Sodium Bicarbonate and Renal Function after Cardiac Surgery: A Prospectively Planned Individual Patient Meta-analysis

ACUTE kidney injury (AKI)^1,2  is a common postoperative complication in cardiac surgery patients and is independently related to increased mortality and long-term morbidity.³  Even minimal serum creatinine increases are independently associated with poorer outcomes⁴  and, as over a million cardiac surgery operations are carried out annually, the associated costs are extensive.^5,6 

Multiple mechanisms of cardiac surgery–associated acute kidney injury (CSA-AKI) have been suggested. These include generation of reactive oxygen species, ischemia reperfusion injury, activation of inflammatory pathways, and hemolysis.^7–10  If such mechanisms exist, the process behind it may relate to the ability of bicarbonate to slow the Haber–Weiss reaction that generates free radicals.^11,12  Bicarbonate may also scavenge other reactive species such as peroxynitrite.¹³  By inducing urinary alkalinization, it may reduce kidney injury mediated by free hemoglobin species, the generation of methemoglobin-containing tubular casts, and ferrous ion–catalyzed production of free radicals.¹⁴  These potential benefits are important because no simple, safe, and effective intervention to prevent cardiopulmonary bypass–associated AKI has yet been found in the broad cardiac surgery population.^15–19 

Meta-analysis is a powerful statistical technique designed to contrast and combine results from individual studies providing increased precision of estimates along with greater generalizability of findings. In a critical care setting, however, the heterogeneous nature of critically ill patients and their treatment complicates the use of standard aggregate meta-analysis techniques and can result in misleading conclusions.²⁰  This problem can be overcome through the use of individual patient data meta-analysis (IPDMA),²¹  which uses individual patient data rather than summary data from each of the participating studies. Unlike standard meta-analysis techniques, IPDMA enables interaction effects to be estimated within trials, thus reducing the potential for confounding bias. IPDMA has the additional benefit of enabling time-to-event analysis, covariate adjustment, definitive subgroup analysis, and a more balanced interpretation of results—all of which are difficult if not impossible to achieve using standard meta-analysis techniques. Despite being the acknowledged gold standard for meta-analyses,²²  IPDMAs are infrequently carried out, primarily because of the lack of consistency between studies and the difficulty in obtaining and combining detailed trial data.^23,24  To avoid such hurdles, the Cochrane Collaboration^* advocates the use of prospectively planned meta-analysis, which enables outcomes, variables, methods, and data collection to be standardized before study commencement, thus ensuring homogeneity from all participating studies. In recent times, there has been a subsequent push toward using this approach in intensive care medicine.²⁵ 

In 2009, a single-center pilot randomized controlled trial enrolling a broad cohort of cardiac surgical patients reported a statistically significant reduction in the incidence of AKI after the infusion of perioperative sodium bicarbonate.²⁶  A recent review identified urinary alkalinization as the single most important drug-based intervention with convincing evidence to prevent AKI.²⁷  To confirm these findings in broader geographic and institutional settings, two additional multicenter randomized clinical trials^28,29  were planned and conducted, facilitating a prospectively planned IPDMA of these three studies.

This study was a prospectively planned IPDMA of the only three published double-blinded randomized controlled trials using a 24-h regimen of bicarbonate infusion (one single-center, two multicenter in design). It was conducted to assess if the administration of sodium bicarbonate as a continuous infusion commenced before cardiopulmonary bypass would result in less postoperative AKI in patients undergoing cardiac surgery. Each individual trial was conducted at university-affiliated hospitals and was registered with ClinicalTrials.gov (NCT00334191, NCT00672334, and NCT00878956). After an interim analysis by the Data Safety Monitoring Board, a decision was made based on a lack of efficacy and potential harm, to halt one of the multicenter studies with recruitment at 70%. In accordance with this finding, a subsequent interim analysis was also conducted at the second multicentre study, and recruitment was halted for a lack of efficacy with 95% of recruitment complete. The Human Research Ethics Committee of each center approved the study, and all sites obtained written informed consent to participate from each patient before surgery. Patient flow for the combined studies is reported in a CONSORT diagram (fig. 1).

Patient eligibility was consistent across the three trials. Patients were eligible for inclusion if they were scheduled for elective or urgent cardiac surgery utilizing cardiopulmonary bypass and were identified preoperatively to be at increased risk for CSA-AKI.

This high-risk group was identified using previously published criteria shown to accurately predict patients deemed at risk of developing AKI.³⁰  Whilst this scoring system developed by the Cleveland Clinic group was originally designed to predict a tiered level of risk associated with AKI requiring dialysis, like Burns et al.¹⁶  we have simplified and customized this scoring process to enable identification of patients with an increased risk in a nondialysis AKI setting, thus facilitating recruitment of an increased number of cardiac surgery patients. Details of inclusion and exclusion criteria are shown in table 1.

At each study center, the hospital pharmacy clinical trials coordinator or unblinded site research coordinator used a Microsoft Excel–based (Microsoft Corp., Redmond, WA) random number generator to create the randomization list using a permuted block strategy with varying blocks sizes. The active study treatment or saline was prepared either in infusion bags contained in shrink-wrapped black plastic bags or 500-ml glass bottles. Both treatments were identical in appearance, ensuring allocation concealment to patients, anesthesiologists, cardiac surgeons, intensive care specialists, bedside nurses, and investigators. Treatment allocation was only revealed after the study had been completed, the database locked, and statistical analysis completed.

Study intervention was similar in all trials, although there were some minor differences in the total dose of bicarbonate given and the total volume of fluid used as diluent. Patients from the pilot study received a loading dose of 0.5 mm/kg body weight (=bolus) over 1 h immediately after the induction of anesthesia followed by continuous intravenous infusion of 0.15 mm kg⁻¹ h⁻¹ over 23 h (24-h total = 4.0 mm/kg). Each patient from the subsequent multicenter studies received a dose of 0.5 mm/kg body weight (=bolus) over 1 h immediately after the induction of anesthesia followed by continuous intravenous infusion of 0.2 mm kg⁻¹ h⁻¹ over 23 h. Bolus and continuous infusion achieved a total dose of 5.1 mm/kg over 24 h, which resulted in a 25% greater dosing regimen than that used in the first study. For the multicenter studies, both bicarbonate and saline were administered as hypertonic (1 mm/ml) solutions giving a total dose per patient of 5.1 mm/kg. For these two studies, a regime that administered the bicarbonate or saline in a total volume of 1,250 ml was used.

Apart from the addition of the infusion of study drug, there were no changes to normal clinical practice. Anesthetic techniques, including preoperative medications, were at the discretion of the attending anesthetist. Surgical approach and cardiopulmonary bypass were conducted according to the standard technique of each institution. Postoperative care, including hemodynamic, fluid, and analgesic management, was at the discretion of the intensivists and nursing staff at each institution.

The primary outcome measure was the number of patients who developed postoperative AKI. This was defined a priori as an increase in plasma creatinine concentration of greater than 25% or 44 μm/l from baseline to its peak value at any time within the first 5 days after cardiopulmonary bypass and was identical in all three studies. This end point had been used for previous studies of CSA-AKI.^16,18 

Renal secondary outcomes included the change in serum creatinine from induction to the 5-day maximum, number of patients who received renal replacement therapy (RRT) during their hospital stay, and number of patients who had an increase in serum creatinine of greater than 50% and greater than 100%. In addition, the number of patients who developed acute kidney dysfunction according to the Acute Kidney Injury Network classifications (AKIN ≥ 1, AKIN stage 1, AKIN stage 2, and AKIN stage 3)³¹  was also used. Secondary clinical outcomes included the duration of mechanical ventilation, length of intensive care unit and hospital stay, and hospital and 90-day mortality.

Two clinically relevant subgroups were considered:

• Preoperative chronic kidney disease (as defined by baseline creatinine of more than 130 μm/l)

• Low-risk versus high-risk surgery defined by elective coronary artery bypass surgery (CABG) versus all other

With a total of 877 randomized patients, this meta-analysis had a 90% power to detect an absolute reduction in the incidence of AKI (as defined by a postoperative increase in serum creatinine concentration of greater than 25% or 0.5 mg/dl (>44 μm/l) within the first five postoperative days) of 13% (50% vs. 37%) or a relative risk reduction of 26% with a two-sided P value of 0.01.

Data were analyzed according to the intention-to-treat principle. All variables were assessed for normality and log-transformed if appropriate. Baseline comparisons were performed using chi-square tests for equal proportion and reported as percentages (n). Continuously normally distributed variables were compared using Student t-tests and reported as mean (standard deviation) while nonnormally distributed data were compared using Wilcoxon rank sum tests and reported as medians (interquartile range). Binomial outcomes were analyzed using logistic regression and reported as odds ratios with 99% confidence intervals (CIs) whilst the change in creatinine was analyzed using linear modeling and reported as differences (99% CI). Continuous secondary outcomes measuring duration (intensive care and hospital lengths of stay and duration of ventilation) were found to be well approximated by log-normal distributions so were log-transformed and analyzed using linear modeling with results reported as ratios of geometric means (99% CI). Survival analysis was performed using Cox-proportional hazards regression and reported as hazard ratios (99% CI). Combined overall survival was reported using a Kaplan–Meier curve with a corresponding log-rank test. Patient survival was censored at 90 days or at hospital discharge if further follow-up data were not available. As this was a prospectively planned IPDMA of three functionally identical studies and the goal was to compute a common effect size, in accordance with Borenstein³²  overall treatment effects were determined using one-stage multivariate analysis with each individual study treated as fixed effects. Heterogeneity across studies and sites was determined by fitting interactions between treatment and study and treatment and sites. To determine if the relationship between treatment and outcome differed significantly between subgroups, interactions were fitted between treatment and subgroup. To further ensure that observed effects could not be attributed to specific hospitals within each study or from baseline imbalances, additional covariate adjusted sensitivity analysis was performed accounting for attending hospital and all potentially imbalanced baseline variables with P < 0.2 (peripheral vascular disease, insulin dependent diabetes, and acute myocardial infarction). Given the large number of outcomes considered, to reduce the chance of type I error and increase robustness of the analysis, a two-sided P value of 0.01 was used to indicate statistical significance with meta-analysis results reported using forest plots (99% CI). All analysis was performed using SAS version 9.3 (SAS Institute Inc., Cary, NC).

Between June 2006 and June 2011, the three trials enrolled a total of 883 patients at five hospitals in New Zealand, Australia, Germany, Canada, and Ireland. A total of six patients were transferred or withdrawn (fig. 1), leaving 877 patients included in the final analysis, with 439 patients randomized to the intervention arm and receiving intravenous sodium bicarbonate and 438 patients receiving saline. Patient and operative characteristics are shown in table 2. While there were no statistically significant differences (P < 0.01) between the two groups at baseline, there were, however, some minor imbalances (P < 0.2) that could potentially confound results. Bicarbonate patients were more likely to have peripheral vascular disease (14% vs. 9%, P = 0.02), insulin-dependent diabetes (7% vs. 4%, P = 0.05), and acute myocardial infarction in the previous 6 months (14% vs. 10%, P = 0.10).

While there were no significant differences in biochemical variables between groups at baseline, at both 6 and 24 h, there were significant treatment-induced biochemical differences between groups (P < 0.0001) with the bicarbonate group having higher levels of plasma bicarbonate, base excess, plasma, and urine pH and lower levels of chloride, potassium, and calcium (table 3).

The primary outcome (creatinine rise > 25% or > 44 μm/l over 5 days) occurred in 199 patients in the bicarbonate group and 183 patients in the control group. There were no significant differences in the proportion of patients developing the primary outcome (Bicarbonate 45% [99% CI: 39–51%] vs. Control 42% [99% CI: 36–48%], P = 0.29, table 4). This result remained unchanged when adjusting for study as a fixed effect (Bicarbonate vs. Control, odds ratio [OR]: 1.15 [99% CI: 0.81–1.64], P = 0.29) or controlling for covariates (Bicarbonate vs. Control, OR: 1.11 [99% CI: 0.77–1.60], P = 0.45, fig. 2). There was no overall difference between groups when considering the need for RRT. There was, however, a highly significant reduction in risk for bicarbonate patients who received elective CABG (lower risk patients) (Bicarbonate vs. Control, OR: 0.38 [99% CI: 0.25–0.58], P < 0.0001, fig. 3). Similarly, when considering a far more severe measure of AKI severity such as the risk of developing AKIN grade = 3, despite there being no overall difference between groups, within the subgroup of patients who received elective CABG surgery, there was a significant reduction in risk for bicarbonate patients (Bicarbonate vs. Control, OR: 0.45 [99% CI: 0.43–0.48], P < 0.0001, fig. 4). These results were further reflected by statistically significant heterogeneity (P < 0.01) between CABG and non-CABG patients for both outcomes with this heterogeneity potentially driven by a significant difference in the cardiopulmonary bypass times (CABG only 2.1 ± 0.9 h vs. non-CABG 2.4 ± 1.2 h, P < 0.001). There were no significant differences in the change in serum creatinine between groups.

When considering other renal outcomes, there was no difference between groups in the risk of developing an AKIN grade = 2 (OR: 0.65 [0.31–1.39], P = 0.15) or an AKIN grade ≥ 1 (OR: 1.03 [0.72–1.47], P = 0.82, fig. 3). There were no differences in intensive care or hospital length of stay or duration of mechanical ventilation (fig. 5).

Twenty-five participants died during hospital stay and a further four died between hospital discharge and day 90 with the median time to death being 19 (8–30) days. Nonsurvivors were significantly older (76 [70–79] vs. 70 [60–76] yr, P = 0.004) spent longer in intensive care (3.4 [1.4–6.2] vs. 1.2 [0.9–2.6] days, P < 0.004), and were more likely to have developed AKIN grade ≥1 (79% [23/29] vs. 40% [335/848], P < 0.0001) or AKIN grade 3 (41% [12/29] vs. 4% [31/848], P < 0.0001). They were also more likely to require postoperative RRT (41% [12/29] vs. 3% [24/844], P < 0.0001), had longer bypass times (3.6 ± 2.0 vs. 2.2 ± 1.0 h, P < 0.0001), were ventilated for longer (2.5 [0.9–6.5] vs. 0.6 [0.5–0.9] days, P < 0.0001), and were more likely to received fresh frozen plasma and packed erythrocytes in the first 48 h (64% [18/29] vs. 25% [214/845], P < 0.0001) and (79% [23/29] vs.46% [390/848], P = 0.0004), respectively.

While the proportion of deaths was higher in the bicarbonate group (4.3% [19/439] vs. 2.3% [10/438], P = 0.09), there remained no statistically significant difference after adjustment for study, attending hospital and covariates (OR: 1.83 [99% CI: 0.63–5.30], P = 0.14). This result was similar when considering survival analysis of time to death (hazard ratio: 1.96 [99% CI: 0.71–5.41], P = 0.09, fig. 6). The pattern of survival between groups can be seen in a Kaplan–Meier survival curve (fig. 7) with the increased number of death in the bicarbonate group most apparent during the first 7 days. However, neither 7-day mortality (Bicarbonate 1.4% [6/429] vs. 0.2% [1/438], P = 0.06) nor overall survival (log-rank P = 0.09) was statistically significant.

The principal cause of death was treatment-resistant cardiogenic shock (Bicarbonate 2.1% [9/439] vs. Saline 1.4% [6/438], P = 0.44). Other causes of multiple deaths include myocardial infarction (n = 3), multiorgan failure (n = 2), ventricular fibrillation (n = 2), and sepsis (n = 2), all of which had insufficient events to be statistically linked to treatment.

In this IPDMA of three high-quality double-blind randomized controlled clinical trials evaluating the efficacy of sodium bicarbonate infusion to reduce the incidence of CSA-AKI, no overall renal benefit was found. We did, however, find renal benefit in a subgroup of patients receiving low-risk surgery (elective CABG) with patients receiving sodium bicarbonate having a significant reduction in the need for RRT and the development of an AKI (AKIN grade = 3). Of particular interest was the tightness of the CIs around these findings, which may be attributed to the small number of trials considered; however, the finding of heterogeneity between high-risk and low-risk patients for these outcomes clearly suggests that the underlying effect of sodium bicarbonate on AKI differs according to patient risk. This may suggest that, in lower risk patients, the primary mechanism of perioperative AKI is related to cardiopulmonary bypass time, whereas in the higher risk patients, the bypass time becomes secondary to the hemodynamic perturbations that occur before, during, and after.

Despite one of the multicenter studies being halted for potential harm, this IPDMA failed to identify a significant increased risk of death associated with bicarbonate. Cox proportional hazards regression further enables integration of the duration of survival for both survivors and nonsurvivors adjusting for study, attending hospital and potentially imbalanced baseline covariates; however, this analysis also failed to show a significant difference between groups.

IPDMA is the acknowledged gold standard for meta-analyses.²²  The ability to design, capture, and combine patient information from three separate trials conducted across five different countries is unprecedented in intensive care medicine. The use of individual patient data over standard meta-analysis techniques facilitates more informative statistical analysis (such as time to event analysis), enables covariate adjustment, allows avoidance of the ecological fallacy,³³  and further enables definitive comparison of subgroups of interest.

Prospective planning enabled homogeneity between the three studies, including patient population, the intervention used and outcomes captured, and using the same definition of AKI, which has been a significant factor limiting the use of IPDMA for the analysis of studies in this field before. The validity of the findings is further strengthened by the fact that enrollment was completed over a period of 5 yr.

The additional statistical power provided by this technique enabled the study of two clinically important subgroups: those with preexisting renal disease and those undergoing complex surgery. This would not have been possible in any of the individual trials and demonstrates the value of prospective planning and standardization of study data collection whenever possible.

Despite the valuable insights that can be obtained from IPDMA, there are several limitations to the interpretation of data. In particular, both multicentre studies were prematurely terminated before completion of planned enrollment. As described in Bassler,³⁴  this has the potential to bias effect estimates, although, as the combined study recruitment of these two trials exceeded 80% of planned enrolment, this bias is unlikely to be large. Additionally, no attempt was made by study investigators to control intraoperative factors, including surgical and anesthetic techniques; however, a large sample size combined with the benefits of randomization into multicentric, double-blinded, and treatment-concealed trials should ensure that confounding from these effects is likely minimized.

Furthermore, the decision to use CABG as a marker of risk for the development of postoperative AKI is limited and may be prone to imbalances in the true baseline risk. While we are hopeful that the principles of blinding and randomization will ensure that the distribution of risk would be similar between groups, we acknowledge that in a truly high-risk population, there may have been differences that we were unable to identify. Similarly, we acknowledge that the decision to use serum creatinine to define preoperative renal insufficiency does have limitations and that the estimated glomerular filtration rate may be better able to define a population at greater risk of AKI. It is further worth noting that, as patient recruitment excluded patients undergoing emergency cardiac surgery, these results are specific to elective cardiac surgery patients and that, as only 61% of eligible patients were randomized, care should be taken when extrapolating results to the larger population.

This study builds on the findings of the three individual trials exploring the relationship between perioperative sodium bicarbonate infusion and AKI. Following the suggestion of benefit in the incidence of AKI in cardiac surgery patients,²⁶  a further two multicenter randomized controlled trials were conducted with both terminated prematurely by data safety management committee decisions based on lack of benefit^28,29  and potential harm.²⁹ 

The above decisions and early trial findings have been borne out in the IPDMA. Through this technique, we can confirm that there is no statistical evidence to suggest that sodium bicarbonate reduces the overall incidence of kidney injury. However, there is some evidence to suggest a likely beneficial effect on renal function in patients receiving CABG. Finally, at the time of this IPDMA, an additional smaller study where bicarbonate infusion was given for only 6 h was also published¹⁹  showing no benefit.

This IPDMA provides additional insights beyond those of each study. The observation that there is a reduced incidence of AKI in patients who undergo elective CABG provides a possible explanation for the initially promising findings of the first single-center study²⁶  and suggests that the hypothesized beneficial effect of bicarbonate on kidney function is likely to exist in low-risk cardiac surgery patients.

Our observations support the value of IPDMA as a statistical technique and argue for greater pretrial harmonization of planned data collection for investigations of similar interventions in different geographical jurisdictions. Finally, the findings of this IPDMA indicate that further clinical use of bicarbonate infusion in unselected cardiac surgery patients may not be justified.

Urinary alkalinization using sodium bicarbonate infusion for cardiac surgery patients does not significantly reduce the overall risk of AKI. However, it provides a degree of renal protection to patients undergoing low-risk surgery. On this basis, we recommend further studies of the use of sodium bicarbonate infusion to reduce CSA-AKI only in lower-risk cardiac surgery patients such as those receiving elective CABG.

The three studies included in this individual patient data meta-analysis were funded by grants from the German Heart Foundation (Deutsche Stiftung für Herzforschung, Frankfurt a. M., Germany); the Else Kröner-Fresenius-Stiftung (Bad Homburg, Germany); the Canadian Intensive Care Foundation (Edmonton, Alberta, Canada); the Intensive Care Foundation (Melbourne, Australia) (Grant no. 2008/04); the Austin Hospital Intensive Care Unit Research Fund (Melbourne, Australia) (Grant no. 2008/02); Green Lane Research and Educational Fund Board (Auckland, New Zealand) (Grant no. 08/07/4058); and A+ Charitable Trust (Auckland, New Zealand) (Grant no. PG-1104-001).

Dr. Bagshaw has previously received funding by a Canada Research Chair in Critical Care Nephrology and Clinical Investigator Award from Alberta Innovates—Health Solutions (Edmonton, Alberta, Canada). All other authors report no conflicts of interest.