of time spent on HDF was included in model 2 as a time-dependent covariate (HDFTime) to take into account the bias effect in model 1 of conversion of longer surviving patients to HDF. Increasing proportion of time spent on HDF was a significant predictor for survival with a patient spending 100% of time on HDF having an HR for death of 0.66 (95% CI 0.47 to 0.92).
Discussion
Online HDF may confer potential advantages over standard and high-flux HD. Our data suggest an association of predominant HDF treatment and improved mortality, compared with high-flux HD. In addition, the proportion of dialysis career time spent being treated with HDF independently predicted survival, despite correction for confounding variables including age, BMI, comorbidity, residual renal function, and dialysis dosage (Kt/VDialysis).
No previous large studies have compared outcomes between patients who were treated with online HDF and high-flux HD. A small, single-center, crossover, randomized trial by Schiffl (15) did not find differences in mortality between online HDF and high-flux HD; however, that study was too small (n = 76) to demonstrate any survival differences and was also confounded by significantly higher Kt/V urea in patients who were on HDF. Canaud et al. (12) demonstrated a 35% reduction in mortality associated with high convection volume HDF over low-flux (standard) HD in a retrospective analysis of Dialysis Outcomes and Practice Patterns Study (DOPPS) data. There was also a benefit of high convection volume HDF over a combined HD group of high-flux (29%) and low-flux (71%) HD. It is not clear whether there was a benefit of HDF over high-flux HD. The numbers treated with high convection volume HDF were relatively small at <5% of the study population. The RISCAVID study (14) prospectively observed outcomes in patients who were on HD, sterile bag HDF, and online HDF. A survival benefit was associated with HDF techniques over standard HD; however, only 4.9% of the HD population (21 patients) used high-flux membranes. Jirka et al. (13) also reported similar results from data collected through the EuCliD network with a mortality reduction of 35.3% compared with an HD group. It was not clear from that report what proportion of the HD group, if any, used high-flux membranes. These studies suggest a mortality benefit associated with HDF over low-flux HD, but data comparing HDF and high-flux HD are lacking. This distinction is important, because there is now some evidence from prospective studies that high-flux membranes confer survival benefit. The Membrane Permeability Outcomes study (8) suggests that, at least for patients with hypoalbuminemia and those with diabetes, high-flux HD is superior to standard HD. Subgroup analysis of the HEMO study had previously suggested a survival benefit of high-flux HD in the those who had undergone dialysis for >3.7 yr (20,21).
In considering the potential benefits of one treatment modality over another, the effect of selection bias needs to be examined critically. Patients who were treated predominantly with HDF tended to be younger and have a higher body weight and were less likely to have malignancy. Conversely, they were more likely to have low levels of residual renal function compared with those on high-flux HD. Further bias may exist when comparing treatment modalities because of the inherent tendency for patients to be crossed over to the treatment perceived to be superior by clinicians. Various statistical methods may be used to try to reduce the effect of this bias (16), but it is impossible to eliminate. We used a method whereby a time-dependent covariate was created for each patient, varying between 0 and 1 at different time points depending on how much time had been spent on HDF. This was possible only because the treatment modality of every session had been prospectively recorded. This technique usually cannot be used in large registry data sets because of lack of data on individual dialysis sessions. We demonstrated in survival model 2 that the mortality benefit associated with HDF persists after taking into account not only factors such as age, BMI, comorbidity, residual renal function, and dialyzer clearance but also the crossover of patients from one modality to another, although the magnitude of the benefit was reduced compared with model 1. Our second survival model demonstrated a reverse association of BMI with mortality in keeping with previous studies (22,23).
Identifying the patient group that stands to benefit most from the addition of a convective component to diffusive clearance is important for both treatment and the design of randomized, controlled trials (RCTs). Prevalent patients (>3.7 yr vintage) in the HEMO study seemed to benefit from the use of high-flux membranes, and there was an association between β2 microglobulin levels and mortality (20,21). Presumably, these patients had very low levels of residual renal function. We previously demonstrated the critical dependence of β2 microglobulin levels on residual renal function (24). The MPO study (8), which randomly assigned incident patients, failed to find outcome differences between those who were treated with high- and low-flux membranes, except in the very-high-risk subgroups. The presence of residual renal function may have overridden the benefit of the added convective clearance in lower risk groups. We hypothesize that HDF may be most beneficial in patients with low levels of residual renal function. Enhanced middle molecule clearance has other benefits. Previous studies demonstrated improved β2-microglobulin clearance by HDF (11,25,26), and one retrospective study suggested that convective dialysis treatments reduce the incidence of carpal tunnel syndrome (27).
There were fewer hypotensive episodes in the HDF group, despite their higher UF requirements. Previous studies demonstrated similar benefits (28,29), although such findings are not universal (30,31). It seems that a cooling effect of the infusate may play a major role (31,32). There has been some concern that using the same dialysis fluid sodium level in online HDF and HD may result in sodium loading in the HDF group. Our data suggest that this may not be a major problem because serum sodium levels and predialysis BP were similar in both groups. There is no consensus in the literature on BP differences in patients who were treated with HD and those who were on convective treatments (29,30,33,34).
We found no advantage of HDF over high-flux HD in anemia management. Although hemoglobin levels were higher in the predominant HDF group at several time points, there was no difference between the groups with respect to erythropoietin resistance index at any time point. Previously, a small (n = 70) crossover design study (35) found a higher hematocrit at a lower erythropoietin dosage while patients were treated with HDF, although the comparison group was standard, not high-flux HD. Our results do seem to conflict with those of Maduell et al. (25), who demonstrated improved hemoglobin levels and reduced erythropoietin requirements after conversion from low-volume HDF to online high-volume HDF. This was attributed to the higher convection volumes, although there were also significant differences in urea clearance. We have not demonstrated a nutritional benefit of HDF over high-flux HD, as suggested by others (29), although our analysis was confined to a comparison of serum albumin levels and BMI; however, our data do suggest a beneficial effect of HDF on inflammatory status, CRP being generally lower in the HDF group, significantly at two time points.
Phosphate levels were higher in patients who were on HDF compared with high-flux HD at all time points, although this reached statistical significance only at 24 mo. This is likely to be a result of lower levels of residual renal function in the HDF group. There were no substantial differences in corrected calcium and PTH between patients who were on high-flux HD and patients who were on HDF. Changes in prescription of calcium-based phosphate binder make it difficult to assess from our data whether patients with HDF may benefit from a higher dialysate calcium (e.g., 1.5 instead of 1.25 mmol/L) as suggested previously by Malberti and Ravani (36).
Our study is mainly limited by its retrospective design and the potential for selection bias in treatment modality. We therefore advise caution in interpreting the results of this study, because the association of HDF with improved survival cannot be extrapolated to conclude a cause–effect relationship. In addition, our method used to adjust for residual renal and Kt/VDialysis in the survival analysis is not commonly used.
A number of RCTs to compare high-flux HD and online HDF are reported as being under way. The Dutch Convective Transport Study (CONTRAST) will randomly assign 800 patients to either low-flux HD or HDF (37). An Italian multicenter study will randomly assign 250 patients to low-flux HD or convective therapy, equally split between HDF and hemofiltration (38). A large French multicenter RCT will compare online HDF with high-flux HD in 600 dialysis patients aged ≥65 yr (39). We eagerly await the results of these trials. Our data suggest that to demonstrate a survival benefit of HDF over high-flux HD will require large numbers of patients unless higher risk subgroups are selected, such as those with low levels of residual renal function.
In the absence of randomized, prospective studies, we believe that it is reasonable to conclude that online HDF is a logical and safe therapy and at least as effective as high-flux HD in maintaining the life and well-being of dialysis patients. We have provided unique comparative data between high-flux HD and online HDF. There are suggestions of survival benefit for HDF, but we recommend caution with interpretation of these retrospective data.
Disclosures
None.
Acknowledgments
We are grateful for the statistics advice provided by Sam Norton (University of Hertfordshire, Hertfordshire, UK) and for the acceptance of an abstract based on this article for the British Renal Society Conference 2009.
Footnotes
Published online ahead of print. Publication date available at www.cjasn.org.
References
- 1.Henderson LW, Colton CK, Ford CA: Kinetics of hemodiafiltration: II. Clinical characterization of a new blood cleansing modality. J Lab Clin Med 85: 372–391, 1975 [PubMed] [Google Scholar]
- 2.Henderson LW, Beans E: Successful production of sterile pyrogen-free electrolyte solution by ultrafiltration. Kidney Int 14: 522–525, 1978 [DOI] [PubMed] [Google Scholar]
- 3.Lornoy W, Becaus I, Billiouw JM, Sierens L, Van Malderen P, D'Haenens P: On-line haemodiafiltration: Remarkable removal of beta2-microglobulin—Long-term clinical observations. Nephrol Dial Transplant 15[Suppl 1]: 49–54, 2000 [DOI] [PubMed] [Google Scholar]
- 4.Ward RA, Schmidt B, Hullin J, Hillebrand GF, Samtleben W: A comparison of on-line hemodiafiltration and high-flux hemodialysis: A prospective clinical study. J Am Soc Nephrol 11: 2344–2350, 2000 [DOI] [PubMed] [Google Scholar]
- 5.Tattersall J: Clearance of beta-2-microglobulin and middle molecules in haemodiafiltration. Contrib Nephrol 158: 201–209, 2007 [DOI] [PubMed] [Google Scholar]
- 6.Kuchle C, Fricke H, Held E, Schiffl H: High-flux hemodialysis postpones clinical manifestation of dialysis-related amyloidosis. Am J Nephrol 16: 484–488, 1996 [DOI] [PubMed] [Google Scholar]
- 7.van Ypersele de Strihou C, Jadoul M, Malghem J, Maldague B, Jamart J: Effect of dialysis membrane and patient's age on signs of dialysis-related amyloidosis. The Working Party on Dialysis Amyloidosis. Kidney Int 39: 1012–1019, 1991 [DOI] [PubMed] [Google Scholar]
- 8.Locatelli F, Martin-Malo A, Hannedouche T, Loureiro A, Papadimitriou M, Wizemann V, Jacobson SH, Czekalski S, Ronco C, Vanholder R: Effect of membrane permeability on survival of hemodialysis patients. J Am Soc Nephrol 20: 645–654, 2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Pizzarelli F, Cerrai T, Dattolo P, Tetta C, Maggiore Q: Convective treatments with on-line production of replacement fluid: a clinical experience lasting 6 years. Nephrol Dial Transplant 13: 363–369, 1998 [DOI] [PubMed] [Google Scholar]
- 10.Canaud B, Bosc JY, Leray-Moragues H, Stec F, Argiles A, Leblanc M, Mion C: On-line haemodiafiltration: Safety and efficacy in long-term clinical practice. Nephrol Dial Transplant 15[Suppl 1]: 60–67, 2000 [DOI] [PubMed] [Google Scholar]
- 11.Wizemann V, Lotz C, Techert F, Uthoff S: On-line haemodiafiltration versus low-flux haemodialysis: A prospective randomized study. Nephrol Dial Transplant 15[Suppl 1]: 43–48, 2000 [DOI] [PubMed] [Google Scholar]
- 12.Canaud B, Bragg-Gresham JL, Marshall MR, Desmeules S, Gillespie BW, Depner T, Klassen P, Port FK: Mortality risk for patients receiving hemodiafiltration versus hemodialysis: European results from the DOPPS. Kidney Int 69: 2087–2093, 2006 [DOI] [PubMed] [Google Scholar]
- 13.Jirka T, Cesare S, Di Benedetto A, Perera Chang M, Ponce P, Richards N, Tetta C, Vaslaky L: Mortality risk for patients receiving hemodiafiltration versus hemodialysis. Kidney Int 70: 1524, author reply 1524–1525, 2006 [DOI] [PubMed] [Google Scholar]
- 14.Panichi V, Rizza GM, Paoletti S, Bigazzi R, Aloisi M, Barsotti G, Rindi P, Donati G, Antonelli A, Panicucci E, Tripepi G, Tetta C, Palla R: Chronic inflammation and mortality in haemodialysis: Effect of different renal replacement therapies: Results from the RISCAVID study. Nephrol Dial Transplant 23: 2337–2343, 2008 [DOI] [PubMed] [Google Scholar]
- 15.Schiffl H: Prospective randomized cross-over long-term comparison of online haemodiafiltration and ultrapure high-flux haemodialysis. Eur J Med Res 12: 26–33, 2007 [PubMed] [Google Scholar]
- 16.Fisher LD, Lin DY: Time-dependent covariates in the Cox proportional-hazards regression model. Annu Rev Public Health 20: 145–157, 1999 [DOI] [PubMed] [Google Scholar]
- 17.Gotch F: Kinetic modelling in hemodialysis. In: Clinical Dialysis, 3rd ed., edited by Nissenson A, Fine R, Gentile D.Norwalk, CT, Appleton & Lange, 1995, pp 156–188 [Google Scholar]
- 18.Depner TA: Hemodialysis adequacy: Basic essentials and practical points for the nephrologist in training. Hemodial Int 9: 241–254, 2005 [DOI] [PubMed] [Google Scholar]
- 19.Galliford JW, Malasana R, Farrington K: Switching from subcutaneous to intravenous erythropoietin alpha in haemodialysis patients requires a major dose increase. Nephrol Dial Transplant 20: 1956–1962, 2005 [DOI] [PubMed] [Google Scholar]
- 20.Eknoyan G, Beck GJ, Cheung AK, Daugirdas JT, Greene T, Kusek JW, Allon M, Bailey J, Delmez JA, Depner TA, Dwyer JT, Levey AS, Levin NW, Milford E, Ornt DB, Rocco MV, Schulman G, Schwab SJ, Teehan BP, Toto R: Effect of dialysis dose and membrane flux in maintenance hemodialysis. N Engl J Med 347: 2010–2019, 2002 [DOI] [PubMed] [Google Scholar]
- 21.Cheung AK, Levin NW, Greene T, Agodoa L, Bailey J, Beck G, Clark W, Levey AS, Leypoldt JK, Ornt DB, Rocco MV, Schulman G, Schwab S, Teehan B, Eknoyan G: Effects of high-flux hemodialysis on clinical outcomes: Results of the HEMO study. J Am Soc Nephrol 14: 3251–3263, 2003 [DOI] [PubMed] [Google Scholar]
- 22.Kalantar-Zadeh K, Kopple JD, Kilpatrick RD, McAllister CJ, Shinaberger CS, Gjertson DW, Greenland S: Association of morbid obesity and weight change over time with cardiovascular survival in hemodialysis population. Am J Kidney Dis 46: 489–500, 2005 [DOI] [PubMed] [Google Scholar]
- 23.Beddhu S, Pappas LM, Ramkumar N, Samore M: Effects of body size and body composition on survival in hemodialysis patients. J Am Soc Nephrol 14: 2366–2372, 2003 [DOI] [PubMed] [Google Scholar]
- 24.Fry AC, Singh DK, Chandna SM, Farrington K: Relative importance of residual renal function and convection in determining beta-2-microglobulin levels in high-flux haemodialysis and on-line haemodiafiltration. Blood Purif 25: 295–302, 2007 [DOI] [PubMed] [Google Scholar]
- 25.Maduell F, del Pozo C, Garcia H, Sanchez L, Hdez-Jaras J, Albero MD, Calvo C, Torregrosa I, Navarro V: Change from conventional haemodiafiltration to on-line haemodiafiltration. Nephrol Dial Transplant 14: 1202–1207, 1999 [DOI] [PubMed] [Google Scholar]
- 26.Shinzato T, Kobayakawa H, Maeda K: Comparison of various treatment modes in terms of beta 2-microglobulin removal: Hemodialysis, hemofiltration, and push/pull HDF. Artif Organs 13: 66–70, 1989 [DOI] [PubMed] [Google Scholar]
- 27.Locatelli F, Marcelli D, Conte F, Limido A, Malberti F, Spotti D: Comparison of mortality in ESRD patients on convective and diffusive extracorporeal treatments. The Registro Lombardo Dialisi E Trapianto. Kidney Int 55: 286–293, 1999 [DOI] [PubMed] [Google Scholar]
- 28.Munoz R, Gallardo I, Valladares E, Saracho R, Martinez I, Ocharan J, Montenegro J: Online hemodiafiltration: 4 years of clinical experience. Hemodial Int 10[Suppl 1]: S28–S32, 2006 [DOI] [PubMed] [Google Scholar]
- 29.Velasco N: Convection with conviction: online hemodiafiltration for all—Single center clinical observations. Hemodial Int 10[Suppl 1]: S67–S71, 2006 [DOI] [PubMed] [Google Scholar]
- 30.Kerr PB, Argiles A, Flavier JL, Canaud B, Mion CM: Comparison of hemodialysis and hemodiafiltration: A long-term longitudinal study. Kidney Int 41: 1035–1040, 1992 [DOI] [PubMed] [Google Scholar]
- 31.Karamperis N, Sloth E, Jensen JD: Predilution hemodiafiltration displays no hemodynamic advantage over low-flux hemodialysis under matched conditions. Kidney Int 67: 1601–1608, 2005 [DOI] [PubMed] [Google Scholar]
- 32.Donauer J, Schweiger C, Rumberger B, Krumme B, Bohler J: Reduction of hypotensive side effects during online-haemodiafiltration and low temperature haemodialysis. Nephrol Dial Transplant 18: 1616–1622, 2003 [DOI] [PubMed] [Google Scholar]
- 33.Altieri P, Sorba G, Bolasco P, Asproni E, Ledebo I, Cossu M, Ferrara R, Ganadu M, Cadinu F, Serra G, Cabiddu G, Sau G, Casu D, Passaghe M, Bolasco F, Pistis R, Ghisu T: Predilution haemofiltration: The Second Sardinian Multicentre Study—Comparisons between haemofiltration and haemodialysis during identical Kt/V and session times in a long-term cross-over study. Nephrol Dial Transplant 16: 1207–1213, 2001 [DOI] [PubMed] [Google Scholar]
- 34.van der Sande FM, Kooman JP, Konings CJ, Leunissen KM: Thermal effects and blood pressure response during postdilution hemodiafiltration and hemodialysis: The effect of amount of replacement fluid and dialysate temperature. J Am Soc Nephrol 12: 1916–1920, 2001 [DOI] [PubMed] [Google Scholar]
- 35.Vaslaki L, Major L, Berta K, Karatson A, Misz M, Pethoe F, Ladanyi E, Fodor B, Stein G, Pischetsrieder M, Zima T, Wojke R, Gauly A, Passlick-Deetjen J: On-line haemodiafiltration versus haemodialysis: Stable haematocrit with less erythropoietin and improvement of other relevant blood parameters. Blood Purif 24: 163–173, 2006 [DOI] [PubMed] [Google Scholar]
- 36.Malberti F, Ravani P: The choice of the dialysate calcium concentration in the management of patients on haemodialysis and haemodiafiltration. Nephrol Dial Transplant 18[Suppl 7]: vii37–vii40, discussion vii57, 2003 [DOI] [PubMed] [Google Scholar]
- 37.Penne EL, Blankestijn PJ, Bots ML, van den Dorpel MA, Grooteman MP, Nube MJ, ter Wee PM: Resolving controversies regarding hemodiafiltration versus hemodialysis: The Dutch Convective Transport Study. Semin Dial 18: 47–51, 2005 [DOI] [PubMed] [Google Scholar]
- 38.Bolasco P, Altieri P, Andrulli S, Basile C, Di Filippo S, Feriani M, Pedrini L, Santoro A, Zoccali C, Sau G, Locatelli F: Convection versus diffusion in dialysis: An Italian prospective multicentre study. Nephrol Dial Transplant 18[Suppl 7]: vii50–vii54, discussion vii59–vii62, 2003 [DOI] [PubMed] [Google Scholar]
- 39.Canaud B, Morena M, Leray-Moragues H, Chalabi L, Cristol JP: Overview of clinical studies in hemodiafiltration: What do we need now? Hemodial Int 10[Suppl 1]: S5–S12, 2006 [DOI] [PubMed] [Google Scholar]