Tuesday Nov 13 2018, 9 pm Eastern
Wednesday Nov 14 2018, 8 pm GMT, 12 noon Pacific
Association of Angiotensin-Converting Enzyme Inhibitor or Angiotensin Receptor Blocker Use With Outcomes After Acute Kidney Injury
Full Text at JAMA IM
This week we discuss an observational study from the Alberta Kidney Disease Network population database. After acute kidney injury (AKI), is use of Angiotensin-Converting Enzyme Inhibitor (ACEI) or Angiotensin Receptor Blocker (ARB) associated with better renal and mortality outcomes? A critically important clinical question that has been much discussed recently – does this study provide evidence that we can apply to routine practice? Join us to discuss
ACE Inhibitors and ARBs are drugs that have transformed clinical care. There is strong randomised trial evidence for improved clinical outcomes from these drugs for patients with heart failure with reduced left ventricular function, and for proteinuric kidney disease. They are also recommended for use for the treatment of hypertension and after ischaemic heart disease. These multiple indications have led to them being one of the most commonly prescribed drug groups. However, in recent years there has been growing concern about their potential for ‘nephrotoxicity’. They are widely believed to be associated with AKI, particularly in patients who are hypovolaemic or septic. Multiple guidelines recommend dose reduction or cessation of these drugs for patients who have developed or are at risk of AKI. However, it is often unclear whether, and when, the drugs should be restarted, particularly when patients are at risk of further AKI. We do know that patients are at high risk of readmission with cardiac failure after AKI but it is not clear how much this is due to cessation of ACEI/ARB drugs as part of AKI management. In addition, AKI itself is associated with increased risk of future adverse outcomes including mortality and development of chronic kidney disease. Do these drugs offer specific benefits after AKI? These issues are at the heart of daily clinical work, and yet there is minimal evidence to help us make these important decisions. Hence the excitement that has greeted this new study by Sandeep Brar to investigate the effects of these drugs after AKI.
This was an observational cohort study using the Alberta Kidney Disease Network population based database, a resource for kidney disease research which uses laboratory data linked to a series of other administrative data sources. The study population was adults over 18 who were admitted to hospital between July 1, 2008, and March 31, 2013, or the first hospitalisation if there were many. Only patients who had AKI during the admission were included and this was defined by comparing the peak hospital creatinine level to a baseline value taken as the mean serum creatinine in the 180 days before admission. AKI was defined as an increase in serum creatinine concentration of 50% or greater, or an increase of 0.3 mg/dL within 48 hours, and/or a need for dialysis during the index hospitalization. Participants who died or whose condition progressed to ESRD (eGFR<15 mL/min/1.73 m2, long-term dialysis, kidney transplant) before or during the index hospitalization were excluded. Included patients were followed up from the discharge date until March 31, 2015 and (note this) with minimum follow-up of two years.
The investigators were able to obtain detailed information to accurately characterize potential confounders from a range of data sources. These included details of the hospital admission including need for ICU admission, comorbidities, medication use and outpatient visits.
The primary outcome was all-cause mortality while secondary outcomes were hospitalization for a renal cause, ESRD, and a composite outcome of ESRD or sustained doubling of serum creatinine concentration (compared to pre AKI baseline). ‘Hospitalization for a renal cause’ was defined as an admission after the index hospitalization until March 31, 2015 (only the first one if there were several), with a ‘most responsible’ diagnosis code of acute renal failure, (note this too) congestive heart failure, hypervolemia, hyperkalemia, or malignant hypertension. ESRD was defined not by coding but by at least 2 consecutive eGFR measurements of less than 15mL/min/1.73 m2 until the end of follow-up.
Right, concentrate here because this bit is tricky. The main analysis was comparing patients who were dispensed at least one prescription for an ACEI or an ARB within 6 months after discharge, to those who weren’t prescribed the drugs. In the secondary analysis patients were categorised as:
no use of ACEI/ARB, (no prescription in the 6 months before or 6 months after the AKI admission)
new use (≥1 prescription within 6 months after discharge from the index hospitalization, with no prescriptions in the 6 months before admission)
prior use (≥1 prescription in the 6 months before admission)
continued use (≥1 prescription in the 6 months before admission and ≥1 prescription within 6 months after discharge).
But ‘hang on – aren’t users of ACEI and ARB are likely to be very different from non-users?’ The authors tried to deal with this by constructing a propensity-score matched group of users and non-users. We have discussed the benefits and pitfalls of propensity score matching in NephJC before here. Having created these smaller but more similar groups the authors went on to use Cox-regression modelling to examine the association between use and non-use of ACEI/ARB after AKI with the mortality and renal outcomes.
In addition, the authors repeated the main analyses within subgroups defined by those with and without proteinuria, baseline eGFR above and below 60mL/min/1.73m2 and the presence or absence of comorbidities: diabetes, hypertension, chronic heart failure, or cardiovascular disease (myocardial infarction or stroke). In a sensitivity analysis, outcomes for patients who started taking an ACEI or ARB within 90 days of discharge were compared with those who started after 90 days.
The study identified 59,951 eligible people who had an episode of AKI. Table 1 in the paper shows the characteristics of potential participants stratified by post-discharge ACEI/ARB use and, as you would expect, finds ACEI/ARB users to be older with a higher proportion of cardiovascular comorbidities, diabetes and cardiovascular causes for hospitalization. After propensity score matching (eTable 2) these differences are much reduced, but some moderate differences remain, for example age, cardiovascular causes of hospitalization and cardiac catheterisation during admission.
A critical aspect of the results is understanding how many of the patients were ACEI/ARB users before their AKI admission, and how many after. The authors report that in the whole cohort 41.1% of the cohort were users of ACEI/ARB who continued after AKI while 38.6% never used the drugs. Only 13.4% of the cohort stopped ACEI/ARB drugs after AKI and 6.9% of previous non-users started taking them. Whether these proportions are the same in the propensity score matched cohort, and in particular the information in Table 3, is unclear. See our interpretation below of how the cohort for the primary analysis (table 2 and 4) and the secondary analysis (table 3) are constructed.
In the main analysis, the adjusted hazard ratio (HR) for mortality associated with ACEI or ARB use after hospital discharge, compared with no ACEI or ARB use, was 0.85 (95% CI, 0.81-0.89) (Table 2). Use of an ACEI or ARB after hospitalization, however, was associated with a higher risk of hospitalization for the composite ‘renal cause’ (HR, 1.28; 95% CI, 1.12-1.46). Of the component end-points, HRs were increased for acute renal failure and hyperkalemia but most markedly for congestive heart failure. No association was found between ACEI or ARB use and progression to ESRD.
When stratifying by categories of drug use, both new ACEI/ARB use and continued use after hospital discharge were associated with lower mortality compared with no ACEI/ARB use but (note this!) stopping use of an ACEI/ARB prescribed before hospital admission was associated with increased mortality (HR, 1.23; 95% CI, 1.17-1.30). Tests for interactions between post-AKI ACEI/ARB use and outcomes depending on baseline comorbidities, to my eye, didn’t show anything informative, and there was no difference in results depending on immediate or later ACEI/ARB prescription post-AKI discharge
Definition of AKI
Reproducing the KDIGO AKI definition in routinely collected data is a hard challenge, since patients aren’t having creatinine levels regularly tested and baseline can be hard to define. This study uses good methodology but there are still limitations. Firstly, imagine the degree of misclassification that can occur from only having 2 creatinines – 1 baseline and 1 in hospital. Patients who are rarely seen in primary care only have blood tests when they are ill, so the creatinine level is likely to be higher than true baseline and this would lead to an underestimate of AKI. Alternately patients having regular outpatient monitoring have routine blood tests and will have a more ‘true’ measure of baseline renal function. Misclassification in epidemiology isn’t too major a problem unless it is related to the exposure but in this study it may well be: prior ACEI/ARB users are likely to have more frequent blood tests and could be more likely to trigger the AKI definition.
Another issue is the patient with progressive outpatient CKD. Here using the mean serum creatinine prior to admission compared to in hospital levels may trigger the AKI definition when in fact there is steady deterioration in renal function rather than AKI. This is of course important when subsequent renal events are an outcome.
Ideally it would be good to see information about the number of creatinine tests that made up the AKI definition. Sometimes sensitivity analyses using last creatinine before admission, or median, or restricting to people with more than a certain number of blood tests can be useful to understand the impact of the AKI definition on the overall study results.
Definition of the exposure
Again defining drug use from routine prescribing data has well-recognised issues. The main one here is that the patients may not be actually taking the drugs. How often have you seen an outpatient who has collected their routine pre-admission prescription after their discharge and you tell them to stop taking one or more of the drugs? Those people would be classified as users in this study – remember they only have had to collect one ACEI/ARB prescription within 6 months after admission to be classified as a user (and similarly before). This means there is bound to be substantial misclassification of whether patients are indeed taking the drugs.
To understand the study more completely it would be useful to know the average number of represcriptions and how many people were classified as AKI users after just one. Drug-use was time updated so that people switched from non-users to users after they were prescribed the drugs. However, a sensitivity analysis adjusted for the duration of post AKI ACEI/ARB use, or time updated for stopping the drugs might have helped show the actual impact of these medications.
Definition of the outcome
Deciding what the outcomes should be in this study must have been difficult. Ultimately, overall mortality is what we care about most as clinicians. However, total mortality is a tricky outcome since we would anticipate that the drugs would impact some but not all causes of death. In a clinical trial ACEI/ARB should reduce cardiovascular mortality, but in a study where you compare users to non-users you might well see that cardiovascular causes of death are higher in ACEI/ARB users due to confounding by indication – a major problem of pharmacoepidemiology. Conversely, ACEI/ARB drugs are often stopped in patients who are frail and approaching death so in observational studies non-cardiovascular mortality is often lower for ACEI/ARB users compared to non-users – but not for causal reasons. Combining all causes of mortality together makes the contribution of these various sources of bias complex to tease apart.
The secondary outcome is similarly not straightforward. It is unclear to me why congestive heart failure was included with other renal causes when heart failure is such an important indication for the drugs outside of renal disease. You would anticipate that congestive heart failure outcomes would be higher in ACEI/ARB prior users compared to non-users and Table 2 suggests that this is true: the adjusted HR for congestive heart failure comparing users to non-users is 1.69 (95% CI 1.18-2.41), a higher effect size than for other components of the renal outcome (but note that here we are getting down to small numbers of outcomes, so this analysis is only adjusted for age and sex).
Finally, note that the post-AKI renal outcomes are defined by coding and not by serum creatinine levels. While this has advantages it also creates potential sources of bias. For example, patients with heart failure taking ACEI/ARB are more likely to have blood tests than non-users, and so may end-up being diagnosed and admitted with AKI than non-users.
There are a couple of other issues worth mentioning. The first is that patients had to survive for two years of follow-up after AKI admission. This is likely to be a way to avoid the results being heavily influenced by drug cessation in the frailest patients, and provide information about long-term outcomes. However, we know that this is a very unwell group of people with high mortality so examining only those who survived for 2 years limits the generalisability of the results to routine practice. Similarly, excluding patients whose eGFR dropped to <15mls/min/1.73m2 in hospital regardless of subsequent recovery limits generalisability to the sickest group of patients we care for.
Another limitation is the lack of information about post AKI renal function. People with poor recovery or high potassium levels are more likely not to have ACEI/ARB commenced or restarted, but also more likely to have poor outcomes. This is not a confounder – it is on the causal pathway – so the authors were right not to attempt to adjust for it. Nonetheless it would be helpful to see information about renal function for post-AKI ACEI/ARB users and non-users, and within the classes of patterns of drug use.
Conclusion: my view
This is an immensely complex paper where the authors have gone to great lengths to make the information clinically relevant and as robust as possible. However, to me it does not provide information that will change my clinical practice.
As mentioned, propensity score matching provides balance for measured confounders but here there is a great potential for unmeasured variables, in particular related to frailty. Therefore, the primary outcome of overall mortality could be explained by those prescribed ACEI/ARB after AKI being less sick than those who are not (supported by the fact that stopping use of an ACEI or ARB prescribed before hospital admission was associated with increased mortality (HR, 1.23; 95% CI, 1.17-1.30)). Similarly, patients who continue to take the drugs after AKI are presumable those at high risk of renal and heart failure outcomes, so it is not surprising that these are increased compared to non-users. The critical information we need – what happens to people where there is equipoise about restarting ACEI/ARB after AKI is not clear from this study, and perhaps can only be addressed by a clinical trial, something where I can firmly agree with the authors.
Finally, here is a thought experiment. As nephrologists we have become obsessed with an arbitrary definition of AKI which is hard to operationalise in routine data. Imagine this study was repeated with admission following a road traffic accident rather than AKI (let’s assume it was minor, with no associated AKI!). Can you convince yourself that the results would be different?
Summary by Laurie Tomlinson, Nephrologist, Brighton and Hove