The RIFLE versus AKIN classification for incidence and mortality of acute kidney injury in critical ill patients: A meta-analysis

The sensitivity and accuracy of the Risk/Injury/Failure/Loss/End-stage (RIFLE) versus acute kidney injury Network (AKIN) criteria for acute kidney injury (AKI) in critically ill patients remains uncertain. Therefore, we performed a systematic review and meta-analysis to investigate the incidence and prognostic value of the RIFLE versus AKIN criteria for AKI in critically ill patients. Literatures were identified by searching Medline, Embase, PubMed, and China National Knowledge Infrastructure (CNKI) database. Nineteen studies with 171,889 participants were included. The pooled estimates of relative risk (RR) were analyzed. We found that the RIFLE and AKIN criteria is different for the incidence of AKI in intensive care unit (ICU) patients (P = 0.02, RR = 0.88), while not for cardiac surgery patients (P = 0.30, RR = 0.93). For AKI-related hospital mortality, the AKIN criteria did not show a better ability in predicting hospital mortality in either ICU (P = 0.19, RR = 1.01) or cardiac surgery patients (P = 0.61, RR = 0.98) compared to RIFLE criteria. Our findings supported that the AKIN criteria can identify more patients in classifying AKI compared to RIFLE criteria, but not showing a better ability in predicting hospital mortality. Moreover, both RIFLE and AKIN criteria for AKI in cardiac surgery patients had better predictive ability compared with the ICU patients.

also proposed to evaluate the hospital mortality and outcome of AKI patients by the two classifications. However, the results are still inconclusive. In the present study, we aimed to perform a systematic review and pool the available data to evaluate the incidence and prognostic value of the RIFLE and AKIN classification for AKI patients.

Results
Flow and Study Characteristics. The literature search yielded 299 articles, 19 studies 12-30 were eligible for inclusion, and flow diagram of included/excluded studies was showed in Fig. 1. Twelve studies for Intensive care units (ICU) with 138,521 patients, 7 studies for cardiac surgery with 33,038 patients. There are 8 studies from Asia, 5 from Europe, both America and Australia have 2 studies, each of Brazil and Bitlis had one study. 15 studies were retrospective, and 4 studies were prospective, all included studies are compare RIFLE with AKIN for AKI patients in ICU or cardiac surgery, the basic characteristics of the included studies are summarized in Table 1. Among these studies, 7 studies were multicenter, 12 studies were from single center, 14 studies use hospital mortality to evaluate mortality endpoint, 1 study uses mortality of ARF, and 1 study with 28-day mortality, 3 studies did not acquire the endpoint information. The quality of the including studies by QUDAS II was listed below (Supplementary figure 1).

The RIFLE VS. AKIN classification for the incidence of AKI in ICU patients. Twelve studies with
138,521 patients were included for the incidence of AKI in ICU patients, 50,054 patients were diagnosed AKI with RIFLE classification and 50,521 patients were diagnosed AKI with AKIN classification. As shown in Tables 2 and 3, the incidence of AKI range from 18.13%-66.67% and 24.14%-76.67% by RIFLE and AKIN classification, respectively. The total incidence of AKI diagnosed by RIFLE and AKIN classification showed a significance difference (RR, 0.88; 95%CI, 0.80-0.98; P = 0.02; Fig. 2), for each stage, the Risk VS. Stage 1 (RR, 0.70; 95%CI, 0.53-0.93; P = 0.02; Fig. 2) and the Injury VS. Stage 2 (RR, 1.29; 95%CI, 1.17-1.43; P < 0.00001; Fig. 2) for the incidence of AKI were showed a significant difference, but the Failure VS. Stage 3 (RR, 0.90; 95%CI, 0.73-1.11; P = 0.34; Fig. 2) for the incidence of AKI in ICU patients were not showed significant difference.  The RIFLE VS. AKIN classification for the hospital mortality of AKI in cardiac surgery patients. Six studies to evaluate the mortality for AKI in cardiac surgery patients, 9,530 patients diagnosed with AKI using the RIFLE classification and 821 patients died in hospital, 837 patients died in hospital with a total of 9,591 AKI patients using AKIN classification. As showed in Tables 3 and 4, the total hospital mortality in cardiac surgery patients by RIFLE and AKIN classification were not showed significant difference, the RR ratio were 0.98, 95%CI, 0.89-1.07; P = 0.61; Fig Area under the receiver operator characteristics (AuROC) curves for the RIFLE and AKIN classification for hospital mortality of AKI. The AuROC curve for incidence was 0.598 for RIFLE classification (95%CI, 0.592-0.603, P< 0.0001) and was 0.594 for AKIN classification (95%CI, 0.589-0.600, P < 0.0001) for hospital mortality of AKI in ICU patients, whereas the AuROC curve was 0.762 (95%CI, 0.743-0.782, P < 0.0001) for RIFLE classification and was 0.761 for AKIN classification (95%CI, 0.741-0.781, P < 0.0001) for hospital mortality of AKI in cardiac surgery patients, although the AuROC curve was not significant between RIFLE and AKIN classification in either ICU patients or cardiac surgery patients, the AuROC of both RIFLE and AKIN classification for cardiac surgery patients had better predictive ability compared with the ICU patients (Fig. 6).
Publication bias. Begg's funnel plot was performed to access the publication bias of the literature. The shapes of the funnel plots revealed some evidence of obvious asymmetry. The funnel plot of the incidence of AKI in ICU patients and the incidence of AKI in cardiac patients were showed in Supplementary  Figure 2 and 3, respectively.

Discussion
Acute kidney injury is very common with high hospital mortality in critically ill patients 31 , epidemiological studies demonstrate the wide variation in etiologies and risk factors, developing into chronic kidney disease and progression to dialysis dependency 32,33 . However, there is lack of a universally accepted and standardized definition for AKI for nephrologists and health care workers. The Acute Dialysis Quality Initiative's RIFLE criteria has been validated in several clinical settings and shown to correlate with important outcomes, such as needing for renal replacement therapy (RRT), length of hospital stay, and mortality 22,[34][35][36] . But it is still imperfect. In 2007, the AKIN convened to refine the RIFLE criteria. A few modifications were added to AKI including the eliminating the change of glomerular filtration rate (GFR) and the outcome categories of Loss and ESRD; the stage 1 was redefined with an absolute increase in creatinine of at least 0.3 mg/dl; patients starting RRT are automatically classified as stage 3. Both the RIFLE and AKIN classifications use the changes of serum creatinine or urine output to establish the clinical syndrome of AKI in 3 severity levels. But whether the sensitivity of AKIN in the diagnosis of AKI in ICU patients has clinical significance, or the new classification can representative the severity of AKI patients and good predictive value for prognosis is not very clear. A number of epidemiologic studies have tried to compare the RIFLE and AKIN criteria for the incidence and in-hospital mortality of AKI in critical ill patients and cardiac surgery patients [12][13][14]25,27,28 . Bagshaw and their college work revealed that the AKIN criteria do not improve the sensitivity and predictive ability of classification of AKI in the first 24 h after admission to ICU compared to the RIFLE criteria 12 . However, Ratanarat et al. 23 found that AKIN criteria improved sensitivity for detection of AKI and prediction of in-hospital mortality was better than that of RIFLE criteria in critically ill patients with multi-organ dysfunction syndrome, Lopes et al. 13 , Zhang et al. 17 and Jiang et al. 16 showed that although AKIN criteria improved sensitivity of AKI diagnosis but does not improve ability in predicting in-hospital mortality of critically ill patients. In cardiac surgery patients, Haase et al. 25 reported that the AKIN classification do not materially improve the clinical usefulness of RIFLE definition, however, Yan et al. 26 found that the AKIN criteria seem not to have greater sensitivity and specificity compared with the RIFLE classification.
In the present study, 19 studies with more than 171,559 participants are included in our meta-analysis. We have consistently confirmed that patients with AKI, the RIFLE and AKIN classification is different for the incidence of AKI in ICU patients, the incidence of AKI diagnosed by AKIN classification is higher than the RIFLE classification, in subgroup analysis, we found that Risk vs. Stage 1 and Injury vs. Stage 2 were also different, but the Failure vs. Stage 3 did not show a statistical significance. This may be explained by the change criteria in stage 1 and the elimination of GFR, thus more patients were diagnosed as AKI by AKIN classification, and there still some patients needs RRT for fluid overload were stratified in Stage 3 without increased serum creatinine which cannot be included in Failure stage by RIFLE. All this contributes to the high incidence of AKI by AKIN criteria. In cardiac surgery patients with AKI, the RIFLE and AKIN classification did not show difference for the diagnosis, only Injury vs. Stage 2 has a significant difference by subgroup analysis while the other stages did not. Comparing to cardiac surgery patients, the AKI incidence is higher in ICU patients, although some cardiac surgery patients were transferred to ICU. ICU is mixed with all kind of critical ill patients such as septic shock, acute respiratory distress syndrome, or hepatic cirrhosis. We also evaluated the in-hospital mortality,   38 showed that AKI incidence was highest according to the KDIGO definition (18.3%) followed by the AKIN (16.6%), and RIFLE (16.1%). In addition, another retrospective observational study of 49,518 admissions indicated that 11.6% were diagnosed with KDIGO criteria, 11.0% were diagnosed with RIFLE criteria, and only 4.8% were diagnosed with AKIN criteria 39 . In critically ill patients, KDIGO was more predictive than the RIFLE criteria, but there was no significant difference between AKIN and KDIGO 40 , Luo et al. found that the incidence of AKI using the RIFLE, AKIN, and KDIGO criteria were 46.9%, 38.4%, and 51%, respectively 40 . Moreover, for in-hospital mortality, only small differences in predictive abilities between RIFLE and KDIGO concerning clinical outcomes at 30 days in acute decompensated heart failure patients 41 . Therefore, all these definitions have their own limitation for wide accepted, new classification was needed to establish an early diagnosis of AKI that would be a simple and useful clinical tool.
The major limitation is that most of the included studies were retrospective, and the AKI incidence was a large range in different medical center, which cause a heterogeneity. Second, some studies were multicenter which increased the weight in the meta-analysis. Third, we only evaluate the AKI in ICU and cardiac patients, the other diseases or syndromes associated acute kidney injury were not included in our meta-analysis. At last, A publication bias may have occurred. The funnel plot shows significant evidence of the bias ( Supplementary Figures 2 and 3).
In conclusion, our study found that the AKIN criteria can identify more patients in classifying AKI in ICU patients compared to RIFLE criteria but not cardiac patients, for the prediction of AKI-related mortality, the AKIN criteria did not show a better ability in predicting hospital mortality in both ICU and cardiac surgery patients compared to RIFLE criteria. But both the RIFLE and AKIN classifications for AKI in cardiac surgery patients had better predictive ability compared with the ICU patients.

Materials and Methods
Data Sources, Search Strategy, and Selection Criteria. We performed a systematic search to identify the studies examined the RIFLE and AKIN criteria for acute kidney injury. Literatures were identified by searching MEDLINE via Ovid, EMBASE, PubMed, and China National Knowledge Infrastructure (CNKI) database. The last updated search was performed on November 1st, 2013. The searching terms were "Risk or Injury or Failure or RIFLE", "AKIN or Acute Kidney Injury Network" and "acute renal failure or acute kidney injury or AKI". We manually searched the references of the identified studies and review articles, and academic congresses on kidney disease with available data were also included. The search was limited to compare the RIFLE and AKIN classification for AKI in ICU and cardiac surgery patients out without restriction on language.
Data extraction. Two authors independently extracted the information from all eligible publications using standard data extraction forms. Disagreement was resolved by discussion between the two authors, or the consultation with a third reviewer. The standardized data form was used for data collection, including first author, year of publication, country of origin, ethnicity of the study. The data of baseline serum creatinine, length of stay and renal replace therapy were recorded when available. All completed studies that compare AKI was defined and classified by the RIFLE criteria and the AKIN criteria in ICU or cardiac patients were eligible for inclusion.   the pooled estimate of relative ratios (RRs) and 95% confidence intervals (CIs) were calculated for AKI patients with the two classifications. We also looked at the RRs comparing the three RIFLE classes Risk, Injury, and Failure with AKIN stages (Risk VS. Stage 1, Injury VS. Stage 2 and Failure VS. Stage 3). Data were combined using a random effects model or fixed effects model according to the I 2 test, I 2 value 25%, 50% and 75% correspond to low, medium and high levels of heterogeneity, Fixed effects model was selected while the I 2 under 50%, otherwise the a random effects model will be used 42 . Analysis was performed with Stata software, version 12. We used QUDAS II to evaluate the quality of the including studies and Asymmetry funnel plots were used to assess potential publication bias by Revman software, version 5.3. Model fit was assessed by the goodness of-fit test, and discrimination was assessed by the area under the receiver operator characteristic (AuROC) curve (SPSS version 20). The p value less than 0.05 was considered as a statistical significance.