Introduction
Acute renal failure (ARF) is as an abrupt decline in kidney function. Although simple to define conceptually, there has long been no consensus on an operational definition of ARF. As reported in a recent survey, more than 35 definitions have been used so far [
1]. Depending on the definition used, ARF has been shown to affect from 1% to 25% of intensive care unit (ICU) patients and has led to mortality rates from 15% to 60% [
2].
Because the lack of a uniform definition is a major impediment to epidemiological research in the field, the Acute Dialysis Quality Initiative Group (ADQIG) [
3] recently proposed consensus definition criteria, namely, the RIFLE criteria based on three grades of increasing severity (Risk of renal dysfunction, Injury to the kidney, and Failure of kidney function) and two outcome classes (Loss of kidney function and End-stage kidney disease) (Table
1). Furthermore, they proposed that the old nomenclature ARF be replaced by the term acute kidney injury (AKI) to encompass the entire spectrum of the syndrome, from minor changes in renal function to need for renal replacement therapy (RRT).
Table 1
RIFLE classificationa
Risk | Increase in serum creatinine ≥1.5 × baseline or decrease in GFR ≥25% | <0.5 ml/kg/hour for ≥6 hours |
Injury | Increase in serum creatinine ≥2 × baseline or decrease in GFR ≥ 50% | <0.5 ml/kg/hour for ≥12 hours |
Failure | Increase in serum creatinine ≥3 × baseline or decrease in GFR ≥75% or serum creatinine ≥350 μmol/L with an acute rise of at least 44 μmol/L | <0.3 ml/kg/hour for ≥24 hours or anuria ≥12 hours |
Loss | Complete loss of kidney function >4 weeks |
End-stage kidney disease | Need for RRT >3 months |
The RIFLE classification is undoubtedly a major advance in that it allows easier comparisons across studies. Overall, it seems to correlate well with patients' outcomes [
4‐
9]. In the ICU setting, only four multiple-center studies using the RIFLE criteria have been published so far [
10‐
13]. All but one [
12] found AKI to be associated with a poor outcome, with some residual heterogeneity regarding both incidence and mortality, however. In addition, estimates of AKI-associated mortality in these studies derived from traditional logistic regression or Cox models, while concerns about their reliability have been raised recently [
14]. Briefly, logistic regression analysis ignores the timing of events and their chronological order, potentially leading to an overestimation of the association between a specific risk factor (for example, nosocomial pneumonia) and mortality [
15]. This problem can be solved to some extent by applying the Cox model, which allows for the consideration of time-dependent covariates. Yet, this model does not deal with the competing risks issue. This issue arises when more than one endpoint is possible [
16]. Typically, "dying in hospital" and "discharge alive" are two competing risks. If "dying in hospital" is the event of interest, the nonfatal competing event "discharge alive" hinders the event of interest from occurring as a first event.
Statistical models able to handle time-dependent covariates and allowing the simultaneous analysis of different endpoints (that is, competing risks) are now available [
15,
17‐
19]. In recent years, these models have engendered growing interest in hospital epidemiology (especially with regard to cancer research) but have rarely been used in the ICU field.
The aim of this study was to further assess the association between AKI defined by RIFLE criteria and in-hospital mortality in critically ill patients by using such an original competing risks approach.
Discussion
The association of AKI with critically ill patients' outcomes has been widely investigated, but very few multiple-center evaluations using the RIFLE criteria have been published so far [
10‐
13]. Our study, carried out in a large cohort of general ICU patients, supports the use of RIFLE as a classification tool and confirms previous evidence that AKI negatively influences patients' outcomes.
The originality of our results lies mainly in the original competing risks approach. This approach has many potential advantages over the commonly used logistic regression and Cox models. Actually, logistic regression has been reported to cause loss of information because it yields a time-independent probability of dying and ignores the timing of events and their chronological order [
27,
28]. While the Cox model may partially alleviate these limits, it has been shown to overestimate the incidence of the event of interest, with most of the overestimation being related to the rate of the competing event [
29]. By contrast, the Fine and Gray model adequately addresses time spent in the hospital as a risk factor for mortality by considering death hazard rates and takes into account the time-varying exposure status, thus avoiding a potential misjudgment in terms of time-dependent bias [
30,
31]. Moreover, it provides a more accurate estimation of mortality because death hazard rates are not confounded by the competing event "discharge alive."
In keeping with the few similar multiple-center evaluations that have used the RIFLE criteria [
10,
11,
13], we found that AKI was an overall predictor of poor outcomes (it must be noted, however, that results regarding crude hospital mortality rates vary considerably from one study to another, indicating residual heterogeneity despite the use of consensual definition criteria) and that mortality differed according to the maximum RIFLE class reached during the ICU stay. Of note, even moderate renal dysfunction (R class) impaired patients' prognosis as previously shown [
10,
13,
32], and, interestingly, the SHRs for I and F classes were similar. These data suggest, similarly to the study by Ostermann
et al. [
13], that the maximum risk of death might be reached as soon as patients are in I class of the RIFLE criteria. Thus, therapeutic and preventive strategies, such as optimization of hemodynamic parameters and avoidance of nephrotoxic drugs, must undoubtedly be in order at an early stage of renal dysfunction to prevent further aggravation and to reduce the risk of death.
Despite its strengths, our study has potential limitations. First, the definition of AKI was not based on the most recent consensus criteria proposed by the Acute Kidney Injury Network (AKIN) group [
33]. The main differences between the AKIN and RIFLE classifications are as follows: a smaller change in serum creatinine level (>26.2 μmol/L) used to identify patients with stage 1 AKI (analogous to the RIFLE Risk class), a time constraint of 48 hours for the diagnosis of AKI and any patient receiving RRT classified as having stage 3 AKI. However, compared to the RIFLE criteria, there is currently no evidence that the AKIN criteria improve the sensitivity, robustness and predictive ability of the definition and classification of AKI in the ICU [
34‐
36]. This is consistent with our finding that maximum renal dysfunction during the ICU stay was reached within a two-day period in most patients. Furthermore, classifying any patient receiving RRT in stage 3 is questionable and may introduce bias because of the lack of uniform recommendations regarding the timing and modalities of RRT.
Second, assessing baseline creatinine values by the MDRD equation as in previous reports may have exposed our study methodology to the risk of inclusion of patients with modest chronic disease not captured by the APACHE II definitions as having end-stage renal disease or RRT dependence. This is a potential source of misclassification bias [
37] and underestimation of the association between AKI and hospital mortality. This issue needs further investigation.
Third, we encountered the same problem as others did [
9,
13]: the 6- and 12-hour urine outputs were not recorded in our database. Therefore, patients were classified according to the GFR criteria only. Patients classified according the GFR criteria seem to be more severely ill and have slightly higher mortality rates than their counterparts classified according to the urine output criteria [
11,
38,
39]. Consideration of both criteria may have resulted in a lowest estimation of the risk of death (and, conversely, a higher incidence of AKI).
Fourth, the potential confusing role of RRT was not evaluated. However, the extent to which RRT interferes with AKI patients' prognosis remains unclear, and practices regarding this technique vary widely from one institution to another. Consequently, considering RRT as a confounder could have led to hazardous conclusions. This issue deserves further specific evaluation.
Fifth, although it is multicentered, our database is not multinational. So, our population may not be representative of ICU patients in other countries. Nevertheless, the baseline characteristics, AKI incidence and proportion of patients receiving RRT were similar to those reported in previous studies [
11,
13].
Finally, we did not have any information as to the exact etiology of AKI, although sepsis was probably the commonest one. Of note, a recent study revealed that RIFLE classification can be used to evaluate the overall prognosis of septic patients, suggesting a close link between AKI and sepsis [
40]. However, AKI often results from a combination of several risk factors whose respective contributions are difficult to determine. Whether any of these risk factors plays a preponderant role (or whether patients' prognosis differs according to the cause of AKI) remains unknown.
Acknowledgements
We are indebted to the persons listed below for their participation in the OUTCOMEREA study group.
Scientific committee:
Jean-François Timsit (Hôpital Albert Michallon and INSERM U823, Grenoble, France), Elie Azoulay (Medical ICU, Hôpital Saint Louis, Paris, France), Yves Cohen (ICU, Hôpital Avicenne, Bobigny, France), Maïté Garrouste-Orgeas (ICU Hôpital Saint-Joseph, Paris, France), Lilia Soufir (ICU, Hôpital Saint-Joseph, Paris, France), Jean-Ralph Zahar (Microbiology Department, Hôpital Necker, Paris, France), Christophe Adrie (ICU, Hôpital Delafontaine, Saint Denis, France), and Christophe Clec'h (ICU, Hôpital Avicenne, Bobigny, and INSERM U823, Grenoble, France).
Biostatistical and informatics expertise:
Jean-Francois Timsit (Hôpital Albert Michallon and INSERM U823, Grenoble, France), Sylvie Chevret (Medical Computer Sciences and Biostatistics Department, Hôpital Saint-Louis, Paris, France), Corinne Alberti (Medical Computer Sciences and Biostatistics Department, Robert Debré, Paris, France), Adrien Français (INSERM U823, Grenoble, France), Aurélien Vesin INSERM U823, Grenoble, France), Christophe Clec'h (ICU, Hôpital Avicenne, Bobigny, and INSERM U823, Grenoble, France), Frederik Lecorre (Supelec, France), and Didier Nakache (Conservatoire National des Arts et Métiers, Paris, France).
Investigators of the OUTCOMEREA database:
Christophe Adrie (ICU, Hôpital Delafontaine, Saint Denis, France), Bernard Allaouchiche (ICU, Edouard Herriot Hospital, Lyon), Caroline Bornstain (ICU, Hôpital de Montfermeil, France), Alexandre Boyer (ICU, Hôpital Pellegrin, Bordeaux, France), Antoine Caubel (ICU, Hôpital Saint-Joseph, Paris, France), Christine Cheval (SICU, Hôpital Saint-Joseph, Paris, France), Marie-Alliette Costa de Beauregard (Nephrology, Hôpital Tenon, Paris, France), Jean-Pierre Colin (ICU, Hôpital de Dourdan, Dourdan, France), Mickael Darmon (ICU, CHU Saint Etienne), Anne-Sylvie Dumenil (Hôpital Antoine Béclère, Clamart France), Adrien Descorps-Declere (Hôpital Antoine Béclère, Clamart France), Jean-Philippe Fosse (ICU, Hôpital Avicenne, Bobigny, France), Samir Jamali (ICU, Hôpital de Dourdan, Dourdan, France), Hatem Khallel (ICU, Cayenne General Hospital), Christian Laplace (ICU, Hôpital Kremlin-Bicêtre, Bicêtre, France), Alexandre Lauttrette (ICU, CHU G Montpied, Clermont-Ferrand), Thierry Lazard (ICU, Hôpital de la Croix Saint-Simon, Paris, France), Eric Le Miere (ICU, Hôpital Louis Mourier, Colombes, France), Laurent Montesino (ICU, Hôpital Bichat, Paris, France), Bruno Mourvillier (ICU, Hôpital Bichat, France), Benoît Misset (ICU, Hôpital Saint-Joseph, Paris, France), Delphine Moreau (ICU, Hôpital Saint-Louis, Paris, France), Etienne Pigné (ICU, Hôpital Louis Mourier, Colombes, France), Bertrand Souweine (ICU, CHU G Montpied, Clermont-Ferrand), Carole Schwebel (CHU A Michallon, Grenoble, France), Gilles Troché (Hôpital Antoine, Béclère, Clamart France), Marie Thuong (ICU, Hôpital Delafontaine, Saint Denis, France), Guillaume Thierry (ICU, Hôpital Saint-Louis, Paris, France), Dany Toledano (CH Gonnesse, France), and Eric Vantalon (SICU, Hôpital Saint-Joseph, Paris, France).
Study monitors:
Caroline Tournegros, Loic Ferrand, Nadira Kaddour, Boris Berthe, Samir Bekkhouche, Sylvain Anselme.
OUTCOMEREA is a nonprofit organization supported by nonexclusive grants from four pharmaceutical companies (Aventis Pharma, Wyeth, Pfizer, and MSD) and by research grants from three publicly funded French agencies (Centre National de la recherche Scientifique [CNRS], Institut National pour la Santé et la Recherche Médicale [INSERM], and the French Ministry of Health).
Authors' contributions
CC designed the study and wrote the manuscript. CC, JFT and MNM performed the statistical analyses. FG, AL, MGO, SJ, DGT, ADD, FC, RHR and EA participated in the collection of data and critically revised the manuscript for important intellectual content. All authors read and approved the final manuscript.