Acute kidney injury is an independent risk factor for pediatric intensive care unit mortality, longer length of stay and prolonged mechanical ventilation in critically ill children: a two-center retrospective cohort study

Acute kidney injury (AKI) is well known to be associated with longer hospital length of stay (LOS), morbidity and mortality in adults [1, 2]. In recent years, it has become apparent that even very small increases in serum creatinine (SCr) levels in hospitalized adults and in children undergoing cardiac surgery are associated with poor hospital outcomes [3, 4]. Most previous research in pediatric AKI was focused on patients requiring acute dialysis. Recently derived AKI definitions now allow us to comprehensively evaluate risk factors for and outcomes of AKI [5, 6], which is a necessary first step to perform prior to initiating clinical trials to reduce AKI incidence and improve AKI outcomes.

Single-center studies have been performed in select pediatric populations using one of two similar standardized AKI definitions: the pediatric Risk, Injury, Failure, Loss, End-Stage Renal Disease criteria (pRIFLE) [5] or the Acute Kidney Injury Network (AKIN) staging system [6]. These studies have reported variable AKI incidence in critically ill children, largely due to variations in illness severity and in inclusion/exclusion criteria of study populations. However, all of these studies have found that AKI is associated with poor outcomes (such as longer LOS or mortality) [4, 7‐9]. Schneider et al. [8] recently published the largest single-center database study of all children admitted to the critical care unit. They reported an overall AKI rate of 10% and found that AKI was independently associated with increased mortality and hospital LOS. This study was extremely important because of its large study population and because of its detailed description of the timing of AKI development.

There is still much to learn about AKI epidemiology and disease description in children. First, all previous studies were single-center studies; therefore, validation of findings from previous studies using large cohorts is necessary. Second, it is still unclear to what extent the use of differing definitions for baseline kidney function affects AKI disease descriptions and AKI associations with outcomes when using large database studies. This is particularly relevant, since baseline kidney function is often unknown in children and the definition of AKI is structured upon examining SCr increases relative to baseline SCr (bSCr) values.

The following were the aims of the current study: (1) to determine the incidence of AKI according to standard AKI definitions and to elucidate AKI risk factors in two tertiary healthcare center pediatric intensive care units (PICUs) for the purpose of validating and comparing findings from other recent studies, (2) to determine to what extent the use of different ways of estimating bSCr levels affects inferences regarding the association between AKI and outcomes and (3) to evaluate how early small increases in SCr might predict the future development of more significant AKI.

We have validated findings from previous single-center reports that AKI is common and independently associated with poor PICU outcomes. In addition, we evaluated two approaches to assessing bSCr which are likely to be used for either prospective or retrospective database studies and found that, regardless of the measurement method used, AKI-outcome associations persist. We also found that in a substantial proportion of children, SCr level does not return to baseline by the time of PICU discharge.

Our finding that AKI occurs extremely early after PICU admission validates findings of other single-center studies [5, 8, 12]. The implication of this finding is of paramount importance. A voluminous amount of research has been devoted to early AKI detection by using novel biomarkers of AKI [17‐21], since SCr is thought to increase late with AKI and, for AKI treatments to be successful, early biomarker application is necessary [22, 23]. However, applying these new biomarkers to the PICU population will clearly be a great challenge if most children already have AKI within 24 to 48 hours of PICU admission. Future PICU AKI clinical trials may need to apply new AKI biomarkers even prior to ICU admission. In a previous study of children undergoing cardiac surgery, we found that early increases in SCr (< 50%) moderately predicted future overt AKI [4]. In the analyses presented in this article, we found that in the subgroup of children who did not have AKI at the time of PICU admission, the percentage increase in SCr had moderate predictive value for determining who would or would not develop AKI in subsequent days (AUC = 0.67). Specifically, a 25% SCr increase predicted future AKI with specificity of 81% but sensitivity of only 33%. Therefore, the presence of a 25% increase may be useful for predicting future AKI development, but lack of a 25% increase does not provide much information. This simple and inexpensive way of predicting AKI events should also be examined in adults.

There is almost no information available on what happens to children with AKI after the AKI event occurs. Yet, in adults, it is known that AKI is a significant predictor of postdischarge mortality and chronic kidney disease [24, 25]. As a first step toward understanding the natural history of PICU AKI, we have shown that SCr does not return to bSCr values at the last PICU SCr assessment in many children who develop AKI and that those with worse AKI are less likely to return to bSCr values. This finding has several potential implications. First, it is possible that children whose SCr levels do not return to bSCr concentrations may be those at highest risk for long-term renal abnormalities. This should be confirmed in prospective studies, and our results provide a starting point for determining the at-risk population that might be studied. Second, this finding may also demonstrate a problem in the process of care which requires future study. Anecdotally, children who leave the PICU with "normal" SCr levels likely do not have follow-up examinations of renal function. However, clearly, a substantial portion of children do not leave the PICU with "normal renal function" compared to bSCr levels.

We evaluated two approaches to estimating bSCr. This follows from our previous research demonstrating that, depending on which bSCr measurement is used, AKI patterns, incidence and outcome associations may differ substantially [14]. However, bSCr is often unknown in children admitted to the PICU; therefore, applying the "true" bSCr to determine AKI status is extremely challenging. Moreover, in future studies, which may be based on provincial or state databases wherein prehospitalization SCr data may not be readily available, it is extremely worthwhile to understand the extent to which alternative methods of estimating bSCr may affect study findings. As shown in the current study, bSCr was unknown in over 50% of the study population. On the basis of our previous research, we evaluated what we felt was the most reasonable and reproducible method of estimating bSCr, which was to use center-specific age- and gender-based normative values in all patients versus only in those with an unknown bSCr. Particularly in children, this is reasonable, given that the prevalence of chronic kidney disease is extremely low. Despite finding evidence of differences in AKIN classification when comparing both bSCr methods, regardless of the bSCr method used, the AKI-outcome associations were maintained. This suggests that using normative estimated bSCr may be a reasonable alternative when "true" bSCr data are unavailable, which is particularly likely to occur in population-based studies.

One of the biggest strengths of our study is that the data were from two different university health care centers. Although both centers are located in the same city, practice patterns and processes of care differ between them, thus increasing the generalizability of our findings and enhancing our validation of findings from previous studies. We were careful to perform our statistical analyses in a targeted fashion in an attempt to study the most parsimonious statistical models and avoid overfitting of the data. This is the first AKI study in pediatric patients to evaluate the return of SCr to baseline values after AKI and the first to explore how early SCr increases may aid in predicting AKI in this population. In addition, it is important to note that the PRISM score does not include an assessment of renal status, thereby alleviating concerns regarding colinearity of AKI and illness severity score in the statistical models. However, the physiologic relationship between AKI and mechanical ventilation is complex, and colinearity between these two elements of the statistical models cannot be completely ruled out. An inherent limitation of this study is its partial retrospective nature. For example, we were unable to include one of the criteria of the AKIN definition for ascertaining AKI status: that the increase in SCr should occur over a period of 48 hours. A positive aspect of this criterion is that it leads to less misclassification of patients who develop a slow SCr increase throughout admission as having AKI. However, in the case of children, almost all AKI occurs within the first few days after PICU admission (Figure 3); therefore, it is unlikely that a substantial proportion of patients with a slow, prolonged increase in SCr were misclassified as having "true" AKI. It will be worthwhile in future studies to specifically evaluate the extent to which this "48-hour increase" criterion leads to differences in AKI diagnoses and study conclusions in pediatric populations, particularly since this criterion is often difficult to ascertain. Another limitation is that because we sought to combine only those variables from both databases which we determined to have been collected in a similar fashion, we were unable to assess other potentially important factors such as the use of nephrotoxic medication or specific diagnoses (for example, sepsis or hematologic and oncologic diseases). Some of our analyses were limited to the MCH center because data were unavailable at the CHUSJ center. Finally, it is important to note that we were unable to evaluate the pRIFLE criteria for AKI in comparison to the AKIN staging system, since we did not have complete data regarding patients' height. Particularly in small children, in whom small changes in SCr may have a large effect on AKI classification, it is possible that conclusions regarding AKI-outcome associations or the effect of age as an AKI risk factor might differ when using the pRIFLE criteria. When we excluded neonates from the analyses evaluating AKI as a predictor of PICU LOS and duration of mechanical ventilation, our results were unchanged. However, in future studies, investigators should strongly consider determining to the extent to which pRIFLE versus AKIN staging may lead to different conclusions, especially in young children.

AKI is an important risk factor for poor outcomes in critically ill children. Future studies should elucidate the effects of AKI on long-term renal and nonrenal outcomes. Clinical trials of AKI are needed, but challenges will be associated with preventing PICU AKI because of early occurrence during admission. While SCr is an imperfect AKI biomarker, early small SCr increases may provide an important clue that critically ill children will develop overt AKI and be at risk for negative effects of AKI, such as fluid overload and medication toxicity.