Background
Early during ICU admission clinicians often find it difficult to predict the long-term outcome of critically ill patients [
1]. Even during the course of an intensive care unit (ICU) admission the prognosis may remain unclear. In order to support decision-making on the continuation or withdrawal of ICU treatment, identifying valid clinical predictors early during ICU admission is particularly relevant [
2].
The first day of ICU admission is critical for prognosis. Its clinical relevance is made clear by the high prognostic value of disease severity scores based on the first day of ICU admission. A multitude of physiologic variables are included in models such as the Acute Physiology and Chronic Health Evaluation (APACHE) and Simplified Acute Physiology Score (SAPS) [
3,
4]. However, only the presence of specific ICU complications such as acute kidney injury (AKI) is taken into account. Worldwide established and detailed classification systems for complications identifying different levels of severity, were not used. A considerable number of critically ill patients develop AKI during their ICU stay [
5]. Over twenty percent of all general ICU patients do so within 24 hours of ICU admission [
6]. Moreover, when AKI does accompany critical illness, it is a risk factor for increased mortality, up until one year after ICU admission [
5‐
14]. Early AKI (eAKI) could be an especially prevalent warning sign of poor long-term outcomes. If so, it could be used as a building block for personalized prognoses.
Multiple studies have investigated predictors and models for the short-term prognosis of the critically ill. As a consequence, scoring systems such as APACHE IV and SAPS 3 have been developed for risk stratification [
3,
4]. These studies focussed on associating predictors and models with hospital mortality. To facilitate decision-making on continuing or withdrawing treatment in the ICU, however, patients and their relatives usually want to be informed about the chances of survival beyond hospital discharge. Often, they want to take the expected quality of life into account. Health-related quality of life (HRQoL) has only been studied scarcely as the outcome in prognostic factor studies [
15‐
17]. Especially in the general ICU population studies the investigation of predictors of HRQoL is rare. Furthermore, prognostic factors of a combination of survival and HRQoL have not yet been studied.
Therefore, the aim of this study was to investigate whether the occurrence and severity of eAKI, which is defined as AKI occurring during the first 24 hours of admission, in a mixed ICU population is independently associated with one-year mortality and HRQoL.
Discussion
This cohort study showed that occurrence of AKI early during the ICU stay was associated with an increased probability of being dead or having low HRQoL one year after ICU admission. When compared to patients without eAKI, patients with increasing eAKI severity were associated with increasing risks of poor outcome one year after the ICU stay. Patients with a RIFLE class, failure, on the first day of admission even had a 25 % significantly increased risk of poor outcome, independent of other measured predictors.
To illustrate the effect of eAKI in the setting of the high overall outcome incidence, we used our full statistical model to calculate the absolute predicted probability of poor outcome for two typical ICU patients. Patient A is a low-risk 40-year-old male patient, without comorbidities, admitted to the ICU after elective surgery and a day of prior hospital stay, without an infection or mechanical ventilation within 24 hours of ICU admission and an Acute Physiology Score (APS) of 10. Patient B is a high-risk 60-year-old female patient, with a Charlson Comorbidity Index of 3, admitted to the ICU for medical reasons after a week of prior hospital stay, with a confirmed infection and mechanical ventilation within 24 hours of ICU admission and an APS of 20. If these patients developed severe eAKI (RIFLE failure) Patient A’s risk of poor outcome would increase from 21 to 26 %, while Patient B’s risk would rise from 58 to 72 % (see Table
4 for the full statistical model).
Table 4
Poisson regression model for poor outcome
Intercept | −2.401 | | | <0.001 |
eAKI | No eAKI | Reference | |
Risk | 0.023 | 1.02 | 0.91, 1.15 | 0.691 |
Injury | 0.133 | 1.14 | 1.01, 1.29 | 0.034 |
Failure | 0.223 | 1.25 | 1.01, 1.55 | 0.042 |
Sex | Male | Reference | |
Female | 0.096 | 1.10 | 1.01, 1.2 | 0.030 |
Admission type | Elective surgical | Reference | |
Urgent surgical | 0.185 | 1.20 | 0.99, 1.45 | 0.057 |
Medical | 0.270 | 1.31 | 1.09, 1.57 | 0.004 |
Mechanical ventilation within 24 hours of ICU admission | No | Reference | |
Yes | 0.086 | 1.09 | 0.93, 1.28 | 0.300 |
Confirmed infection within 24 hours of ICU admission | No | Reference | |
Yes | 0.116 | 1.12 | 1.02, 1.24 | 0.017 |
Age (transformed) | 1.615 | 5.03 | 3.63, 6.96 | <0.001 |
Charlson Comorbidity Index | 0.050 | 1.05 | 1.03, 1.07 | <0.001 |
Pre-ICU hospital length of stay | 0.005 | 1.01 | 1.00, 1.01 | 0.050 |
Acute Physiology Score (transformed) | 0.324 | 1.38 | 1.26, 1.52 | <0.001 |
Patients, family members and clinicians desire more prognostic information about an ICU patient’s survival in conjunction with the expected HRQoL than is currently available [
2,
33,
34]. Furthermore, long-term quality of life is conditional on long-term survival. When patients base decisions made during the ICU stay on predicted HRQoL, they need information which also takes into account the condition of long-term survival. We decided to tackle this form of conditionality by creating a composite outcome that is clinically relevant at the time of major ICU treatment decisions. To our knowledge, this is the first study to specifically address this clinically relevant composite endpoint of poor outcome.
So, in respect to the results of previous studies, only the result on the separate constituents of this composite outcome can be compared. The association between (e)AKI and mortality described here is supported by current literature. A recent systematic review described studies of survival for 6 months after ICU discharge. The included studies all reported a large and significant decrease in survival probability in the AKI failure group when compared to all other AKI or no AKI groups [
35]. Three studies have reported on the association between (e)AKI and HRQoL in long-term ICU survivors and support the findings presented here. When comparing those survivors who had suffered from (e)AKI and survivors without (e)AKI, there was no significant association with any HRQoL classification [
12,
36,
37]. Based on another recent systematic review, the presented study population is by far the largest one to date [
38]. Additionally, none of the prior AKI and HRQoL studies took into account the conditionality of HRQoL on survival [
12,
36‐
38]. Finally, with respect to the contribution of survival and HRQoL to the composite endpoint, the increased risk of eAKI for poor outcome seemed to be mainly caused by an increased risk of death within one year after ICU admission.
Different from these previous studies, HRQoL was analysed dichotomously in this study. Aside from this being necessary in order to determine whether a patient suffered from a poor composite outcome, a qualitative interpretation of HRQoL (“low” versus “high” or “severely impaired HRQoL” vs. “not or mildly impaired HRQoL”) was constructed. Choosing a threshold was, and still is, not straightforward. The EQ-5D index itself contained minimal qualitative interpretation: its guidelines merely indicated that a score of 1 corresponds to “full health” and scores below zero to equal states of living valued worse than death [
22]. We therefore decided to set a threshold value based on the average EQ-5D index value measured in patients with severe physical, cognitive and/or psychiatric disabilities [
23‐
25]. Still, after classifying patients as such, patients with a low HRQoL might not have considered themselves to be (severely) disabled. However, based on the EQ-5D index formula it can be shown that patients with an EQ-5D index below 0.4 all experienced extreme problems on at least one of the EQ-5D dimensions [
22]. Altogether we assumed this threshold therefore corresponded to a clinically relevant major disability or impairment of HRQoL one year after ICU admission.
A strong feature of this study is that we measured and defined RIFLE classification in high detail using an algorithm for routinely collected data. In this study, as originally proposed by Bellomo et al., the RIFLE classification was based on both serum creatinine changes and urine output per hour [
5]. As a result, this study distinguished itself from those studies using only serum creatinine changes and/or 24 hour urine output when classifying AKI [
35,
39].
Another strength of this study is the way attrition was handled. In cohort studies with lengthy follow up non-response occurs frequently, but seldom completely at random. Consequently, not properly dealing with non-response may lead to bias in any direction by selective loss to follow up [
27‐
30]. In order to minimize the risk of this bias, multiple imputation techniques were used. Additionally, the internal structure of the EQ-5D index was maintained by using these techniques to replace the missing EQ-5D dimensions in survivors who did not respond to the EQ-5D questionnaire, instead of the EQ-5D index value.
However, potential limitations also have to be acknowledged. One limitation of this study is potential unmeasured relevant predictors of poor outcome, and effect modification. In particular, frailty before ICU admission [
40] and cardiac or respiratory complications during early ICU admission [
41] have recently been suggested as being closely related to, and possibly reducing or altering, the association between AKI and long-term outcomes. As we did not collect data on these variables, it was not possible to account for these factors in our analyses. We did study the predictive values of eAKI in different subgroups (see Additional file
3). These analyses suggested no effect modification or only slight effect modification. Future prognostic studies could study this phenomenon in more detail by accounting for effect modification and frailty in their models.
Another limitation, is that these results apply to the first day of admission only. This might have resulted in an attenuated estimate. The estimate of the effect of eAKI could have been decreased due to patients without early AKI then experiencing AKI later during admission. Data to verify or reject these shortcomings were not available at this time, and this was not the goal of this study. Future research will be aimed at predictors of outcome during the later days of ICU admission.
In clinical practice, some patients and doctors will base their decision for treatment continuation on survival predictions alone, while others decide to incorporate the expected quality of life as the main argument for their treatment wishes. In the process of shared decision-making and accurately informing patients and families, clinicians will then want to provide relevant information [
1,
2], without relying on a single predictor for a single outcome. So, given its strong independent association with survival and the composite, poor outcome, which incorporates HRQoL, the severity of eAKI should be considered as a candidate predictor in the future development of multivariable and personalized decision support models, to be used during ICU admission.
Abbreviations
95 % CI, 95 % confidence interval; AKI, acute kidney injury; APACHE, Acute Physiology and Chronic Health Evaluation; APS, Acute Physiology Score; eAKI, early AKI; EQ-5D, EuroQoL 5D-3LTM; HRQoL, health-related quality of life; ICU, intensive care unit; IQR, interquartile range; IRB, Institutional Review Board; RIFLE, risk, injury, failure, loss, end-stage renal failure; RR, relative risk; SAPS, Simplified Acute Physiology Score
Acknowledgements
The authors would like to acknowledge the work of W. Pasma, DVM, for his support in the managing the study data and his advice on the early AKI algorithm.