Intermediate and Long-term Outcomes of Survivors of Acute Kidney Injury Episodes: A Large Population-Based Cohort Study

1Editorial, p. 3

2Editorial, p. 3

3Acute kidney injury (AKI) occurs in 1 in 7 hospital admissions. Even those with small increases in creatinine levels have 4-fold greater hospital mortality than those with no creatinine level increase. Overall, for those who survive to hospital discharge, the prognosis remains poor. However, although 2 previous systematic reviews reported long-term increased mortality and long-term renal replacement therapy (RRT) after AKI (vs no AKI), there was substantial heterogeneity in both outcomes (I2 > 85%)., To discuss risk and plan care for individual patients requires a better understanding of what drives variation, which patients are at elevated risk, and how long risk remains elevated.

4Clinical guidelines recognize that the long-term prognostic factors for AKI are poorly understood., The KDIGO (Kidney Disease: Improving Global Outcomes) guideline advocates follow-up of all patients with AKI, and UK guidelines recommend surveillance for at least 2 to 3 years. Monitoring of all patients may be prudent, but in clinical practice, some patients are prioritized over others, only a minority see a nephrologist, and follow-up may be brief., Without more detailed prognostic studies, it is difficult for clinicians to communicate individual risks and prioritize high-risk patients for an appropriate duration.

5Potential drivers of variation in AKI prognosis include the limited availability of pre-AKI (baseline) data in previous AKI definitions, differences in the severity of AKI, level of baseline kidney function, degree of subsequent recovery of kidney function, and variation in the prognostic role of AKI as follow-up time progresses.,,,, These factors are potentially quantifiable in all patients with AKI, but have not been systematically explored in any one study. In particular, the changing prognostic role of AKI at different levels of baseline function and at different follow-up times has received little attention, and the relevance of prior AKI episodes has not previously been studied. Furthermore, the time at which recovery of kidney function is assessed has varied in previous studies, and it is unclear how often this will have reclassified patients.

6The Grampian Laboratory Outcomes Morbidity and Mortality Study II (GLOMMS-II) is a population cohort linking national and regional data sources in a single UK health authority. Uniquely, all biochemistry data are obtained by a single laboratory service regardless of clinical location (inpatient, outpatient, and community), thus minimizing the loss of baseline and follow-up data. We have previously exploited GLOMMS-II to study different approaches to using kidney function data to define AKI in clinical practice and prognostic research.,

7To guide decisions and discussions with patients about the implications of an AKI episode, we have used GLOMMS-II to study how the prognosis after isolated AKI episodes changes during the course of a long follow-up from intermediate (30-364 days) to long term (1-10 years). We hypothesized that the adverse prognosis of AKI would lessen over time and the role of AKI prognostic factors (baseline, severity, and prior AKI episodes) would change as follow-up progressed.

8The GLOMMS-II was developed through novel data linkage of regional biochemistry results (1999-2009) to hospital episode data, the local renal information management system, and the Scottish Renal Registry for long-term RRT, morbidity, and outcomes., It includes all patients with abnormal kidney function test results (estimated glomerular filtration rate [eGFR] < 60 mL/min/1.73 m2) and a 20% sample of those with normal kidney function. It has been extensively used in renal research, including AKI.,,, All serum creatinine measurements are isotope-dilution mass spectrometry−aligned and processed by a single biochemistry service. Data linkage enables population follow-up without formal patient recruitment, which minimizes selection biases. Information Services Division (ISD) Scotland refreshed the linkages using the community health index, a unique identifier for all residents in Scotland, to connect each AKI episode to individual hospital admissions. There were no patients without a community health index indexed in the ISD population “spine,” meaning that all records were linkable. The ISD reports precision of 99.9% for record linkages. We obtained approval from the Privacy Advisory Committee (study number XRB14137) and the Regional Ethics Committee (reference 14/NW/1371), which waived the requirement for informed consent for this study. Data were hosted and managed by Grampian Data Safe Haven.

9The exposure was the first hospital admission with AKI in 2003. We excluded those with AKI who were not admitted to the hospital within 7 days. This was to ensure that both the exposed and comparator groups contained only admitted patients and because KDIGO-based AKI criteria perform differently in patients who are not admitted to the hospital. We defined AKI using a modification of the National Health Service (NHS) England AKI “e-alert” algorithm, derived from the KDIGO AKI definition (Box 1). The first result meeting AKI criteria was recorded as the start of an AKI episode, and the corresponding “look-back” result from the criteria was used as the baseline.

10We have demonstrated the performance of the NHS England algorithm elsewhere, but because it does not organize blood tests into discrete AKI episodes, we developed a modified version that was capable of distinguishing AKI episode severity and prior and subsequent AKI episodes. The original algorithm uses an 8- to 365-day look-back to estimate the median baseline creatinine level, but a prior AKI episode could result in a falsely high estimate of baseline from tests in the past year leading to underdetection of recurrent AKI. Therefore, as illustrated in Fig S1 (available as online supplementary material), we modified the algorithm criteria to a 1.5-fold or greater increase from median creatinine level within 8 to 90 days in those with an available test, and further look-back to 365 days only in those without a more recent test (criterion 1, Box 1). To ensure clinical relevance for those using the original NHS England algorithm, we compared diagnostic agreement and 30-day mortality in a sensitivity analysis.

11Definitions are summarized in Box 1. We determined AKI severity stages 1 to 3 using the peak creatinine level in the AKI episode relative to the baseline creatinine level. We counted the number of prior AKI episodes that occurred in the preceding 91 to 1,095 days (ie, 3 years). Given the varying definitions of recovery and the KDIGO recommendation of reassessment at 90 days,,,,, we determined how an earlier assessment (7 days) compared to a later assessment (90 days) with regard to recovery status and completeness of repeat testing. Because our interest was in the classification of those with complete return to baseline, those receiving RRT were classed as nonrecovery rather than grouped separately.

12Comparators were patients admitted to the hospital in 2003, who had a blood test for kidney function during admission, and did not have AKI. To ensure that they had the opportunity to develop AKI in 2003, the last admission in 2003 was used. We used their hospital admission creatinine as baseline.

13Outcomes were mortality and long-term RRT (dialysis or transplantation).

14Follow-up was from the date of initial hospital admission to 2013. Follow-up was through data linkage to the national register rather than direct patient contact. Migration from Grampian to beyond Scotland was negligible for the period and age-mix of the cohort. Those not registered as dead during follow-up were therefore assumed to be still alive.

15We used International Classification of Diseases, Tenth Revision codes for Charlson comorbid conditions from the 5 years prior to admission as previously described and validated., We also determined diagnostic categories (eg, circulatory system and respiratory system) related to the acute admission by using the admission categories grouped and recorded in International Classification of Diseases, Tenth Revision, as previously utilized elsewhere. We used the CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) creatinine equation to describe baseline kidney function in 4 eGFR groups: normal, ≥60; mild decrease, 45 to 59; moderate decrease, 30 to 44; and severe decrease, <30 mL/min/1.73 m2. We did not include measures of proteinuria because only a minority of the population had been tested. We also recorded whether the admission was an emergency readmission and the specialty involved during the hospital admission with the following priority when more than 1 was involved: critical care, surgical, care of the elderly, medical, and “other” (eg, obstetrics and psychiatry).

16We reported the incidence of hospital AKI in those with and without baseline decreased kidney function. Due to cohort sampling, those with normal baseline function and no AKI were under-represented. This means that the incidence of AKI among those admitted with normal baseline function cannot be taken directly from the data presented in Table 1 because this would overestimate its occurrence. Therefore, the incidence of AKI was calculated directly for those with baseline decreased kidney function and estimated by multiplying out the sampled fraction for those with normal baseline function. We described overall characteristics and outcomes stratified by baseline kidney function, AKI severity stage, and history of prior AKI episodes. We compared patient recovery of kidney function status at 7 and 90 days, including the proportion of patients who improved, deteriorated, and died. We reported crude mortality and long-term RRT and plotted 10-year mortality by AKI stage using 1 − Kaplan-Meier curves. We assessed mortality in intervals of 0 to 30 days, 31 to 90 days, 91 to 182 days, 183 to 364 days, 1 to less than 3 years, and 3 to 5 years. Within each interval, we computed mortality rates by AKI stage stratified by baseline eGFR group. Using Poisson regression, we calculated age- and sex-adjusted mortality rate ratios (RRs) within each period by AKI stage and prior AKI episodes. We then determined long-term (1-10 years) mortality among those alive at 1 year using multivariable Cox regression adjusted for age, sex, and covariates of interest. Based on previous literature, we included an interaction term between baseline eGFR and AKI on mortality., In a series of sensitivity analyses, we also calculated hazard ratios (HRs) for intermediate outcomes (30-364 days), and when analysis was restricted to those younger than 75 years, to patients with AKI without complete recovery to baseline, and those who also had prior AKI episodes. All analyses were conducted using Stata/SE 13.0 (StataCorp LP).

17The GLOMMS-II is outlined in Fig 1. Of 17,630 hospitalized patients, 3,426 (19.6%) had AKI. The estimated incidence of AKI in the hospital was 17.6% among patients with baseline decreased kidney function and 8.4% among patients with normal baseline function. The 10-year observation period extended from 2003 until 2013 and comprised 114,696 person-years (median follow-up, 9.0 years).

18Patient characteristics for those with and without AKI are described in Table 1. Data are stratified by baseline kidney function to allow for population sampling. The left panel describes patients with and without AKI further stratified by AKI severity. Of those with normal baseline function, patients with AKI were older, received more critical care, and had more comorbid conditions than patients without AKI, but age and comorbid conditions varied little with AKI severity. The pattern was similar for those with baseline decreased kidney function, though all groups were older and had more comorbid conditions. Additional data for patient characteristics in eGFR subgroups are provided in Table S1.

19The Table 1 right panel focuses on the 3,426 patients with AKI, by history of prior AKI episodes. There were 688 (20.1%) patients who had prior AKI episodes, of which 38.4% had occurred within the last 1 year. Patients with prior AKI episodes had more comorbid conditions than those without prior AKI (eg, congestive heart failure and diabetes).

20As reported in Fig S2, the recovery status of many patients with AKI changed between 7 and 90 days. Of the 3,081 of 3,426 patients with AKI still alive at 7 days, by 90 days, 20.4% had improved, 9.9% had deteriorated, 20.5% had died, and 49.1% were unchanged from 7 days. At 7 days, a substantial proportion (360 of 3,081 [11.7%]) had also not yet had a repeat blood test. For these reasons, subsequent analysis used recovery of kidney function at 90 days.

21Tables 2 and S2 summarize crude mortality and long-term RRT. Irrespective of baseline function, mortality was higher for AKI versus no AKI, even at 10 years (Table 2). In those with normal baseline kidney function, long-term RRT was rare, irrespective of AKI episode, severity, or duration (Table S2). In those with baseline decreased kidney function, long-term RRT was most frequent in patients with AKI stage 3. It was also more frequent in those with prior AKI episodes. Further data for long-term RRT in eGFR subgroups are available in Table S3.

22Regardless of baseline eGFR, the more severe the AKI, the greater the initial mortality (Fig 2). However, after 1 year, the mortality curves no longer diverged. Figure 3 describes the change in mortality rates over the first 5 years in greater detail. Figure 3A to D describes mortality rates in patients at risk in each of 6 intervals (0-30 days, 31-90 days, 91-182 days, 183-364 days, 1-<3 years, and 3-5 years) by AKI severity and stratified by baseline eGFR. Figure 3E describes mortality rates by prior AKI episodes. Age- and sex-adjusted RRs and 95% confidence intervals (CIs) are reported. At all levels of baseline eGFR, early mortality rates were higher in those with AKI than without AKI and highest in those with AKI stages 2 to 3. The early (0-30 days) mortality RRs for AKI (vs no AKI) were greater in those with higher baseline eGFRs: RRs of 6.80 (95% CI, 5.24-8.82), 4.22 (95% CI, 3.00-5.96), 2.63 (95% CI, 1.82-3.79), and 1.93 (95% CI, 1.20-3.10) for AKI stage 1 at eGFRs ≥ 60, 45 to 59, 30 to 44, and <30 mL/min/1.73 m2, respectively. In the same eGFR subgroups, the association of AKI (vs no AKI) with mortality also diminished over time: RRs at 1 year of 1.84 (95% CI, 1.55-2.17), 1.89 (95% CI, 1.49-2.39), 1.47 (95% CI, 1.12-1.93), and 1.01 (95% CI, 0.67-1.53), respectively. Those with prior AKI episodes (vs no prior episodes) had similar early mortality, but greater mortality from 6 months onward.

23Table 3 describes the relationship between AKI and long-term mortality (conditional on surviving 1 year) in adjusted models (age, sex, baseline comorbid conditions, and acute circumstances). There was effect modification by baseline eGFR (P for interaction = 0.01). Among 1-year survivors, the adjusted mortality HRs for AKI (vs no AKI) were lower at lower baseline eGFRs: HRs of 1.44 (95% CI, 1.31-1.58), 1.25 (95% CI, 1.09-1.43), 1.21 (95% CI, 1.03-1.42), and 1.08 (95% CI, 0.85-1.36) for eGFRs ≥ 60, 45 to 59, 30 to 44, and <30 mL/min/1.73 m2, respectively. As demonstrated by a comparison to those who had no AKI and baseline eGFRs ≥ 60 mL/min/1.73 m2 (Table 3), AKI made little difference among those with lower baseline eGFRs because their mortality was high irrespective of AKI.

24In a series of sensitivity analyses for mortality outcomes (Table S4), adjusted mortality HRs were higher when calculated for intermediate outcomes (30-364 days): 2.48 (95% CI, 2.15-2.88), 2.50 (95% CI, 2.04-3.06), 1.90 (95% CI, 1.51-2.39), and 1.63 (95% CI, 1.20-2.22) for eGFRs ≥ 60, 45 to 59, 30 to 44, and <30 mL/min/1.73 m2, respectively. HRs were similar when limited to those younger than 75 years or when limited to patients with AKI without complete recovery to baseline at 90 days. Although those with prior AKI episodes had higher mortality HRs, this could be explained by adjusting for comorbid conditions and acute circumstances.

25In this study, we modified the widely used NHS England AKI warning algorithm to enable the grouping of blood tests into discrete AKI episodes and the identification of prior AKI episodes. There was substantial overlap with the original algorithm with 99.2% agreement in diagnosis and a κ statistic of 0.95. Moreover, 30-day AKI mortality differed by only 0.2% (20.1% modified algorithm, 19.9% original algorithm).

26The relationship between a discrete episode of AKI and outcome is complex. This large population study with long follow-up demonstrates the diminishing prognostic role of AKI over time, with risk only modestly increased among 1-year survivors, especially those with baseline decreased kidney function. Both intermediate- and long-term mortality with AKI were modified by baseline kidney function, with the greatest association of AKI with mortality in those with normal baseline function. In contrast, although AKI was not associated with significantly increased long-term mortality among those with baseline eGFRs < 30 mL/min/1.73 m2, this group of patients still had the highest absolute mortality risk. Over time, there were also different prognostic roles for AKI severity and prior AKI. Severe AKI had greater short-term mortality, but AKI severity was less relevant as follow-up progressed. Patients with prior AKI episodes had greater long-term mortality, although this could be explained by comorbid conditions and acute circumstances. Finally, although patients with AKI received more long-term RRT, RRT was a rare outcome in the absence of baseline decreased kidney function.

27This analysis is consistent with and extends previous work. Previous studies have also noted effect modification by baseline function,, and a diminishing role of AKI after hospital discharge., This analysis now shows that the diminishing role of AKI continues for up to 1 year after AKI and is present at all levels of baseline kidney function. The description of prior AKI as a prognostic factor is also novel and important. One in 5 patients with AKI had prior AKI episodes, with worse long-term mortality. Prior AKI was particularly common in those with heart failure, which is consistent with a recent study that reported increased recurrent AKI in heart failure but did not describe the long-term consequences. Finally, the 2-fold increased incidence of hospital AKI in those with baseline decreased kidney function (vs normal baseline function) is in agreement with recent reports elsewhere.

28A strength of this analysis is our use of a large unselected population all served by a single biochemistry service, and laboratory data capture all patients throughout the time course of observation, pre– and post–hospital admission. The unique situation ensures the completeness of the data over a prolonged follow-up and overcomes the shortcomings of other studies wherein selective laboratory testing or access to tests would lead to missing baseline or follow-up data. We also based our AKI definition on an existing e-alert algorithm. Our analysis is therefore particularly relevant for clinicians who are evaluating patients with AKI e-alerts or are developing similar AKI e-alert systems. However, we recognize and have reported elsewhere that some misclassification of CKD occurs.

29A relative limitation, which may be also a strength for real-world generalizability, is the fact that laboratory data were not protocolized with respect to intervals of data collection. Blood testing performed at clinical discretion rather than in standardized testing intervals introduces an ascertainment bias. This is not unique to our study and reflects real-life practice, but it is possible that we have missed some AKI cases. In addition, long-term RRT was a rare outcome. Although this prevented more detailed subgroup analyses, the paucity of long-term RRT among people without baseline decreased kidney function was still striking.

30A key message of this study is that risk after AKI is not static, but changes over the course of a long follow-up. The diminishing association of an AKI episode with mortality is accompanied by greater importance of long-term factors (baseline kidney function and history of prior AKI episodes) over acute factors (AKI severity). Future work should now explore whether these factors at different time points can be used to form prognostic prediction tools for use in clinical practice.