Acetate-containing bicarbonate dialysate increases all-cause mortality more than acetate-free bicarbonate dialysate containing citrate in hemodialysis patients

Background No studies have compared mortality between acetate-containing bicarbonate dialysate (ABD) and ace-tate-free bicarbonate dialysate containing citrate (AFD) in hemodialysis (HD) or online hemodiafiltration (OHDF). We therefore compared mortality between ABD and AFD in each modality. Methods This retrospective observational study included 738 patients who were receiving super high-flux (SHF) or SHF albumin-leaking HD ( n = 310: ABD 235 and AFD 75) or OHDF ( n = 428: ABD 321 and AFD 107) at our institution between 1 April and 1 July 2017. Three-year all-cause mortality was compared between ABD and AFD in the HD or OHDF groups using a propensity score matching model. Kaplan–Meier survival curves were compared using the log-rank test, and then Cox regression analysis with adjustments was performed for some covariates that remained significant. Results After propensity score matching, mortality on ABD was not significantly different from that on AFD in the HD group [ n = 75; hazard ratio (HR) 2.271, 95% confidence interval (CI) 0.863–5.981, P = 0.087] or in the OHDF group ( n = 107; HR 1.944, 95% CI 0.585–6.458, P = 0.269) without adjustments. However, with adjustments using some covariates, mortality was significantly higher on ABD than on AFD (adjusted HR 4.501, 95% CI 1.434–14.125, P = 0.010) in the HD group, but not in the OHDF group. Conclusions These findings suggest that ABD worsens mortality more than AFD in patients on SHF and SHF albu-min-leaking HD.


Introduction
Acetate dialysate induces peripheral vasodilation and depression of myocardial contractility, leading to hemodynamic instability [1].Therefore, acetate-containing bicarbonate dialysate (ABD) is widely used instead of acetate dialysate because the trace amount of acetate in bicarbonate dialysate prevents the crystallization of calcium and magnesium compounds.However, even a trace amount of acetate may not be tolerated because of the activation of nitric oxide synthase [2].Acetate-free bicarbonate dialysate containing citrate (AFD) has also been used because the trace amount of citrate prevents crystallization of calcium and magnesium compounds.A population-based retrospective multicenter observational study demonstrated that all-cause mortality on AFD was not significantly different from that on mixed dialysate of ABD and acetate-free bicarbonate dialysate containing hydrochloric acid (AFD-HCl) in a mixed population of patients on hemodialysis (HD) or online hemodiafiltration (OHDF) [3].To date, no studies have directly compared prognosis between ABD and AFD.
Correction of metabolic acidosis in patients on HD is associated with modest improvements in nutritional status.The National Kidney Foundation Kidney Disease Outcome Quality Initiative guidelines for maintenance dialysis recommend a predialysis serum bicarbonate level of ≥ 22 mEq/L for nutrition [4] and bone metabolism [5].In Japan, several types of ABD and only one type of AFD are available.Given that the bicarbonate and total alkali levels are higher in AFD than in ABD, there may be advantages and disadvantages, such as improvement of metabolic acidosis and excessive metabolic alkalosis, depending on the alkalizing agent used.
Mortality has been reported to be significantly lower when using a dialyzer with β 2 -microglobulin (β2MG) clearance of ≥ 70 mL/min than when using a dialyzer with β2MG clearance of < 70 mL/min [6].In addition, mortality on super high-flux (SHF) albumin-leaking HD [β2MG clearance of ≥ 70 mL/min and sieving coefficient (SC) for albumin of ≥ 0.03] was significantly lower than that on SHF HD (β2MG clearance of ≥ 70 mL/min and SC for albumin of < 0.03) [7].Although mortality was not significantly different between Japanese-style predilution OHDF (pre-OHDF) and postdilution OHDF (post-OHDF) [8], mortality was significantly lower on OHDF with high albumin leakage than on OHDF with low albumin leakage [9].
Therefore, we compared mortality between ABD and AFD in patients on SHF HD or SHF albumin-leaking HD and in patients on OHDF, with elimination of the influence of albumin leakage.

Patient selection
As shown in Fig. 1, 738 of 944 previously described patients on maintenance dialysis with HD (β2MG clearance rate ≥ 70 mL/min) or OHDF at our institution and registered in our medical records database as of 1 July 2017 [8] were recruited to prepare a propensity score matching (PSM) model.The patients were divided according to whether they were on HD (n = 310) or OHDF (n = 428).The exclusion criteria were as follows: age younger than 20 years, a blood purification method other than HD or OHDF, dialysis frequency of fewer than three sessions per week, dialysis time less than 3 h, substitution volume < 60 L for pre-OHDF or < 8 L for post-OHDF, missing covariate values, and pregnancy or lactation.Patients whose dialysis conditions (dialysis method, dilution method, substitution volume, and/ or membrane material) at the start of the study on 1 July 2017 were different from those on 1 April 2017 were also excluded.Patients receiving HD or OHDF were defined as those confirmed annually to have received the same dialysis method for 3 years (1 July 2017 to 1 July 2020).Switching between groups was censored in the Kaplan-Meier survival analysis.The dialysis modality was chosen at the physician's discretion.Two types of ABD and one type of AFD dialysate were used at our facilities during the study period (Table 1).Because our facility has ten dialysis rooms (ABD A used in four rooms, ABD B used in three rooms, and AFD used in three rooms), we confirmed that each patient had been using the same dialysate annually for 3 years.Switching dialysates between ABD and AFD was censored in the Kaplan-Meier survival analysis.Blood test results were extracted from the medical records.

Preparation of propensity score-matched pairs
HD with ABD (n = 235) versus AFD (n = 75) and OHDF with ABD (n = 321) versus AFD (n = 107) were compared for all-cause mortality.Propensity scores were matched for 75 pairs of patients receiving HD and 107 pairs receiving OHDF.The following 13 items were used to calculate the propensity score for comparing patient survival outcomes: age, dialysis vintage, presence or absence of diabetes mellitus (DM), Kt/V, albumin leakage, body mass index (BMI), normalized protein catabolism rate (nPCR), albumin, high-sensitivity C-reactive protein (HS-CRP), systolic blood pressure (SBP), hemoglobin (Hb), corrected calcium (Ca), and phosphorus (P).The duration of each dialysis session was 4 h, and both dialysate flow rate (QD) in HD and total QD (QD plus the substitution volume) in OHDF were fixed at 500 mL/min.
To calculate the propensity score for each patient, multivariable logistic regression analysis was performed using the two patient groups with the different dialysates to be compared as the dependent variable and 13 covariates as independent variables, followed by logit transformation.The propensity scores were calculated to 14 decimal places.Regardless of the number of cases, patients in the two groups were paired by nearest available matching at a ratio of 1:1 within a caliper of 0.18978 for ABD in HD and 0.227488 for ABD in OHDF (0.2 multiplied by the standard deviation of the logit value) for all patients in both groups [10].

Adjustment for variables in standard Cox proportional hazards regression analysis
All-cause mortality was compared between ABD and AFD in the HD or OHDF group by Cox proportional hazards regression analysis with adjustments for some covariates that remained significant even after matching.In model 1, the data were adjusted for age, sex, dialysis vintage, presence or absence of DM, and presence or absence of cardiovascular disease (angina pectoris, myocardial infarction, atrial fibrillation, heart failure, stroke, peripheral artery disease, and limb amputation).In model 2, the data were adjusted for the variables in model 1, Kt/V, β2MG, and albumin leakage.In model 3, the data were adjusted for the variables in model 2 and markers of nutritional status and chronic inflammation (BMI, nPCR, albumin, and HS-CRP).In model 4, the data were adjusted for the variables in model 3, SBP, Hb, Ca, and P.

Statistical analysis
Survival was determined from the medical records, which include information on deaths and transfers to other hospitals.A daily survival analysis was performed for the two groups, including censored cases, using the Kaplan-Meier method.Between-group differences were examined for statistical significance using the log-rank test.Cox regression analysis was used to calculate hazard ratios (HRs).All analyses were performed using SPSS Statistics for Windows, version 26 (IBM Corp., Armonk, NY).A two-tailed P value < 0.05 was considered statistically significant.

Comparison of patient survival outcome between ABD and AFD in the HD group
Table 2 shows a comparison of the variables recorded before and after PSM.After PSM, Hb was significantly lower for ABD than for AFB.Three-year all-cause mortality was not significantly different between ABD and   AFB [HR 2.271, 95% confidence interval (CI) 0.863-5.981,P = 0.087, log-rank test] (Fig. 2).Although all-cause mortality was not significantly different between ABD and AFD on univariate analysis, mortality on ABD was significantly higher than that on AFD after adjustments for model 1 (HR

Comparison of patient survival outcome between ABD and AFD in the OHDF group
Table 3 shows comparisons of the variables recorded before and after PSM.After PSM, Hb was significantly lower and dialysis vintage, β2-MG, and HS-CRP were significantly higher for ABD compared with AFB.Three-year all-cause mortality was not significantly different between ABD and AFB (HR 1.944, 95% CI 0.585-6.458,P = 0.269, log-rank test) (Fig. 3).
In univariate analysis, all-cause mortality was not significantly different between ABD and AFD even after adjustments for models 1, 2, 3, and 4 as shown in Table 4(b).

Discussion
This study is the first to demonstrate that mortality is significantly higher on ABD than on AFD in HD patients.The Hb levels could not eliminate bias between the groups in PSM, and no significant difference was detected in the comparison of mortality using the logrank test (P = 0.087).However, a significant difference was found between groups after further adjustment for covariates in the Cox proportional hazards model.In model 4, the upper limit for the 95% CI of the HR was extremely large, so the results of model 3 were adopted (adjusted HR 4.501, 95% CI 1.434-14.125,P = 0.010).It seems that the significant difference would have emerged in PSM if the number of patients were increased, considering that a significant difference could be detected even after adjustment using model 1.On the other hand, no significant differences were found between ABD and AFD by the PSM and Cox proportional hazards model with adjustments in the OHDF group.
Although the trace amount of acetate in ABD is likely to be metabolized rapidly, it can activate nitric oxide synthase and secretion of inflammatory cytokines in dialyzed patients with or without acetate intolerance.Dialysate solution containing acetate 4 mmol/L was found to increase the activity of nitric oxide synthase by 2.0 ± 0.5-fold, resulting in relaxation of vascular smooth muscle, whereas acetate-free solution produced no effect [2].Postdialysis hyperacetatemia due to ABD can be prevented by using AFD [11].
The anticoagulant effect of citrate is attributable to the low calcium environment in blood; many important enzymatic steps in the coagulation cascade are calcium dependent, and citrate chelates calcium ions to form calcium-citrate complexes [12].Although citrate has the potential risk of inducing hypocalcemia, calcium ion levels have been found to remain stable within the normal range during both HD and OHDF [13].Furthermore, it has been reported that the calcification propensity, as measured by the change in T50, is improved to a significantly greater extent by AFD than by ABD [14].
Acetate-free biofiltration (AFB) is a specific HDF modality performed with a base-free dialysate and simultaneous injection of nonpyrogenic sodium bicarbonate solution.A significantly greater decrease in cytokine production was found with AFB than with ABD, suggesting that an absence of acetate in the dialysate has the effect of preventing dialysis-related complications [15].Furthermore, unlike ABD, AFB does not cause activation of polymorphonuclear neutrophils and monocytes [16].Therefore, it is assumed that even a trace amount of acetate in the dialysate can cause microinflammation.However, AFB and AFD have only been reported to improve clinical symptoms and to show in vitro effects, and there  has been no report that these dialysates improve survival compared with ABD.It has been reported that survival on AFD was not significantly different from that on a mixed dialysate of ABD and AFD-HCl in a mixed population of HD and OHDF patients [3].Since the use of AFD-HCl is associated with premature aging of concentrate piping and damage to the dialysis machine, it was not available in Japan.This study is the first to directly compare mortality between AFD and ABD in HD or OHDF.The results suggest that survival is better on AFD compared with ABD in HD but not in OHDF.High albumin leakage improves survival [17], suggesting that the worsening of survival observed for ABD in HD with albumin leakage of 2.8 ± 1.5 g/session was canceled out in OHDF with albumin leakage of 5.2 ± 2.4 g/session.
In patients with chronic kidney disease, metabolic acidosis is associated with increased protein catabolism, decreased protein synthesis, and a negative protein balance that improves after bicarbonate supplementation.Metabolic acidosis also induces insulin resistance and a decrease in elevated serum leptin levels [18].A 2-year cohort study in 56,385 patients on maintenance HD demonstrated that a predialysis bicarbonate level of > 22 mEq/L was associated with lower mortality, whereas a predialysis bicarbonate level < 22 mEq/L was associated with greater mortality [19].Moreover, a cross-sectional observational study showed a significant negative correlation between the coronary artery calcification score and the predialysis bicarbonate level [20].
Data from the Dialysis Outcomes and Practice Pattern Study demonstrate that both high (> 27 mEq/L) and low (≤ 17 mEq/L) predialysis serum bicarbonate levels are associated with increased risks for mortality and hospitalization [21].Furthermore, postdialysis metabolic alkalosis as a result of high bicarbonate in the dialysate may contribute to adverse outcomes [22].Patients with metabolic alkalosis defined as a predialysis serum bicarbonate level ≥ 26 mEq/L were found to have double the incidence of cardiac arrhythmias and a 20% greater reduction in intradialytic blood pressure when compared with control patients who had a serum bicarbonate of 19-23 mEq/L [23].However, the 2008-2009 data from the Japanese Society of Dialysis Therapy Renal Data Registry (JRDR), which includes 15,132 patients on dialysis, demonstrated that the postdialysis bicarbonate level increased by only 2.4 mEq/L and the bicarbonate pre-post difference increased by only 0.6 mEq/L when the bicarbonate concentration in the dialysate was increased from 30 to 35 mEq/L, suggesting that the dialysate bicarbonate concentration itself may play a key role in limiting the postdialysis bicarbonate level.It is also likely that a negative feedback system involving production of organic acids during dialysis contributes to the postdialysis pH and bicarbonate level, suggesting that these two parameters are not reliable indices of the degree of postdialysis overalkalosis.Therefore, predialysis pH may be the most appropriate reference for accurate correction of metabolic acidosis in patients on dialysis, considering that a predialysis pH ≥ 7.40 has been associated with a significantly elevated risk of all-cause and cardiovascular mortality, but the predialysis or postdialysis bicarbonate level or the postdialysis pH has not [24].Each report has limitations in terms of the research methods used, including adjustment factors and observation periods.Furthermore, the date of measurement in Japan is the first day of the week, whereas in other countries it is measured mid-week, so a simple comparison cannot be made.Therefore, there is still no consensus regarding the appropriate level of bicarbonate and pH predialysis and postdialysis.
The main limitation of this study is the small number of patients compared with a nationwide database.We consider it reliable to set a caliper value of 0.2 multiplied by the standard deviation of the logit transformed value of the propensity score for all cases [10].When using the caliper value for a nationwide database such as JRDR, we found no significant between-group differences in the observed variables because of larger number of patients in the database.However, the quality of healthcare management and the removal of largemiddle or large molecules that affect survival were not included in variables.Thus, there are likely significant differences in the unobserved variables.Our corporation, which includes seven facilities, has created a unified basic healthcare management policy that covers aspects such as dialysis conditions, dry weight, chronic kidney disease-related mineral and bone disorders, chronic kidney disease-related anemia, and vascular access, with the aim of eliminating any differences among the facilities according to the dialysis room rounding guidelines.In addition, we have added albumin leakage to the variables.Therefore, even if there are significant differences in observed variables, the results when using PSM with the present caliper values are considered similarly reliable to the JRDR results.In fact, the significant difference in the variables disappeared and mortality was not significantly different between the groups when the caliper value was reduced to 0.015 in the HD group (n = 66 after matching; P = 0.167, log-rank test) and to 0.010 in the OHDF group (n = 90 after matching; P = 0.146, log-rank test).Finally, we did not have data on residual kidney function, although the dialysis vintage for patients receiving HD or OHDF was more than 3 months.A randomized controlled trial is needed to confirm our findings.

Conclusion
This study is the first to suggest that mortality on ABD is 4.5 times higher than that on AFD based on the Cox proportional hazards model with adjustments in SHF and SHF albumin-leaking HD.On the other hand, the worsening of survival observed on ABD in the HD group is suggested to be canceled out in the OHDF group because of its high albumin leakage.

Fig. 1
Fig. 1 Flow diagram showing the patient selection process

Fig. 2
Fig. 2 Comparison of patient survival outcome between acetate-containing bicarbonate dialysate (ABD) and acetate-free bicarbonate dialysate (AFD) in the hemodialysis group.CI, confidence interval; HR, hazard ratio

Fig. 3
Fig. 3 Comparison of patient survival outcome between acetate-containing bicarbonate dialysate (ABD) and acetate-free bicarbonate dialysate (AFD) in the online hemodiafiltration group.CI, confidence interval; HR, hazard ratio

Table 2
Comparison of variables before and after propensity score matching between acetate-containing bicarbonate dialysate (ABD) and acetate-free bicarbonate dialysate (AFD) in the hemodialysis group Phosphorus, mg/dL 5.0 ± 1.

Table 3
Comparison of variables before and after propensity score matching between acetate-containing bicarbonate dialysate (ABD) and acetate-free bicarbonate dialysate (AFD) in the online hemodialysis group

Table 4
Comparison of all-cause mortality between acetatecontaining bicarbonate dialysate (ABD) and acetate-free bicarbonate dialysate (AFD) by Cox proportional hazards regression analysis Model 1: adjusted for age, sex, dialysis vintage, presence or absence of diabetes mellitus, and presence or absence of cardiovascular disease.Model 2: adjusted for variables in model 1, Kt/V, β2-microglobulin, and albumin leakage.Model 3: adjusted for variables in model 2, body mass index, normalized protein catabolic rate, serum albumin, and high-sensitivity C-reactive protein.Model 4: adjusted for variables in model 3, systolic blood pressure, hemoglobin, corrected calcium, and phosphorus