Treatment Policy rather than Patient Characteristics Determines Convection Volume in Online Post-Dilution Hemodiafiltration

Both mortality and morbidity remain unacceptably high in hemodialysis (HD) patients [1]. As retention of toxic middle molecular weight molecules (5-40 kDa) has been implicated in the pathogenesis of the uremic syndrome [2,3,4,5], removal of these substances by convective therapies may improve prognosis [6]. However, neither the HEMO [7] nor the MPO study [8] demonstrated a clear advantage of high-flux over low-flux membranes.

Recently, hemodiafiltration (HDF) has gained renewed interest [9]. In HDF, diffusion, which is the main removal mechanism in low-flux HD, is combined with convection. Whereas the estimated amount of convective transport during high-flux HD is <10 liters/session [10], in online post-dilution HDF 25 liters or more can be achieved [11].

While several observational studies suggested a survival benefit of HDF [12,13,14,15], two recent randomized controlled trials with comparable design, namely the CONvective TRAnsport STudy (CONTRAST) [16] and the Turkish OL-HDF Study (THDFS) [17], did not find an overall difference between post-dilution HDF and HD. Interestingly, post hoc analyses of both studies revealed that the lowest mortality risks were observed in patients with the highest convection volumes per session (mean >22.0 liters in CONTRAST and >19.7 liters in THDFS). Lately, a third large randomized controlled trial (ESHOL) [18] showed that the overall mortality risk in HDF patients was 30% lower than in HD patients. In this study, mean convection volume was 23.7 liters. Sub-analysis of ESHOL confirmed the relation between convection volume and mortality risk. Altogether, these findings support the concept of a dose-response relationship between convection volume and survival [12].

Therefore, from a clinical point of view, it appears crucial to define patient- and treatment-related factors that restrict or facilitate the magnitude of convection volume, and their relative contribution. In a preliminary analysis, we found that blood flow and treatment time are important determinants [19]. As it is currently unclear whether the presence of an arteriovenous (AV) fistula [20] and/or use of particular dialyzers are required for high-volume HDF, in the present analysis special attention was paid to these issues. To minimize the probability of dose-targeting bias, we assessed whether subjects achieving high-volume HDF are characterized by a favorable baseline profile.

The present study is a cross-sectional analysis of CONTRAST (NCT00205556) in HDF patients who completed 6 months of follow-up. In CONTRAST, a total of 714 chronic HD patients from 29 centers (26 Dutch, 2 Canadian and 1 Norwegian) was randomly assigned to online post-dilution HDF (n = 358) or low-flux HD (n = 356) and compared with respect to all-cause mortality and cardiovascular events. As 44 patients did not have a recorded value of post-dilution convection volume at 6 months, 314 patients were eligible for the present analysis. A period of 6 months was chosen to ensure enough time to adapt to HDF on the one hand and to avoid potential dropouts, and hence censoring from analysis, on the other.

Details of CONTRAST are described elsewhere [16,21]. The study was conducted in accordance with the Declaration of Helsinki and approved by the medical ethics review boards of all participating centers. Written informed consent was obtained from all patients prior to enrolment.

HDF was performed in the online post-dilution mode. Patients who were temporarily treated with pre-dilution HDF were excluded from analysis (n = 4). A target convection flow rate of 6 liters/h was based on a filtration fraction (FF) between 25 and 33% of an extracorporeal blood flow rate between 300 and 400 ml/min. As guidelines were absent when CONTRAST was started, these targets were mainly based on the operating instructions of the manufacturers.

The following dialysis machines were used: 4008S and 5008 with ONLINEplus™ (Fresenius Medical Care, Bad Homburg, Germany), AK 100/200™ ULTRA S (Gambro AB, Lund, Sweden), DBB-05™ (Nikkiso Co. Ltd, Tokyo, Japan) and Integra™ (Hospal-Gambro AB, Lund, Sweden). The following synthetic high-flux dialyzers were used (online suppl. table 1; for all online suppl. material, see www.karger.com/doi/10.1159/000362108): FX80: 25%, FX100: 12% and Optiflux F200NR: 11% (Fresenius); Polyflux 170H: 21% and Polyflux 210H: 29% (Gambro), and others: 2%. Ultrapure dialysis fluid, defined as <0.1 colony-forming units and <0.025 endotoxin units per milliliter, was used for all treatments.

Treatment times were fixed at baseline and could only be increased if spKt/V_urea was <1.2. Routine patient care was performed according to the guidelines of the Quality of Care Committee of the Dutch Federation of Nephrology.

Demographics, medical history and medication were recorded at baseline, while various clinical, treatment and laboratory parameters were collected both at baseline and every 3 months afterwards. Body mass index (BMI) was calculated as was body surface area (BSA) using the formula of Gehan and George [22]. Systolic and diastolic blood pressures were registered as the average of pre-dialysis values on 3 consecutive dialysis days. Blood samples for routine laboratory measurements were taken before the start of a dialysis session and analyzed in the local hospitals by standard techniques. Serum albumin measured with bromcresol purple method was converted to bromcresol green with the formula: bromcresol green = bromcresol purple + 5.5 (g/l) [23].

Convection volume, collected at 6 months, is defined as the sum of the intradialytic weight loss and the amount of substitution fluid, and reported as the mean of three consecutive sessions. Convective flow rate represents the convection volume in milliliters/minute. FF was calculated by dividing convective flow rate by blood flow rate and is reported as a percentage of blood flow.

Data are reported as proportions, means with standard deviations or standard errors, or medians with interquartile ranges when appropriate. Differences between groups were examined by one-way ANOVA, Mann-Whitney test or Pearson χ² test. To compare differences between variables at baseline and 6 months, paired t tests were used.

To study the independent relationship between each variable (obtained at baseline for demography, medical history and type of vascular access, and at 6 months for clinical, laboratory and treatment parameters) and convection volume, both uni- and multivariable regression analyses were used. First, all variables that showed a univariable relation with convection volume using a cut-off value of p < 0.20 were entered into the multivariable model. Next, a backward regression analysis with a cut-off p value of 0.20 was performed adding the variables hematocrit, blood flow rate and treatment time up front. Sensitivity analyses were performed in two ways: first by re-running the model with exclusion of patients from one center at a time and second, by re-running the analysis using data on variables and convection volume at months 3 and 12. In all models the magnitude, direction and significance of the relation with convection volume remained stable.

To assess whether patients who achieved the highest convection volumes represented a favorable group at baseline, patients were stratified into tertiles of convection volume. The three groups were compared in terms of patient- and treatment-related characteristics. In addition, the Cardiac Risk Score (CRS) was applied. In this score system, the hazard ratios of previous cardiac disease, BMI, dialysis vintage and phosphate are summed [24].

Results were considered statistically significant when p < 0.05 (two-sided). Adjustment for multiple comparisons was made using the Holm-Bonferroni method. All calculations were made by use of a standard statistical package (SPSS for Windows Version 18.0.1; SPSS, Inc., Chicago, Ill., USA).

Data at baseline are presented in table 1.

Data are shown in table 2. Mean convection volume per patient was 19.7 ± 4.4 liters per treatment (range 6.7-32.6) and mean FF was 26 ± 4% (range 8-37). The pre-defined target convection flow rate of 6 liters/h was reached in 74 patients (24%). Mean convection volume per center ranged from 12.9 ± 0.9 to 25.6 ± 0.6 (±SE) liters per treatment. In 11 of the 29 centers, target convection flow rate was reached in none of the patients. Figure 1a and b shows the convection volumes and FFs per center. In figure 1c and d the variation of blood flow and treatment time within and between centers is depicted. The absence of a box-and-whisker plot in 6 centers in figure 1c and 8 centers in figure 1d suggests that blood flow and treatment time, respectively, were fixed within these facilities.

As depicted in table 3, univariate regression analysis showed that male gender, non-Caucasian race, BSA, dialysis vintage, hematocrit and albumin were related to convection volume. In the multivariable regression model dialysis vintage, BSA (borderline significant, p = 0.07) and albumin remained positively related, while an inverse relation was found with hematocrit. No marked differences were found between patients with an AV fistula, graft or CVC. Figure 2 shows the distribution of convection volume per type of vascular access adjusted for confounders. When the crude relation between convection volume and vascular access type was studied, results were similar: AV fistula 19.7 ± 4.4, graft 19.3 ± 4.4, and CVC 21.9 ± 4.2 liters per treatment; p for difference between groups = 0.10.

As depicted in table 3, uni- and multivariate regression analyses showed that blood flow rate (B = 0.05 liters/ml/min; 95% CI 0.05-0.06) and treatment time (B = 0.08 liters/min; 95% CI 0.07-0.10) were positively related to convection volume (online suppl. fig. 1a, b). Neither membrane surface area (range 1.7-2.2 m²) nor ultrafiltration coefficient (range 56-85 ml/h/mm Hg) of the dialyzers showed a linear relationship with convection volume (online suppl. fig. 2a, c) or FF (online suppl. fig. 2b, d).

Characteristics of the HDF population stratified by tertiles of convection volume are depicted in table 4. Neither age, nor gender, history of cardiovascular disease, diabetes, type of vascular access and CRS differed across tertiles. Of the clinical characteristics at baseline, only BSA was greater in the group of patients with high convection volumes. Medication use (antihypertensives, phosphate binders, lipid-lowering drugs, erythropoietin and anticoagulants) did not differ (data not shown).

Increasing evidence indicates that the magnitude of convection volume is related to the clinical outcome in HDF patients [16,17,18]. The present analysis was performed to assess whether convection volume depends on individual patient characteristics, such as body size and cardiovascular risk profile, rather than center-specific treatment policies, such as fixed treatment times.

Apart from BSA and dialysis vintage, none of the demographic or clinical variables was related to convection volume. In this respect it should be mentioned that the clinical effect of BSA on convection volume is limited: an increase of 10% (18 dm² at a mean of 1.83 m²) would yield only 18 × 0.01 = 0.18 liters extra (table 3).

In line with both observational [25] and prospective studies [17,19], hematocrit was negatively and albumin positively linked to convection volume, albeit to a limited extent. During post-dilution HDF, ultrafiltration favors hemoconcentration, which alters intradialyzer blood rheology. These changes may impede convective transport capacity by increasing in- and post-filter viscosity, leading to fiber clotting and pressure alarms [26]. A high serum albumin may promote plasma refilling during treatment, thus allowing higher ultrafiltration rates.

Most investigations in HDF excluded patients with CVCs or malfunctioning shunts [17,18,27]. In the present study, no relation was found between the magnitude of convection volume and the type of vascular access. In this respect, it should be mentioned that many patients with a high convection volume came from a single facility with a high CVC use, but also a higher mean blood flow rate. Nevertheless, it appears that the presence of an AV fistula, as recently advocated by the EuDial group [20], is not a prerequisite for high-volume HDF. Moreover, the presence of a CVC appears neither a contraindication for HDF nor a drawback for obtaining high convection volumes.

Treatment time and blood flow rate appeared by far the most powerful predictors of the magnitude of convection volume. According to the present model, a 30-min increase in time or a 50 ml/min increase in blood flow would raise the convection volume with approximately 2.5 liters per treatment, provided that these parameters are within the range of our study. As the above-mentioned parameters BSA, hematocrit and albumin were only mildly related to convection volume, prescription of adequate blood flow and treatment time appears of paramount importance for obtaining high-volume HDF. As time was fixed at the beginning of the study, dose-targeting bias seems unlikely [28]. The finding that the within-center variation of both blood flow and treatment time was often very low or even nil supports the idea that center policy rather than individualized patient prescription determined convection volume in this population.

The type of dialyzer could theoretically influence the magnitude of convection volume as well. Yet, within the range of dialyzers used, neither membrane surface area nor ultrafiltration coefficient showed a linear relationship with the achieved convection volumes. At this point, however, it should be mentioned that the heterogeneity of the dialyzers used, their co-aggregation within centers and the non-standardized operating conditions prevent solid conclusions.

FF may also greatly influence the magnitude of the convection volume, as it quantifies the relation between convective flow rate and blood flow rate. However, as FF was most often not a pre-set parameter, it was calculated afterwards and considered not appropriate to analyze its relation with convection volume. Nonetheless, FF varied markedly within and between centers. Although a FF of 25% is considered safe in post-dilution HDF [20], this value was not reached in 106 patients (34%). Theoretically, it is conceivable that individual patients would have tolerated and benefited from an optimized, pre-targeted higher FF, as demonstrated in studies using automatic pressure control [29]. When CONTRAST was designed, however, maximization of convection volume was not a pre-specified goal.

Finally, it is conceivable that needle type and size, mode and dose of anticoagulation, and experience with HDF influence the magnitude of convection volume. However, as these data were not prospectively collected, the present analysis cannot give an answer to the relative importance of these items.

Proportions of traditional cardiovascular risk factors, including diabetes and previous cardiovascular disease, and uremia-related issues, such as vascular access type and uremic markers, were not different between tertiles of convection volume. To further assess whether patients in the highest tertile had a more favorable clinical profile than patients in the other groups, the CRS was applied, which, in fact, was similar [24]. Together with data from the regression analysis, our results suggest that the magnitude of convection volume is largely independent of individual patient characteristics. Hence, confounding by indication seems highly unlikely.

The large sample size, the prospective follow-up and the concise data collection are important strengths of this study. Patients who were temporarily treated with pre-dilution HDF were meticulously traced and excluded from analysis. Albumin assays were re-checked and if measured by bromcresol purple converted to bromcresol green. This study also has limitations. First, the post hoc character of the analyses allows only inferences on relationships, not on causality. Second, vascular access was recorded at baseline and may have changed during follow-up. Third, determinants of convection volume at 6 months may not be representative for a longer period of time. However, as the model was consistent at different time points and a close relation was found between convection volumes at 3 and 12 months (data not shown), the chosen time point appears appropriate. Finally, as the CRS used in our study takes only the cardiac risk profile into account, non-registered clinical conditions, such as malignancies and chronic obstructive lung disease, may differ between groups.

Among the various patient- and treatment-related parameters that may influence convection volume in post-dilution HDF, treatment time and blood flow rate are by far the most powerful determinants. Patient characteristics, such as BSA, hematocrit and albumin, play only a minor role in this respect. As both demographic and clinical variables, as well as prognostic markers, did not differ across tertiles of convection volume, our data do not support the idea that the beneficial effect of high-volume HDF [16] is due to a more favorable clinical profile beforehand. Hence, center policy rather than patient characteristics appears to be most decisive for the amount of convection volume achieved. An AV fistula does not seem to be a prerequisite and a CVC not a drawback for achieving high convection volumes. Although the type of dialyzer was not related to the magnitude of convection volume, its co-aggregation within centers and the non-standardized operating conditions prevent solid conclusions. Whether FF can be recommended as a prescription tool to maximize the convection volume is currently under investigation in the Feasibility Study (NCT01877499).

CONTRAST was financially supported by a grant from the Dutch Kidney Foundation (Nierstichting Nederland, grant C02.2019) and unrestricted grants from Fresenius Medical Care (The Netherlands) and Gambro Lund AB (Sweden). Additional support was received from the Dr. E.E. Twiss Fund, Roche Netherlands; the International Society of Nephrology/Baxter Extramural Grant Program, and the Dutch Organization for Health Research and Development (ZonMW, grant 17088.2802). The authors are grateful to all patients, nursing staff and collaborators who participated in this project.

The authors have no conflicts of interest to disclose.

I. Chapdelaine and I.M. Mostovaya contributed equally to this work.Part of this work was presented at the ERA-EDTA Congress, Istanbul, May 2013.