A worldwide multicentre evaluation of the influence of deterioration or improvement of acute kidney injury on clinical outcome in critically ill patients with and without sepsis at ICU admission: results from The Intensive Care Over Nations audit

This was a substudy of the Intensive Care Over Nations (ICON) audit, a multicentre, worldwide audit conducted between 8 and 18 May 2012 to collect data on characteristics, including organ dysfunction, infection and outcomes, of adult critically ill patients worldwide. Full details of the methodology have been published previously [15] and are provided in Additional file 1. Briefly, participating centres were recruited by open invitation, through national scientific societies, international meetings and/or individual contacts. A list of participating centres is provided in Additional file 2. Participation was voluntary with no financial reimbursement. In each institution, the study was approved by the institutional research ethics committee in accordance with local ethical regulations. Informed consent was not required due to the observational and anonymous nature of the data collection. Participating centres prospectively collected data from all adult patients (> 16 years) admitted to the ICU during the study period, except those admitted for less than 24 hours for routine postoperative surveillance. Readmissions of previously included patients were not considered. For the purposes of this substudy, only patients with creatinine and urine output data were included. Patients who developed sepsis during the ICU stay (≥ 2 days after admission) were excluded because we wanted to evaluate the impact of sepsis on ICU admission on AKI outcomes.

Sepsis was defined as the occurrence of at least one failing organ (Sepsis-related Organ Failure Assessment (SOFA) score > 2 for respective organ system) combined with the presence of infection, as defined according to the International Sepsis Forum [16]. The presence of AKI was evaluated using the Acute Kidney Injury Network (AKIN) criteria [17]. AKI stages were determined according to the AKIN criteria, using the largest increase between two values of serum creatinine obtained maximally 48 hours apart within a maximum period of 72 hours after ICU admission or the urine output for the 24-hour period after ICU admission. As urine output was only recorded per 24-hour period in ICON, we adjusted the stage 1 and 2 urine output criteria (Additional file 3: Table S1). The criterion (urine output or serum creatinine) that led to the worst possible AKIN classification was used in each case. We chose to use the AKIN criteria rather than the more recent KDIGO criteria, because in the KDIGO definition of AKI the creatinine criterion relies on the percentage change over a period of 7 days (rather than 48 hours for the AKIN criteria) and our aim was to demonstrate the effect of deteriorating or resolving AKI during the first 7 days after ICU admission on outcome. To assess AKI deterioration or improvement, the last serum creatinine or urine output data available before discharge or death up to day 7 were used to define the AKIN stage, again using the criterion (urine output or serum creatinine) that led to the worst possible AKIN classification in each case. Patients who died were not penalized to a maximum stage. This AKIN stage was compared to the initial stage, and patients were identified as having deteriorated, improved or remained the same. Complete recovery was defined as improvement from AKIN stage 1, 2 or 3 to the No AKI category. The inotropic score was calculated to express the use of vasoactive/vasopressor agents, as described elsewhere [18].

Patients with comorbid chronic renal failure (CRF), indicated on the original ICON case report form by the ICON investigator, were analysed as a separate subgroup.

All statistical analyses were conducted in the Department of Intensive Care of Erasme Hospital. For the purpose of this study, the world was divided into nine geographical regions, and individual countries were classified into three income groups in accordance with the gross national income (GNI) per person [14]. Data are expressed as mean and standard deviation (SD), median with interquartile range (IQR, first–third quartiles) or numbers and percentages. For continuous variables, normality assumption checking was performed by inspection of residual and normal plots and by using the Kolmogorov–Smirnov test. Difference testing between groups was performed using the generalized linear models procedure, Kruskal–Wallis test, Student’s t test, Mann–Whitney test, χ² test or Fisher’s exact test, as appropriate.

To describe the prevalence of patients receiving RRT up to day 7, data were imputed by the last observation carried forward in the case of discharge or death. If a patient died before day 60, the ICU and hospital length of stay (LOS) was each set to 60 days.

Data were censored at day 60. In addition, discharge of a patient was considered a competing risk factor for the occurrence of death. A competing risk regression model [19] was used to estimate crude and adjusted hazard ratios (HRs) and their 95% CIs for ICU and hospital mortality according to the AKI stage or CRF group. To determine the adjusted relative risk of in-ICU or in-hospital death, we developed a multivariable competing risk proportional hazard regression model, stratified according to the presence or not of sepsis. Other confounding variables considered included age, sex, Acute Physiology and Chronic Health Evaluation (APACHE) II score without age and renal components, type of admission, source of admission, reason for admission, the need for mechanical ventilation or RRT on admission to the ICU, comorbidities and the inotrope score. We also adjusted for ICU and hospital-related organizational factors including type of hospital, ICU specialty, total number of ICU patients in 2011 and number of staffed ICU beds. Geographic region and GNI were also considered. Collinearity between variables was excluded before modelling and the time-dependent covariate method was used to check the proportional hazard assumption of the model. The cumulative incidence functions of death according to AKIN stage within the first 72 hours of admission or CRF were plotted and Gray’s test was used to test cause-specific death differences [20, 21]. To quantify the association between the direction of the evolution of the AKI and sepsis, a cumulative link mixed-effects ordinal response model with logit link function was fitted. The random intercept model was used. The longitudinal ordinal response variable, AKIN stage, was the main dependent. The explanatory variables were the time points of AKIN stage assessment (admission/follow-up), sepsis (yes/no) and their interaction [22, 23]. Two-sided p < 0.05 was considered statistically significant. Data were analysed using IBM® SPSS® Statistics software, version 23 for Windows (IBM, Armonk, NY, USA) and R software, version 2.10.1 (CRAN project).

By the 7-day follow-up, sepsis patients without AKI within the first 72 hours of ICU admission were less likely to have developed AKI compared to the non-septic population (n = 325 (52%) vs n = 1582 (60%), p < 0.0001). Sepsis patients admitted with AKIN stage 1 or 2 were more likely to have complete recovery of AKI, compared to patients without sepsis for the same AKIN stage (n = 206 (36%) vs n = 522 (26%) for AKIN stage 1 and p < 0.0001; n = 65 (30%) vs n = 130 (21%) for AKIN stage 2, p = 0.005, respectively). However, sepsis patients with AKIN stage 3 were less likely to have recovered to a lower AKIN stage by day 7 than non-septic patients with AKIN stage 3 (n = 108 (21%) vs n = 258 (32%), p < 0.0001). The random intercept model using time points of AKI measurements (admission/follow-up), sepsis (yes/no) and their interaction showed that the estimated parameter for interaction was – 0.62 ± 0.08, p < 0.001, confirming that sepsis patients with less severe AKI were more likely to recover renal function within the first week after admission than non-septic patients, whereas sepsis patients with more severe AKI (AKI stage 3) were less likely to recover renal function than non-sepsis patients with AKI stage 3.

The use of RRT was greater in the septic patients than in the non-septic population (n = 380 (20%) vs n = 326 (5%), p < 0.0001; Additional file 4A). During the first 4 days, haemofiltration was the most frequent form of RRT in sepsis patients and haemodialysis in non-septic patients (Additional file 3: Table S2). RRT use was greatest in patients admitted with AKIN stage 3 in patients with and without sepsis (Table 3). In patients admitted with AKIN stage 3, RRT use was more frequent in patients who remained in stage 3 than in those who improved (Additional file 4B). In patients with AKIN stage 3, RRT was more often initiated in upper-middle and high income-class countries compared to low and lower-middle income-class countries (Additional file 3: Table S3). RRT treatment modality for each region and income class is presented in Additional file 3: Table S4.

ICU and hospital LOS was longer for patients admitted with sepsis compared to patients without sepsis (8 (4–60) vs. 3 (2–6), p < 0.001 and 38 (14–60) vs 11 (6–32), p < 0.001, respectively) and increased with increasing AKI severity (Table 3). ICU and hospital LOS for each region and income class is presented in Additional file 3: Table S5.

Crude mortality rates were higher in patients receiving RRT compared to those without RRT (ICU mortality, sepsis n = 150 (40%) vs n = 341 (22%) and no sepsis n = 91 (29%) vs n = 564 (10%); in-hospital mortality, sepsis n = 180 (49%) vs n = 71 (32%) and no sepsis n = 105 (34%) vs n = 764 (14%), all p < 0.001). Of those receiving RRT, in-hospital mortality was higher in non-septic patients receiving haemofiltration compared to haemodialysis, whereas in the septic group there were no differences between the modalities (Additional file 3: Table S6).

Crude in-hospital mortality rates for patient with and those without sepsis were 2–3-fold higher in patients with AKIN stage 3 compared to patients without AKI and were higher for each AKIN stage compared to no AKI in patients with sepsis (Table 3). In patients with AKIN stage 3, in-hospital mortality was comparable for each income class (Additional file 3: Table S3).

In patients with sepsis, crude in-hospital mortality was greatest in those admitted with AKI stage 3 and AKI stage 2 who had not recovered by day 7 (Additional file 5); in patients without sepsis, this relationship was only present in patients with AKI stage 3. AKI stage 3 patients who fully recovered renal function had a 13% (septic patients) and 20% (non-septic patients) lower in-hospital mortality compared to those who remained in stage 3 at day 7. However, in-hospital mortality in these patients was still twice as high compared to patients recovering from AKI stages 2 and 1 (Additional file 3). Cumulative survival curves are shown in Fig. 2. AKI stages 2 and 3 were associated with significantly higher hazard ratios of crude ICU and hospital mortality in septic and non-septic patients compared to no AKI (Table 4). When adjusted for covariates, AKI stage 3 was still associated with ICU and in-hospital mortality in both patient groups (with and without sepsis) and AKI stage 2 was associated with in-hospital mortality in patients without sepsis (Table 4).

Of the 845 patients with a medical history of CRF, 263 (31%) had sepsis on admission. Baseline demographic and clinical characteristics are presented in Additional file 3: Tables S7 and S8. Use of RRT and ICU and in-hospital LOS and mortality are presented in Additional file 3: Table S9. Patients with CRF had lower ICU and in-hospital mortality rates than those with AKI stage 3 patients (sepsis p = 0.017; non-sepsis: p < 0.001) (Fig. 2). When adjusted for covariates, these differences disappeared in patients with sepsis, but remained in patients without sepsis (Table 4).

We observed that septic patients with less severe AKI were more likely to recover renal function during the first week after admission than non-septic patients with the same initial degree of AKI, whereas septic patients admitted with AKIN stage 3 were less likely to recover to a lower AKIN stage. Importantly, sepsis and non-sepsis groups are not necessarily similar at baseline, as patients with sepsis are more likely to have early, evolving AKI, and so changes in illness trajectory may reflect a temporal bias, rather than the influence of the underlying disease. Also, if the bar is set at the creatinine level at admission, it may be easier for patients who have a lower creatinine level pre admission to get back to this level. Clearly, there also are differences in the distinct pathogenesis of septic versus non-septic AKI. Although a variety of mechanisms are involved in the development of AKI [25], AKI in patients with sepsis may predominantly be the consequence of the dysregulated inflammatory response to an infection, resulting in renal inflammation, microcirculatory dysfunction and cell bioenergetic adaptive responses [13]. Clearing underlying infection may be more likely to help AKI resolution in milder cases of sepsis-associated AKI than in more severe stages. Conversely, non-septic AKI may be related more to extensive damage to the micro-architecture of the kidney, requiring prolonged periods for repair and renal recovery. In support of this suggestion, autopsy investigations did not reveal widespread tubular damage or cell death in septic AKI [26].

In this study, RRT was mostly started within the first 2 days after ICU admission and predominantly in patients admitted with AKIN stage 3 who did not recover during the first week after admission. It is currently unclear whether or not early initiation of RRT may improve patient outcome [27, 28] despite two large clinical trials [29, 30] and further study, such as the ongoing STARRT-AKI trial (ClinicalTrials.gov NCT02568722), is necessary to elucidate the effects of RRT timing on outcomes and to guide clinical decisions related to the initiation of RRT.

Although in-hospital mortality was lower in patients with renal recovery, septic and non-septic AKIN stage 3 patients who had recovered full renal function by day 7 still had a 2-fold higher in-hospital mortality rate than patients recovering from AKI stage 2 or 1 or patients without AKI. In addition, our data confirm that mortality is higher in patients who receive RRT compared to those who do not [31]. Our hypothesis-generating data could encourage a novel approach to risk stratification in AKI for current therapy, future investigations and clinical therapeutic trials. For example, in the absence of life-threatening complications, it may be prudent to delay RRT, as a large number of patients, especially in the lower AKIN stages, recovered full renal function. Indeed, use of RRT may delay recovery of renal function [32]. While resuscitation strategies appear not to influence the development of AKI in patients with septic shock [33], secondary prevention, by mitigating further toxic and haemodynamic renal insults, may represent an important therapeutic opportunity in this population. Also, models of ICU mortality and prognostic scores could likely be refined by considering the dynamic changes in renal function over time.

Our study has several limitations. Firstly, we lack information about secondary or post-day 7 renal insults that may have occurred as a result of hypotension, adverse drug reactions or inappropriate blood pressure targets. These factors can play an important role in the deterioration or improvement of AKI. Secondly, renal function was assessed using the AKIN criteria, a widely accepted and validated classification system for AKI in the critically ill [17], which we adjusted because urine output was only available per 24-hour period. As a consequence of using AKIN categories and not absolute creatinine values, smaller changes could be missed. Importantly, patients admitted to the ICU with ongoing AKI may be missed using these criteria and/or the severity of AKI was not correctly identified because pre-ICU admission reference creatinine values were not available, as is frequently the case in clinical ICU practice. This may limit the determination of the clinical consequences of deterioration or resolution of AKI, and may hamper comparison between sepsis and non-sepsis patients. Thirdly, we have no data enabling analysis of longer term consequences of AKI and progression to chronic kidney disease or AKI relapse in patients who had recovered normal renal function by day 7. These long-term consequences, stratified by evolution of AKI, represent an important area for future investigations. Fourthly, although the worldwide inclusion of patients provides valuable insight into the global burden of AKI, the causes of AKI vary by country and economic status [34]. Also, resources may differ in various regions around the world, which could influence the clinical outcome of AKI patients. However, we showed that although the use of dialysis in AKIN stage 3 patients was higher in upper-middle and high-income countries compared to low and lower-middle-income countries, in-hospital mortality rates were comparable across income classes in these patients. Finally, some centres included a limited number of patients, possibly leading to sampling bias, and the voluntary participation of ICUs may have resulted in compliance bias. Nevertheless, this is a prospective study in which a large number of patients was included from all over the world with daily assessment of renal function, enabling a useful comparison of the prevalence and evolution of AKI and its consequences among patients with and without sepsis.