Hospital acquired Acute Kidney Injury is associated with increased mortality but not increased readmission rates in a UK acute hospital

Acute Kidney Injury (AKI) is a frequent complication of acute illnesses requiring hospitalisation, affecting approximately 10% of patients [1] depending on how it is measured. It is associated with increased morbidity, a doubling of length of stay (LoS) in hospital [2, 3] and an excess in-hospital mortality 6–9 times greater than patients without AKI, independent of established co-morbidities [4, 5]. Episodes of AKI have also been associated with increasing risks of chronic kidney disease (CKD), stroke and other cardiovascular events [6, 7].

The data on mortality after surviving an episode of AKI in unselected inpatient population studies, is limited [8, 9]. A study from the US Veterans Administration reported that excess mortality risk persists after discharge and at 1 year is double that of patients without AKI [1]. However, 95.1% of this cohort was male. Other studies have generally been limited in size and to specific disease settings [10‐14]. The data on readmission rates after an episode of AKI is even more limited with studies largely in specific diseases, showing increased readmission rates [4, 15‐18]. To the best of our knowledge there is no large scale study examining hospital readmission rates from European healthcare systems.

With the increasingly widespread use of electronic patient records (EPR), it may be possible to identify patients with, or at risk of, AKI as well as associated and potentially reversible factors [19]. We have previously shown that systemic inflammation, as determined by C-reactive protein (CRP), is a major determinant of AKI [20]. To date no study has examined the relationship between in-hospital CRP and outcomes post-AKI.

Acute kidney injury is expensive [2] and potentially preventable [21]. It is therefore a legitimate target for modification relevant to both improved patient and health economic outcomes. Despite this, few patients are subject to specialist review following discharge. There is a need for evidence to inform whether targeted post-discharge interventions based upon a history of AKI would add value to patient care. This includes evidence regarding readmission and mortality within the first 90 days post-discharge, as arguably this is a period during which intervention might reasonably be focussed. This would have resource implications and evidence from a contemporary European population is needed to assess whether intervention might reasonably be expected to influence outcome.

Our principle questions were therefore whether AKI in patients not previously followed by a nephrologist, had similarly high rates of deaths associated with AKI and whether there were identifiable avoidable deaths. We also wanted to understand whether readmission rates in survivors were higher in patients with AKI where targeted resource allocation may be helpful.

We, therefore, analysed a large, unselected population in a single UK centre looking at emergency admissions, in conjunction with parameters available from the EPR to determine in-hospital mortality, as well as readmission rates and mortality in the year post discharge after an episode of AKI. We also examined the causes of death, including malignancy within 90-days of discharge in patients who had suffered an episode of AKI.

We studied all adult emergency admissions to an urban, 1200 bedded tertiary academic referral centre in the UK, between April 2006 and March 2010. The institution admits approximately 42,000 emergency admissions per year. It is the regional trauma centre and has 100 Intensive Care (ICU) beds; it does not have paediatric or obstetric and gynaecology services.

Hospital acquired AKI was studied defined on first and highest creatinines after admission by the proportionate KDIGO classification, parameters that should be available in all UK hospitals. Baseline kidney function, which is currently unknown in many emergency admissions in the UK, was not included in this study. AKI was defined as a delta change from the first creatinine taken in the emergency admission. In addition to patients who did not have AKI on creatinine measurement, patients were assumed to have no AKI if they had no creatinine or a single creatinine measured.

All emergency admissions were included except where patients were previously known to the renal service, or were transferred directly into ICU from another centre (Fig. 1). Patients known to nephrologists were excluded on the basis that their higher risk for AKI is known and their care already under renal specialists. Readmission and admission data was analysed for emergency admission only, no elective admission data was included.

Data collated from EPR included: demography, time-stamped drug administration (Additional file 1), ICD-10 diagnoses coded at discharge (for definitions of all diagnoses via ICD-10 see Additional file 2, infection is an aggregate of several ICD-10 codes), routine biochemical analyses, parenteral radiological contrast administration, the need for renal replacement therapy (RRT), length of stay, death in hospital and post-discharge (to census date 30/9/2011). Readmission data was extracted from the hospital systems (from Lorenzo, the Patient Administration System) by the Informatics Department, this data is used to report hospital activity nationally. Data on patient death was extracted from the NHS Spine (national death files), also via Informatics.

The timeframe for data collection was 24 h prior to admission until discharge, except in those meeting the criteria for hospital-acquired AKI in which case the C-reactive protein concentration (CRP) was censored on the day of the peak creatinine. CRP is not adjusted for renal function in our laboratories. Creatinines were measured using a Jaffe method on a Cobas 8000 machine with calibration traceable to isotope dilution mass spectrometry. Patient survival and readmission was analysed following an index admission. The index admission was the first in the study period for all patients. Time to readmission was analysed from both admission and discharge date to readmission (in separate analyses – available on request). CRP was stratified in deciles to 100 mg/dL after which it was stratified in 50 mg/dL up to 400 mg/dL, based on prior clinic consensus and previous published data. This ensured adequate event rates and numbers of patients per stratum [20, 22].

Using data within the EPR, we subsequently examined subsets of patients who had AKI in greater detail. In order to establish whether there were any common factors amenable to manipulation to reduce the mortality in the patients dying in the first 90 days post discharge, we reviewed the case notes of 153 cases of patients with AKI, manually. Cases were reviewed by three experienced clinicians using a semi-structured proforma to look for cause of death, diagnosis of AKI and any factors that could have changed outcome as previous NCEPOD reports have done [21].

In the 4 year study period, 73,016 individual patients had one or more emergency admissions, in total, 112,366 admissions. Men represented 52.8% of the population and the mean age was 56.8 years (Table 1). No creatinine measurement was available in 15.4% of index admissions, and 30, 913, (42.3%) patients had only one creatinine measured.

There were 1635 (2.2%) patients in whom the index admission was associated with an episode of hospital-acquired AKI (stage 1: 998 (61%), stage 2: 440 (27%), stage 3: 197 (12%). Of note many published studies quote AKI incidence as a percentage of the patients who had two or more creatinines measured, in which case the incidence of AKI would be 5.3%. We used the whole population; pragmatically, treating clinicians would have to assume that patients with unmeasured delta renal function did not have AKI.

As expected, patients with AKI were older and had a greater co-morbidity than those without hospital-acquired AKI.

Index admission creatinine measured >177 μmol/L (2 mg/dL) in 2290 (3.1%) of patients and 0.51% had an admission creatinine >354 μmol/L (4 mg/dL).

There is considerable concern about AKI, and increasingly policy guidelines suggest this is frequent, increasing in incidence and a cause of avoidable death. From the current literature it is also expensive [2] and is associated with high mortality. The actual incidence of AKI is difficult to assess in the UK until there are robust electronic patient records (EPR) in all UK hospitals. Data from the Hospital Episode Statistics (HES) database [2], report the incidence of AKI to be 2.43% but others report this to be much higher [24, 25]. The data presented in this paper, where AKI was identified using electronic systems to detect proportional changes in creatinine, conclude the incidence on admission in a large urban hospital, to be 2.2% (or 5.3% of the population where renal function was measured more than once). We included patients who had only one, or no creatinines measured, in the patients without AKI as these patients would not be identified as having AKI in hospital. Patients known to renal services were also excluded on the basis that they were already under nephrology care and therefore renal risk factors were likely to be closely managed in this group.

Mortality associated with AKI is high compared to a population without AKI and this has been previously documented for in-hospital mortality [4, 5]. In this study in-hospital mortality associated with AKI, was 10 times that of the non-AKI population and worsened with severity of kidney injury. Thirty-three percent of patients who had AKI, died during their hospital admission.

This association between mortality and AKI has previously been reported to persist after discharge [26‐28]. Dialysis dependent AKI is associated with increased mortality. There are a few studies which report increased mortality after discharge in specific disease states [28‐32] and after ICU admission [33, 34]. A recent study looked at this in a multivariable model for survival post 90 days [1]. They found an adjusted mortality risk of 1.41 for patients who had had AKI not requiring dialysis, and this trend persisted to 4 years follow up.

There are few reports looking at mortality after AKI in unselected hospital populations in the first 90 days post discharge [8, 9] although some studies look at specific patient groups post discharge. In the LaFrance study [1] the risk of dying did persist after discharge and was high within the first 90 days. 42% of patients diagnosed with AKI in hospital had died by 90 days post discharge. Adjusted mortality risk in our study up to 90 days post discharge, was 1.5 for patients who had had inpatient AKI compared to those who did not have AKI. As in previous literature, this mortality risk persists beyond 90 days and up to 1 year [35]. In our data the association between AKI identified in hospital and death was persistent after 90 days follow up.

Risk of dying increased with severity of AKI in hospital, in the first 90 days and post 90 days discharge and this is consistent with previous studies [36‐38].

Acute kidney injury in this data was an independent factor for death when adjusted for co-morbidities. The current assumption is that AKI is causative in the high mortality rates with which it is associated, and has attracted considerable attention as being a reversible factor, that is preventable, with lives saved as a result [39]. However, the current NICE AKI guideline states that “No studies were identified for prognostic outcomes with KDIGO for adults, children or young people." [29]. In our study we found that high CRP was an important predictor of death and when combined with a diagnosis of AKI increased the odds ratio of death in hospital to more than 10 times that of patients without AKI. Why AKI and specifically AKI with inflammatory states might be an independent risk factor for death is likely to depend on the effects that these changes have on the entire body state, and experimental models already show examples of changes in the heart and lungs [40‐42]. The CRP may represent risk in AKI over and above its association with SIRS/SEPSIS, but as far as we are aware there is no current literature examining the relationship between AKI, CRP and mortality. What is clear from this data is that the combination of high CRP in the setting of AKI greatly increases mortality risk.

We were also interested in defining mortality within the first 90 days as a potential timeframe within which it may be possible to target quality improvement based upon early post-discharge review. Of those 1085 with hospital-acquired AKI who survived to discharge, 13.9% died in the first 90 days post discharge and 22.8% in the first year. Age and malignancy were dominant risk factors for early post discharge mortality in this population. The presence of malignancy, specifically advanced malignancy established by case note review was present in >40% of deaths in this group of patients.

These findings suggest that a simple approach targeting those with hospital-acquired AKI for intervention to prevent early mortality, may have limited utility. Any such intervention would need to account particularly for patients with advanced malignancy. Although this is a single centre study there is a wide range of disease represented and there was no preponderance of patients with a diagnosed malignancy to suggest that the current analysis is particularly unbalanced. This is an important finding in relation to the debate about reduction in death and morbidity due to AKI. There may be utility in understanding the combination of AKI and incurable disease which may serve to target therapies in palliation but it is unlikely that this will reduce overall mortality. This is not to say of course that the presence of malignancy per se means AKI follow-up is unhelpful, patients may have curable cancers and nephrology involvement may influence chemotherapy treatments. In order to target patients where early intervention will reduce mortality related to AKI, it is likely that that this much smaller subset is easier to define as we understand the description of the larger group. The preventable deaths are hidden within this larger subset.

Thirty day readmission is a traditionally collected measure and represents ‘failed discharges’. In terms of health economics, and resource, readmissions alone, independent of deaths, are an important resource model. In this study in cox regression, diabetes, ischaemic heart disease, heart failure, vascular disease, malignancy, infection, liver disease and hypovolaemia were all predictive of readmission; AKI was not; perhaps due to the high mortality rate in this group.. As a result, in patients who survive, those with AKI are less likely to be re-admitted probably as a combination of survival advantage, and potentially for some, because of palliative care packages which reduce readmission. In our study AKI did not increase 30 or 90 day readmission rates, with death included or excluded. Overall rate of readmission was not increased for the whole period of follow up.

Interventions which may impact death in the early follow up period and readmission have not been extensively researched but it may be relevant for nephrologists to be involved in early post-discharge care in patients who have had AKI. This may influence drug management, early identification of deterioration with rapid access to dialysis and more meticulous fluid management particularly relevant as early pulmonary oedema has been linked to readmission [43].

There is little literature discussing AKI and readmission rates but AKI has previously been reported as contributing to readmission rates with increased risk of readmission at 30, 60 and 90 days. However, the rate is much attenuated after adjustment and is only present at 30-days in the matched validation cohort. Additionally in Silver et al. the reported mortality rate at the index admission is only 2%, very low compared to data in our study and therefore may mean fundamentally different cohort. All patients were excluded in which a creatinine nadir was unable to be calculated.

Definition of AKI in the literature is variable since the considerable interest in this area has generated scoring systems which have been developed as this has been studied. We define AKI on the basis of the creatinine changes for KDIGO, superseding AKIN and RIFLE. Any definitions of AKI must include ability to easily identify patients using EPR which are likely become more ubiquitous with rapid adoption across the UK. Baseline creatinine is not necessarily available in acute admissions and we aimed to understand the impact of available hospital information at the point of patient treatment. Most patients have creatinine measured, some institutions use more complex measurements of renal functions, estimated GFR by various methods, CKI-EPI and others, but these are not yet standardised. Similarly urine output is not reliably collected in the majority of emergency admissions.

There are several limitations to this study. It is a single centre study, and although it is based in a large urban hospital, some of the hospital services are tertiary and therefore will over-represent relatively rarer disease states such as advanced liver disease. Although we have found AKI to be an independent risk factor it is often actually a marker of disease severity and may have as yet unidentified confounders within this study. Sicker patients are likely to have more creatinine measurements and therefore may skew the data. We have assumed that patients with one or less creatinine measurements did not have AKI. CKD rates are low, which is likely to be due to the exclusion of patients known to the renal service. CKD is a significant risk factor for AKI and patients who are sicker therefore be under nephrology and excluded from this analysis. This may result in fewer patients in this study with heavy comorbidity burdens, who may have been more likely to die or be readmitted after an AKI.

The study data collection ends in 2011, AKI has been increasingly discussed and management strategies therefore may have changed since the conclusion of this data collection. Readmission rates do not include patients admitted to other hospitals. KDIGO definition of AKI includes urine criteria and this is not utilised here as it is unreliably recorded in EPR. The impact of this is difficult to study as in practice urine output is poorly recorded in acute medical settings outside of high dependency beds.

In a large, unselected hospital inpatient population, AKI identified during hospital admissions is associated with a high in-hospital mortality and this persists into the period up to 90 days post discharge and beyond. AKI is an independent risk factor in hospital and up to 90 days post discharge even when adjusted for co-morbidities, age, and CRP, and is associated with mortality after 90 days. There is a progressive relationship between CRP concentration, AKI and death which is poorly understood and there is a case for studying this further. Readmission rates were found to be unaffected by AKI.

It is important to note that many of the patients who are affected and who die with AKI are already under palliative care services and many have malignancy. This study may support the concept of interventions aimed at preventing AKI such as early nephrology review but post discharge interventions in terms of preventing mortality or 90 day readmission would need to be carefully targeted.