Prognostic robustness of serum creatinine based AKI definitions in patients with sepsis: a prospective cohort study

There is growing evidence that Acute Kidney Injury (AKI) is an independent predictor of mortality rather than an innocent bystander. In cardiac surgery it has clearly been demonstrated that even small serum creatinine (sCr) increases are associated with increased mortality risk [1‐4]. It remains unclear if this also applies to sepsis.

There is continuing debate on the interpretation of the proposed AKI classification criteria [5‐8], and several interpretations of the same AKI definitions continue to appear in the literature [9‐12]. The most important issue in this debate relates to the way we calculate the sCr increase. The currently proposed practice to use the highest creatinine value over a certain time span after ICU (Intensive Care Unit) admission does not take into account the evolution of sCr as a response to therapeutic interventions. Although this might not be problematic when "AKI" is used as a diagnostic classifier, it can have an influence on the prognostic value of the label "AKI".

The prognostic value of AKI, using a 0.3 mg/dl cut off value for sCr increase, as proposed by AKIN (Acute Kidney Injury Network), KDIGO (Kidney Disease Improving Global Outcomes) and ERBP (European Renal Best Practice) [5, 7, 8], has not previously been validated in a prospective cohort of exclusively sepsis patients. These patients are particularly prone to capillary leak and fluid accumulation which influences the distribution volume of sCr, potentially resulting in a delay in sCr increase due to dilution [13]. This might result in even smaller increases of serum creatinine being associated with mortality in sepsis.

One hundred and ninety five consecutive adult patients (age ≥ 17 years) with sepsis admitted to the intensive care unit (ICU) of the Ghent University Hospital between 12/01/2010 and 27/03/2011 were included in this prospective cohort study. Sepsis, severe sepsis and septic shock were defined according to the American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference guidelines [14]. Exclusion criteria were: 1) ICU stay less than 24 h or withdrawal of therapy, 2) no bladder catheter, 3) patients treated with chronic hemodialysis, 4) patients with RRT need due to AKI upon ICU admission, 5) Age < 17 years, 6) a history of organ transplantation, 7) obstructive AKI and 8) no central line or arterial catheter. Survival status was assessed at ICU, at 3 months, 1 year and 2 years by JV, either by checking hospital records or by telephone interview with the family practitioner.

The study was approved by the ethical committee of the Ghent University Hospital. Written informed consent was obtained from the patient or their next of kin.

Three algorithms were used to calculate the sCr increase: 1) “ΔHIS”, defined as the highest value within 24 h after ICU admission minus the value of a pre-admission historical baseline; 2) “ΔEST”, defined as the highest value 24 h after ICU admission minus an estimated baseline value obtained by solving the MDRD (Modification of Diet in Renal Disease) equation assuming a GFR (Glomerular Filtration Ratio) of 75 ml/min/1.73 m², as suggested by ADQI [6] and 3) “ΔADM”, defined as the value at 24 h after ICU admission minus the value at ICU admission. We also used different values of sCr increase as threshold for AKI diagnosis (0.1, 0.2, 0.3, 0.4, and 0.5 mg/dl), to assess the robustness of the 0.3 mg/dl sCr increase criterion in sepsis.

Blood samples for measuring sCr were collected at the moment of study inclusion (D0T0), four hours later (D0T4), the next morning at 6 AM (D1) and daily at 6 AM for the next four days. during the first five days after ICU admission, centrifuged immediately and frozen at −80° for later batch analysis, using an Isotope Dilution Mass Spectroscopy traceable method (Roche Diagnostics®). Once patients were on dialysis, serum creatinine values within study protocol were no longer measured.

Severity of illness was assessed by the APACHE (Acute Physiology and Chronic Health Evaluation) II score during the first 24 h after admission, with or without the renal score.

The 24 h fluid balance was registered for all patients to account for bias due to fluid dilution.

All sepsis patients admitted to ICU between 6 AM and 18 PM were included on the day of ICU admission. Patients admitted to ICU after 18 PM were included the following day at 6 AM. For all patients, ‘Day 1’ (D1 = the day following the inclusion day) starts at 6 AM following the day of inclusion. (Fig. 1) Since the time interval between D1 and ICU admission was the same in each individual patient for the three algorithms, this issue is not likely to have influenced our results. The mean time interval also approximates 24 h (mean = 26 h). Additionally, if anything, having to deal with different time intervals between patients would have lowered our chances of demonstrating that including the evolution of serum creatinine in the AKI definition enhances the predictive performance of the label ‘AKI’, The latter because in those patients with a time interval less than 24 h, there might not have been sufficient time to evaluate the response to fluid resuscitation yet.

This study demonstrates that the predictive value of an increase in serum creatinine of 0.3 mg/dl, on mortality in sepsis, depends on how this sCr increase is calculated. Only if the evolution of sCr after ICU admission, rather than a peak sCr value over the same time span, is taken into account, an association with mortality is found. An increase in sCr of 0.3 mg/dl according to the ICU admission value is the lowest robust value still associated with mortality, confirming previous data in the cardiac surgery setting. This 0.3 mg/dl cut off remained robust even after adjustment for 24 h fluid balance. However, after adjusting for severity of illness, a sCr increase of 0.3 mg/dl was no longer associated with mortality.

Although the definition of AKI has become progressively more uniform since the introduction of RIFLE [6], uncertainty and debate on the methodology to calculate the sCr increase remains, and different interpretations are still appearing in the literature [9‐12].

Most guidelines advocate to use the difference between the highest sCr value over a certain time span, and a pre admission historical baseline value [15, 16] or an estimated sCr value by solving the MDRD formula assuming a GFR of 75 ml/min/1,73 m² [17‐20]. It has been demonstrated that the use of surrogate baseline sCr values can lead to misclassification by either under- or overestimation of AKI [15]. Siew et al. investigated the impact of using different surrogate baseline values in a large cohort of 4863 adults, on AKI diagnosis and outcome [16]. As in our cohort, they found that the incidence of AKI decreased by using the ICU admission sCr value as baseline versus a pre admission historical sCr value. They also demonstrated that mortality rates were different according to which surrogate value was used [16]. As in our study, mortality rates were highest when AKI was defined according to the admission value. However, Siew et al. used a cohort containing mainly non-critically ill patients (only 19 % of patients were admitted to ICU) and only in-hospital and 60 days mortality were evaluated, without adjustment for severity of illness [16].

The strategy of using the peak sCr value over a certain time span after admission does not allow to study the impact of the evolution of sCr after starting therapeutic interventions such as fluid resuscitation. In our cohort, AKI was associated with mortality, but only if based on the difference between the value 24 h after admission and the value at ICU admission and not if based on the peak value over the same time span compared to a historical or estimated baseline value. These findings suggest that the evolution of sCr in the first 24 h after ICU admission, and thus the potential response to fluid resuscitation, is most predictive for outcome which is in line with previous observations [21]. By taking into account the evolution of serum creatinine after ICU admission, those patients who experience a decrease in serum creatinine under fluid resuscitation are no longer classified as having AKI which translates in a lower number of AKI cases compared to using the peak serum creatinine value over the same time period. It allows us to differentiate patients who are generally sicker and have more severe AKI and hence a worse outcome, from those with less severe AKI, responding to fluid resuscitation.

It is generally accepted that AKI is associated with increased mortality [22‐24]. Bagshaw et al. performed a retrospective interrogation of prospectively included patients with sepsis, septic AKI and non-septic AKI. Sepsis, septic AKI and non-septic AKI were all found to be significantly associated with poor outcome. However, authors only looked at short term mortality (28 days and hospital mortality) and no data on RRT need were available, so it is unclear whether the effect on mortality was mainly driven by RRT need [23].

Despite this generally accepted association between AKI and mortality, the topic is still also a matter of debate, even in the non-critically ill. In a population based study of AKI, no association between AKI and outcome was found [25]. A higher risk for chronic dialysis need, but not mortality, was found in a large cohort of ICU survivors [26]. However, several reports indicated that the duration of AKI is highly predictive of mortality [27], an aspect that was not evaluated in the current study. This might explain why in our cohort the diagnosis of AKI was no longer independently associated with mortality after adjustment for severity of illness. Our data indicate that the response to treatment in the first 24 h after ICU admission also influences prognosis, probably by unveiling those who positively respond to volume resuscitation. Several older and more recent studies also demonstrated that the most important predictors for mortality were already present at admission to the ICU and included advanced age, the presence of infection, a past history of chronic diseases and the presence of other failing organs [28‐32].

Although recent literature shows that even small increases in sCr are independently associated with mortality, these results are mainly obtained in cohorts of cardiac surgery patients, after coronarography and after myocardial infarction [1‐4, 33‐35]., In cardiac surgery, the relation between small increases in sCr and mortality might be amplified by the observation that in patients without AKI, sCr levels are expected to decrease in the first 24 h after surgery because of fluid loading perioperatively, diluting the sCr value [36]. The association between small increases in sCr and poor outcome cannot necessarily be extrapolated to other conditions, such as sepsis. Sepsis patients are particularly vulnerable to capillary leak and fluid accumulation. Thus theoretically even smaller increases of sCr could be associated with mortality because of dilution of sCr, and a delay in diagnosis [13]. However, according to the results of the current study, even after adjustment for 24 h fluid balance, a 0.3 mg/dl increase in sCr 24 h after ICU admission compared to the ICU admission value, seems to be the lowest robust threshold for increased risk in sepsis patients.

The strong points of this prospective study are the availability of longer term outcomes and the detailed patient information.

Our study is the first to consider that the evolution of sCr after start of therapy, rather than an absolute highest value over a certain time span is important with regard to outcome.

A limitation of this study is its observational nature and the fact that it is a single centre study describing a relatively small cohort of patients which enhances the chance of a type 2 error. We acknowledge that our results need to be validated in a larger cohort and are not necessarily generalizable to sepsis cohorts including patients with different severity of illness. However, the finding that taking into account the evolution of serum creatinine demonstrates a better association with mortality compared to only relying on a peak serum creatinine value over a certain time span for AKI diagnosis, does make sense from a pathophysiological point of view. Taking into account the evolution of serum creatinine allows for the identification of those who have more sever AKI, are more ill and consequently have a worse outcome.

By capturing AKI during the first 24 h after ICU admission we could incorporate the early response to fluid resuscitation in the ∆ADM algorithm. Although by doing so we missed the AKI cases occurring after 24 h this is not likely to change our general message that AKI classification criteria should incorporate the evolution of sCr in response to fluid resuscitation. Also when we used a 48 h interval, our findings did not change (see Additional files 2, 3 and 4).

We did not include the urinary output as a prognostic criterion in this study. We demonstrated before that taking into account urinary output improved the diagnostic performance of RIFLE [37]. In several other studies, it was demonstrated that urinary output is associated with mortality in the critically ill [38, 39]. In the current study, however, we wanted to evaluate the independent impact a change in the sCr criterion, as this is the most widely used criterion for prognostic modelling, especially in large (administrative) database cohorts [35, 40, 41].

Mortality rates in our cohort are low compared to what was reported in a recent study [42]. However, in general, mortality rates in sepsis have been reported to decline [42‐45]. Mortality rates in our cohort of patients with sepsis might also be lower than in previous reports because we have a very good system in place in our ICU to alert physicians of pending AKI. Most of these alerts are induced by reduced urinary output [46]. It has been demonstrated before that AKI defined by oliguria has a better prognosis as AKI defined by the creatinine criterion [24].

The prognostic value of a 0.3 mg/dl increase in sCr, on mortality in sepsis, depends on the algorithm to calculate this sCr increase. Only if the evolution of serum creatinine over the first 24 h after ICU admission is taken into account, an association with mortality is found. This 0.3 mg/dl increase criterion remains robust, even after adjusting for 24 fluid balance. After adjustment for severity of illness, a sCr increase of 0.3 mg/dl is no longer associated with mortality in sepsis.