Statin use and the risk of acute kidney injury in older adults

We used a previously described population-based database [11], which incorporates data from Alberta Health (AH; the provincial health ministry) such as demographics, physician claims, hospitalizations, ambulatory care utilization and Alberta Blue Cross Coverage for Seniors drug data; the Northern and Southern Alberta Renal Programs and the clinical laboratories in Alberta. All people registered with the Alberta Health Care Insurance Plan (AHCIP) were included in the database; all Alberta residents are eligible for the AHCIP and > 99% participate in coverage. For eligible individuals aged 65 years and older, prescription drugs dispensed from community pharmacies are adjudicated through Alberta Blue Cross. There are no premiums for this coverage and the individual shares the cost of the prescription with AH by paying 30% to a maximum of $25 for each prescription dispensed. We used the database to assemble a cohort of individuals 66 years of age and older with an incident statin prescription between 2002 and 2013 who were not previously receiving renal replacement therapy nor had a kidney transplant. We excluded people insured by Health Canada’s, First Nations and Inuit Health, Non-Insured Health Benefit Program rather than AH due to lack of drug data.

Participants entered the cohort on the date of their first statin prescription (i.e., index date) on or after their 66th birthday between May 1, 2002 and March 30, 2013. Prescriptions for any of atorvastatin, fluvastatin, lovastatin, pravastatin, rosuvastatin, or simvastatin were considered. By limiting entry to occur at least one year after eligibility for Alberta Blue Cross Coverage for Seniors commenced, we could properly assess previous statin use, using a one-year washout period. We excluded individuals who had ≥1 statin prescription during the one year period prior to their first statin prescription; we considered the remaining participants as incident statin users and included them in the main analyses (Additional file 1: Figure S1).

We deemed exposure to statins to have commenced on the date of the first statin prescription and reassessed exposure every 30 days of follow-up. We considered individuals as statin users for the entire 30-day period if they were dispensed at least one statin tablet. Incident statin users were grouped according to the intensity of statin [12] they were receiving and could move back and forth between statin use and non-use as well as between different types of statins. The non-use periods of the time-varying statin exposure variable formed the reference group in analyses. We classified statin intensity as high, medium or low according to the National Institute for Health and Clinical Excellence (NICE) clinical guidelines on lipid modification [12] (Additional file 1: Table S1). If a non-use period was ≤90 days and was followed by a period of statin use, we assumed that these individuals were statin users during this “non-use” period. We assumed individuals were adherent with statin prescriptions if they continued to have prescriptions for statins dispensed. If in a 30-day period an individual was dispensed more than one type of statin, we classified that individual as receiving only one agent for the entire period (the agent which accounted for the majority of tablets during that period). Usage was based on dispensing events and number of days supplied (Additional file 1: Figure S2).

We used the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [13] and a standardized serum creatinine assay to estimate the baseline estimated glomerular filtration rate (eGFR) for each participant. The most recent outpatient serum creatinine on or within 1 year prior to the index date was selected if available. We assessed albuminuria using the most recent outpatient measurement (or the median of all multiple measurements available) on or within 1 year prior to the index date and categorized it as: normal [albumin:creatinine ratio (ACR) < 3 mg/mmol, protein:creatinine ratio (PCR) < 15 mg/mmol or urine dipstick negative], moderately increased (ACR 3–30 mg/mmol, PCR 15–50 mg/mmol or urine dipstick trace or 1+), severely increased (ACR > 30 mg/mmol, PCR > 50 mg/mmol or urine dipstick ≥2+), or not measured/available. We used ACR as the primary measure of albuminuria; if ACR was unavailable we used PCR; if PCR was unavailable we used dipstick urinalysis. We used validated algorithms based on claims and hospitalization data to classify participants regarding the baseline presence of 30 comorbidities [14]. We also classified individuals with respect to any use of ACEI, ARB or loop diuretics (oral furosemide or ethacrynic acid; as captured through dispensing events) within 1 year of the index date.

We followed participants from their index date until the outcome of interest, date of death, initiation of chronic renal replacement, outmigration from the province or the study end (March 31, 2013). The primary study outcome was hospitalization with acute kidney injury (AKI) based on International Classification of Diseases (ICD9 or ICD10) codes obtained from Alberta Health data [15] (Additional file 1: Table S2).

We reported baseline descriptive statistics as percentages or medians and interquartile ranges (IQR) as appropriate. We used Cox proportional hazards models to assess the relationship between statin intensity and the primary outcome of interest. We treated statin use categorized by intensity as a time-varying exposure. Covariates (eGFR, albuminuria, comorbidities, medication use) were also time-varying; they were updated for each 30-day interval and defined in the same manner as at baseline. We tested the proportional hazards assumption by assessing log-log plots.

To test our secondary hypothesis, we examined whether the relationship between statin exposure and hospitalization with AKI was modified by a) ACEI/ARB use at baseline; or b) loop diuretic use at baseline. We also examined other potential dichotomous effect modifiers as follows: c) age 66–75 years versus age > 75 years; d) sex; e) diabetic status at index; f) chronic kidney disease (CKD) status at index (where individuals were classified as having CKD if either their baseline eGFR was less than 60 ml/min/1. 73m² or they had “moderately increased” or worse albuminuria); and g) albuminuria status at baseline (where individuals had albuminuria if they had “moderately increased” or worse albuminuria). Finally, to test whether AKI apparently due to statin use was actually due to cardiovascular events or interventions that were associated with both initiation of statins and hospitalization with AKI, we stratified on the presence of i) cardiovascular events. Cardiovascular events of interest included cerebrovascular vascular accident (CVA), myocardial infarction (MI), coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI), or coronary angiogram. The presence or absence of these cardiovascular events was classified during the entire follow-up period. Analyses a) to i) were done by including an interaction term between the stratification variable and statin exposure in each model.

We did several sensitivity analyses, including a) repeating the primary analyses in individuals without previous cardiovascular disease as of the index date; and b) assessing the relation between statin intensity and hospitalization for AKI requiring dialysis (based on ICD 9 or ICD 10 codes in Additional file 1: Table S2); and c) repeating the primary analyses based on defining statin exposure in terms of medication possession ratio [16]. We defined cardiovascular disease as any of: acute myocardial infarction, percutaneous coronary intervention, coronary artery bypass grafting, coronary catheterization, heart failure or stroke/transient ischemic attack. We assessed medication possession ratio (MPR) as a percentage at each 30-day interval and was defined as the “number of tablets” divided by “days in period”. We categorized MPR into three categories as follows: MPR ≥ 75%, MPR 1 to 75%, and non-use (where as before, these were the periods where no tablets were taken). If a non-use period was ≤90 days and was followed by a period of statin use, the last available MPR value was carried forward into this period.

We also compared the characteristics of statin users to an age- and sex-matched cohort of randomly selected non-users. First, random index dates were generated for the non-users according to the distribution of index dates of the statin users; the statin users were each matched to a non-user (if a match was available) based on age (in 5 year increments) and sex. Characteristics between statin users and non-users were compared using standardized differences [17].

We did statistical analyses using Stata 13.1 MP software (www.​stata.​com). The institutional review boards at the Universities of Calgary and Alberta approved the study (REB14–0884, Pro00053469) with a waiver of consent granted.

Compared to non-use, use of any statin (all types; all doses) was associated with a significant increase in the unadjusted and adjusted risks of hospitalization with AKI [hazard ratio (HR) 1.22 (95% confidence interval CI 1.16, 1.28), and HR 1.11 (95% CI 1.06, 1.17)], respectively.

High-intensity regimens were associated with a significantly increased unadjusted risk of hospitalization with AKI [HR 1.35 (95% CI, 1.28, 1.43) vs non-use]. Similarly, the unadjusted HR for the risk of hospitalization with AKI for medium-intensity users versus non-use was also significant [HR 1.09 (95% CI 1.03, 1.16)] (Table 2). After adjustment, the adjusted hazard ratios (aHR) for the risk of hospitalization with AKI were attenuated but remained significantly increased (high-intensity versus non-use aHR 1.16 (95% CI 1.10, 1.23); medium-intensity versus non-use aHR 1.07 (95% CI 1.01, 1.13). There were no differences in the risk of hospitalization with AKI between low-intensity users and non-use (Table 2).

In sensitivity analyses, we searched for clinical characteristics that might modify the association between statin regimen intensity and the risk of hospitalization with AKI. Of 8 characteristics considered, sex, ACEI/ARB use and loop diuretic use significantly modified these associations (Fig. 1). Specifically, the dose-adjusted risk of hospitalization with AKI associated with statin use was substantially greater in women than in men (p for interaction 0.002). While the absolute risk of hospitalization for AKI in women was smaller than in men, compared with non-use, the risk of hospitalization with AKI associated with high-intensity statin use among women was 1.29 (95% CI 1.19, 1.39) but only 1.07 (95% CI 0.99, 1.14) in men. Similarly, the increased risk of hospitalization with AKI associated with high-intensity statin use was 1.26 (95% CI 1.19, 1.34) among those using ACEI or ARB, but only 1.10 (95% CI 0.99, 1.22) among those not using the latter (p for interaction 0.01). There was also a significant interaction between diuretic use and the risk of hospitalization with AKI associated with statin use (p for interaction 0.04). Among those using a diuretic at the index date, the risk of hospitalization with AKI with high-intensity statin use was 1.35 (95% CI 1.22, 1.49) and only 1.16 (95% CI 1.09, 1.23) among those not using a diuretic.

Our study has important strengths, including its use of a large population-based database from a setting with universal health care coverage, its use of validated algorithms for ascertaining the presence or absence of hospitalization with AKI as well as baseline comorbidity, and its rigorous analytical methods. However, our study also has several potential limitations that should be considered when interpreting results. First, like all studies using administrative data, residual confounding is possible by unmeasured characteristics including use of over-the-counter medications such as non-steroidal anti-inflammatory medications, intercurrent illness or the underlying risk of AKI among study participants. Confounding by indication is a particular concern, but the similar findings in a sensitivity limited to individuals with known cardiovascular disease at baseline should increase confidence in our findings. Second, not all individuals had baseline eGFR or albuminuria measurement available to assess CKD status. This would not have been expected to affect our key findings regarding the association between statin use and AKI, although it might have reduced statistical power to show effect modification by baseline CKD status. It is worth noting that low eGFR does not affect blood statin levels appreciably, since these medications primarily undergo hepatic metabolism rather than renal excretion [24]. Third, we studied only people older than 66 years. Although such people account for most statin users in Alberta, whether our findings apply to younger people requires confirmation. Fourth, our algorithm for identifying AKI has limitations, including a sensitivity of approximately 35%, although it is approximately 98% specific [15], as compared with a clinical gold standard. On balance, these limitations may have led to under-recognition of milder forms of AKI (perhaps reducing statistical power), although they are unlikely to have led to bias. Fifth, we were not able to identify the cause of AKI, although it is unlikely to have been due to rhabdomyolysis, given how infrequently this event occurs [1]. Sixth, like most pharmacoepidemiological studies, our data source captured dispensing of medications from pharmacies and not actual medication use. Participants may not have taken medications that were dispensed, and dates of dispensing may not correspond to actual doses taken during each 30-day interval. Finally, we studied people from a single Canadian province and our findings may not apply to other settings.