Introduction
Acute kidney injury (AKI) is a common clinical syndrome in the intensive care unit (ICU) and associated with an increase in morbidity, mortality and length of stay [
1]. The Risk, Injury, Failure, Loss and End-Stage Renal Disease (RIFLE) classification system developed in 2004 by the Acute Dialysis Quality Initiative (ADQI) [
2,
3] is a consensus definition for the diagnosis of AKI. The severity grades risk, injury and failure are defined on the basis of the changes in serum creatinine (SCr) or urine output where the worse of each criterion is used (Table
1). If a reliable baseline SCr is unknown, ADQI suggests the calculation of a theoretical baseline value by the modification of diet in renal disease (MDRD) equation [
4]. RIFLE is the first widely accepted AKI definition, validated in over half a million patients worldwide [
5‐
7]; however, the urine criteria are frequently discarded [
8‐
16]. Notably, transient oliguria occur frequently in ICU patients and its use often identifies a higher percentage of AKI patients compared to SCr alone [
17‐
19].
Table 1
Risk, Injury, Failure, Loss and End-stage Kidney (RIFLE) classification [
2]
Risk | ↑ SCr ≥1.5 × from baseline | <0.5 ml/kg/h ≥6 h |
Injury | ↑ SCr ≥2 × from baseline | <0.5 ml/kg/h ≥12 h |
Failure | ↑ SCr ≥3 × from baseline or an acute ↑ SCr ≥44 μmol/l from baseline SCr ≥354 μmol/l | <0.3 ml/kg/h ≥24 h or anuria ≥12 h |
Loss | Complete loss of kidney function >4 weeks |
End-stage | End-stage kidney disease >3 months |
We hypothesized that discarding the urine criteria not only decreases the estimated incidence of AKI but also increases the time to AKI diagnosis.
We determined the time to reach AKI diagnosis (AKI-0) in a heterogeneous ICU population admitted to the ICU for more than 48 hours using both RIFLE methods (with and without urine output). Additionally, we assessed the impact of these two RIFLE methods on the incidence and grading of AKI.
Materials and methods
Study design and setting
We performed anonymous analysis of routinely collected clinical data. The Medical Ethics Review Committee of our institution waived the need for informed consent. The study was carried out between April 2009 and August 2009 in the ICU of the Academic Medical Center, a major university hospital in Amsterdam with a 28-bed general, multidisciplinary closed format ICU. During the study period all patients receiving ICU treatment for more than 48 hours were eligible for enrolment. Patients with known end-stage renal disease or receiving renal replacement therapy were excluded.
Data collection
Demographic data, clinical history (including the lowest documented SCr within six months of ICU admission), and severity of illness were recorded on ICU admission. For each patient the lowest documented SCr within six months of hospital admission was recorded (pre-ICU admission SCr). The estimated SCr baseline was calculated from the MDRD equation assuming a GFR of 75 ml/min/1.73 m
2 (MDRD
75) [
4]. Urine output was measured hourly by visual readings of the amount of urine accumulated in a urine metre. Fluid balance, SCr and the presence of renal replacement therapy (RRT) were documented daily. We did not record details of type of fluid administration, use of diuretics and other medications.
Assessment of acute kidney injury
During the first seven days of ICU treatment patients were scored daily for AKI based on RIFLE using SCr and urine output criteria (RIFLE
SCr+UO) and based on the RIFLE SCr criteria only (RIFLE
SCr)
. The lesser of pre-ICU admission SCr and ICU admission SCr served as baseline renal function. If pre-ICU admission SCr was unknown the baseline was taken as the minimum between the MDRD
75 based and ICU admission SCr [
20].
For each patient we determined the number of days elapsed until AKI was first diagnosed (AKI-0) according to the two RIFLE methods. In addition, we classified patients into four grades according to their maximum RIFLE grade: no AKI, risk, injury and failure. Patients receiving RRT therapy were classified as having failure.
Statistical analysis
Statistical analyses were performed in the statistical environment R version 2.10.1 (R Foundation for Statistical Computing, Vienna, Austria) [
21] and we used the "boot" library for performing the bootstrap procedures. Data are presented as number and percentage, mean ± SD, or median and quartiles. The baseline characteristics of the patients with and without a pre-ICU admission baseline SCr were compared using the t-test (for normally distributed quantities) or the Mann-Whitney U-test and the proportion test (for proportions). We tested differences between the two RIFLE methods for the following outcomes:
a) Difference in the distribution of first day on which AKI was diagnosed
b) Difference in the distribution of the maximum RIFLE grade
To measure the differences in the distributions a) and b), we calculated the 95% confidence interval (CI) around the two-sample Kolmogorov-Smirnov D statistic and the
P-value associated with the null-hypothesis that D = 0, that is, that there are no differences between the methods for the two distributions. To obtain D's 95% CI we used the standard bootstrap procedure [
22] with 3,000 bootstrap samples. A bootstrap sample has the same size as the original dataset and is obtained by random re-sampling, with replacement, from the original dataset. To obtain
P-values for D we use a permutation test in which we construct 3,000 permutation re-samples and calculate the proportion of times in which the Kolmogorov-Smirnov statistic for the permutation was larger than D.
c) Difference in incidence of AKI and AKI associated mortality
To determine the difference in AKI incidence, that is, having or developing AKI in the first seven days of hospital stay, we again used the basic bootstrap procedure with 3,000 samples [
22]. This allowed us to obtain the incidence and variance of AKI in the whole sample in the first seven days of the hospital stay. To test the difference in mortality rate we used the proportion test.
d) Difference in diagnosis of AKI in individual patients
To test differences in concordance between RIFLE methods in diagnosing AKI in individual patients we used the McNemar test. The following example illustrates the difference between incidence of AKI in a sample and AKI diagnosis in individual patients: if method M1 diagnoses three patients as "AKI", "non-AKI" and "AKI", and method M2 diagnoses these same patients as "AKI", "AKI" and "non-AKI" respectively, then the incidence of AKI in both methods is equal (two out of three), but the individual diagnoses are different for the second (non-AKI, AKI) and third (AKI, non-AKI) patient. The diagnoses are hence concordant only in the first patient.
e) Difference in fluid balances
To test differences in fluid balance in patients classified according to both RIFLE methods, we calculated fluid balance both on the first day of AKI diagnosis and cumulative (from ICU admission up to the first day of AKI diagnosis). For comparison we used the Mann-Whitney U-test.
For all analyses, P < 0.05 was considered to indicate statistical significance.
Discussion
The RIFLE classification is the first widely accepted definition for AKI; however, many studies have applied RIFLE incorrectly without the use of urine output [
7]. We performed a prospective observational study and compared AKI diagnosis based on RIFLE
SCr+UO with that based on RIFLE
SCr. The two RIFLE methods resulted in statistically significantly different outcomes for incidence of AKI, diagnosis of AKI for individual patients, time to diagnosis of AKI and maximum RIFLE grade. Discarding the RIFLE urine output criteria for AKI diagnosis significantly underestimated the presence of AKI on admission and during the first ICU week (
P < 0.001), and significantly delayed the diagnosis of AKI (
P < 0.001). In our study, the use of RIFLE
SCr instead of RIFLE
SCr+UO resulted in fewer patients diagnosed with mild AKI (AKI-risk and AKI-injury) and more patients having no AKI. A total of 102 (39%) patients never had AKI during the first ICU week according to RIFLE
SCr, while these patients were indeed diagnosed as having AKI based on RIFLE
SCr+UO. The question arises of whether at least some of the oliguric patients without an increase in SCr actually did have AKI, or whether they were oliguric for some other reason (for example, their hydration status) [
23,
24]. In our patients, AKI-0 was diagnosed based on a decrease in urine output without a rise in SCr in 132 (51%) patients. In 9 (7%) of these patients CVVH was subsequently started before a rise in SCr while in 24 patients (18%) SCr rose in the next one to three days reaching the RIFLE
SCr criteria. Eight (6%) had persistent oliguria and died without a rise in SCr and 91 (69%) patients recovered and never reached the RIFLE
SCr Risk criteria. The majority (83%) of patients diagnosed with AKI based on RIFLE
SCr+UO had positive fluid balances on the day AKI was diagnosed.
These findings suggest that for mild AKI the patient's urine output criterion does not match well with the patient's respective creatinine criterion. Our findings confirm prior observations [
19,
25]. In the small (
N = 75) prospective observational study by Macedo
et al., 28% of patients were diagnosed with AKI based on the SCr criteria only, in comparison to 55% when using only the urine output criteria [
25]. In the recent multicentre observational study by Prowle
et al., AKI diagnosis based on SCr was infrequent, while oliguria was relatively common [
19].
In the present study, the applied RIFLE method also affected the time to diagnosis of AKI. In comparison with RIFLE
SCr+UO, the use of RIFLE
SCr increased the time to AKI diagnosis and resulted in fewer patients with AKI on admission: 210 (81%) patients had AKI during the first week of ICU according to RIFLE
SCr+UO while only 108 (42%) patients had AKI according to RIFLE
SCr. Of note, on the day of ICU admission 63 (24%) patients had AKI according to RIFLE
SCr while 116 (45%) patients had AKI according to RIFLE
SCr+UO. According to RIFLE
SCr, 45 patients developed AKI after ICU admission and in 53% of these patients AKI would have been diagnosed at least one day earlier based on the RIFLE urine criteria. Our findings are congruent with the recent prospective study by Macedo
et al. in 317 critically ill surgical patients, showing that oliguria diagnosed AKI earlier in comparison with the SCr criterion [
26].
Our findings are not surprising. Different definitions lead to different answers. An important factor is why most studies did not apply the recommended consensus urine output criteria [
3]. The catalyst for the changes in SCr in the consensus definition came from Chertow's paper: a solid statistical argument [
27]. In contrast, the urine output criteria arrived via expert opinion; however, there is always the possibility that it is wrong. In addition, measuring urine output is tedious and it is still unclear how the hourly criteria should be applied (continuously or for each six-hour period of the day), with or without diuretics. Many studies omitted the urine criteria because they retrospectively applied the RIFLE criteria to existing databases that did not capture either any urine output criteria or only captured urine output data in a form that cannot be applied. The big question remains - does it really matter and why? We need to know whether defining AKI with or without including urine output actually leads to a difference in AKI-outcome associations. In the present study, ICU mortality in patients with AKI was significantly higher when AKI was diagnosed by RIFLE
SCr (38%) compared to that based on RIFLE
SCr+UO (24%)
. Similar differences are also suggested by two large multicenter epidemiologic studies by Hoste
et al. (AKI based on RIFLE
SCr+UO) and Uchino
et al. (AKI based on RIFLE
SCr) [
16,
20]. In these two studies, baseline mortality in non-AKI patients was comparable; however, mortality in the AKI-risk, -injury and -failure group was much higher in the cohort studied by Uchino
et al., despite the fact that the latter was a hospital-wide population and the former a general ICU population [
16,
20]. Similarly, the systematic review by Ricci
et al. showed that the relative risk for death among studies that used RIFLE
SCr+UO was lower than in those using RIFLE
SCr [
6]. In the present study, mortality in the Risk and Injury groups was higher when AKI was based on RIFLE
SCr, while in the Failure group mortality was higher when AKI was based on RIFLE
SCr+UO. AKI-associated mortality, however, was not part of our primary hypothesis and the small number of patients in each RIFLE stratum keep us from any conclusions.
In addition to its effect on AKI-associated mortality, the nonuse of the urine criterion may also influence the diagnostic accuracy of new biomarkers for AKI, including neutrophil gelatinase-associated lipocalin (NGAL) and cystatin C [
11,
28‐
31]. Serum cystatin C was found to be a good predictor for AKI (without urine criteria) in the study by Herget-Rosenthal [
11], while cystatin C was a poor predictor for AKI (with urine criteria) in the study by Royakkers
et al. [
31]. In addition to case mix, the opposite findings of both studies may also be caused by the application of two different RIFLE methods (with and without urine output criteria).
To apply the SCr criteria of RIFLE information on prior renal function is needed. When a pre-ICU admission SCr is not available, ADQI suggest that the baseline SCr be estimated from the MDRD formula [
2]. Zavada
et al. showed that estimating baseline SCr may over- or underestimate AKI [
32]; however, in another study by Bagshaw
et al. [
33], estimating baseline by the MDRD equation appeared to perform reasonably well for determining the RIFLE categories as long as the pre-ICU admission GFR was near normal. In our study, a pre-ICU admission SCr was available in 101 (39%) patients and in these patients the difference between pre-ICU SCr and estimated SCr was not statistically significant (90 ± 34 μmol/L versus 88 ± 12 μmol/L,
P = 0.39. However, SCr level on ICU admission was significantly higher than the pre-ICU admission SCr (125 ± 121 μmol/L versus 90 ± 34 μmol/L,
P < 0.01). Of note, in the present study, 81 (51%) out of the 159 patients with an unknown prior SCr had lower SCr at ICU admission compared with the estimated SCr. Although this issue is not discussed by the ADQI, in these patients we used the lower SCr level as suggested by Hoste
et al. [
20].
In the present study patients receiving CVVH were classified as Failure as suggested by the acute kidney injury network (AKIN) [
3]; however, in the original RIFLE system introduced by the ADQI, renal replacement therapy was not included as a distinct stage [
2]. Indeed, it may be questionable to classify patients as Failure if they did not achieve the specific RIFLE score. In our study, using RIFLE
SCr+UO, 67% of the patients had Failure and 33% had Injury at the start of CVVH. In contrast, using RIFLE
SCr, only 39% of the patients had Failure, 27% had Injury, 16% had Risk and 18% had no AKI. Given the variability in the timing of renal replacement therapy worldwide, it may be more appropriate to always report the AKI stage at the start of renal replacement therapy in future epidemiologic studies.
Our study is the first study comparing the effects of two RIFLE methods (with and without urine output criteria) on time to AKI diagnosis as well as AKI incidence, AKI associated mortality and maximum AKI grade. We, however, recognize the limitations of our study. First, our study is single-centre, including a limited number of patients. Second, SCr was measured daily, while urine output was measured hourly. More frequent SCr measurements may result in earlier detection of AKI. Third, although we recorded fluid status, we did not evaluate whether our patients received diuretics. However, although the use of diuretics is common practice worldwide, their use is not explicitly addressed in the RIFLE criteria. Fourth, we did not correct SCr for hemodilution. A positive fluid balance may cause dilution of SCr and, therefore, a delay in the diagnosis based on RIFLE
SCr [
18]. Two studies showed that hemodilution of SCr may affect AKI diagnosis [
18,
34]. The basis for the development of the RIFLE classification, however, was Chertow's paper [
27] showing that a small rise in SCr increased mortality, and this paper did not correct for hemodilution. In addition, estimating the dilution factor in critically ill patients is notoriously difficult. Fifth, we did not specifically evaluate patients with chronic kidney disease because this subgroup was too small in our sample. Last, our results were statistically significant; however, future research will need to study the clinical significance in more detail.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
KW helped in acquisition of data, performed the statistical analysis, interpreted the results and drafted the manuscript. AAH supervised and performed parts of the statistical analysis, interpreted the results and was involved in critically revising the manuscript. MS participated in acquisition of data and revision of the manuscript. RC and MJS were involved in critically revising the manuscript. CB conceived the study, participated in the design of the study and acquisition of data, interpreted the results and drafted the manuscript. All authors read and gave final approval of the version of the manuscript to be published.