Acute kidney injury (AKI) is one of the most common and serious complications in critical care patients [
]. Renal replacement therapy (RRT) provides cornerstone management in severe AKI. While RRT is initiated promptly for life-threatening indications (e.g., severe hyperkalemia), there is controversy as to whether earlier initiation is beneficial in the absence of urgent indications [
]. Two recent randomized trials reached opposite conclusions as to whether early initiation is beneficial [
]. The Artificial Kidney Initiation in Kidney Injury (AKIKI) trial used AKI stage 3 and conventional indications for RRT initiation. Early initiation was not superior to a conservative approach, and the lowest mortality was observed in patients who never received RRT. In the standard RRT group, 49% of the patients spontaneously recovered [
]. In the Effect of Early vs Delayed Initiation of Renal Replacement Therapy on Mortality in Critically Ill Patients with Acute Kidney Injury (ELAIN) study, however, AKI stage 2 plus plasma neutrophil gelatinase-associated lipocalin (NGAL) more than 150 ng/mL were used as inclusion criteria. Patients in the standard group had a 90% chance of RRT. This shows that using a biomarker could be an important prognostic criterion for the prediction of RRT requirement.
Unfortunately, it is difficult to predict the requirement for RRT, and clinicians vary widely in their decision making regarding when to initiate therapy. Recently, the furosemide stress test (FST) has been validated in patients with AKI stage 1 and 2 by the Kidney Disease Improving Global Outcomes (KDIGO) criteria as a novel test for the prediction of progression to AKI stage 3, the need for RRT, and in-hospital death [
]. FST also outperformed several novel biomarkers for the prediction of adverse outcomes [
]. Therefore, FST might be suitable to risk stratify AKI patients in guiding the decision to initiate RRT.
However, before embarking on a large trial to test alternate strategies for RRT initiation, we sought to determine whether FST could be used in a clinical trial setting to stratify AKI patients and to determine the feasibility of using FST in this setting. We therefore conducted a pilot study comparing early or standard initiation of RRT in FST-nonresponsive AKI patients. We also examined 28-day mortality and other clinical outcomes, although we did not power our study for these endpoints.
Trial design and oversight
The FST trial was funded by the Kidney Foundation of Thailand. The study was a prospective, multicenter, open-label, two-group randomized trial conducted in five intensive care units (ICU) in Thailand from March 2016 to July 2017. The trial was registered at clinicaltrials.gov (NCT02730117). The institutional ethics boards of all participating centers approved the protocol. The investigators informed patients or their surrogates about the trial both orally and with a written document. Informed consent was obtained from participating patients or their substitute decision-makers before the FST was performed. Co-investigators at each participating site were responsible for enrolling patients, ensuring adherence to the protocol, and completing the case record form. All analyses were performed by an independent statistician in accordance with the International Conference on Harmonization Good Clinical Practice Guidelines.
All adult patients (≥ 18 years old) admitted to the ICU were screened. Patients with AKI at any stage (defined by KDIGO criteria) [
] were assessed for additional inclusion criteria (both of the following needed to be fulfilled): 1) clinical diagnosis of acute tubular necrosis (e.g., the presence of granular or epithelial cast, fractional excretion of sodium ≥ 1%, fractional excretion of urea ≥ 50%, plasma NGAL ≥ 150 ng/mL); 2) in the opinion of the treating team, the patient was well resuscitated and euvolemic; and 3) in the opinion of the treating team, the patient had neither an emergent indication nor a contraindication to RRT (Additional file
: Appendix 1). We excluded patients with any of the following criteria: 1) baseline serum creatinine ≥ 2 mg/dL (male) or ≥ 1.5 mg/dL (female) [
]; 2) history of renal allograft; 3) known pregnancy; 4) allergy or known sensitivity to loop diuretics; 5) moribund patients with expected death within 24 h or whose survival to 28 days was unlikely due to an uncontrollable comorbidity (i.e., end-stage liver or heart disease, untreatable malignancy); 6) patients with advanced directives who issued the desire not to be resuscitated; 7) prior treatment with RRT within 30 days; 8) serum albumin < 2 g/dL; and 9) patients receiving extracorporeal membrane oxygenation or circulatory assistance (Additional file
: Appendix 1). We considered the patients to be provisionally eligible if all the inclusion criteria were met, and no exclusion criteria were present.
Furosemide stress test (FST)
FST was performed by giving intravenous furosemide 1 mg/kg to naive patients or 1.5 mg/kg to patients with a history of furosemide use within 7 days. Urine output was measured hourly and, if the urine output exceeded 200 mL for the subsequent 2 h, the patient was considered to be FST responsive. Patients with a urine output less than 200 mL in 2 h were considered FST nonresponsive and underwent randomization. Additional data are provided in Additional file
: Table S1 [
We randomized patients 1:1 to early or standard RRT initiation using a randomly permuted block of four, stratified by center and type of ICU. Patients randomized to early RRT were to receive RRT within 6 h of randomization [
]. The 6-h period was for the establishment of vascular access and RRT initiation. In the standard RRT group, RRT was initiated only if one of the following criteria were met: blood urea nitrogen ≥ 100 mg/dL, serum potassium > 6 mmol/L, serum bicarbonate < 12 mmol/L or pH < 7.15, PaO
ratio < 200, or chest radiograph compatible with pulmonary edema.
The starting RRT modality was continuous venovenous hemofiltration (CVVH) using integrated machines (The Prismaflex® system, Gambro, Sweden or the Aquarius™ system, Nikkiso, Japan) with high-flux hemofilters (AN69 or HF12) and pre-filter replacement fluid of 25–30 mL/kg/h. The blood flow target was 150–200 mL/min. Regional citrate anticoagulant was the first-line of the anticoagulation strategy, followed by heparin in patients requiring systemic anticoagulation, and no anticoagulation in patients with contraindications to citrate or coagulopathy. Once the patients became hemodynamically stable (a decreasing dose or no longer needing the use of inotropic drugs) or were transferred out of the ICU, the RRT modality could be switched to prolonged intermittent RRT (PIRRT) or intermittent hemodialysis according to the judgment of the treating physician. RRT was continued until death, patient withdrawal, or renal recovery. The anticoagulant of choice in our study was intravenous heparin. However, if the patients had a contraindication to heparin, we used a no anticoagulant strategy. Renal recovery was defined as spontaneous diuresis more than 1000 mL/day or 2000 mL/day with diuretics with resolution of electrolyte or acid-base abnormalities and did no requirement to resume RRT for at least 7 days [
Follow-up and data collection
We followed all patients for 28 days from randomization or until hospital discharge and serially assessed the severity of illness, laboratory data, and physiological data. RRT use and prescription details were recorded. We collected blood samples at days 0, 3, and 7 to measure: serum N-terminal prohormone of brain natriuretic peptide (NT-proBNP), a marker of volume overload [
]; plasma NGAL, a marker of kidney injury [
]; and serum angiopoietin-2, a marker of endothelial dysfunction [
The primary outcome was feasibility as judged by: 1) compliance with the study protocol for > 90% of patients; 2) the ability to use FST to differentiate the RRT rate in FST responders and standard group of nonresponders 50%; 3) successful randomization of FST nonresponders; 4) separation of timing of RRT initiation between the early and standard RRT groups for at least 24 h; and 5) < 10% lost to follow-up.
Secondary outcomes included: 1) 28-day all-cause mortality; 2) 7-day fluid balance; 3) ICU-free days; 4) mechanical ventilator-free days; 5) RRT-free days; 6) length of ICU stay and hospital stay; 7) renal recovery; 8) dialysis requirement on day 28; 9) the proportion of patients free from RRT on days 0, 3, and 7; 10) nonrenal Sequential Organ Failure Assessment (SOFA) score on days 0, 3, and 7; and 11) RRT-related adverse events and vascular access-related adverse events (Additional file
: Appendix 2). Exploratory endpoints included biomarkers (plasma NGAL and angiopoietin-2, and serum NT-proBNP) on days 0, 3, and 7.
Blood samples were drawn in pyrogen-free vials and plasma was separated by centrifugation and frozen (−80 °C). Blood samples for NGAL, angiopoietin-2, and NT-proBNP were determined at trial inclusion, and day 3 and day 7 after randomization. Plasma biomarkers (NGAL and angiopoietin-2) were tested using enzyme-linked immunosorbent assay (R&D Systems, USA). Serum NT-proBNP was determined using electrochemiluminescence immunoassay analysis (Cobas assay; Roche Diagnostics, Mannheim, Germany).
Sample size determination
To detect a 10.0% difference in 28-day mortality rate between early and standard RRT with a power of 80% and a 5% significance level on the basis of a previous report, approximately 900 patients would be needed [
]. As a feasibility study, we aim to use FST to risk stratify patients who would need RRT and not need RRT. At least thirty patients were required to detect a 50% absolute difference in the proportion of RRT between FST responders and FST nonresponders (standard group) with a power of 80% (β = 0.2) at a 5% significance level (α = 0.05). The trial was terminated on 31 July 2017 after recruiting 162 patients (44 FST responders and 118 FST nonresponders).
All analyses adhered to the intention-to-treat principle. Categorical data are described as numbers and percentages and compared between treatment groups using Chi-square or Fisher’s exact test. Continuous variables are described as means (with standard deviations (SD)) or medians (with interquartile range (IQR)) and compared between each group using unpaired
t test in normally distributed data or Wilcoxon rank sum test for non-normal data. Overall survival for all patients was estimated by the Kaplan-Meier method. A log-rank test was used to compare time to death between treatment arms and secondarily among patients undergoing RRT versus no RRT, and for patients with positive versus negative FST. The univariate Cox proportional hazard regression model was used to determine factors associated with RRT requirement in the standard arms using
p values < 0.10, and a multivariate model was analyzed using significant factors from the univariate model, including gender. Data from all the patients were censored at the time of death or at day 28. Severity score, laboratory data, and physiological data between days 0, 3, and 7 were computed using repeated measures analysis of variance (ANOVA) for differences within groups and generalized estimating equations for differences between groups. All analyses were performed using Stata 14.0.
In this pilot randomized controlled trial (RCT), we demonstrated the feasibility and safety of conducting a trial comparing early versus standard RRT using FST as an initial triage strategy. The results of the present study demonstrate that FST was easy to administer in the context of a clinical trial (100% compliance) and provided excellent predictive ability for the subsequent use of RRT; nonresponsive patients had an RRT rate of 75% versus 13.6% for FST-responsive patients (Additional file
: Table S1). Compliance with other aspects of the study protocol was also excellent, with > 95% of patients receiving the intervention they were randomized to receive. Randomization was successful in that few baseline differences were seen between intervention arms and the separation of timing of initiation (early versus standard) approached, but did not quite achieve, 24 h. Finally, we achieved excellent follow-up, with 100% of patients available for survival analysis.
We did not encounter any safety issues using the FST, and the only adverse events encountered with early initiation were increased rates of hypophosphatemia and dialysis catheter issues (Additional file
: Table S4). Our sample may have been too small to detect some adverse events. For example, there was more hemodynamic instability (34.5% versus 20%) with early initiation while less cumulative fluid removal was seen at 7 days with standard RRT (1.2 versus 1.7 L), but neither of these differences were significant. However, there was a significant difference in fluid accumulation from randomization to initiation of RRT (4.8 versus 8.7 L,
= 0.02) favoring early initiation.
Our final sample size was insufficient to test whether timing of initiation of RRT impacted 28-day survival. However, we did not observe any differences in survival between early and standard RRT (Table
). Other secondary outcomes including renal recovery rate, ICU-free days, mechanical ventilator-free days, 7-day fluid balance, and dialysis dependence rate were not significantly different between both treatment arms (Table
), although we were unpowered for many of these endpoints. For example, ventilator-free days and ICU-free days were both greater with early initiation but with very wide confidence intervals. Similarly, the small differences observed in ICU and hospital length of stay (difference of 1.5 and 2.5 days, respectively) are clinical relevant but would have required a much larger trial to detect.
The optimal timing to initiate RRT in AKI patients remains to be established [
]. Two recently published RCTs examining timing of RRT initiation reached different conclusions. The AKIKI multicenter trial in France investigated early initiation (within 6 h after documentation of KDIGO stage 3) versus a “wait and see” strategy (as per conventional indications). Sepsis, severe sepsis, or septic shock were present in 80%. Mortality at 60 days was not different between the two strategies [
]. Conversely, a single center RCT in Germany (ELAIN study) defined early RRT as AKI KDIGO stage 2 plus plasma NGAL > 150 ng/mL and delayed RRT as AKI stage 3. Early initiation of RRT significantly reduced 90-day mortality compared with delayed initiation (39.3% versus 54.7%) [
]. The other ongoing study, Standard versus Accelerated initiation of Renal Replacement Therapy in Acute Kidney Injury (STARRT-AKI), also uses the higher cut-off level of plasma NGAL (≥ 400 ng/mL) as one of the three inclusion criteria along with a twofold rise in serum creatinine and oliguria [
]. By using plasma NGAL as a screening biomarker to filter patients, the ELAIN trial was able to select 90% of patients in the standard arm who required RRT. On the contrary, 49% of the patients in the standard indication arm of the AKIKI trial, which used only AKI staging as a screening tool, showed spontaneous recovery, which implied that RRT could also be avoided in some patients in the early indication arm had there been screening tools for selection of high-risk patients. Therefore, a pure clinical strategy may not be enough to analyze early versus standard initiation strategy and prevent unnecessary RRT. In our study, we identified 44/162 (27.2%) FST-responsive patients with only a 13.6% rate of RRT. RRT was averted in 86.4% of FST responders. We were able to select a group with a 75% RRT rate in the standard RRT group of FST nonresponders. In the standard arm, plasma NGAL was also a significant predictor for spontaneous recovery. This suggests that while FST is an excellent strategy to select patients who would recover, combining FST nonresponsiveness with plasma NGAL might be an even more suitable strategy to predict patients who are likely to require RRT.
There are some limitations in our study. First, due to the nature of the study, this was an unblinded RCT. The robust protocol for initiation of RRT and high compliance rates minimizes the risk of bias in RRT initiation. Second, the numbers of participants were rather small (60 in each arm) leading to insufficient power for secondary endpoints. However, as a pilot study, our results support the feasibility and safety of this approach for a definitive trial in the future. We were likely underpowered for our exploratory biochemical analysis as well. The incidence of hypophosphatemia was higher in the early RRT group, and severe hypophosphatemia is known to be associated with respiratory failure and weaning failure [
]. Plasma NGAL, NT-proBNP, and angiopoietin-2 levels in the early intervention arm were not significantly different from standard RRT. However, there were wide confidence intervals and important differences could have been missed.
We thank the staff, fellows, nurses, and research coordinators at Chulalongkorn University, Vajira Hospital, Nakhon Ping Hospital, Vajira Phuket Hospital, Bhumibol Adulyadej Hospital. We also thank Miss Sasipha Tachaboon, M.Sc., the medical technologist and Miss Pimnara Peerawaranun for statistical analysis.
This study is a part of a project supported by the Excellence Center for Critical Care Nephrology, King Chulalongkorn Memorial Hospital, and is funded by the National Kidney Foundation of Thailand.
Availability of data and materials
All data generated and/or analyzed during this study are included in this published article.
Ethics approval and consent to participate
The study was approved by the Institutional Review Board of every participating center.
Consent for publication
The manuscript has been read and its submission approved by all co-authors. Patients were prospectively included with informed consent.
JAK reports grant support and consulting fees from Baxter and NxStage, unrelated to this study. The remaining authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.