Three-year mortality in 30-day survivors of critical care with acute kidney injury: data from the prospective observational FINNAKI study

Background The role of an episode of acute kidney injury (AKI) in long-term mortality among initial survivors of critical illness is controversial. We aimed to determine whether AKI is independently associated with decreased survival at 3 years among 30-day survivors of intensive care. Results We included 2336 30-day survivors of intensive care enrolled in the FINNAKI study conducted in seventeen medical–surgical ICUs in Finland during a 5-month period in 2011–2012. The incidence of AKI, defined by the Kidney Disease: Improving Global Outcomes criteria, was 34.6%, and 192 (8.3%) commenced RRT. The 3-year mortality among AKI patients was 23.5% (95% CI 20.6–26.4%) compared to 18.9% (17.0–20.9%) of patients without AKI, p = 0.01. However, after adjustments using Cox proportional hazards regression, AKI was not associated with decreased 3-year survival (HR 1.05; CI 95% 0.86–1.27), whereas advanced age, poor pre-morbid functional performance, and presence of several comorbidities were. Additionally, we matched AKI patients to non-AKI patients 1:1 according to age, gender, presence of severe sepsis, and a propensity score to develop AKI. In the well-balanced matched cohort, 3-year mortality among AKI patients was 136 of 662 (20.5%; 17.5–23.6%) and among matched non-AKI patients 143 of 662 (21.6%; 18.5–24.7%), p = 0.687. Neither AKI nor RRT was associated with decreased survival at 3 years in the sensitivity analyses that excluded patients (1) with chronic kidney disease, (2) with AKI not commenced renal replacement therapy (RRT), and (3) with estimated pre-admission creatinine, chronic kidney disease, or AKI stage 1. Conclusion AKI was not an independent risk factor for 3-year mortality among 30-day survivors. Increased 3-year mortality among patients with AKI who survive critical illness may not be related to AKI per se, but rather to advanced age and pre-existing comorbidities. Electronic supplementary material The online version of this article (doi:10.1186/s13613-016-0218-5) contains supplementary material, which is available to authorized users.

AKI has been significantly associated with increased short-term mortality [3,4,6,8], but results about its impact on increased long-term mortality among initial survivors of AKI and critical illness are few and conflicting. A large cohort study revealed the 1-year mortality of 30-day survivors of AKI defined by RIFLE (risk, injury, failure, loss, end-stage renal disease) creatinine (Cr) criteria to range from 20.5 to 23.8% with increasing severity of AKI, compared to 10.7% of the patients without AKI [5]. A 10-year follow-up study among critically ill patients found stage 1 AKI patients to have worse crude survival rate than patients without AKI [9]. The difference was significant also among 28-day survivors, but became nonsignificant after adjusting for confounders [9]. Additionally, 30-day survivors commenced renal replacement therapy (RRT) did not have worse 3-year outcome compared to their propensity-matched control group of hospitalized patients without AKI [10]. Moreover, among all ICU patients, the short-term mortality has been related mostly to the type and severity of acute illness, whereas long-term mortality among initial survivors has been primarily determined by patient age and comorbidities [11], which are also well-known predisposing factors for the development of AKI [1].
Thus, we hypothesized that critically ill patients with AKI have decreased long-term survival compared to patients without AKI, but that pre-existing comorbidities and advanced age could be explanatory factors. Therefore, in this analysis among 30-day survivors of intensive care, we aimed to determine whether AKI is independently associated with increased 3-year mortality.

Methods
We performed a prospective cohort study in 17 ICUs in Finland between September 1, 2011, and February 1, 2012 [3]. The Ethics Committee of the Department of Surgery at Helsinki University Hospital gave approval for the study protocol and the use of deferred consent with written informed consent obtained from the patient or proxy as soon as possible. The Finnish National Institute of Health and Welfare approved collection of data of deceased patients from medical records if an informed consent could not be obtained.

Patients
All adult (>18 years) patients with an emergency admission to the study ICUs of any duration, or an elective admission expecting to last over 24 h, were included to this Finnish Acute Kidney Injury (FINNAKI) study [3]. Patients who (1) were on chronic dialysis, (2) were readmitted and had received RRT during their previous admission, (3) were organ donors, (4) had insufficient language skills for giving informed consent or were not permanent Finnish residents, (5) were transferred from another ICU and had already been included for the study data collection period of five days, or (6) who were intermediate care patients were excluded from the study. We followed the 2901 included patients [3] until 3 years. For this current analysis, we excluded all 30-day non-survivors (N = 548), those who had previously received any organ transplant (N = 16) or had acquired immune deficiency syndrome (AIDS) (N = 1). As a further explanatory analysis regarding potential survival bias, we present data from the entire FINNAKI cohort (n = 2901) without exclusions.

Data collection
We prospectively collected data on previous and present medical status, patient demographics, ICU severity scores, presence of sepsis, possible risk factors for AKI, and existing comorbidities with study-specific case report form and from the database of the Finnish Intensive Care Consortium (Tieto Ltd, Helsinki, Finland). Data regarding ICU treatment were collected until day 5. Finnish Population Register Centre provided the survival data and the Finnish Registry for Kidney Diseases data on need for chronic dialysis at 3 years.

Definitions
We measured plasma Cr concentration daily and urine output hourly and used the Kidney Disease: Improving Global Outcomes (KDIGO) [1] criteria to define and stage AKI, using both Cr and urine output criteria. As the baseline Cr, we used the latest value from previous year excluding the previous week. If it was unavailable, we estimated it by using the Modification of Diet in Renal Disease (MDRD) equation assuming a glomerular filtration rate (GFR) of 75 ml/min/1.73 m 2 [12]. End-stage renal disease (ESRD) was defined by need for maintenance dialysis at least for three months [13]. We defined sepsis according to the American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference [14].

Cox model
We used Cox proportional hazards model to adjust for confounders related to survival at 3 years. We tested the validity of the proportional hazards assumption for Cox using the cox.zph method of the R survival package (hhtp://R-project.org) and found it to be valid in all models. We adjusted for gender, type of admission, pre-morbid functional performance, presence of severe sepsis, or comorbidities (arteriosclerosis, chronic obstructive pulmonary disease (COPD), CKD, diabetes mellitus, hypertension, liver failure, malignancy, rheumatoid diseases, systolic heart failure, and thrombophilia), Simplified Acute Physiology Score (SAPS) II points without age and renal components, and use of vasoactive drugs in addition to the presence of AKI. These covariates were chosen based on their distribution between AKI and non-AKI groups. Patients with missing data were assumed not to have the chronic condition. We also performed three sensitivity analyses by excluding (1) patients with CKD, (2) patients with AKI who did not receive RRT, (3) those with an estimated pre-admission Cr, or pre-existing CKD, or AKI stage 1. As a further explanatory analysis, time-stratified Cox models were generated as suggested [15].

Matching
As age and comorbidities are risk factors for both AKI [3,6] and increased long-term mortality [11], we performed a matched analysis between AKI and non-AKI patients. We matched patients 1:1 according to (1) age (caliper width ± 5 years), (2) sex, (3) presence of severe sepsis in ICU, and (4) the logit of propensity score for developing AKI (caliper width 0.2 SD) at random and without replacement. The logistic regression model (presented in Additional file 1: Table S1) used to construct the propensity score for AKI included variables previously reported to be associated with development of AKI [1,3,16] and the outcome as recommended [17]. We calculated standardized differences between the matched groups to assess the post-matching balance and considered standardized differences less than 10% indicative of good post-matching balance [18].

Statistical analysis
We present continuous data as median with interquartile range (IQR) and categorical data as absolute number and percentage. We used Chi-square test for categorical values and Mann-Whitney U test for continuous data in comparisons. We considered two-sided p value <0.05 as significant. In the matched sample, we compared categorical data with the McNemar test. We calculated the 95% CI for the difference in the 3-year mortality in the matched groups with Newcombe's method [19]. We used SPSS version 23 (SPSS, Armonk, NY, USA) and R (http://R-project.org) for data analysis.

Matched cohort
We found 662 matches to 662 AKI patients (81.9% of all 808 AKI patients). The groups were well balanced after matching (Table 3; Additional file 2: Figure 1) pre-morbid functional performance and more comorbidities than matched AKI patients (Additional file 1: Table S2). Figure 3 presents the Kaplan-Meier plot of the matched and unmatched patients according to AKI status.

Explanatory analyses
In the whole FINNAKI cohort (n = 2901) without 30-day non-survivors excluded, 252 of the 1141 (22.0%) patients with AKI had treatment restrictions (do not resuscitate, do not escalate the level of treatment, withholding of intensive care) compared to 164 of the 1596 (9.3%) patients without AKI, p < 0.001. When the whole cohort was included in a time-stratified model (Additional file 1: Table S6), AKI adjusted for confounders was a significant risk factor for mortality in the time period from ICU admission to 30 days, but not thereafter.

Discussion
In this follow-up of the prospective, multicenter FIN-NAKI study, we found that the crude 3-year mortality among 30-day survivors was significantly higher among AKI patients compared to those without AKI. However, after adjusting for confounders known to affect longterm mortality, such as age and pre-existing comorbidities, AKI was not independently associated with decreased survival in the 3-year follow-up. The result was confirmed in a propensity-matched analysis and several sensitivity analyses.
While the role of AKI in increasing the short-term mortality is evident [3,4,6,8], results regarding its role  in long-term mortality are conflicting. In contrast to our findings, AKI has been found to associate with increased 1-year mortality among 30-day ICU survivors in a retrospective database analysis [5]. However, the mortality among patients without AKI was notably lower than in our study that excluded low-risk patients with uncomplicated elective post-surgical admissions, while mortality among AKI patients was corresponding, which might explain the difference between the results. Another large retrospective analysis among post-surgical hospital survivors found AKI defined by RIFLE Cr criteria to associate with decreased 10-year survival [20]. Both of these studies [5,20] used the RIFLE Cr criterion to define AKI, and therefore, patients with oliguric AKI only were included in the control group. The long-term survival of patients with oliguric AKI without positive Cr criterion has been shown to be worse than of those without AKI but better compared to patients who fulfill both urine output and Cr criteria [21]. Therefore, the different methods of defining AKI may partly explain the controversy of these previous results to our study. A large propensity-matched database analysis found AKI to associate with decreased survival among hospitalized patients after a median follow-up of 3 years [22]. In line with our findings, AKI was not a significant risk factor in the subgroup of critically ill patients [22]. Among

Table 3 Characteristics of the matched patients with acute kidney injury (N = 662) and their propensity-matched controls (N = 662)
Categorical data are presented as an absolute number/count (percentage) and continuous data as median with IQR AKI acute kidney injury, ICU intensive care unit, non-op non-operative, post-op postoperative, SAPS Simplified Acute Physiology Score cardiac surgical patients, the risk of death was higher at 1 year than at three [23], and possibly the risk of death varies over time. Corresponding to our findings, AKI treated with RRT was not associated with an increased risk of 3-year all-cause mortality among 30-day survivors compared to matched hospitalized patients without AKI [10].
A plausible explanation to our findings is that post-90-day mortality among general ICU patients is mainly determined by age and pre-existing comorbidities instead of characteristic of the acute illness [24]. In our previous sequentially matched analysis, we found the excess mortality attributable to AKI at 90 days to be 8.6% [16], which implies that AKI substantially affects the outcomes of acute illness. Since AKI and long-term mortality share the same risk factors [3,20,24], the higher crude 3-year mortality of AKI patients observed in the current analysis could be explained by higher age and number of comorbidities making the patients more susceptible for AKI as well as subsequent increased long-term mortality. Accordingly, instead of AKI, we found COPD, chronic liver failure, CKD, rheumatoid diseases, poor pre-morbid functional performance, and advanced age to associate with worse long-term outcome. Notably, poor pre-morbid functional status and presence of malignancy were strongly associated with decreased survival at 3 years. Our results corroborate a Scottish study that found underlying CKD among patients commenced RRT to predict long-term mortality rather than severity of acute illness [25]. Finally, our results were robust to excluding patients with pre-existing CKD, those with estimated baseline Cr, and those AKI patients who did not receive RRT.
Albeit our analysis implies that AKI is not an independent risk factor for 3-year mortality among initial 30-day survivors, it does not mean that these patients would not be in an increased risk of other adverse outcomes. An episode of AKI has been shown to be associated with elevated risk of developing CKD and ESRD [7]. Additionally, several animal models suggest that AKI can cause permanent damage on vasculature in kidney as well as outside of it [26]. Among hospitalized patients, de novo CKD developing after an episode of AKI has been shown to be an important explanatory factor for the increased long-term mortality of AKI survivors [22]. The risk of de novo CKD was found highest at three months after acute rise in Cr and to persist up to 5 years, although reducing in time [23]. The risk factors for developing de novo CKD after AKI included advanced age, pre-existing hypertension, and high Charlson comorbidity score [22], all of which are also risk factors for long-term mortality. Taken together, our analysis does not exclude the possible increase in post-3-year mortality due to potentially developing CKD and its subsequent complications such as increased cardiovascular mortality.
An obvious strength of our study was its multicenter nationwide setting, detailed prospective data collection, and a complete follow-up. Unlike other studies, with a comparably long study period including a selected cohort of patients [9,10], our study included a heterogeneous cohort of critically ill patients, improving the external validity of our results. However, this study has some important limitations. First, inherent to all observational studies, it is impossible to adjust for unmeasured factors. For example, we had no data of pre-existing neurological conditions. We included, however, the score for pre-morbid functional performance, which reflects also aspects of neurological problems, such as dementia. Second, we did not have data about de novo CKD developing after an episode of AKI. However, the presence of CKD has been found to attenuate the mortality risk of AKI patients among hospitalized patients [22]. We found no difference between survival of AKI and non-AKI patients despite not adjusting for de novo CKD. Thus, we consider our findings representative and reliable. Third, the sample size of the matched cohort has approximately of 69% power to show a 5% difference assuming a 3-year mortality of the non-AKI group of 21.6%, and therefore, the result of no significant difference must be interpreted with caution. Additionally, despite a statistically insignificant comparison, the number of patients in different subgroups, such as those treated with RRT, is inadequate to refute a clinically significant difference in 3-year mortality due to large CI (0.86-1.70).
Finally, when interpreting our results, a potential selection bias of AKI occurring in more severely ill patients must be kept in mind. First, patients with AKI and greater illness severity more frequently had treatment restrictions, thus possibly forming the group of AKI survivors toward a healthier cohort. Second, as AKI generally occurs in the more severely ill, those who survive AKI may constitute a selected group of patients with lower risk of further adverse outcomes. Third, as encountered in other studies [10,22], a small proportion of AKI patients (18.1%) in the current analysis was left without a match because severely ill patients without AKI could not be found. The survival curves of both the whole cohort after initial separation and the matched patients, however, are very collinear. Therefore, we consider our results to be informative in the case of corresponding baseline possibility of developing AKI and/or surviving intensive care regarding the outcome in the coming 3 years. We believe this result is of value, for example, when informing patients with AKI or their family.

Conclusion
Among critically ill patients surviving over 30 days, AKI was not an independent risk factor for mortality at 3 years. AKI patients had higher crude 3-year mortality than non-AKI patients, but the difference was not robust for adjustments for a number of relevant confounders. Our findings imply that increased long-term mortality at 3 years among patients with AKI who survive critical illness is not related to AKI per se, but rather to advanced age and pre-existing comorbidities.
records of patients who deceased in the ICU if an informed consent could not be obtained.