Concurrent use of diuretics, angiotensin converting enzyme inhibitors, and angiotensin receptor blockers with non-steroidal anti-inflammatory drugs and risk of acute kidney injury: nested case-control study

Data sources

This study was carried out using the Clinical Practice Research Datalink (CPRD), previously known as the General Practice Research Database, and the Hospital Episodes Statistics repository from the UK.

The CPRD is the world’s largest computerised database of longitudinal records from primary care. The data recorded in the CPRD since 1987 include demographic information, prescription details, clinical events, specialist referrals, and deaths.23 In addition, the CPRD collects information on lifestyle variables such as body mass index and data on smoking and excessive alcohol use. The Read code classification is used to enter medical diagnoses and procedures, and a coded drug dictionary based on the UK Prescription Pricing Authority Dictionary is used for recording prescriptions. Recent reviews of all validation studies found that medical data in the CPRD are of high quality.24 25

In addition, patients in the CPRD can now be linked individually and anonymously to the UK national registry of hospital admissions (Hospital Episode Statistics repository). The linkages are made by using the unique National Health Service identifier, date of birth, sex, and postcode of patients. Since 1997, the repository contains dates of hospital admission, discharge diagnosis (coded using ICD-10 (international classification of diseases, 10th revision)), and related procedures (coded using ICD-10 and OPCS-4 (Office of Population Censuses and Surveys classification of interventions and procedures, 4th version)).

Study population

Within the CPRD population, we assembled a cohort of patients who received antihypertensive drugs between 1 January 1997 and 31 December 2008, with follow-up until 31 December 2010. For the purposes of this study, cohort entry was the date of a first prescription for any antihypertensive drug (diuretics, angiotensin converting enzyme inhibitors, angiotensin receptor blockers, calcium channel blockers, α and β blockers) during the study period. Patients were required to have at least one year of up to standard medical history in the CPRD before that first prescription. For patients with less than one year of registration with the general practice at the time of their first prescription, we moved their cohort entry date forward in time to allow a minimum one year medical history before the cohort entry. We excluded patients with a history of any cancer, renal disorders (history of chronic kidney disease, acute kidney injury, or both, including dialysis related procedures), hepatitis, systemic connective tissue diseases, rheumatoid arthritis, crush injury, HIV infection, and any drug misuse at any time before cohort entry. We followed all patients who met the study inclusion criteria until a first ever hospital admission for acute kidney injury, the occurrence of one of the exclusion criteria (including hospital admission for acute kidney injury in secondary diagnosis position, or dialysis related procedures when they were not followed (within 30 days) by a diagnosis of acute kidney injury in primary position), death from any cause, end of registration with the general practice, or end of the study period (31 December 2010), whichever occurred first.

Selection of cases and controls

We defined cases by the first ever occurrence of a hospital admission related to acute kidney injury during follow-up, using the Hospital Episodes Statistics database (identified by using the ICD-10 diagnostic codes N17, N17.0-2, N17.8-9 in primary position). Among these cases, we also identified those that required dialysis during their hospital admission.26 27 As dialysis may precede or follow a diagnosis of acute kidney injury, we searched for procedure codes within 30 days before or after the admission date (ICD-10 codes Z99.2, Z49.0, Z49.1; OPCS-4 codes X40.1,3,8,9, X41.8,9, X42.8,9, X43.1); the index date was the earliest date of either the admission for acute kidney injury or the dialysis procedure.

Up to 10 controls were randomly selected and matched to each case on year of birth (age), sex, calendar year of cohort entry, prevalent user status (that is, patients for whom cohort entry was moved forward in time to allow at least one year of registration in the general practice), and duration of follow-up. Thus, all controls were alive, not previously diagnosed as having acute kidney injury, registered with their general practice when matched to a given case, and had an equal duration of follow-up at the risk set date. The date of the risk set was the index date for the controls.

Exposure assessment

For all cases and controls, we obtained information on all drugs prescribed between cohort entry and the index date, including diuretics, angiotensin converting enzyme inhibitors, angiotensin receptor blockers, and NSAIDs. We defined a double therapy combination as the concurrent use of either diuretics, angiotensin converting enzyme inhibitors, or angiotensin receptor blockers with NSAIDs. We defined a triple therapy combination as the concurrent use of diuretics and angiotensin converting enzyme inhibitors or angiotensin receptor blockers, along with NSAIDs. To be considered a double or triple therapy combination, the drugs of interest had to have been prescribed on the same day or, alternatively, their specified durations of use had to overlap each other for at least one day during follow-up. We calculated this by using the number of prescribed tablets combined with dosing instructions and daily dose. We divided the number of prescribed units by the daily dose to obtain the expected duration of drug use. For the antihypertensive drugs, we defined continuous use as when the duration of one prescription overlapped the date of the next prescription, allowing for a 90 day grace period between prescriptions (a grace period determined by both the pharmacokinetics and prescribing patterns of these drugs in general practice28). For NSAIDs, continuous use was similarly defined, but the grace period consisted of the half life of the drug. In the special situation in which patients were prescribed more than one NSAID, we used the one with the longest half life to determine the overlap with the antihypertensive drugs.

On the basis of the above, we classified patients into one of three mutually exclusive groups: “current use,” defined as a double or triple therapy combination overlapping the 90 days immediately before the index date; “past use,” defined as a combination after cohort entry but ending before the 90 days before the index date; and “never use,” defined as the absence of a double or triple therapy combination at any time between cohort entry and index date.

Potential confounders

We adjusted the risk estimates for comorbid clinical conditions, measured before the index date, known to be associated with acute kidney injury, which could also influence the choice of antihypertensive and NSAID treatment. We thus adjusted all models for the indication for antihypertensive drug use (hypertension, heart failure, and coronary heart disease measured at cohort entry), rhythm disorders, valvular disease, diabetes, number of hospital admissions (any time before the index date), excessive alcohol use, smoking status, and body mass index (all based on the last measure before the index date).6 7 22 29 30 31 To minimise the effect of severity of hypertension, we also adjusted the models for the blood pressure value recorded at cohort entry. We defined hypertension as “mild” when systolic or diastolic blood pressure was between 140 and 159 mm Hg or 90 and 99 mm Hg, “moderate” when systolic or diastolic blood pressure was between 160 and 179 mm Hg or 100 and 109 mm Hg, and “severe” when they were 180 mm Hg or 110 mm Hg or higher.16 Finally, we also adjusted the models for the use of other antihypertensive drugs (α blockers, β blockers, and calcium channel blockers), digoxin, clopidogrel, anti-arrhythmics, statins, antibiotics, immunosuppressive drugs, paracetamol, and corticosteroids, all measured in the year before the index date.

Statistical analysis

We estimated the overall incidence rate of acute kidney injury by dividing the total number of cases that occurred during the study period by the total person years of follow-up, with 95% confidence intervals based on the Poisson distribution. We used conditional logistic regression to compute odds ratios of acute kidney injury, which, for the time matched nested case-control method used here, provides unbiased estimates of the rate ratio and 95% confidence interval.32 In addition to age, sex, calendar year of cohort entry, prevalent user status, and duration of follow-up on which the logistic regression was conditioned, we adjusted for the confounders described above.

In the primary analysis, we compared current users of a double therapy combination, consisting of diuretics or angiotensin converting enzyme inhibitors or angiotensin receptor blockers with NSAIDs, with patients currently using these antihypertensive drugs but without NSAIDs. Likewise, we compared current users of a triple therapy combination, consisting of diuretics and angiotensin converting enzyme inhibitors or angiotensin receptor blockers with NSAIDs, with patients currently using these antihypertensive drugs without NSAIDs. We chose these reference groups to minimise confounding by indication.33

In secondary analyses, we assessed whether the risk of acute kidney injury associated with double and triple therapy combinations varied according to half life of NSAIDs and duration of use (that is, length of exposure to double or triple therapy combination). Thus, patients deemed to be current users of such combinations were further categorised according to whether they used NSAIDs with short half lives (<12 hours: aceclofenac, acemetacin, diclofenac, etodolac, fenbufen, fenoprofen, flurbiprofen, ibuprofen, indometacin, ketoprofen, mefenamic acid, tiaprofenic acid, aspirin) or long half lives (≥12 hours: azaprazone, meloxicam, nabumetone, naproxen, piroxican, sulindac, tenoxicam), as well as according to duration of exposure (≤30, 31-60, 61-90, >90 days). We assessed effect modification by half life of NSAID and duration of use of double and triple therapy combinations by including interaction terms between those categories and the relevant exposure groups.

We also did several sensitivity analyses to evaluate the robustness of the results. The first consisted of repeating all analyses by restricting the cases to the subset requiring dialysis. In a second analysis, we repeated the analyses by additionally matching cases and controls on general practice to further control for potential residual confounders from practice level prescribing patterns and social deprivation (we did not match on general practice as part of the primary analysis because of the exclusion of 573 (26%) cases). In a third analysis, we verified whether the introduction in the UK of new quality outcome framework guidelines over the study period, concerning the recording and treatment of diabetes and chronic kidney disease,34 35 had affected the accuracy of both diagnoses of acute kidney injury and chronic kidney failure in the CPRD. Thus, we stratified cases and matched controls on calendar period (1998-2002, 2003-07, 2008-10) and diabetes diagnosis (as these patients are prone to develop chronic and acute kidney diseases 36 37). In a fourth analysis, we tested the definition of current use by varying the width of the exposure time window and grace period to 60 and 30 days. In a fifth analysis, we stratified cases and matched controls on prevalent user status (this was a matching variable) to assess the effect of including prevalent users in the main models. In a sixth analysis, we assessed the impact of over the counter use of NSAIDs by restricting the cases and matched controls to patients with osteoarthritis, a population for whom the chronic use of NSAIDs is likely to involve physicians’ prescriptions. Finally, because the CPRD does not include information on drug exposure in the inpatient setting, possibly leading to immeasurable time bias,38 we assessed the effect of this bias by repeating the analyses after excluding patients who had been admitted to hospital during the current time window. We used SAS version 9.2 for all analyses.