The optimal time of initiation of renal replacement therapy in acute kidney injury: A meta-analysis

Background The impact on the timing of renal replacement therapy (RRT) initiation on clinical outcomes for patients with acute kidney injury (AKI) remains controversial. Materials and methods We searched the Cochrane Library, EMBASE, Global Health, MEDLINE, PubMed, the International Clinical Trials Registry Platform, and Web of Science. Results We included 49 studies involving 9698 patients. Pooled analysis of 5408 critically ill patients with AKI showed that early RRT was significantly associated with reduced mortality compared to late RRT [odds ratio (OR), 0.40; 95% confidential intervals (CI), 0.32 - 0.48; I2, 50.2%]. For 4290 non-critically ill patients with AKI, there was no statistically significant difference in the risk of mortality between early and late RRT (OR, 1.07; 95% CI, 0.79 - 1.45; I2, 73.0%). Early RRT was markedly associated with shortened intensive care units (ICU) length of stay (LOS) and hospital LOS compared to late RRT in both critically ill and non-critically ill patients with AKI. Conclusions Early RRT probably reduce the mortality, ICU and hospital LOS in critically ill patients with AKI. Inversely, early RRT in non-critically ill patients with AKI did not decrease the mortality, but shortened the ICU and hospital LOS.


INTRODUCTION
Acute kidney injury (AKI) is increasingly common and associated with adverse clinical outcomes, including excess mortality and morbidity, and prolonged hospital length of stay (LOS) [1][2][3][4]. Renal replacement therapy (RRT) is the cornerstone for the treatment of severe AKI. Although RRT provokes a considerable escalation in the complexity of therapy, the optimal timing of initiation of RRT in patients with AKI has been the focus of those debates [5,6]. Conflicting results from clinical trials and systematic reviews have not resolved the debates, leaving clinicians to select the timing of initiation of RRT based on suboptimal evidence.

Meta-Analysis
allow time for the stabilization of a patient's condition before RRT and may even avoid the RRT [12,[21][22][23]. Gaudry et al. showed that the mortality was lower in patients who never received RRT than those received RRT early or late (37.1% vs. 48.5% or 61.8%), and the patients with late RRT were the most severely ill at baseline [13]. Thus, we hypothesized that the different severity of illness for patients with AKI who received early RRT may produce distinct effects on mortality. Therefore, we firstly performed a meta-analysis according to the severity of illness for patients with AKI to investigate the opportunity of RRT initiation.
3 earlier meta-analyses (Seabra et al. [24] identified 23 studies, Karvellas et al. [25] identified 15 studies and Wang et al. [26] included 51 trials) showed that early RRT could confer a survival benefit. 11 trials performed before 1985 in Seabra et al. and Wang et al. were excluded, and the addition of 10 recently published studies have been included in the present meta-analysis. However, a recent meta-analysis found no significant difference in mortality between early and late RRT [27], but included only nine "high-quality" studies. Furthermore, the included studies were limited with high heterogeneity. In the present study, we firstly made a definition of early RRT based on timebased cutoffs for patients with AKI to investigate the optimal timing of initiation of RRT. Figure 1 outlines the process for study selection. 49 studies including 9 RCTs [10,12,13,15,16,19,[21][22][23] and 40 observational studies [7-9, 11, 14, 17, 18, 20, 28-59] were included in our meta-analysis. The eligible studies were conducted from 1985 to 2016 with 9698 patients evaluated the timing of initiation of RRT in patients with AKI. The characteristics of the articles were listed in Table 1, and the details of risk of bias for RCTs were showed in Figure 2.
Subgroup analysis of critically ill patients was firstly conducted in the present study by using the definition of early according to time criteria versus biochemical indicators. The significant association between early RRT and reduced mortality was also found under the studies that defined early by time criteria [early RRT within 12 hours (OR, 0.28; 95% CI, 0. 16 Table 2).

Sensitivity, meta-regression analyses
Statistically similar results were obtained after omitting each study of critically ill patients with AKI, and the results of the sensitivity analyses were robust. Sensitivity analyses showed that Elsevivrs et al. [20] was the main source of heterogeneity for the studies of noncritically ill patients with AKI, and the heterogeneity was significantly decreased by omitting the study. For noncritically ill patients with AKI, there was no statistically significant difference in the risk of mortality between early   (Figure 4).
With the meta-regression, we did not find a correlation between patient mortality and study design (RCT vs. observational), RRT modality (CRRT, IHD vs. Mixed), study quality score, severity of illness [Sequential Organ Failure Assessment (SOFA) score], ICU admission type (surgical vs. mixed medical admissions). However, we find a correlation between patient mortality and sample size (n ≥ 100 vs. n < 100, P = 0.001) in critically ill patients with AKI.

Publication bias
No potential publication bias was observed in non-critically ill patients with AKI (P = 0.347 for the   Begg test, and P = 0.169 for the Egger test). Publication bias was seen in critically ill patients with AKI (P = 0.001 for the Begg test, and P = 0.000 for the Egger test, Figure 5).

DISCUSSION
We identified 49 studies reported data on the timing of RRT initiation among 9698 patients with AKI, and we found that early RRT significantly reduced the mortality compared to late RRT in critically ill patients with AKI. In addition, no significant survival benefits associated with early RRT were seen in non-critically ill patients with AKI. Early RRT was markedly associated with shortened ICU and hospital LOS compared to late RRT in both critically ill and non-critically ill patients with AKI.
Regardless of the definition of early RRT (according to time criteria or biochemical indicators), ICU admission type (surgical vs. mixed) or RRT modality (CRRT, IHD vs. Mixed), subgroup analyses of critically ill patients with AKI did reveal survival benefits from early RRT. Furthermore, subgroup analyses of non-critically ill patients with AKI showed that no evidence of survival advantage in early RRT.
In the present study, we firstly performed the meta-analysis according to the severity of illness and definition of early RRT based on time-based cutoffs for patients with AKI to investigate the time of RRT initiation. We accepted a broad definition of "critically ill patients with AKI" based on AKI with multiple-organ dysfunction syndrome [60], septic shock [40], RIFLE criteria (failure, loss of function, and end-stage kidney disease) [37,43,44], AKIN stages 3 [41,42,46] or Kidney Disease: Improving Global Outcomes (KDIGO) stage 3 [12,61].
By the meta-regression, we found sample size was one of the sources of heterogeneity. In contrast to previous meta-analyses, we found a lower heterogeneity among studies on this topic, especially in the subgroup.  We noted those critically ill patients in early RRT within 12 hours (I 2 , 44.8%), 24 hours (I 2 , 0.0%), 48 hours (I 2 , 30.8%), and 72 hours (I 2 , 48.2%) showed the lower heterogeneities, indicating that the heterogeneity may be partially explained by the definition of early RRT timing. However, we could not account for the observed heterogeneity by meta-regression according to study design, RRT modality, the study quality score, severity of the illness, and ICU admission type. Thereby, the heterogeneity observed is most likely explained by the differences in definitions for early RRT timing, the inability to account for heterogeneity in clinical practice and critical care patterns, many confounding factors that affect the mortality, publication bias, sample size and the inclusion of retrospective, prospective and RCTs.
The present systematic review has some limitations. Firstly, definitions for AKI are to some extent different in the included studies. Secondly, the definition of early RRT based on various arbitrary cutoffs for time, which ultimately downgraded the strength of evidence. Thirdly, there were publication bias and significant heterogeneity in the present study. Many confounding factors affect the mortality, and metaregression may not be enough to verify this issue. Lastly, the association with mortality is largely dependent on observational studies and might have been affected by allocation or selection bias. Thus, further high-quality RCTs focused on mortality according to the optimal time for starting RRT are necessary to fully understand the effects of early RRT for patients with AKI.

Participants, interventions and outcome measures
We included studies that evaluated the timing of initiation of RRT in patients with AKI. For the review, early and late RRT were defined based on criteria used by the authors in their studies. early and late RRT were defined as extended time-based cutoffs (arbitrary cutoffs for time from ICU admission or development of a biochemical "start time"), or biochemical indicators [serum creatinine, serum urea, RIFLE (risk, injury, failure, loss of function, and end-stage kidney disease) classifications, Acute Kidney Injury Network (AKIN) stages, urine output, and fluid balance]. Late RRT criteria also included conventional RRT indications (hyperkalemia, acidosis or fluid overload) and expectant care (no RRT initiated). The primary outcome was mortality, and the secondary outcomes were ICU and hospital LOS.

Searching strategies
We searched the Cochrane Library, EMBASE, Global Health, MEDLINE, PubMed, the International Clinical Trials Registry Platform, and Web of Science from January 1985 to November 2016. Owing to a low likelihood of relevance to modern RRT and critical care practices, studies published before 1985 were excluded in the present study. Keywords include acute renal failure/ acute kidney injury/renal insufficiency, mortality, renal replacement therapy/renal dialysis/ hemodialysis/dialysis. The related research references were also reviewed.

Inclusion and exclusion criteria
The inclusion criteria were as follows: (1) randomized clinical trials (RCTs) and/or observational cohort studies; (2) studies evaluating the timing of initiation of RRT in patients with AKI with direct effect on mortality; (3) complete data available to calculate odds ratio (OR) or mean difference (MD) with 95% confidence interval (CI); (4) clear definitions of AKI stated. Exclusion criteria were as follows: (1) data from the studies could not be extracted and analyzed; (2) duplicate publications; (3) non-human experimental studies.

Study selection and data extraction
Two investigators (Kaiping Luo and Shufang Fu) independently performed the study selection. All the disagreements were resolved by discussion. Data extraction included first author, year of publication, country, study design, sample size, age, sex, RRT modality, mortality, ICU LOS, hospital LOS, and definitions of early and late RRT.
Dr. Gaudry and colleagues [13] showed that the mortality was lower in the patients who never received RRT than those received RRT early or late. Patients who received RRT late were the most severely ill at baseline, and patients who never received it were less ill at baseline. More than 50% mortality in critically ill patients with AKI received RRT was confirmed by many randomized controlled trials [1,3,4,60]]. Thus, we hypothesized that critically ill patients with AKI who receive early RRT may decrease mortality, non-critically ill patients with AKI may confer survival benefits without early RRT. Subjects were identified as being of "critically ill patients" if the late RRT group with high mortality rates (≥ 50%), or "non-critically ill patients" if the late RRT group with low mortality rates (< 50%).

Quality assessment
The Cochrane Collaboration Risk of Bias tool was used to assess RCTs [62]. This tool consists of 6 domains and assesses 5 specific biases. A judgment of low risk, unclear risk, or high risk was provided for each domain. The Newcastle-Ottawa Scale (NOS) was used in the assessment of quality of cohort studies [63]. NOS quality assessment scale ranges from 0 to 9 stars. The star evaluates 3 main categories: selection, comparability, and outcome. www.impactjournals.com/oncotarget

Statistical analysis
Statistical analysis was performed using Review Manager (version 5.3) and STATA statistical software (version 12.0). We calculated OR with 95% CI for dichotomous data and MD with 95% CI for continuous data. Statistical heterogeneity of the data was quantified using the I 2 test, and the I 2 > 50% indicated significant statistical heterogeneity. Sensitivity analysis, metaregression analyses and subgroup analysis were conducted to investigate the potential sources of heterogeneity. Publication bias was assessed by constructing a funnel plot and using the Egger regression test and the Begg rank correlation test. A P value less than 0.05 was considered statistically significant.

CONCLUSIONS
Our data suggest that early RRT probably reduce the mortality, ICU and hospital LOS in critically ill patients with AKI. Inversely, early RRT in non-critically ill patients with AKI did not decrease the mortality, but shorted the ICU and hospital LOS.