Background
Acute kidney injury (AKI) is now recognized as a major public health problem affecting millions of patients worldwide [
1]. Critically ill patients are at high risk of developing AKI, with its incidence during intensive care unit (ICU) stay varying from 36% to 67% [
2,
3]. During the last few years, acute-onset disturbance of kidney function has been a subject of avid scientific discussion, which has led to the definition of “acute kidney injury.” AKI identification was based on changes in serum creatinine (SCr) compared with baseline levels before the disease onset and changes in diuresis. Scoring systems for AKI quantification have been developed at consensus conferences. These included the RIFLE [
4] and AKIN [
5] criteria for AKI. Most recently, the AKIN criteria were revised and clarified as the Kidney Disease Improving Global Outcomes (KDIGO) criteria for AKI [
6].
Although significant advances have been achieved in AKI research following this classification, potential pitfalls can be identified in clinical practice. Intuitively, the shorter the amount of time during which a determined SCr change occurs, the greater the AKI severity. For example, going from a SCr of 1 to 1.5 mg/dl within 12 h signifies a worse glomerular filtration rate (GFR) fall than going from a SCr of 1 to 1.5 mg/dl within 48 h (see Additional file
1 for illustrative examples); however, if the same urinary output is maintained in both situations, AKI severity will be classified similarly. Also, even considering the difficulty in ascertaining a baseline SCr, the AKI score systems do not consider previous underlying chronic kidney disease (CKD) and its possible prognostic implications. To exemplify, a patient whose SCr varied from 0.8 to 1.2 mg/dl has the same AKI severity as another patient whose variation was from 2 to 3 mg/dl, although the GFR is clearly more severely reduced in the second case. Finally, as suggested by Waikar and Bonventre [
7] and demonstrated by our group [
8], an SCr kinetic model can be superior to AKI classification systems in patients with previous CKD.
Assessing the GFR is problematic when the SCr is changing quickly. In severe AKI and anuric patients, it is a consensus to consider that the GFR is < 10 ml/min/1.73 m
2. However, a less reduced GFR may also affect management and impact patient survival. Recently, the nonsteady-state (kinetic) estimated glomerular filtration rate (KeGFR) has been advocated in AKI and renal recovery assessment [
9,
10]. The formula is derived from the initial SCr, the distribution volume, the creatinine production rate, and the quantitative difference between consecutive SCr over a given period. Taking these variables into account, KeGFR yields the measured creatinine clearance (CrCl) rate for that period between two SCr measurements. Thus, the KeGFR results in the same interpretation of a measured CrCl level, but without the need for collecting urine and measuring urinary creatinine levels. Using this approach, we can estimate the GFR in a determined time interval, regardless of whether Scr is slowly increasing as described in the abovementioned examples, where a patient whose SCr levels increased from 1 to 1.5 mg/dl in 12 h had a worse KeGFR in comparison with another whose SCr level also increased from 1 to 1.5 mg/dl in 48 h.
In the present study, we hypothesized that a worse KeGFR could add clinical and prognostic information in critically ill patients beyond the current AKI classification system, mainly regarding the need for renal replacement therapy, hospital mortality, and 1-year survival.
Methods
Multiparameter Intelligent Monitoring in Intensive Care II database and data collection
The Multiparameter Intelligent Monitoring in Intensive Care (MIMIC)-II project, maintained by the Massachusetts Institute of Technology Laboratory for Computational Physiology, contains data on patients hospitalized in an ICU at Beth Israel Deaconess Medical Center from 2001 to 2008 [
11]. The database is freely available so that any researcher who accepts the data use agreement and has attended “protecting human subjects’ training” can apply for permission to access the data. This study was approved by the institutional review boards of Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center and was granted a waiver of informed consent.
We included all patients with an ICU length of stay (LOS) lasting more than 48 h with at least three SCr measurements taken. Patients with known end-stage renal disease (ESRD), previous renal transplantation, those who underwent renal replacement therapy (RRT) before ICU admission, and those with admission SCr > 4 mg/dl were excluded.
Data collection
All data were extracted from the MIMIC-II database (v2.6) and included demographic information (e.g., age, gender) and clinical information from the admission notes. The following admission data were collected: admission body weight, admission type (elective or emergency), care unit type (medical, coronary unit, surgery, or cardiac surgery), sepsis diagnosis as described by Angus et al. [
12], admission SCr, and disease severity as assessed by the Simplified Acute Physiology Score (SAPS) II [
13] and Sequential Organ Failure Assessment (SOFA) [
14] scores. In the first 7 days of ICU stay, we also recorded daily SCr measurements, and the need for vasoactive drugs and mechanical ventilation.
Estimated kinetic glomerular filtration rate
The KeGFR was calculated during the first 7 days of ICU stay according the following equation:
$$ KeGFR=\frac{baseline\; SCr\;X\; eGFR}{Mean\; SCr} $$
$$ x\left[1\hbox{-} \frac{24\mathit{\mathsf{x}}\varDelta \mathsf{SCr}}{\varDelta Time(h)\mathit{\mathsf{x}} Ma\mathit{\mathsf{x}}\varDelta SCr/ day}\right] $$
where eGFR = estimated glomerular filtration rate using baseline SCr, mean SCr = mean of two consecutive SCr measurements, ΔSCr = change in SCr, ΔTime(h) = interval in hours between two consecutive SCr measurements, and MaxΔSCr/Day = the maximal change (increase) in SCr that can occur per day if renal function is completely lost.
The KeGFR was derived from the initial SCr, the distribution volume, the creatinine production rate, and the quantitative difference between consecutive SCr measurements over a given time. We included all SCr levels measured at least 6 h and no more than 48 h apart. KeGFR was calculated by taking each interval between two consecutive creatinine measurements. The volume of distribution for creatinine does not need to be equated with total body water, but can be expressed as a function of the creatinine production rate. The amount by which a known creatinine production rate can increase the creatinine concentration if all excretion has ceased (i.e., near-zero GFR) informs us about the volume of distribution. Since there is only creatinine addition and no subtraction, this situation describes the maximum increment of SCr in 1 day. To obtain it, we identified 94 patients with anuria and two SCr measurements apart with no RRT in this interval. The mean SCr incremental corrected for 24 h was 1.47 ± 0.44 mg/dl for men and 1.41 ± 0.49 mg/dl for women. We used these values instead of a fixed value of 1.7 mg/dl per day, as described by Chen [
9].The other necessary variables for this formula included baseline SCr. Because the MIMIC-II does not provide any laboratory information prior to ICU admission, the lowest SCr available during ICU stay before RRT initiation was used as the baseline renal function and baseline CrCl was calculated using the CKD-EPI formula [
15]. Any SCr measurement after RRT was not considered. After that, to exclude any influence from previous CKD or out-of-ICU-acquired AKI, we performed a complete sensitivity analysis using only patients admitted to the ICU with an eGFR > 70 ml/min/1.73 m
2.
AKI definition
AKI was defined according to the KDIGO criteria [
6]. We classified patients based on the KDIGO maximum stage achieved within the first 7 days of ICU stay. Because we used RRT as an outcome, we did not apply it as a rule to patients that commenced RRT before achieving AKI stage 3. Urinary output was collected in fixed blocks of 6 h beginning at ICU admission. To be acceptable, the maximum gap between two actual values was 3 h. To stage a patient based on urine output (UO), a minimum of 6 h of data were required. Since Kellum et al. [
16] have recently described that the risk of death over the index hospital stay and over the following year is greatest for patients that meet both UO and SCr criteria for AKI, and to exclude the fact that KeGFR is the only other approach to describe SCr increment, we performed a second sensitivity analysis with SCr-based KDIGO criteria only.
Estimated GFR using steady-state SCr formula
Although KeGFR theoretically provides better estimation of GFR than formulae that were developed to be used with steady-state SCr, we evaluated the capacity of the most recently proposed formula (CKD-EPI) [
15] using the highest SCr during the first 7 days of ICU stay to predict the main outcomes.
Outcomes
Patient outcomes included the need for RRT during ICU stay, hospital mortality, and survival up to 1 year.
Statistical analysis
Patients were categorized in groups according to the worst achieved KeGFR. Variables were assessed for normality using the Kolmogorov-Smirnov test. Parametric variables were compared using a t test and nonparametric variables using the Mann-Whitney test. Categorical variables were compared using the chi-square test. We built a logistic regression model to assess the association between categorized KeGFR and hospital mortality according to each AKI KDIGO stage. We defined a priori that the following variables would be included in the logistic regression model for both outcomes: age, gender, SAPS II score, SOFA score, main comorbidities (hypertension, congestive heart failure, cardiac arrhythmias, chronic pulmonary obstructive disease, diabetes mellitus, lymphoma, metastatic cancer, liver disease, obesity), type of admission (clinical or surgical), vasoactive drugs, the need for mechanical ventilation, and baseline eGFR. A Cox model was performed to access survival by AKI severity and lowest KeGFR after adjusting for comorbidities, baseline eGFR, and age.
Discussion
In this study, the performance of KeGFR in critically ill patients was evaluated for the first time. We found that the worst achieved KeGFR within the first 7 days of ICU stay was associated with several short- and long-term outcomes, such as the need for RRT, hospital mortality, and 1-year survival. Moreover, the worst KeGFR appears not to substitute for, but adds prognostic information to the current AKI classification.
Although significant advances have been made in the diagnosis and prognosis of AKI since the development of the consensus classification system, several questions remained when evaluating patients in this setting. First, as stated in the introduction section, current AKI classifications are not able to discern prognosis between patients with pure AKI or acute-on-chronic kidney disease [
8]. Another possible pitfall concerns the time patients take to fully develop AKI severity, as exemplified in the introduction section when one patient had an increment of 50% of baseline SCr within 12 h and another within 48 h, but both were classified as KDIGO stage 1. Theoretically, calculating KeGFR even when SCr changes acutely can avoid these gaps found in the AKI classification system.
Our results demonstrate several important findings. First, we disclosed a disagreement between AKI severity and the worst achieved KeGFR. Several patients had AKI KDIGO stage 3, but maintained KeGFR greater than 70 ml/min/1.73 m
2. This can be explained by a slow increment of SCr over time. For example, one patient had a baseline SCr of 0.6 mg/dl and it increased only approximately 0.3 mg/dl each 48 h, going up to 1.8 mg/dl after 7 days. This patient was classified as AKI stage 3, but his KeGFR was never lower than 70 ml/min/1.73 m
2. On the other hand, other patients had severely reduced KeGFR but no or only a minor AKI stage. Clearly, some of these patients already had reduced eGFR at baseline. However, when evaluating only those patients admitted with eGFR above 70 ml/min/1.73 m
2 (Table
3) we can identify that most of these patients had no eGRF reduction at baseline. In these cases, the increase in SCr was not so great, but occurred within a short time interval (for example, an increment of 0.3 mg/dl in two consecutive SCr measurements, obtained 8 h apart can reduce the KeGFR to less than 30 ml/min/1.73 m
2, but this patient will be classified as only AKI stage 1).
Although the great majority of SCr measurements in the present study had an interval between them of 20 to 28 h, we maintained all measurements with an interval between 6 and 48 h, making it possible to evaluate the patients earlier, within the first 12 h after ICU admission, when SCr can already be increasing. At this time, it is possible there is not enough time for SCr to increase for the KDIGO system to achieve even AKI stage 1, although KeGFR can already be severely reduced.
Secondly, and perhaps most importantly, both the AKI classification system and KeGFR seem to be complementary in predicting outcomes. For example, almost 45% of patients with AKI stage 3 and KeGFR < 30 ml/min/1.73 m2 needed RRT in comparison with less than 10% of patients with AKI stage 3 but less severe KeGFR reduction and less than 5% of those patients with KeGFR < 30 ml/min/1.73 m2 but only AKI stage 1/2.
In relation to hospital mortality, a stepwise reduction in the worst achieved KeGFR conferred an incremental risk of death to each AKI stage in both uni- and multivariate analyses. It is already known that AKI classification systems are not as good at predicting events in patients with previous CKD [
7,
8]. Moreover, it has been recently suggested that different AKI patterns in relation to SCr trajectory (resolving/nonresolving) imply different prognoses [
17]. Analyzing the KeGFR equation, it contains two important pieces of information not contemplated in the AKI classification systems: baseline eGFR and the speed of SCr increase. In part, it is probable that KeGFR adds prognostic information because it can identify patients with previous CKD. However, our data suggest that to correctly quantify renal injury in critically ill patients we must take into consideration not only the SCr increment degree but also the speed at which this increment occurs, as suggested in the introductory section of this manuscript. Supported by the groups shown in Table
3, it is important to highlight that we do not propose substituting the AKI classification with KeGFR, but we think both must be used together—the first to evaluate the magnitude of the acute injury and the latter to measure the effects of AKI on GFR.
Because it is clear that oligoanuric patients had GFR close to zero and it is more difficult to ascertain eGFR in these patients who maintain UO, we performed a sensitivity analysis excluding those patients with AKI KDIGO stage 3 according to the UO. Generally, the results were maintained, mainly when evaluating the need for RRT (almost 50% of patients with both AKI stage 3 according to the Scr criterion and a KeGFR < 30 ml/min/1.73 m2).
Finally, we also evaluated long-term mortality. Except for groups 1 and 2, there was a clear separation in survival lines according to the classification by AKI KDIGO stage/worst KeGFR. These results emphasize that AKI severity alone does not determine long-term outcome but that an interaction between baseline GFR and AKI severity and the speed of onset of AKI are important to assess both short- and long-term prognosis.
Our study has several and important limitations. First, and most important, we did not have access to previous SCr measurements thus making it impossible to identify patients with actual previous CKD. To overcome this fact, we performed a sensitivity analysis including only patients with an eGFR > 70 ml/min/1.73 m
2. We considered the lowest SCr available during ICU stay as the baseline, although this approach can inflate the AKI incidence, indicating that such a level is often lower than the most recent outpatient creatinine value [
18]. Another limitation is that it is difficult to determine the actual maximal increase in SCr when eGFR is near zero in critically ill patients and, consequently, the total body water volume as described in the methodology section. Although it has been suggested to limit this increment to 1.7 mg/dl a day [
9,
10], we analyzed a subset of patients with anuria and no RRT to determine the mean value of daily SCr increment (a real measure of maximal SCr increment) and used different means for men and women, although we acknowledge this can change according to obesity status, age, and other factors. While we have identified KeGFR as a prognostic tool in risk stratification regarding the need for RRT and survival, identifying patients at high risk and highlighting the importance of implementing measures that prevent/limits further renal damage [
19], we recognize that, regarding the further practical use of KeGFR for drug dose correction, validation studies using standard GFR measurements (by measuring CrCl or using exogenous substances such as inulin, iohexol, and others) are warranted.