Background

Acute kidney injury (AKI) is a common complication in the critically ill patients and associated with a substantial morbidity and mortality [1,2,3]. Severe AKI may be associated with up to 60% hospital mortality [4]. Over the years, renal replacement therapy (RRT) has emerged as the mainstay of the treatment for AKI. Intermittent hemodialysis (IHD), peritoneal dialysis (PD) and continues renal replacement therapy (CRRT) are various modalities to conduct RRT. Early initiation of RRT helps in the removal of uremic toxins, allow fluid and electrolyte balance and prevent life threatening complications such as metabolic encephalopathy, hyperkalemia, pulmonary oedema [5].

The timing of initiation of RRT for better patient outcome is still debatable with conflicting data from randomized controlled trials [6,7,8,9]. Two meta-analysis concluded that early RRT improves survival in critically ill patients [10, 11] . However, a recent meta-analysis [12] concluded that “early” initiation of RRT in critical illness complicated by AKI does not improve patient survival or confer reductions in intensive care unit (ICU) or hospital length of stay (LOS). This meta-analysis included both RCTs, and cohort studies. Moreover, after publication of this meta-analysis, two large studies have been published. We conducted an updated systematic review including RCTs and Quasi-RCTs (no observational studies) to support or refute the earlier evidence on the initiation of early versus late RRT. We have also performed a robust subgroup and sensitivity analysis as well as graded the quality of evidence and strength of recommendations by using GRADE approach which is lacking in previous systematic reviews and metanalysis.

Objective

To evaluate the impact of “early” versus “late” initiation of RRT.

Methods

The review has been registered at the PROSPERO register: CRD42016043092

Type of studies

Randomized controlled trials and quasi-randomized trials (RCTs) were included.

Participants

Hospitalized patients with AKI were included. Patients with preexisting chronic kidney disease (estimated glomerular filtration rate [GFR] <30 mL/min) on long term dialysis, previous renal replacement therapy, AKI resulting from vascular malformations (occlusion of the renal artery), glomerulonephritis, interstitial nephritis, vasculitis, post-renal obstruction, hemolytic uremic syndrome (HUS) or thrombotic thrombocytopenic purpura, post renal transplant AKI and confirmed or suspected pregnancy, malignancy and HIV were excluded.

Interventions

The interventions consist of administration of early/prophylactic or as and when required/late RRT in patients with AKI. The definition of early and late RRT was taken as described in the individual study.

Types of outcome measures

Primary outcomes

  1. 1.

    Mortality rate

  2. 2.

    Dialysis Dependence at 3 month

Secondary outcomes

  1. 1.

    Length of ICU stay

  2. 2.

    Length of hospital stay

  3. 3.

    Recovery of renal function

  4. 4.

    Adverse events

Search methods for identification of studies

Cochrane Central Register of Controlled Trials (CENTRAL), PubMed/MEDLINE, Google Scholar, Cochrane renal group were searched from 1970 to October 2016. Following search strategy was applied: (((((((((renal replacement therapy) OR Renal Dialysis) OR dialysis) OR Hemodialysis) OR Hemodiafiltration) OR Hemofiltration)) AND (((((acute kidney injury) OR Acute Renal Injury) OR Acute Renal Insufficiency) OR Acute Renal Failure) OR Acute Kidney Failure)) AND (((((((timing) OR time) OR Initiation) OR start) OR early) OR Earlier) OR Late)) AND ((randomized controlled trial) OR Controlled Clinical Trial). To identify unpublished trial results, we searched the US National Institutes of Health, Department of Health and Human Services trials registry (http://www.clinicaltrials.gov/) and the WHO International Clinical Trials Registry Platform (ICTRP) trial registry.

Data extraction

Data was extracted using a pilot tested data extraction form. Two authors independently extract data including author, type of participants, exposure and intervention (modality of RRT, timing), results (clinical outcomes and adverse events).

Risk of bias (quality) assessment

Two review authors (GC and RD) independently assessed the methodological quality of the selected trials by using Cochrane risk of bias tool [13].

Strategy for data synthesis

The data from various studies was pooled and expressed as mean difference (MD) with 95% confidence interval (CI) in case of continuous data, and risk ratio (RR) with 95% CI in case of categorical data. P-value <0.05 was considered significant. Heterogeneity was assessed by I-squared statistics. In case of high level heterogeneity (>50%), we tried to explore the cause. A fixed effects model was initially conducted. If, significant heterogeneity was found between trials, potential sources of heterogeneity were considered and where appropriate, a random effects model was used. RevMan (Review Manager) version 5.3 was used for all the analyses.

Subgroup analysis

We performed the following subgroup analysis:

  1. 1.

    Surgical versus mixed medical admission.

  2. 2.

    Modality of RRT

  3. 3.

    Severity of illness

  4. 4.

    Risk of bias for allocation concealment

Publication bias

This was looked by construction of the inverted funnel plot as suggest by Egger et al.[14].

Grade of evidence

For assessment of the quality of evidence we used GRADE Profiler software (version 3.2). The software uses five parameters for rating the quality of evidence. The parameters used were - limitations to design of randomized controlled trials, inconsistency of results or unexplained heterogeneity, indirectness of evidence, imprecision of results, and publication bias. The rating was done as – no, serious, and very serious limitation.

Results

Description of the studies

Of the 547 citation retrieved, full text of 44 articles was assessed for eligibility (Fig. 1). Of these a total of 10 RCTs with 1,672 participants were included. Thirty three studies were excluded due to following reasons: Non RCT/review articles (n = 18), comparing different modalities of RRT (n = 09), comparing different doses/drugs during RRT (n = 05); Ongoing studies (n = 02). All the trials were open label with most of the trials having unclear or high risk of bias for allocation concealment. There was a variable definition of early versus late in different studies. Thus, the definition of early or late was taken according to individual study definition. A summary of the studies is provided in Table 1.

Fig. 1
figure 1

PRISMA flow diagram

Full size image
Table 1 Characteristics of the included studies
Full size table

Primary outcome measure

Overall Mortality: Ten studies with 1672 participants reported 662 deaths. Compared with the patients assigned to late RRT, patients assigned to early RRT had 7% reduction in mortality rate. However, pooled results showed no significant difference between the two groups (RR, 0.93;95% CI: 0.75, 1.15) (Fig. 2). Since there was a significant heterogeneity (I2 = 50%;p = 0.17), we tried to explore the heterogeneity based on pre-specified subgroups analysis such as: Surgical versus mixed patients, severity of illness, modality of RRT and risk of bias for allocation concealment. We also performed a period wise mortality analysis to address the heterogeneity in the included trials.

Fig. 2
figure 2

Forest plot showing overall mortality

Full size image

Day 30 mortality: This was reported in 6 trials [5,8,, 7–9, 15, 16] with 1301 participants. The pooled results showed 8% decrease in mortality with early initiation of RRT. However, there was no significant difference between the early and late RRT (RR, 0.92;95% CI: 0.68, 1.06], (Additional file 1: Figure S1 a)

Day 60 mortality: This was reported in 3 trials [8, 9, 15] with 1075 participants. The pooled results showed no significant difference between the two strategies (RR, 0.94; 95% CI:0.78, 1.14) (Additional file 1: Figure S1a).

Day 90 mortality: This was reported in three trials [9, 15, 17] with 555 participants. The pooled results showed no significant difference between the two strategies (RR, 0.94; 95% CI:0.67, 1.33) (Additional file 1: Figure S1a) .

Overall ICU mortality: Overall ICU mortality was reported in 3 trials [7, 15, 17]. Pooled mortality showed no significant reduction in ICU mortality with initiation of early RRT (RR, 1.08; 95% CI:0.84, 1.39) (Additional file 1: Figure S1a).

Overall hospital mortality: This was reported in 6 trials and the pooled results showed no significant difference between the mortality rates between the two groups (RR, 1.07; 95% CI: 0.81, 1.42) (Additional file 1: Figure S1a).

Dialysis Dependence at Day 90: 3 trials [6, 9, 17] reported dialysis dependence at 90 days in the two groups. Pooled data showed no significant difference between the two groups (RR, 1.06 95% CI: 0.53, 2.12) (Additional file 1: Figure S1c)

Subgroup based on Surgical versus mixed patients

Overall 30 day mortality: 2 trials reported this outcome [15, 16]. Overall there was no significant difference in overall 30 day mortality (RR,0.51;95% CI: 0.09, 3.08.) (Additional file 1: Figure S1b)

Overall 60 day mortality: Only 1 trial [15] reported this outcome. Overall there was no significant difference between the two groups (RR, 1.14;95% CI: 0.83, 1.58)

Overall 90 day mortality: 1 trial [15] reported this outcome without any significant difference between the two strategies.

Overall ICU: There was no significant difference between ICU (RR,1.11;95% CI:0.82, 1.52) or hospital mortality (RR,1.01;95% CI:0.74, 1.36) in the surgical patients undergoing early versus late initiation of RRT.

Subgroup analysis based on severity of illness

There was no significant difference in overall day 30 mortality (5 trials, 1257 patients;RR,0.91;95% CI:0.73, 1.15);day 60 mortality (3 trials, 1075 participants; RR, 0.90; 95% CI:0.64, 1.27); day 90 mortality (3 trial, 555 participants, RR, 0.90; 95% CI:0.49, 1.64), hospital mortality (RR, 1.12; 95% CI: 0.76, 1.65) and ICU mortality (RR, 1.12;95% CI:0.75, 1.68) in critically ill undergoing early RRT as compared to late RRT.

Subgroup analysis based on modality of RRT

There was no significant difference in overall day 30 mortality in the patients undergoing continuous renal replacement therapy (CRRT) (3 trials, 413 participants; RR, 0.90;95% CI:0.65, 1.26); patients undergoing intermittent hemodialysis (IHD) (1 trial,44 participants; RR, 0.16;95% CI:0.02, 1.17); patients undergoing either CRRT or IHD (1 trial,620 participants; RR, 0.95;95% CI:0.79, 1.14): day 60 mortality in patients undergoing CRRT (2 trials, 455 participants; RR, 0.93;95% CI:0.62, 1.38) or patients undergoing either CRRT or IHD (1 trial,620 participants; RR, 0.97;95% CI: 0.82, 1.14): day 90 mortality in patients undergoing CRRT (2 trials, 455 participants; RR, 0.92;95% CI:[0.56, 1.50) or patients undergoing either CRRT or IHD (1 trial,100 participants; RR, 1.03;95% CI:0.62, 1.71): Overall hospital mortality in the patients undergoing CRRT (2 trials, 330 participants; RR, 1.14;95% CI:0.88, 1.48) or IHD (1 trial,44 participants; RR, 0.16;95% CI:0.02, 1.17) and overall ICU mortality in the patients undergoing CRRT (2 trial, 330 participants; RR, 1.13;95% CI: 0.85, 1.49) or patients undergoing either CRRT or IHD (1 trial, 100 participants; RR, 0.88;95% CI: 0.47, 1.63) (Fig. 3)

Fig. 3
figure 3

Forest plot showing subgroup analysis based on modality of RRT

Full size image

Subgroup analysis based on risk of bias for allocation concealment

Six trials have low risk of bias for allocation concealment [6, 8, 9, 15, 17] while 4 [5, 7, 16, 18, 19] have unclear or high risk of bias for allocation concealment. There was a significant reduction in overall mortality in the patients assigned to early RRT in the studies with high or unclear risk of bias (RR, 0.74; 95% CI:0.59, 0.91) as compared to those with low risk of bias for allocation concealment (RR, 1.02;95% CI:0.89, 1.17) (Fig. 4)

Fig. 4
figure 4

Forest plot showing subgroup based on risk of bias for allocation concealment

Full size image

Secondary Outcomes

Length of ICU stay: Six studies reported this outcome [8,16,, 9, 15–17]. Out of these, 5 trials reported this outcome as median (interquartile range) [7,8,9, 15, 17] and found no significant difference between ICU stay in the two groups. Another trial [16] reported a significant reduction in ICU stay in the patients undergoing early RRT as compared to late RRT (MD,-45.87;95% CI:-75.54,–16.20).

Length of Hospital stay: Seven trials reported this outcome [8,16,, 9, 15–17] and 5 reported them as median (Interquartile range). In 4 trials there was no significant difference in the length of hospital stay between the two groups while 1 trial has shown significant difference between hospital stay in patients receiving early RRT. Two trials [7, 18] have given this outcome as mean (SD) and were pooled. The pooled results no difference in the length of hospital stay (MD,-3.62; 95% CI :-8.91, 16.16).

Recovery of renal function by day 90: 2 trials reported recovery of renal functions on day 90 [9, 15]. Pooled data showed no significant difference between two groups (RR, 1.04;95% CI:0.80, 1.35) (Additional file 1: Figure S1d).

Adverse events

Bleeding: 7 trials with 1520 participants reported this outcome [6,7,8,9, 15, 17, 18]. On pooling the data no significant difference in the adverse event was observed between the two groups (RR, 0.92;95% CI:0.67, 1.25) (Fig. 5).

Fig. 5
figure 5

Forest plot showing adverse events

Full size image

Catheter related complications: Four trials reported this outcome [6, 9, 15, 17]. The pooled results showed no significant difference between the two groups (RR, 1.41: 95% CI: 0.59, 3.37) with point estimate favouring late strategy (Fig. 5).

Thrombocytopenia: 3 trials reported this outcome [8, 9, 17, 18] and the pooled results showed no significant difference between the two strategies (RR,1.20:95% CI:0.87, 1.65) (Fig. 5).

Arrhythmias: 4 trials reported this outcome [8, 9, 17, 18] and the pooled results showed no significant difference between the two groups (RR, 0.91;95% CI: 0.70, 1.19) (Fig. 5).

Hypotension: 4 trials reported this outcome [6, 9, 15, 17] reported this outcome. Pooled results showed no significant difference with point estimate favouring late RRT (RR, 1.18; 95% CI: 1.00, 1.38) (Fig. 5).

Electrolyte abnormalities: There was no significant difference between the two strategies with respect to hypokalemia (RR, 1.02;95% CI: 0.51, 2.03), hyperkalemia (RR, 0.80;95% CI: 0.45, 1.41) and hypocalcaemia (RR, 0.89;95% CI:0.51, 1.54). Hypophosphatemia was seen more in patients undergoing early dialysis (RR, 1.51; 95% CI: 1.05, 2.18) (Fig. 5).

Publication bias

To assess whether there was a bias in the published literature, funnel plot was constructed using the MD and 1/SE values obtained from trials measuring one of the primary outcome (overall mortality). In the absence of a publication bias, such a plot is expected to have a shape resembling an inverted funnel. From the funnel plot generated, the possibility of publication bias in the analysis is less (Fig. 6).

Fig. 6
figure 6

Funnel plot

Full size image

Grade of evidence

The evidence generated was of “low quality” for all the primary outcomes (GRADE Table 2).

Table 2 Grade of evidence for primary outcomes
Full size table

Discussion

Summary of Evidence

After an extensive search of literature we could find 10 trials to be eligible for inclusion. Our results indicates that in patients with AKI there is no benefit of early initiation of renal replacement therapy on overall mortality, dialysis dependence on day 90, length of ICU or hospital stay and renal recovery on day 90. There was no significant difference in the adverse events between early and late group except for hypophosphatemia which was seen more common in the patients undergoing early RRT. The grade evidence generated was low grade for most of the outcomes.

Studies exploring the initiation strategies for RRT have shown conflicting results. Early initiation of RRT, theoretically may allow for better control of fluid and electrolyte status, fasten removal of uremic toxins and prevent complications like gastric hemorrhage and metabolic encephalopathy [20]. On the other hand a delayed strategy of RRT initiation may give sufficient time for spontaneous patient recovery and may avoid the need for RRT, thus minimizing risk associated with RRT [9].

Two recent trials [8, 9] have also shown conflicting results regarding timing of initiation of RRT. Zarbock et al. [9] (ELAIN trial) reported a significant reduction of mortality over 90 days in critically ill patients with AKI undergoing RRT while in the study by Gaudry et al. [8] (AKIKI trial) the authors found no significant reduction in mortality in patients assigned to early RRT as compared to late RRT. This difference may be due to different patient’s characteristics such as inclusion of more ill patients in the ELAIN trial as compared to that in AKIKI trial (SOFA 16 versus SOFA 11) [21]. Another difference was the use of RRT modality in the two studies. In the AKIKI trial 55% of the patients received intermittent hemodialysis as RRT modality while all the patients received CRRT in ELAIN trial. However, we have done a subgroup analysis based on the modality of RRT, severity of illness and type of patients and found no difference in mortality rates among the two groups. A recent systematic review has shown a benefit of early RRT on reduction of all cause mortality [22]. However, greater heterogeneity in the studies and a combined analysis of both RCTs and non- RCTs together may have overestimated the effect. Further, on subgroup analysis based on the type of studies (RCTs versus non RCTs), authors found no statistically significant decrease in the mortality rate in RCT group.

On subgroup analysis based on risk of bias for allocation concealment we found a significant reduction in mortality (26%) in the patients assigned to early RRT. Previous studies have also shown that treatment inadequate allocation concealment may exaggerate treatment effect by 40% and unclear allocation concealment may exaggerate treatment effect by 30%.

The strength of present systematic review is:1) we have included both randomized and quasi-randomized controlled trials to strengthen the present evidence2) we have done sensitivity analysis by excluding trials with unclear and high risk of bias for allocation concealment3) we also assigned GRADE evidence to further grade the quality of evidence and recommendations.

Conclusion

This updated meta-analysis showed no added benefit of early initiation of RRT for patients with AKI with respect to all cause mortality, dialysis dependence, and recovery of renal functions or hospital stay. The grade evidence generated was of “low quality” and there was high heterogeneity in the included trials. We need more good quality RCTs in different patient subgroups including children to further strengthen the evidence.